|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
MODELING GROWTH AND COMPOSITION OF
PERENNIAL TROPICAL FORAGE GRASSES
STUART J. RYMPH
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
Stuart J. Rymph
I wish to express my sincere thanks and appreciation to Dr. Kenneth Boote (my
supervisory committee chair) for introducing me to plant physiology and crop modeling,
and especially for the conversations that went a little off-topic and tied the theory back to
the field. I would also like to thank my committee members Dr. Lynn Sollenberger for
his advice on both agronomy and academia; Dr. Charles Staples for our conversations on
animal nutrition; a needed break from agronomy and a reminder of how I enj oy working
with dairies; Dr. Jim Jones for making engineering fun and showing me that there is
agriculture outside of the United States; and Dr. Tom Sinclair, the "sounding-board", for
practical discussions and providing a more skeptical point of view. I'd also like to
recognize the late Dr. Bill Kunkle for our many conversations on farming and ruminant
nutrition: it was like going home.
Kudos go to Dr. Jean Thomas for the expertise, labor, and conversation that made
the growth study possible and also enjoyable. Special thanks are offered to Dr. Paul
Mislevy at the Florida Range Cattle REC in Ona, for providing datasets and also for his
friendship, encouragement, and continued efforts to involve me in the practical side of
tropical forage production.
Special thanks go out to friends for their support and encouragement. Dr. John
Moore helped to convince me to take on this endeavor. Dr. Kenny Woodard provided
practical advice on growing forages, and the voice of a farmer, firmly grounded in reality
(something that only enhances a modeling project). He and Tony Sweat provided
comradery, lots of humor, and a view of Florida that I would otherwise never have seen
and would have been the worse for missing it.
Extra special thanks are due my wife, Dr. Mary Beth Hall, for her emotional and
unending technical support which included sharing her expertise in SAS, also a warm
dinner and a break from dissecting bahiagrass tillers. I also thank her for teaching me to
not discard theory if it doesn't have an immediately apparent use in the field. Finally, I
thank my mother and father, Thelma and Dr. Donald Rymph for instilling in me a healthy
curiosity, a strong work ethic, and the belief that practical experience can be one of the
TABLE OF CONTENTS
ACKNOWLEDGMENT S ............. ......___ .............. iii...
LI ST OF T ABLE S ............. ...... ._ .............. viii...
LIST OF FIGURES .............. .................... ix
AB STRAC T ......__................ ........_._ ........xi
1 INTRODUCTION ................. ...............1.......... ......
2 LITERATURE REVIEW .............. ...............5.....
Bahiagrass ............... ... .. ......... ............. .............
Perennating Organs: Rhizomes and Stolons ................. ...............7............ ...
Dorm ancy .............. ...............8.....
Photosynthesis .............. ...............12....
T he CROPGRO Model ............. ...... .__ ...............19..
Model Evaluation............... ...............3
3 BAHIAGRAS S GROWTH STUDY .....__.....___ ..........._ ...........3
Introducti on ............. ...... ._ ...............35...
M materials and M ethods .............. ...............36....
Results and Discussion .............. ...............42....
Plant Growth............... ...............42.
Photosynthesi s ................ ...............47.................
4 DEVELOPMENT OF CROPGRO SPECIES FILE PARAMETERS FOR
B AHIAGRA S S ............. ...... ._ ...............61..
Introducti on ............. ...... ._ ...............61...
M materials and M ethods .............. .. .... .......... .. .. ........6
Description of Data Sets Used to Fit Parameters ............. .. ...__...........63
Preparation of Datasets ................. ...............65........... ....
Results and Discussion .............. ...............67....
Photosynthesis Parameters .............. ...............67....
Root Parameters.................... .. ...............7
Carbon and Nitrogen Mobilization Parameters ....._____ .........__ ..............73
Vegetative Partitioning Parameters .............. ...............74....
Leaf Growth and Senescence Parameters............... ...............7
Phenology Parameters .............. ...............75...
Testing of Literature-Based Parameters ............_...... .__ .........._......76
Optimization ............ .... ..__ ...............79....
Testing of Optimized Parameters .............. ...............80....
5 ADAPTING CROPGRO TO MODEL PERENNIAL TROPICAL GRASSES:
STRUCTURAL CHANGES TO THE MODEL ......____ ..... ... ._ ...............97
Introducti on ............ ..... .._ ...............97...
M materials and M ethods .............. ...............99....
Results and Discussion ............ ..... ._ ...............103..
Storage Or gan ............ ..... ._ .............. 103....
Dorm ancy .............. ...............107....
Freeze Damage ............ ..... ._ .............. 114...
Photosynthesis ............ _.. ........_ .... .._ ... .... ... ........ 1
Overall Model Performance Herbage Mass and N Concentration .................1 23
6 SUMMARY AND CONCLUSIONS ................ ...............151...............
Bahiagrass Growth Study .............. .. .... ........ ..........5
Development of Species File Parameters for Bahiagrass .................. .. ............... ...152
Adapting CROPGRO to Model Perennial Tropical Grasses: Structural Changes to
the Model .............. .... ...............153.
Implications of the Research ........._._.. ....... ...............155...
Future Research ........._... ...... ___ ..............._ 156...
A CROPGRO CSM PARAMETER CODE DEFINITIONS ................ ................. .158
B SPECIES, CULTIVAR, AND ECOTYPE FILES FOR THE UNMODIFIED CSM
VERSION OF CROPGRO ............ ......__ ...............162..
C SPECIES, CULTIVAR, AND ECOTYPE FILES FOR THE FORAGE VERSION
OF CROPGRO .............. ...............183....
D NEW PARAMETER CODE DEFINITIONS FOR THE FORAGE VERSION OF
CROPGRO .............. ...............198....
E DORMANCY AND STOR CODES AND DEFINITIONS FOR DATA.CDE FILE203
F CODE ADDITIONS AND CHANGES IN THE FORAGE VERSION OF
CROPGRO .............. ...............206....
LIST OF REFERENCES ............ ..... ._ ...............307...
BIOGRAPHICAL SKETCH ............ ..... ._ ............... 15...
LIST OF TABLES
3-1 Schedule of sampling and harvest activities. ............. ...............52.....
3-2 Weekly averages of daily temperatures and daily solar radiation and total weekly
rainfall + irrigation water applied to bahiagrass grown at the Irrigation Park,
Gainesville, FL -2001 .............. ...............52....
3-3 Results of statistical comparison of treatment effects on plant growth and
photosynthesis. Period means are least squares means. Signifieance determined
by ANOVA for Period and orthogonal contrast for Week and Per X Week
interaction. ............. ...............53.....
4-1 Bahiagrass parameter values for the CROPGRO species Eile. Preliminary values
were derived from the literature. Optimized values were derived from
optimization runs made based on the preliminary values. .............. .........._....85
4-2 Evaluation of the performance of CROPGRO with literature-based and optimized
species Eiles, with and without winter photosynthesis reduction. ..........................87
5-1 Summary of performance of the forage version of CROPGRO in simulating mass
of below-ground plant organs (kg DM hal. )................... .... ...........12
5-2 Summary of performance of the CSM (unmodified) and forage version of
CROPGRO in simulating five experiments to predict herbage mass, herbage N
concentration, and herbage N mass. ............. ...............129....
LIST OF FIGURES
2-1 Modular structure and summary of model components of the DSSAT-CSM
cropping systems model............... ...............22.
3-1 Sod core as removed from the soil............... ...............54..
3-2 Example of a separated sub sample of bahiagrass after removing roots. ...............54
3-3 Total plant mass for established bahiagrass grown at Gainesville, FL from 18 July
to 12 Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001. .................. ...............55
3-4 Root mass for established bahiagrass grown at Gainesville, FL from 18 July to 12
Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001. ............. .....................5
3-5 Stolon mass for established bahiagrass grown at Gainesville, FL from 18 July to
12 Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001. ............. .....................5
3-6 Stem mass for established bahiagrass grown at Gainesville, FL from 18 July to 12
Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001. ............. .....................5
3-7 Leaf mass for established bahiagrass grown at Gainesville, FL from 18 July to 12
Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001. ............. .....................5
3-8 V-stage for established bahiagrass grown at Gainesville, FL from 18 July to 12
Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001. ............. .....................5
3-9 Leaf area index (LAI) for established bahiagrass grown at Gainesville, FL from
18 July to 12 Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001. .........................58
3-10 Specific leaf area (SLA) for established bahiagrass grown at Gainesville, FL from
18 July to 12 Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001. .........................58
3-11 Leaf + Stem (green) area index (GrAI) for established bahiagrass grown at
Gainesville, FL froml8 July to 12 Sept. (-) and 12 Sept. to 7 Nov. (- -),
2001 ..............._ ...............59......_......
3-12 Leaf photosynthetic rate for established bahiagrass grown at Gainesville, FL from
18 July to 12 Sept. (m) and 12 Sept. to 7 Nov. (0), 2001............... ..................5
3-13 Canopy gross photosynthetic rate adjusted to 1500 Cpmol Par m-2 S-1 (Pl500) for
established bahiagrass grown at Gainesville, FL from 18 July to 12 Sept. (m) and
12 Sept. to 7 Nov. (0), 2001. ............. ...............60.....
4-1 Observed herbage mass (m), predicted herbage mass (--), water stress (-
-), and N stress (-) of bahiagrass grown with 468 kg N ha-l yr- at Ona, FL,
using the literature-based species file and the leaf-level photosynthesis option,
with a) No adjustment to winter growth, or b) 70% reduction in photosynthetic
rate and partial defoliation (frost) over the winter. ................ ..................8
4-2 Observed bahiagrass herbage mass (m) and predicted bahiagrass herbage mass
using the preliminary (literature-based, non-optimized) species Eile and the leaf-
level photosynthesis option (-), or daily canopy photosynthesis option (-).
4-3 Observed bahiagrass herbage N concentration (m) and predicted bahiagrass
herbage N concentration using the preliminary (literature-based, non-optimized)
species Eile and the leaf-level option (-), or daily canopy option (-). For
bahiagrass grown at Ona, FL with 468 kg N ha-l yr- .............. ..................9
4-4 Observed bahiagrass herbage N concentration (m) and predicted bahiagrass
herbage N concentration using the preliminary (literature-based, non-optimized)
species Eile and the leaf-level option (-), or daily canopy option (-). For
bahiagrass grown at Eagle Lake, TX with 168 kg N ha-l year-. ............ ...............91
4-5 Observed herbage mass (m), predicted herbage mass (--), water stress
(-), and N stress ( ) of bahiagrass grown with 468 kg N ha-l yr- at Ona,
FL, using the optimized species Eile with a winter defoliation, 70% reduction in
winter photosynthetic rate, and a) the leaf -level photosynthesis option or b) daily
canopy photosynthesis option. .............. ...............92....
4-6 Predicted vs. observed herbage mass of bahiagrass grown with 468 kg N ha-l yr-
at Ona, FL, and 168 kg N ha-l yr- at Eagle Lake, TX, using a) the leaf-level
photosynthesis option, or b) daily canopy photosynthesis option. ........................93
4-7 Observed herbage N concentration (m) and predicted herbage N concentration of
bahiagrass grown with 468 kg N ha-l yr- at Ona, FL, using the optimized species
Eile and the leaf-level option (-) or daily canopy photosynthesis option (-).94
4-8 Observed herbage N concentration (m) and predicted herbage N concentration of
bahiagrass grown with 168 kg N ha-l yr- at Eagle Lake, TX, using the optimized
species Eile and the leaf-level option (-) or daily canopy photosynthesis option
( ) .. .. ...............95
4-9 Predicted vs. observed herbage N concentration (g kg- ) of bahiagrass grown with
468 kg N ha-l yr- at Ona, FL, and grown with 168 kg N ha-l yrl at Eagle Lake,
TX, using a) the leaf-level photosynthesis option, or b) the daily canopy
photosynthesis option............... ...............96.
5-1 Schematic of daily partitioning of new growth among vegetative tissues for the
forage version of CROPGRO. ............. .....................130
5-2 Schematic of the calculation of potential mobilization of CH20 from leaf, stem,
root and STOR in the forage version of CROPGRO............. .._.........___....131
5-3 Schematic of the calculation of potential mobilization of N from leaf, stem, root
and STOR in the forage version of CROPGRO. ............. .....................132
5-4 Predicted vs. observed stolon mass for bahiagrass grown in the field in
Gainesville, FL in 2001(), and in temperature and CO2 gradient greenhouses at
360 CLL CO2 L^ (*), and 700 CLL CO2 L^ (0). ............. ...............133....
5-5 The a) controlling functions and b) seasonal expression of the predicted effect of
daylength on incremental (increase above baseline) partitioning to STOR(--)
or mobilization from STOR (- -) in the forage version of CROPGRO. .........134
5-6 Mobilization factors in the forage version of CROPGRO that affect mobilization
from STOR as a function of a) vegetative N status and b) LAI. ......................... 135
5-7 Predicted (m) vs. observed bahiagrass herbage mass for late-season harvests at
Ona, FL in the 1993-1994 and the 1995-1996 growing seasons using the
modified leaf-level photosynthesis option in the forage version of CROPGRO. 136
5-8 Schematic of freeze damage to leaves and stems and cold-hardening of STOR
tissues in the forage version of CROPGRO. ......____ ........__ .................1 37
5-9 Predicted CO2 COmpensation point for the CSM version of CROPGRO (
-) and for two hypothetical C4 Species with a Ci/Ca of 0.4 and a CO2
concentrating factor of either 3 (-) or 10 (-) predicted using the forage
version of CROPGRO .........._ _... .... ._ ...............138..
5-10 Relative CO2 COncentration effect on a) CO2 factor for LFMAX and b) QE for a
C3 Species in the CSM version (- -) and for a C4 Species in the forage version
of CROPGRO using a Ci/Ca of 0.4 and CCNEFF of 3 (--). .......................... 139
5-11 Relative temperature effect on a) CO2 factor for LFMAX and b) QE for a C3
species in the CSM version (- -) and for a C4 Species in the forage version of
CROPGRO using a Ci/Ca of 0.4 and CCNEFF of 3 (-) .............. ..............140
5-12 Predicted growth of bahiagrass components under 350 CLL CO2 L^ atmospheric
CO2 Stolon (- -), root (*****), and herbage (-) relative to predicted growth
under 700 CLL CO2 L^ atmospheric CO2 Stolon (- -), root (*****), and herbage
(-) using the forage version of CROPGRO. ................. ..................141
5-13 Observed bahiagrass herbage mass (m), predicted stolon (- -), root (*****), and
herbage mass ( ) for bahiagrass grown in a) Ona, FL with 468 kg N ha-l yr or
b) Eagle Lake, TX with 168 kg N ha-l yr-. ............ ...............142.....
5-14 Predicted stolon (- -), root (*****), leaf (--), and stem (-) growth for
bahiagrass grown with 468 kg N hal yil in Ona, FL in 1997 using the forage
version of CROPGRO ..........__....... .__ ...............143..
5-15 Observed herbage mass (m), predicted herbage mass (--), water stress ( )
and N stress (-) of bahiagrass grown with 468 kg N hal yil at Ona, FL.
Predicted using the leaf-level photosynthesis option in a) the forage version of
CROPGRO or b) the unmodified CSM version of CROPGRO. .........................144
5-16 Predicted (m) vs. observed bahiagrass herbage mass for five experiments, using
the modified leaf-level photosynthesis option in the forage version of CROPGRO.
5-17 Observed/predicted bahiagrass herbage mass for a) the 0 kg N hal yrl (O and
-), 468 kg N hal yil (O and -), and 942 kg N hal yrl (A and *****)
treatments at Ona, FL and b) the 0 kg N ha' yi' (O and -), 168 kg N ha' yr'
(O and -), and 336 kg N hal yr (A and *****) treatments at Eagle Lake, TX.
5-18 Observed bahiagrass herbage N concentration (m) and predicted bahiagrass
herbage N concentration using the forage version of CROPGRO and the leaf-level
option (-), or daily canopy option (-). For a) bahiagrass grown at Ona, FL
with 468 kg N hal yr or b) bahiagrass grown at Eagle Lake, TX with 168 kg N
hal yeail. ........... ...............147.....
5-19 Predicted (m) vs. observed bahiagrass herbage N concentration for five
experiments, using the modified leaf-level photosynthesis option in the forage
version of CROPGRO ..........__....... .__ ...............148..
5-20 Observed/predicted bahiagrass herbage N concentration for a) the 0 kg N hal yil
(O and -), 468 kg N hal yil (O and -), and 942 kg N hal yil (A and
*****) treatments at Ona, FL and b) the 0 kg N hal yil (O and -), 168 kg N
hal yrl (O and -), and 336 kg N hal yil (A and *****) treatments at Eagle
Lake, T X .. ............ ...............149.....
5-21 Predicted (m) vs. observed bahiagrass herbage N mass for five experiments, using
the modified leaf-level photosynthesis option in the forage version of CROPGRO.
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
MODELING GROWTH AND COMPOSITION OF
PERENNIAL TROPICAL FORAGE GRASSES
Stuart J. Rymph
Chair: Kenneth J. Boote
Major Department: Agronomy
In addition to their role as feedstuffs, perennial tropical forage grasses such as
bahiagrass (Paspalum notatum Fliigge) can play a maj or role in nutrient management on
livestock farms; recycling N from fertilizers and manure to produce feed and reduce the
importation of other feeds, while lowering potential N leaching. Balancing feed quality,
feed quantity, and nutrient recovery can be difficult. A computer model capable of
simulating forage growth, composition, and N-dynamics could be a useful management
tool. Farmers and consultants could test management practices virtually, then implement
those showing the most promise. Our obj ective was to develop a tool to predict the
growth and composition of bahiagrass that responds to environmental and management
Bahiagrass sod cores were dug weekly for two 8-week regrowth periods (18 July to
12 September, and 12 September to 7 November). Plants were separated into leaf blades,
stem, stolon, and roots. Leaf and canopy photosynthesis were measured in each period.
Leaf photosynthetic rate was not different between periods. Leaf and stem growth and
rate of development of new leaves was reduced in the second period; however, stolon
mass increased dramatically starting in mid-October.
This information aided the development of species-specific parameters required for
simulating bahiagrass in the model, CROPGRO. In the process, limitations in the model
structure that prevented the prediction of realistic growth patterns were identified.
Despite the limitations, prediction of herbage mass was good, having an index of
agreement of 0.85, with slightly lower accuracy predicting herbage N concentration.
To address the model's limitations, we modified the CROPGRO source code to
include a storage organ (STOR), equivalent to a stolon, and added dormancy functions to
increase partitioning of growth to STOR and reduce mobilization from STOR and roots
under short daylengths. The freeze-kill function was modified, allowing gradual death of
leaves. The Rubisco specifieity factor in the leaf-level photosynthesis option was
modified for C4 photosynthesis. Model performance was improved, predicting realistic
seasonal growth patterns. Excessive N stress was predicted frequently, but the cause was
not identified. The forage version of CROPGRO performs realistically but should be
tested under cooler temperatures and finer-textured soils.
Perennial tropical grasses such as bahiagrass (Paspalum notatum Flugge) and
bermudagrass [Cynodon dactylon (L.) Pers.] have long been the basis of beef production
in Florida. As environmental concerns have come to the fore over potential movement of
nitrate into Florida groundwater from dairy farms, there is a growing interest in use of
these grasses as part of the nutrient management systems on Florida dairies. In both beef
and dairy operations, the success of a forage program is measured by several criteria: it
must generate enough mass to feed the herd, it may need to utilize a minimum amount of
manure nutrients, and the forage produced must meet a significant proportion of the
nutritional needs of the animals consuming it. Managing forages to meet all of these
goals is a complex trade-off of yield and quality. Emphasizing forage quality generally
requires harvesting at young stages of maturity, with shorter regrowth periods and lower
yields, whereas managing for forage mass produces forage with reduced concentrations
of digestible nutrients.
Current management strategies for perennial tropical forages rely on harvesting at
regular intervals with no allowance for changing weather conditions. A 3 to 5 week
harvest interval will nearly always produce high quality forage (Staples, 1995) but often
the yields are unacceptably low. Delaying harvest when forage growth rates are low
increases yield but at what cost to quality? A system that could compensate for changing
temperature, rainfall, and soil fertility might allow farmers to harvest forage of more
consistent quality in reasonable quantities.
Such a system also could improve N cycling on the farm and reduce N leaching to
groundwater. The high growth potential and expansive root system of tropical grasses
such as bahiagrass and bermudagrass allow these species to not only uptake large
quantities of N from the soil but they may alter the seasonal pattern of N leaching as well
(Woodard et al., 2002). Properly managing the timing of fertilizer and manure
applications as well as the timing of forage harvests can reduce nutrient losses to the
environment while producing additional feed for the herd.
A crop model capable of predicting forage yield and composition along with N
leaching dynamics could allow evaluation of management practices (including harvest
strategies) before implementing them in the Hield. To provide output with the desired
level of detail, the model must have the ability to respond to a variety of environmental
(temperature, rainfall, daylength, soil moisture, etc.) as well as management (fertilizer,
irrigation, harvest schedule, etc.) inputs. It would also be advantageous if the model
could be easily adapted to simulate other forage species to avoid rewriting the model for
each new species to be modeled. The CROPGRO model meets both of these
CROPGRO is a mechanistic or process-oriented model. That is, it predicts plant
growth by simulating many of the underlying biological processes. Some of the plant
processes simulated include photosynthesis, transpiration, senescence/mobilization of
plant tissue, and root uptake of water and nutrients. In addition several supporting
processes such as water infiltration and evaporation in the soil, soil organic matter and N
dynamics mineralizationn, nitrification, denitrification, and leaching), and biological N
Eixation are also simulated. The structure of the model also allows it to be readily
adapted to new plant species.
The model code itself is quite generic, including a minimum of crop-specific
relationships in the source code. Rather, parameters defining the species-unique
responses of the various processes are read from a set of species-specific input files. By
providing alternate parameter files, the behavior of the model can be adapted to simulate
other plant species. Parameter files have already been created to allow CROPGRO to
simulate several crop species including soybean (Glycine max L.), peanut (Arachis
hypogaea L.), dry bean (Pha~seolus vulgaris L.), faba bean (Vicia faba L.), and tomato
(Lycopersicon esculentum Mill.) (Scholberg et al., 1997; Boote et al., 1998a, 1998b,
2002). Modeling perennial tropical grasses poses challenges not encountered when
modeling annual grain or fruit crops. Perennial forages must persist from one year to
another, regrow each spring, and go dormant each fall. Additionally, forages are not
usually allowed to reach physiological maturity and may be harvested several times in a
single growing season. The current CROPGRO structure can accommodate each of these
challenges to some degree. However, some code changes are needed to achieve the
desired level of model performance for perennial forages.
The goal of this proj ect is to adapt the CROPGRO model code and species file to
simulate the growth and composition of bahiagrass over multiple, consecutive growing
seasons. Parameter values will be obtained from the existing literature when available,
directly measured in experiments when practical, and estimated by calibration when
necessary. The Cropping Systems Model or CSM version of CROPGRO which is part of
the Decision Support System for Agrotechnology Transfer (DSSAT) version 4 model
will be evaluated to quantify its ability to predict and determine any limitations for
modeling perennial grasses. Finally, portions of existing code will be changed and some
new code added to create an independent perennial forage version of CSM CROPGRO.
The build date for the code used in our proj ect is 11 July 2003. This is a pre-
release version, but the relevant code is substantially the same as the final version
released as DSSAT v4 (Hoogenboom et al., 2003). To be compatible with the release
version, the PESTCP module was updated, changing the forage cutting code from
MOWE to MOW and changing the stubble mass units from g (leaf+stem) DM m-2 to kg
(leaf+stem) ha- While efforts will be made to maintain compatibility with the existing
CROPGRO model and its input-output structure, the perennial forage version will be
considered to be a "new" model rather than a new version of the current model.
Native to South America (Ward and Watson, 1973) and northward to Mexico
(Scott, 1920), bahiagrass (Paspalum notatum Flaigge) is a perennial tropical grass that
spreads by seed and vegetative stolons. First planted at the Florida Agricultural
Experiment Station in May of 1913 (Scott, 1920), bahiagrass use has spread to over one
million hectares (2.5 million acres) in Florida alone (Chambliss, 2002). Initial reports
praised the grass for its palatability to cattle and especially for its persistence and capacity
to spread and form a sod even under heavy grazing pressure (Scott, 1920). This
persistence along with its ability to tolerate a wide variety of management systems has
been key to the popularity of this species.
While used primarily for grazing beef cattle, bahiagrass is also harvested as hay.
Annual hay yields of near 8000 kg ha-l may be expected when harvested every 4 wk
(Johnson et al., 2001) and may exceed 11000 kg ha-l when harvested every 6 to 8 wk
(Blue, 1973). Bahiagrass responds well to N fertilization, yielding 2700 kg ha-l yr- with
no fertilizer N (Beaty et al., 1964) and increasing to nearly 14000 kg ha-l yr- when
fertilized with very high levels of N and cut to a low stubble height (Pedreira and Brown,
1996a). The most prolific biomass growth occurs below the common cutting height of 5
to 10 cm. The prostrate growth habit of 'Pensacola' bahiagrass produces a large amount
of leaf low in the canopy, with reports of as much as 5600 kg ha-l of leaf DM ( Beaty et
al., 1964) and a leaf area index (LAI) in excess of 1.6 m2 leaf m-2 land remaining after
harvest (Pedreira and Brown, 1996b). Beaty et al. (1968) reported that nearly 40% of the
total forage produced by Pensacola bahiagrass was present in the bottom 2.5 cm of the
canopy. Depending on the fertility and age of the stand, an additional 9800 kg ha-l
(Beaty et al., 1964) or more of rhizome or stolon (the terms rhizome and stolon are used
interchangeably in the literature when referring to bahiagrass) mass were present at the
soil surface. Bahiagrass also has an extensive fibrous root system which may produce as
much as 19700 kg ha-l DM in just the top 15 cm of the soil (Beaty et al., 1964), although
a level near 4500 kg ha-l root DM (Burton et al., 1954) is more common. Pensacola
bahiagrass roots have been reported to extend more than two meters below the soil
surface (Burton et al., 1954; Chambliss, 2002). Although other cultivars may exhibit a
more erect growth habit or lower stubble mass in the seedling year than Pensacola, even
these "improved" cultivars may still have a harvest index of 10% or less (harvested
forage as a proportion of total above-ground biomass present at harvest) after the field
has been established for more than two years (Pedreira and Brown, 1996b). Such an
investment in biomass close to the ground and below-ground encourages the persistence
and quick sod formation that the species is noted for. It also can support the crop through
periods of stress, allowing it to adapt to a variety of management practices.
The most popular bahiagrass cultivar is Pensacola bahiagrass (Chambliss, 2002), a
West Indies-type bahiagrass, thought to have been introduced in ballast offloaded at the
Perdido Wharf sometime prior to 1926 and promoted after 1939 by Escambia County
Cooperative Extension county agent Ed Finlayson for its aggressive sod formation,
persistence, drought tolerance, and palatability to cattle (Finlayson, 1941). Pensacola is
distinguished by long, narrow leaves and tall seedheads. Aside from its persistence,
Pensacola bahiagrass is also more cold-tolerant than many other cultivars (Ward and
Watson, 1973; Chambliss, 2002).
The productive season for bahiagrass is April to November in North Florida. The
season starts in early March in South Florida but growth still slows in October
(Chambliss, 2002), suggesting that the decline in growth may not be solely temperature
dependent. Indeed, recent work by Sinclair et al. (2001; 2003) and Gates et al. (2001)
demonstrate a role of daylength in initiating winter dormancy.
Perennating Organs: Rhizomes and Stolons
While bahiagrass reproduces by seed, the plants spread and form a dense sod
through the growth of rhizomes (equivalent to stolons) (Ward and Watson, 1973;
Chambliss, 2002). Stolons allow the plants to spread and occupy more land area with
nodes on the stolons providing new growing points for additional tiller growth. New
tillers form at an axillary bud in response to the flowering of a nearby culm (Sampaio and
Beaty, 1976), maintaining tiller density and rejuvenating the stand. New tillers may also
form in response to changing light (daylength and light quality) with the onset of spring
or after a harvest (and the possible removal of apical dominance). This new growth
allows the stand to persist from year to year and under grazing or hay management
sy stem s.
The stolons also promote persistence by acting as a storage vessel for C and N
reserves, providing nutrients for growth under stressful conditions and promoting rapid
regrowth after winter or harvest. As was previously mentioned, the mass of stolons may
be several times the mass of leaves, and the reserves may last two to three years
(Chambliss, 2002), providing for new growth. Partitioning to, and mobilization from,
stolons, may at times account for the maj ority of the nutrient flows within the plant.
Studies of carbon allocation and movement in bahiagrass using 14CO2 have shown 50 -
60% of the 14C partitioned to stolons, with only 10% partitioned to new leaf growth
(Beaty et al., 1974). Patterns of mobilization are also affected by harvesting. The same
study reported that re-mobilization of 14C fTOm the surviving tissue ceased within 3 to 6 d
after a severe defoliation. No more than 20% of the 14C assimilated was re-mobilized, of
which only 10% moved to the leaves of the main plant while 20 to 40% went to the
stolons. Such isotope studies characterizing patterns of mobilization and partitioning to
new growth during spring regrowth have not been conducted, but general studies of fall
and spring plant growth suggest that there are regulatory roles for two additional factors:
daylength and temperature. Late and early season growth can be slowed by the cooler
temperatures associated with the fall, winter, and spring months. This is also the period
of the year when daylength is near its minimum.
Forage production from tropical grasses like bahiagrass and bermudagrass
[Cynodon dactylon (L.) Pers.] drops dramatically in the late summer and fall months in
the southeastern United States despite continued warm temperatures. Gates et al. (2001)
quantified this reduction in yield in bahiagrass by measuring forage growth through two
successive fall and winter seasons concurrently at Ona, FL and Tifton, GA. Mean
temperatures were 60C cooler in Tifton where multiple freeze events occurred while there
was only one night in the two growing seasons where temperatures dropped below 00C at
Ona. The pattern of seasonal forage production was similar for both sites; however,
production at Tifton was lower than for Ona. At Ona daily forage production (leaf and
stem mass recovered above the mowing height) decreased from 36 kg DM ha-l dl to 8 kg
DM ha-l dl between 23 September 1993 and 8 November 1993, a 78% decrease in
growth rate. The growth rate remained between 5 and 15 kg DM ha-l dl through 7
March, then increased to 35 kg DM ha-l dl by 6 April. Cooler temperatures in the
second growing season (1994-1995) resulted in a similar pattern of forage growth but
lower overall yields and a more rapid increase in growth rate in the spring. Although
there was no difference in growth rates between the three cultivars tested (Pensacola,
Tifton 9, and RRPS Cycle 18) on dates when minimum growth rates were realized,
regrowth of Pensacola in September and April was considerably lower than that of the
newer, higher yielding cultivar, RRPS Cycle 18. No information on the cause of the
seasonal reduction could be discerned from the study other than the generally lower
yields at lower temperatures.
Growers in Georgia have attributed the low productivity of tropical grasses in the
fall to a lack of rainfall and fertilizer (Burton et al., 1988). Investigating this hypothesis,
Burton et al. (1988) compared the forage yields of well-fertilized, irrigated and non-
irrigated Coastal bermudagrass (another C4 graSs that exhibits winter dormancy)
harvested every 24 d from 1 April through 27 October for three consecutive years. A
seasonal decrease in forage yield was observed in all treatments during the fall and
winter. The magnitude of the drop was quite dramatic with yields for the September -
October harvest being only slightly more than 1/3 the yield of the May June harvest
period despite slightly higher temperatures in the fall. Correlation analyses of the data
showed only a moderate relationship between yield and temperature (r-0.46) or growing
degree days (GDD) (r=0.37), thus, discounting the roles of temperature, rainfall, and
fertilizer in the seasonal yield reduction pattern. Two variables that did show a strong
correlation with yield were daylength (r=0.95) and daily solar radiation (r=0.93). It was
not possible to separate these two effects with this dataset. Stepwise regression was also
employed using individual growth period data for all years. The single variable model
with the best fit was daylength (r2= 0.64, 0.69, and 0.61 for all yields, irrigated yields
only, and non-irrigated yields, respectively). When daylength was excluded from the
model, the best fit for a single variable was rainfall + irrigation (r2=0.45). Adding solar
radiation to the rainfall model only increased the R2 to 0.57, while the daylength +
rainfall gave an R2 Of 0.73 for all yields and 0.77 for irrigated yields only. A model with
daylength and water deficit (pan evaporation rainfall) had an R2 Of 0.71 when fit to the
non-irrigated yield data. Stepwise regression did not compensate for the confounded
effects of daylength and total solar radiation but did lend more weight to the daylength
Marousky et al. (1992) attempted to separate daylength effects from solar radiation
effects in a study of daylength effects on turf-type bermudagrass cultivars. Long days
were simulated using the same 9-h daytime as the short-day treatment but the dark period
was interrupted with 4 hours of light from incandescent lamps. Interruption of the dark
period with light causes the reversal of many red/far-red phytochrome reactions; in effect,
simulating a much longer day than the total time exposed to light. Long-day plants
exhibited a greater degree of leaf extension but no change in plant dry weight or number
of stolons produced. The authors' rationale for the apparent conflict with the results of
Burton et al. (1988) was that cultivar differences resulted in different responses between
the long-leaved, forage-type Coastal bermudagrass and the fine-leaved, turf-type
Another possible explanation might be found in work by Britz et al. (1958) who
studied short-day (late-season) accumulation of starch in leaves of another tropical grass,
Digitaria decumbens Stent.. This accumulation of starch was associated with a decrease
in translocation of assimilate under short days. The authors used several methods to elicit
a long-day response. Gradual extension of the daylength from 7-h (short-day) to 14 h
(long-day) with full-spectrum, full-irradiance light (400-600 Cpmol m-2 S-1 photosynthetic
photon flux density) resulted in no decrease in starch accumulation with 9 h daylength
but a dramatic drop between 11 and 12 h of light with starch concentrations after 12 h of
light equal to levels at 14-h daylength. In contrast, interrupting the dark period with 0.5 h
or 2.0 h of full-spectrum light resulted in a partial (~40% and ~60%, respectively)
reduction in starch accumulation relative to a 14 h day. The full long-day response was
achieved when the 7 h light period was shifted from the first half of the 24 h period to the
second half. These results implied that timing of the light periods may be more important
than total time of exposure. To separate the effects of daylength and total solar radiation,
plants were exposed to 14 h of amber light (589 nm), having a good photosynthetic
spectra but lacking photomorphogenetically-active wavelengths. Photosynthate
production under long days of amber light was similar to that of the long-day plants
under full-spectrum light but starch partitioning was similar to that of the short-day
plants. This helped to solidify the conclusion that the starch accumulation was a
photoperiodic response but also indicated that it involved a more complex mechanism
than a single red/far-red phytochrome response. The different responses of the long day
and interrupted night treatments may indicate a similarly complex mechanism in the
short-day response of bermudagrass.
Sinclair et al. (2003) avoided these complications by extending the natural
daylength over field-grown Pensacola bahiagrass, 'Florakirk' bermudagrass, 'Tifton' 85
bermudagrass (Cynodon spp.), and 'Florona' stargrass (Cynodon nlemfuensis Vanderyst)
to 15 h using halogen lamps during the short daylength months. Light levels under the
lamps were equivalent to less than 2% of full sunlight and, thus, not considered to
contribute significantly to total daily radiation. While all four species had higher forage
production under the extended daylength treatment, bahiagrass showed the greatest
response. Extended daylength yields were frequently more than twice those of the
natural daylength treatment with yields six times greater on one harvest date. Despite the
increases, mid-winter yields under extended daylength were still quite low compared to
summer yields. Although forage growth increased, below-ground biomass production
was not affected by extending the light period. Another factor potentially related to
dormancy, total nonstructural carbohydrate (TNC) concentration in the below-ground
biomass (root + stolon), decreased during the short-daylength months in both the natural-
and extended-daylength treatments. All this paints a picture of fall dormancy in tropical
grasses being triggered, at least in part, by daylength. The exact mechanisms have not
been described but the relatively low forage production even under extended daylength as
well as the continued loss of TNC during the winter, point to a complex of factors
contributing to the response rather than a simple phytochrome-mediated response.
The large DM production of bahiagrass can largely be attributed to it being a
tropical grass species expressing the C4 photosynthetic pathway. Temperate grasses
express the C3 photosynthetic pathway wherein CO2 is fixed by ribulose 1,5-bisphosphate
carboxylase/oxygenase (Rubisco) in the chloroplasts to produce 2 molecules of 3-
phosphoglyceric acid (3-PGA; a 3-carbon acid) and, eventually carbohydrate via the
Calvin cycle. Rubisco is also an oxygenase, capable of fixing 02 aS well as CO2. When
Ol is fixed, one molecule of 2-phosphoglycolate and only one molecule of 3-PGA are
produced. Not only is there half as much 3-PGA produced when Oz is fixed, two
molecules of glycolate can be metabolized to release a molecule of CO2, hence the term
photorespiration referring to oxygenase activity. Both CO2 and 02 COmpete for binding
sites on Rubisco, with higher relative concentrations of 02 in the chloroplast resulting in
higher rates of photorespiration and reduced photosynthetic efficiency. In C4
photosynthesis, CO2 is fixed in the mesophyll cells (cells near the leaf surface) by
phosphoenolpyruvate carboxylase (PEPCase) to form oxaloacetate (OAA; a 4-carbon
acid). The OAA is subsequently converted to either malate or aspartate, depending on
the type of C4 pathway expressed in the species, and transported to the bundle sheath
cells isolated deep within the leaves. In the bundle sheath cells, the malate or aspartate is
decarboxylated near the chloroplast where Rubisco re-fixes the CO2 to produce 2
molecules of 3-PGA as in C3 photosynthesis. Energetically, C4 photosynthesis is more
expensive with the additional cost of 2 ATPs to fix HCO3- with PEPCase on top of the
cost of 3 ATPs and 2 NADPHs to fix one mole of CO2 in the C3 pathway (Kanai and
Edwards, 1999). However, PEPCase has no oxygenase activity and, thus, can be more
efficient than Rubisco in fixing CO2 in the presence of Oz. Also, by isolating Rubisco in
the bundle sheath and "shuttling" the CO2 to it, CO2 is concentrated around the Rubisco
and photorespiration is minimized. Under atmospheric conditions (~21% 02 and 370
ppm CO2), photorespiration in C4 plants may be on the order of 3% of the net rate of CO2
fixation, compared to rates of 54% observed in the C3 Species wheat (Triticum aestivum
L.) (Kanai and Edwards, 1999). Reduced photorespiration not only increases the
efficiency of Rubisco carboxylation, it also increases the light level at which light
saturated photosynthesis occurs, lowers the CO2 COmpensation point, increases quantum
efficiency (QE), changes the temperature sensitivity of both QE and photosynthesis, and
allows high photosynthetic rates at relatively low concentrations of leaf N or
The higher carboxylation rate resulting from CO2 Saturation of Rubisco and the
reduced photorespiration enables C4 plants to potentially attain higher photosynthetic
rates at high light. While C3 photosynthesis becomes light saturated at relatively low
light levels, the enhanced capacity of the C4 System can tolerate very high light levels
without becoming light saturated. Ludlow and Wilson (1971) compared leaf net
photosynthetic rates of tropical grasses (C4) with tropical legumes (C3) OVer a range of
illuminances. Consistent with the capacity to respond to higher light levels, the C4
grasses had net photosynthetic rates (37.8 Cpmol CO2 m-2 S-1 ) that were approximately
double those noted for the C3 legumes (17.7 Cpmol CO2 m-2 S-1). The light response
curves of the legumes reached a plateau at around 4000 5000 foot-candles of light
while the curves for the C4 graSses were only approaching saturation at the highest
illuminance of 10 000 foot-candles. Boote et al. (1999) measured leaf net photosynthesis
(Pn) near 28 Cpmol CO2 m-2 S-1 in established bahiagrass under atmospheric conditions
compared to approximately 20 Cpmol CO2 m-2 S-1 for the C3 legume rhizoma peanut
(Arachis glabrata Benth.). When the CO2 COncentration was doubled from 350 CLL L^1 to
700 CLL L- bahiagrass responded with only a 20% increase in Pn, about half of the 36%
increase measured for rhizoma peanut (Boote et al., 1999).
Higher photosynthetic efficiency also allows the CO2 COmpensation point (CCMP)
(atmospheric CO2 COncentration where the rate of CO2 uptake by photosynthesis equals
the rate of CO2 OffluX) Of C4 plants to be considerably below that of C3 plants. Bolton
and Brown (1980) recorded CO2 COmpensation points of 4-14 CIL L^1 for the C4 graSS
Panicum maximum, much lower than the values (47-59 CIL L^1) measured in tall fescue
(Festuca arundinacea Schreb.), a C3 graSs. Several other C4 graSses have been shown to
have CCMPs near 0 pIL L^1 including Vetiveria zizanoides (0 to 5 CIL L^)~, and a variety of
Cymbopogon species (0 to 3 CIL L^1) (Rajendrudu and Das, 1981).
Efficiency of light utilization may also improve with reduced photorespiration.
Quantum efficiency or quantum yield is the efficiency of leaf photosynthesis when
measured at low light and is generally expressed as Clmol CO2 ClmOl-1 absorbed photons.
This describes the initial slope of the photosynthetic response to light (light-limited
region). A frequently cited QE value for C3 Species is 0.05241Clmol CO2 ClmOl-1
absorbed photons, the average QE measured at 330 CLL CO2 L^ for seven C3 Species
(Ehleringer and Bjoirkman, 1977). Values for C4 Species are generally higher than those
for C3 Species and may range from 0.046 for Sorgha~strum nutans (Monson et al., 1982)
to 0.075 for Saccharum spontaneum (Meinzer and Zhu, 1998). Differences in efficiency
exist between the three types of C4 pathways (NAD-ME, NADP-ME, and PCK-type)
(Ehleringer and Pearcy, 1983) as well. Ehleringer and Pearcy (1983) found that species
expressing the NADP-ME pathway, [e.g. bahiagrass and maize (Zea mays L.)] exhibited
the highest QE with eight monocot species averaging 0.065 Clmol CO2 ClmOl-1 absorbed
photons. Reports of QE for other NADP-ME species range from 0.062 to 0.075 Clmol
CO2 ClmOl-1 absorbed photons for various sugarcane (Saccharum) species at 350 ppm
CO2 (Meinzer and Zhu, 1998). Quantum efficiencies for other C4 pathway types
recorded by Ehleringer and Pearcy (1983) were 0.064 Clmol CO2 ClmOl-1 absorbed
photons for five monocot species exhibiting the PCK-type pathway, and 0.060 Clmol CO2
Clmol-l absorbed photons for three NAD-ME species. It should be noted that both the
Ehleringer and Pearcy (1983) and Ehleringer and Bjiirkman (1977) studies measured QE
at 330 ppm CO2 rather than the customary 3 50 ppm CO2. The narrow range of values for
a given pathway when measured across many species suggests that QE may be a
characteristic that is highly conserved across species. This would imply that values for a
particular pathway type may be generally applied to other species exhibiting that
photosynthetic pathway and that the QE for bahiagrass is near 0.065 Clmol CO2 ClmOl-1
While the QE of C4 plants is generally greater than that of C3 plants at higher
temperatures, the QE of C3 plants may exceed those of C4 plants at temperatures below
300C (Ehleringer and Bjiirkman, 1977; Ku and Edwards, 1978; Monson et al., 1982).
The QE of C4 plants is temperature insensitive, decreasing very little as temperatures
increase while the QE of C3 plants falls dramatically as temperature increases (Ehleringer
and Bjiirkman, 1977; Ku and Edwards, 1978; Monson et al., 1982; Ehleringer and
Pearcy, 1983). The decrease in C3 QE at higher temperatures is attributed to increased
photorespiration (Ehleringer and Bjiirkman, 1977; Monson et al., 1982). The solubility
of CO2 decreases relative to that of 02 aS temperature increases, creating a condition
where Ol COncentration around Rubisco may be enhanced, resulting in increased
oxygenase activity relative to carboxylation. This effect alone cannot account for the
drop in QE of C3S at high temperatures (Ehleringer and Bjiirkman, 1977; Ku and
Edwards, 1978), rather it is likely the combined effects of the changing relative gas
solubilities and a changing affinity of Rubisco for CO2 and 02 (JOrdan and Ogren, 1984)
that favors 02 at higher temperatures. The CO2-COncentrating mechanisms of the C4
pathways provide a high concentration of CO2 relative to 02 in the bundle sheath
chloroplast such that photorespiration effects are not evident (Jordan and Ogren, 1984).
Consistent with a higher QE at high temperatures, the temperature optimum for
photosynthesis in C4 plants is generally about 100C higher than for C3S (Long, 1999).
Conversely, at lower temperatures, C4 photosynthetic rates may be below that of
comparable C3 plants. The higher temperature optimum may be explained by the lack of
photorespiration in the C4S, but the mechanism behind the decreased performance at
lower temperatures has been more elusive. In an extensive review of C4 photosynthesis
at low temperatures, Long (1983) explored the effect of low temperatures on several of
the steps in the C4 pathway(s). No single step could be shown to be the limiting factor at
low temperatures. Two enzyme steps were identified to be the most likely to limit C4
photosynthetic rate at low temperatures: pyruvate Pi dikinase (PPDK) activity or Rubisco
activity. Pyruvate Pi dikinase was suspected for its relatively low activity at all
temperatures and the dramatic increase in its activation time under cold conditions.
Rubisco limitation might occur due to the low quantities of the enzyme in C4 leaVCS along
with a possible inhibition of the CO2 COncentrating mechanism under cold temperatures.
In a more recent review of responses of C4 photosynthesis to various environmental
factors, Long (1999) revisits his investigation of low temperature effects, this time
refuting the argument for PPDK limitation of photosynthetic rate and concluding that
"there is no inherent flaw in C4 photosynthesis that prevents efficient operation at low
temperatures" (p. 240) and citing M~iscanthus as an example of a C4 Species capable of
persisting at high altitudes/low temperatures. Recent work by Pitterman and Sage (2000)
using two ecotypes of Bouteloua gracilis Lag., adapted to high or low elevations
implicates Rubisco capacity as limiting photosynthesis at temperatures below 170C.
They also identify PPDK activity or ribulose 1,5-bisphosphate (RuBP) regeneration as
potentially limiting photosynthesis between 200C and the optimum temperature (~370C),
and PEPCase capacity as potentially limiting photosynthetic rate at temperatures greater
than the optimum. That C4 photosynthesis could have a rate lower than the C3 rate at low
temperatures if Rubisco was limiting in both pathways was attributed to the low Rubisco
content of C4 leaVCS coupled with the added cost of carboxylation of PEP in the C4
The low Rubisco content of C4 leaVCS (Pittermann and Sage, 2000) contributes to a
lower leaf N concentration and smaller proportion of soluble leaf N than is typically
found in C3 plants (Slack and Hatch, 1967; Crespo et al., 1979; Sugiyama and Hirayama,
1983; Usuda et al., 1984). While Rubisco may contribute over 40% of the soluble leaf N
in a C3 leaf, it represents only about 5-28% of soluble N in C4S (Slack and Hatch, 1967;
Sugiyama and Hirayama, 1983). The added N present in PEPCase (4-10% of soluble leaf
N (Slack and Hatch, 1967)) does not match the relative decrease in Rubisco
concentrations, resulting in a lower total quantity of photosynthetic enzymes (PEPCase
plus Rubisco) and lower total N concentration in C4 leaVCS compared to C3S (Slack and
Hatch, 1967; Brown, 1978; Raghavendra and RamaDas, 1993). Leaf N concentration can
be increased by increasing N supply to the plant. Sugiyama and Hirayama (1983) found
that Rubisco, PEPCase, and PPDK quantities all increased with increasing N supply.
However, while the concentration of PEPCase and PPDK increased (as a proportion of
DM), the concentration of Rubisco decreased relative to C3 plants even as leaf N
concentration increased with N supply. The high photosynthetic rates at low leaf N
concentrations result in a high N-use efficiency, a factor critical to the productivity of
many tropical grasses in low input systems. Conversely, their low crude protein (CP)
concentration coupled with high cell wall concentrations (largely a consequence of the
extensive vascular system associated with the Kranz anatomy and bundle sheath) are
primary contributors to the relatively low forage quality of most tropical grasses.
The CROPGRO Model
The goal of this section is to familiarize the reader with the maj or features of
CROPGRO pertinent to adapting the model for perennial tropical forages. Discussion
will be limited to a review of the pattern of information flow and summaries of some of
the plant-related subroutines. More complete descriptions of CROPGO have been
published by Boote et al. (1998a, 1998b) with an extensive review of the hedgerow
photosynthesis approach by Boote and Pickering (1994).
One of the primary obj ectives in developing CROPGRO was to have a model that
could easily be adapted to simulate the growth of different plant species. CROPGRO
was created as a way to consolidate the existing SOYGRO, PNUTGRO, and BEANGRO
models into a single program (Boote et al., 1998a). The three programs shared much of
their code, so merging the models primarily involved moving the parameters describing
species and cultivar traits from the actual code to external input files. This structure
allows other species to be modeled by creating new parameter (input) files.
CROPGRO was developed in the early 1990s as a stand-alone model but could also
be run under the Decision Support System for Agrotechnology Transfer (DSSAT) shell
(ICASA, 1998), allowing it to be linked to other crop modeling programs as well as
graphics programs to automate presentation of results. After DSSAT version 3.1 was
released in 1996, the CROPGRO code was reorganized into a modular structure. Code
for simulating different plant and soil processes was organized into individual
subroutines for each process. The new subroutines were designed to be executed in four
common steps initializationn, rate calculation, integration and Einal/summary) called by
the main model. The modular structure was designed to allow users to add new code or to
link code from other programs by inserting new modules into the CROPGRO structure.
This approach has been applied to the DSSAT shell such that CROPGRO is an integral
component of the Cropping Systems Model (CSM) (Jones et al., 2003). CROPGRO
serves as the crop template module (Hoogenboom et al., 2003; Jones et al., 2003); a
universal interface for modeling several different species. An overview of DSSAT
version 4, which incorporates the CSM version of CROPGRO, can be found in the
DSSAT v4 documentation (Hoogenboom et al., 2003) and in Jones et al. (2003).
The modular version of CROPGRO was the first version to include the soil organic
matter (SOM) transformation module based on the CENTURY model (Gij sman et al.,
2002). The only SOM transformation option available in earlier versions was an
adaptation of the PAPRAN model (Godwin and Jones, 1991). Both options address
mineralization from SOM as well as immobilization, nitrification, and denitrifieation of
N. The CENTURY option adds the capability of simulating decay of surface residues
and the movement of those nutrients into the soil profie. Pasture and other perennial
forage systems are often low-input systems relying heavily on recycled nutrients from
plant residues for continued productivity. Omission of surface litter nutrient pools from
the PAPRAN option may dramatically reduce the available soil N and C pools,
potentially resulting in chronic underprediction of plant growth especially over the multi-
year simulation periods typical for perennial forage systems. Thus, the CENTURY
option is better suited to modeling perennial forages.
CROPGRO and CSM incorporate several modules for simulating environmental
and management responses, soil N transformations, soil water availability, etc. (Figure 2-
1). Individual modules depicting different processes are executed once within each of the
four steps of the modular structure. This modular structure should enhance the future
development of these programs as it allows others to add new features by inserting their
own modules. Also, by separating the rate and integration steps, the order of execution of
the modules within each step is generally less critical, again, facilitating further
In the initialization step, parameter values and simulation control information are
read from various input Hiles and initial values set for state variables (variables
representing the state of the system at the end of the day, variables that have a quantity
such as mass of roots or number of leaves). This step is run once per simulation,
although there are provisions to re-run each simulation for multiple years where the
initialization step is run once per repetition.
Several input files are used to set the parameters for a species and control the
execution of the simulation. Three Hiles contain the plant parameters: a species Eile, a
cultivar fie, and an ecotype Eile. The experiment file, or "X-file" controls the simulation
and is supplemented by weather files, a soil information file, and a pest file. Comparison
of predicted data to measured experimental results is automated, with the program
P rim a ry M od ule s *Environmental Modifications
MIain Weather Havstn
Pro gra m
Start I ILFertilizer Application
InitiaiRuzation LIod Ini Ma ag me t ol~yamc
ca II I Soll T em p era tu re
Seasonal I IISollWater
Soll Nltrogen & Carbon
Rate (TeadntTemplate Crop Models
S Calculations MhoeduMe InsP rda Soyea
to perform each step
of processing and In Peanut
tu rn calls each of th e
Integration I IPrim ary Mo d ule s ) Dyba
Soll Plant -Atmosphe re
Output I IIOther crops
Pest& Disease Damage-
CROPGRO Plant Template Plant Modules
End CERES Wheat
1 Plant CERES Rice
Figure 2-1. Modular structure and summary of model components of the DSSAT-CSM cropping systems model (Jones et al., 2003, p.
reading the measured data from a time-course field data file or "T-file" or a field average
observational data file or "file A".
The species file contains species-specific parameters describing the response of the
crop to the environment as well as parameters describing growth and photosynthesis.
These parameters are set during model development and are not generally altered by the
user. Some parameters may be adjusted to reflect differences in behavior of different
cultivars and ecotypes via additional parameters in the cultivar and ecotype files.
Parameters in these files include factors such as differences in physiological time
between growth stages, relative differences in photosynthetic rate, and leaf size, among
Other files, such as the X- and T-files, must be created by the user. A file X
contains information describing the simulation, including what weather files to use, what
soil type is present (along with site-specific soil profile information), location of the
experiment site, management information (planting date, fertilization and irrigation
schedules, harvest date, etc.), and simulation controls such as which photosynthesis
option to use, whether to predict potential growth (assume no stresses), or water-limited
growth, or water- and N-limited growth, and other options that determine modules of the
model to be used.
If the simulation results are to be compared to an actual experiment, measured data
for each treatment can be listed in a time-course field data file or "T-file" or the field
average observational data file (file A). Data from this file can be read by a graphics
program to plot the predicted variables against the measured data. The T-file also can be
used to list amounts of pest damage and the dates on which the damage is to be imposed.
Based on the pest codes used in the T-file, the Pest file determines what type of damage
to impose. The Pest file is generally written during model development, tailoring the
codes to the specific crop being modeled. Because the "harvest" date listed in the X-file
signals the termination of simulation in CROPGRO, periodic forage harvests must be
simulated using the pest code, "MOW" that is entered in the T-file. The user specifies
the amount of stubble mass to remain after a harvest or grazing event and CROPGRO
reduces the amount of leaf and stem proportionally to leave that amount of stubble at a
reduced V-stage. As with any pest damage, the leaf and stem removed by MOW are lost
from the simulated system, with the assumption that they were exported from the field.
While annual species may be simulated without enabling the Pest option, all forage
simulations must use the Pest option to create multiple harvests.
Other site-specific information that must be provided by the user is daily weather
information and soil profile information. Daily solar radiation, rainfall, and minimum
and maximum temperature data are listed in the weather files which are named according
to their location and the year for which they contain data. The soil file (SOIL.SOL)
contains soil profile information on the specified soil types. The user must add
information to create the soil for the simulation in a specified format. CROPGRO
requires all of the above mentioned files except the T-file and A-file to initialize and run
a simulation, unless the simulation involves multiple harvest, in which case the T-file is
required as well.
Initialization of a simulation begins on the simulation start date, which may be well
in advance of the planting date. Planting date can be the day that seed is sown or the day
that transplants are placed in the field. The transplant option can be used to initialize the
simulation with an established stand of grass. Sowing date triggers the start of plant
growth simulation. Separating the beginning of the simulation from the sowing date
allows the model to predict soil conditions on sowing day, which may be helpful if the
user has limited information about the actual conditions existing at the site on that day.
The rate step is run once on each day of the simulation and calculates the rate
variables (variables representing the amount of change in a state variable occurring over a
specified period of time usually one day or less) for the current day of the simulation.
For each simulated day, prior to running the CROPGRO plant template, the date, daily
minimum and maximum temperatures, daily rainfall, and total daily solar radiation are
read from the weather file, and the X- and T-files are checked for any management
operations (e.g. planting, irrigation, harvest, or pest damage) for the day. The weather
and management information is fed into the soil processes module, predicting the rates of
change of available soil water, NH4 and NO3-. The weather data along with plant and
soil information from the end of the previous day is used to calculate daily
Predicting plant phenology or stage of maturity is a key component of CROPGRO.
Several parameters such as partitioning of new growth between plant organs (leaf, stem,
roots), and daylength sensitivity of development are dependent on or may vary with the
stage of maturity. Stage of maturity is a function of accumulated physiological time
which is a combination of time and temperature. Cardinal temperatures (minimum,
optimal, and maximum) for various stages of development of each crop are described in
the SPE file. Generally, there is no development at temperatures below the minimum or
above the maximum cardinal temperatures whereas plants mature at a faster rate as air
temperature nears the optimal range. This rate of change/progression towards the next
stage can be altered by daylength and water stress. Progress during vegetative growth is
typified by progressive increase in the number of leaves per plant (V-stage) and,
beginning with floral induction, by progression through a series of reproductive stages
The daily rate of pest damage is also calculated in the rate step. Information on the
extent of damage is listed by the user in the "T-file" and CROPGRO uses the previous
day's plant mass and leaf area to calculate "actual" damage. Codes in the Pest fie allow
the user to specify the amount (kg DM ha l) or proportion of leaf and stem to be removed
or the reduction in assimilate production due to disease or pest damage. As with all other
rate variables, the pest damage rates are not deducted from the existing plant mass until
the integration step which signals the end of the current day. This ensures that all new
rates are based on the same conditions the plant mass at the end of the previous day.
The calculation of photosynthetic rate presents an interesting contrast with animal-
based models. Many animal nutrition models are sink-driven, predicting nutrient intake
as a function of body weight and animal performance while, in CROPGRO, the supply of
assimilate, a function of absorbed solar radiation and photosynthesis, determines plant
weight and performance (source-driven). One exception is that some plant species may
exhibit a "juvenile" period where limited seedling demand can feed back on
photosynthate production. CROPGRO offers two options for predicting photosynthesis:
a daily canopy photosynthesis option and an hourly leaf-level photosynthesis option.
The daily canopy photosynthesis option is the simpler of the methods and uses an
asymptotic exponential response to daily solar radiation to calculate the potential daily
photosynthetic rate. This pattern of daily photosynthetic response to light is defined by
two parameters from the species file; the maximum canopy photosynthesis rate and the
amount of PAR at which photosynthesis is 63% of maximum. Estimated light
interception of the canopy is a function of the predicted LAI on the previous day and the
canopy light extinction coefficient specified in the SPE file. The hourly leaf-level
photosynthesis option is more mechanistic (based on processes and stoichiometry) but
more mathematically complex.
The leaf-level option uses the hedgerow approach described in Boote and
Pickering (1994) to estimate potential hourly photosynthetic rates for sunlit and also
shaded leaf area portions of the canopy. The hedgerow approach uses parameters
describing canopy shape, height and width, leaf angle, row width and direction, latitude
of the site, day of year and time of day along with the predicted LAI to estimate light
absorption by sunlit versus shaded leaves. Hourly distribution of solar radiation and
temperature are estimated from the daily values provided in the weather file and further
divided into direct and diffuse components. Potential hourly leaf photosynthetic rate is
calculated using an asymptotic exponential equation where quantum efficiency (initial
photosynthetic response at low light) sets the initial slope of the response and the
maximum potential leaf photosynthetic rate is the asymptote. The photosynthetic rates
for the sunlit and shaded leaves are multiplied by their respective LAls and summed each
hour to calculate the hourly canopy photosynthetic rate. The 24 hourly rates are
integrated to yield the daily photosynthetic rate. In both the daily canopy and hourly leaf
photosynthesis options, the potential photosynthetic rate may be adjusted for cultivar
differences, temperature, leaf N concentration, leaf thickness (specific leaf weight, SLW),
atmospheric CO2 COncentration, and an incomplete canopy (light absorption).
In the integration step, updated maturity stages are calculated by computing the
daily rate of change for each day and adding this to the previous day's stage rating. All
other rates are processed similarly to calculate the new day's state variables. Much of the
mechanics of the integration step is an exercise in carbohydrate (CH20) accounting and
Potentially available CH20 from stored reserves is calculated and added to the
daily photosynthate production to determine the maximum amount of CH20 available for
the day. The day's maintenance respiration costs and any assimilate loss due to pest
damage are subtracted from the total, the remainder being the amount of CH20 available
for nutrient uptake and growth. Potential CH20 demand for seed and shell growth, CH20
cost per g of new vegetative growth (based on proportions of leaf and stem predicted by
the new V-stage), as well as potential N demand to "refill" N that has been mobilized
from old tissue are calculated and subtracted from the remaining available CH20.
The amount of new growth that can be produced from the available CH20 depends
on the cost of that new growth which is a function of its composition. Composition of
total new growth is, in turn, determined by the partitioning of new growth between the
different plant organs (leaf, stem, roots, seeds, etc.) and the protein, carbohydrate
(comprised of cell wall and starch), lipid, lignin, organic acid, and ash concentration of
each organ's new growth. Parameters describing partitioning to organs and organ
composition are listed in the species file. Coefficients developed by Penning de~rries et
al. (1974) describing the cost to assemble each of these components (both direct cost of C
for C-skeletons as well as the energy used in the biochemical pathways to form them)
expressed in glucose equivalents are used along with the composition parameters to
calculate the glucose cost of new tissues. Crude protein or N concentrations of new leaf,
stem, and root growth can vary within a range of values set by three parameters; a
maximum concentration for new growth, a "normal" growth concentration and a residual
concentration left after senescence. If there is adequate CH20 available but available N
is limiting, new growth can occur at reduced N concentrations.
Nitrogen demand can be met by two different sources; "Actual" N uptake by the
plant, and if this uptake cannot meet the N demand for new growth, N mobilized from
vegetative tissues. Once the balance of N uptake and mobilization are calculated, the
CH20 cost of N uptake and any N mobilization is subtracted from the remaining
available CH20 to update the energy budget. If the crop is a legume, N-fixation is
estimated, the N added to the available N pool, and the CH20 cost subtracted from the
available CH20. The remaining CH20 is allocated to new growth; first to seed and shell,
then to vegetative growth.
The increased root length associated with the day's predicted new root growth is
calculated. The total root length and its distribution are used in determining nutrient and
water uptake on the succeeding day. Losses are also calculated; the day's predicted
senescence of leaves, stems, and roots are estimated as well as any damage due to
Finally, all of the gains in new growth and losses due to senescence, pest damage,
and frost are added to the previous day's pools of tissue mass and a new total is
calculated representing the plant mass present at the end of the current day. This ends the
"day" and after outputs are printed to their respective files, the integration step ends and
the rate step is repeated the next day.
The rate and integration steps are repeated daily until the simulation is terminated
on the harvest date specified in the X-file or when a killing freeze occurs. At this time
the final step is executed to finish printing the output files. These are the basic mechanics
and information flow within CROPGRO. The structure has been shown to work well
with annual grain crops and legumes as well as tomato. In adapting the model to
simulate the growth of perennial tropical grasses, accurate parameters must be developed
for the species file. Any plant processes unique to these plants that are not already
included in the model may require re-definition of some of the parameters or even
changes and additions to the model code.
The ultimate measure of a model's performance is the user' s satisfaction with both
the accuracy of predictions and overall utility of the model. Understandably, such a
measure is difficult to quantify and is relevant only to the user that generated the rating.
Statistical approaches to quantify the accuracy of model predictions provide standardized
measures of model performance. Unfortunately, even these methods do not provide
completely clear-cut conclusions about the accuracy of model predictions. Use of vague
terms like "fairly close" in instructions for interpreting various measures impart an air of
skepticism around the use of some of these methods. Given these caveats, the use of
several different measures of performance to evaluate a model may present a more
complete picture of model performance than any single measure and allow the user to
weight individual results according to their priorities.
Two measures that are commonly reported in the literature are the sample
correlation coefficient r and coefficient of determination T2. The correlation coefficient
provides a measure of the linear relationship or closeness between predicted and observed
values. Interpretation ofr is quite general. An r of 1.0 indicates perfect prediction by the
model with positive values of r indicating some level of a positive correlation between
the predicted values (Pi) and observed values (Oi). Conversely, an r of 0.0 indicates no
correlation of the model to reality whatsoever and negative values indicate an inverse
relationship. The coefficient of determination is informally described as the proportion
of the variance of the observed values that can be accounted for by the model. This
measure has more utility in that it presents an idea of how thoroughly the model
represents the system. Statistical analyses demonstrating the level of significance of r
only proves that a linear relationship with a non-zero slope exists between Pi and Oi
(Snedecor and Cochran, 1989). The validity of this conclusion can come into question if
Pi and Oi do not meet the underlying assumptions required for the particular analysis used
(Willmott, 1981). In spite of their popularity, these measures provide little detail to
characterize the relationship between Pi and Oi.
A simple method of visualizing the relationship between Pi and Oi is plotting a
scatterplot of Pi (Y-axis) and Oi (X-axis), relative to a line designating a 1:1 relationship.
While not quantifiable, some relationships (e.g. consistent underprediction) become
apparent. Scatterplots also provide a common sense check for more sophisticated
methods of evaluation. If results of a test do not appear consistent with the results of the
scatterplot, the test should be re-evaluated. The relationship between Pi and Oi presented
in the scatterplot can be quantified using linear regression. The slope of the regression
line (a) and its Y-intercept (b) may provide evidence of systematic error in the model,
providing quantities that can be compared across models. A slope of 1.0 with a Y-
intercept equal to 0.0 indicates perfect fit of the model predictions. These results along
with the means ( P and O ) and standard deviations of the predicted values and observed
values should be considered for their own merit as well as their use in calculating other
measures when evaluating model performance.
Difference measures, derived from the fundamental quantity (P1-Ox) (Willmott,
1982), build on the statistical measures listed above to quantify bias and average error.
Root mean squared error (RMSE) describes the average difference between Pi and Oi.
RMSE = '1 -- (Eq. 2-1)
Also, RMSE can be readily compared against the mean of the observed values for
comparison of relative error. Both RMSE and its square (mean square error or MSE) can
be subdivided into systematic (RMSEs and MSEs) and unsystematic (RMSE, and MSEu)
components (Willmott, 1981):
f(P,-O,)" 1(Pi-Ox)1 i(P,-Pi)
MSE = MSs= S1 (Eq 2-2)
n n n
where n= the number of pairs of predicted and observed data, and Pi=aOl+b When the
systematic component is minimized, the model is predicting at its maximum possible
accuracy and the primary source of error is not model-related. An alternative
presentation is offered by Roseler et al. (1997) where mean square prediction error
(MSPE), which has the same mathematical definition as MSE (Neter et al., 1990; Roseler
et al., 1997), is considered as the sum of three components: mean bias (O-P) line
bias Sllb~ and random variation around the regression line [S21], where
Sp2 and So2 ar the variances of the predicted and observed values. These measures
provide insight not just on the magnitude of error but also hint at the potential sources of
Willmott (1981; 1982) proposed another measure of model performance that he
called an "index of agreement". This is referred to elsewhere as the d-index. The d-
index describes the degree to which the observed data are accurately estimated by the
predicted data. More formally, it specifies the degree to which the deviation of the
observed data around O corresponds with the deviation of the predicted data around O,
both in magnitude and sign.
1 (P -O1)2
d=1-~ (Eq. 2-3)
where P'1=P1-O and O'1=01-O Potential values of d range from 0 to 1, with 1.0
indicating perfect agreement between predicted and observed data and 0.0 indicating
complete disagreement. The sole assumption is that O is free of error so that all error is
contained in P'1 and O't The equation can be rewritten as
d=1- (Eq. 2-4)
for simplified calculation when MSE is known. The innovation of the d-index is that it
responds to both differences between predicted and observed data as well as some
changes in proportionality (Willmott, 1981). The d-index is an improvement on the
simple "r"; still, it is not an absolute measure of performance. As with the
aforementioned methods, the d-index should be evaluated in the context of knowledge of
natural variations in the system being modeled, the capabilities of the model, and an
awareness of the amount of potential error in the observed values used in the comparison.
No one of these approaches will be best in all situations, reviewing several of these
measures together will provide a more complete description of model performance. The
results should also be viewed in the context of the intended use of the model. If the
model is to be used to demonstrate the response to a change in the environment to a class
of students, a model that predicts a response of the correct direction but severely under-
or over-predicts the magnitude may be preferable to a more accurate model if the latter is
more difficult for the students to use. Users must decide for themselves what level of
performance is acceptable. Likewise, individuals will have their own views of which
approach is most appropriate to their interests.
BAHIAGRASS GROWTH STUDY
There has been a resurgence of interest in the cause of winter dormancy in tropical
perennial grasses, particularly bahiagrass (Mislevy, 1998; Gates et al., 2001; Sinclair et
al., 2003). To date, the primary emphasis of the research has been to identify the
conditions triggering dormancy. Traditionally, dormancy was thought to be related to the
cooler temperatures of fall and winter; however, daylength has recently been implicated
as the triggering condition (Gates et al., 2001; Sinclair et al., 2003). Along with
identifying the cause, quantifying the effects of dormancy may help us identify specific
characteristics associated with dormancy to both aid in identifying non-dormant
individuals as well as help develop management strategies to promote higher yield and
longevity of both current and new, non-dormant bahiagrass varieties that may be
Sinclair et al. (2003) presented growth and composition data at the organ level
(leaf, stem, and below-ground material) characterizing relative differences in growth and
composition between plants grown under normal or extended photoperiods. Their data
were quantified for each harvest, at 4 to 5 week intervals. Information quantifying
changes within a regrowth period, however, is scarce. More detailed growth analyses
conducted during the transition into dormancy may help identify some of the mechanisms
involved in the reduction of herbage growth associated with dormancy. Our obj ective for
this study was to document, in detail, weekly patterns of plant growth in late summer and
fall regrowth periods with concurrent measurements of leaf and canopy photosynthesis.
The purpose of this information is to help "fill in" some of the detail missed in other
studies, and needed for perennial forage crop model development.
Materials and Methods
This research was conducted at the Plant and Soil Science Field Teaching
Laboratory at the campus of the University of Florida, Gainesville (290 38' N, 820 22'
W) on an established bahiagrass sod during the summer and fall of 2001. Based on the
age of the stand, fine leaf texture and abundant seedhead production in June, the variety
ofbahiagrass was assumed to be Pensacola. The soil was an Arredondo fine sand
(loamy, siliceous, hyperthermic Grossarenic Paleudult). The experimental design was a
randomized complete block four replications each being 190-m2 plOts (PLOT).
Treatments were two 8-wk regrowth periods (PER) (18 July 12 September or 12
September 7 November). The crop was harvested to a stubble height of 10 cm every 8
wk starting 21 May. The 18 July harvest served as the staging harvest, establishing the
initial conditions and base stubble mass for the first growth period. A single sod core
sample was taken weekly (WEEK) from each of the four plots beginning on 20 July
(Table 3-1). Plant height (distance from the soil surface to the point where the leaves
curved over and began to hang down) was measured at six locations within each plot on
the same days that the sod cores were sampled.
While the bahiagrass had been established for several years, it had not been
fertilized or irrigated regularly in recent years. During our study all plots were fertilized
with a commercial blended 16-4-8 fertilizer including trace nutrients and slow-release N
at 78 kg N ha- 10 kg P ha- and 37 kg K ha-l every 8 wk beginning on 13 April.
Irrigation was provided as needed to prevent water limitation of plant growth via portable
impact sprinklers (see Table 3-2 for combined rainfall + irrigation). Weather data (total
daily PAR, minimum. maximum, and average air temperature and total daily rainfall)
were recorded on an automatic datalogger (CR10, Campbell Scientific, Inc. Logan, UT)
and are summarized in Table 3-2.
The sod cores measured 20 cm by 35 cm by 15 cm deep and were dug by hand
using a pair of narrow-bladed shovels (Figure 3-1). Loose soil was shaken by hand from
the sod core and any loose bits of roots and other plant material were recovered. All
material was placed in a plastic bag and immediately placed in a cooler for transport to
the laboratory. At the laboratory, each sample was thoroughly washed with a garden
hose over a 2-mm sieve to dislodge soil. Rinsed samples were placed in sealed plastic
bags and refrigerated until processed.
Roots were trimmed from the stolons using hand clippers, placed in a paper bag
and transferred to a 550C forced-air oven. A subsample of the remaining plant material
was set aside for detailed analyses. Both the remaining sample and subsample were
separated into live leaf, stem, stolon, and dead leaf components. Live leaves were
separated from stems at the ligule (if there was one) or where the leaf emerged from the
stem (if the leaf had no ligule). Using this separation methodology, the leaf sheath is
included in the stem fraction (Figure 3-2). Stems were separated from stolons at a point
where they naturally broke by hand. Dead leaves were peeled from the tillers and, thus,
included dead sheath material.
Material in the subsample was analyzed for leaf and stem areas (one sided only)
using a LI-COR model 3100 leaf area meter (LI-COR Inc., Lincoln, NE). Stems were
scanned intact (Figure 3-2), and not dissected into individual immature leaves. The
number of stems (tillers) per stolon was counted and the number of leaves per tiller was
recorded as the vegetative stage (V-stage) of each tiller. The presence of seedheads in
the subsamples were noted but as there were very few, no additional measures of
reproductive stage were recorded.
Upon completion of separation and measurements, the plant components were
dried until reaching a constant weight in a 550C forced-air oven. Care was taken to
remove loose sand before weighing. Leaf, stem, stolon, and root dry matter (DM) mass
(kg DM ha- ) was calculated from the combined sample and sub sample masses of leaf,
stem, stolon, and root, respectively, and the land area of each core. Specific leaf area
(SLA) (m2 leaf kg-l leaf) was calculated from the measured leaf area and leaf mass for
each subsample. Leaf area index (LAI) (m2 leaf m-2 land) was then calculated by
multiplying the SLA by the total leaf mass m-2 fTOm each core. A "green area index"
(GrAI) representing the total photosynthetic area per area of soil surface was calculated
using the sum of the leaf and stem area indices. The V-stage was calculated for each plot
by averaging the V-stage of all of the tillers in the subsample for each plot. The net
accumulation of each component as well as net change in V-stage (A Leaf, Stem, Stolon,
and Root mass, A V-stage) was calculated for each period by subtracting the WEEK 0
(stubble) mass from the WEEK 8 (final) mass.
Statistical analyses of the growth data were performed using the Mixed procedure
of SAS (SAS Institute Inc., 1987) with the model:
Yijk = CI + Ai + Bj + AiBj + Ck + AiCk + BjCk +Cijk
where C1 was the population mean, A was PLOT, B was PER, C was WEEK, and e was
the residual error for i=4, j=2, k=8. The net accumulation data were analyzed using a
simplified form of the same model; Yij = CI + Ai + Bj + eij. PLOT, and all of its
interactions were assumed to be random effects and therefore appeared in the SAS
random statement in the order presented above. Growth period (PER), WEEK, and their
interaction (PER x WEEK) were fixed effects. Means separation for PER was directly
from the ANOVA. Orthogonal contrasts were used to qualify significant responses to
WEEK and PER x WEEK. The oc=0.10 level was selected as the threshold for
determining the significance of all effects and contrasts.
Concurrent with the growth measurements, leaf and canopy photosynthesis
measurements were recorded four times during each growth period. Due to weather
constraints, these measurements were not evenly allocated throughout each growth
period, nor did they occur at the same time during each period. Leaf and canopy
photosynthesis measurements were made at midday using a LI-COR LI-6200 portable
photosynthesis system (LI-COR Inc., Lincoln, NE). Leaf photosynthetic (or carbon
exchange rate CER) measurements were made on fully expanded, healthy leaves under
full sun conditions (PAR >1600 Cpmol m-2 S-1) using a 0.25-L chamber. Photosynthesis
was measured for three leaves per plot on each sampling date. Two 15-sec measurements
of carbon exchange rate (CER) (Cpmol CO2 m-2 leaf s )~, stomatal conductance (mol m-2
leaf s^)~, and internal CO2 COncentration (CLL L^1) were recorded for each leaf. For canopy
photosynthesis measurements, the leaf chamber was placed "open" inside an aluminum-
frame, clear plastic enclosure. The frame enclosed a land area of 0.56 m2, with a total
volume of 0.49 m3. Canopy CER (Cpmol CO2 m-2 land s- ) measurements were made
under four levels of light varying from full sun to dark. The light level in the chamber
was regulated by placing cloths of varying opaqueness over the chamber. Approximate
light levels were: PAR > 1500 Cpmol m-2 S-1 (full sun), 600-800 Cpmol m-2 S-1, 200-400
Cpmol m-2 S-1, and 0 Cpmol m-2 S-1 (dark). Three 16-second measurements of carbon
exchange rate (CER) (Cpmol CO2 m-2 land s- ) were recorded at each light level. The
canopy enclosure was opened between each light level measurement to let the humidity
and [CO2] in the enclosure equilibrate with the atmosphere. Net CER in full darkness
was considered to represent canopy + root + soil (dark) respiration. Gross canopy
photosynthesis (Cpmol CO2 m-2 land s- ) for each light level was calculated by adding the
absolute value of the dark respiration to the measured net photosynthesis for each light
We fit the canopy light response data to the asymptotic exponential model (Boote
et al., 1985):
P = Pmax [1- e(-E*P4R/Pm,)] (Eq. 3-1)
using TableCurve 2D v4 software (Jandel Scientific Software, 1996), where P = canopy
gross photosynthetic rate (Cpmol CO2 m-2 S-1), P;;;a = maximum photosynthetic rate in
saturating light (Cpmol CO2 m-2 S-1), QE = quantum efficiency or initial slope of the CO2
assimilation : incident PAR response (Clmol CO2 ClmOl-1 absorbed photons), and PAR =
photosynthetically active radiation (Cpmol photons m-2 S-1). We solved for Pmax and QE
and used the resulting values to estimate gross canopy photosynthetic rate at a light
intensity of 1500 Cpmol photons PAR m-2 S-1. By expressing photosynthetic rate for a
common light intensity we could compare the different treatments and days without the
variation due to changing light levels as the days and seasons progressed.
We also attempted to predict a light-saturated leaf photosynthetic rate (Asat) from
the canopy gross photosynthesis data. The measured leaf and stem areas as well as
canopy gross photosynthesis and corresponding PAR measurements from the canopy
light response measurements were input into the hedgerow photosynthesis model of
Boote and Pickering (1994), programmed in SAS, then PROC NLIN in SAS (SAS
Institute Inc., 1987) was used to solve for Asat using an asymptotic exponential function
and outputs from the hedgerow model.
Statistical analyses of the leaf photosynthesis data were performed using the Mixed
procedure of SAS (SAS Institute Inc., 1987) with the model:
Yijk = CI + Ai + Bj + AiBj + C + AiC + BJC + AiBjC + Dk + AiDk + BjDk + AiBjDk
CDk + AiCDk + BjCDk + ijk
where C1 was the population mean, A was PLOT, B was PER, C was day of period or
re growth (DAY), D was leaf number (LEAF) three leaves were measured in each plot ,
and e was the residual error for i=4, j=2, and k=3. PLOT and all PLOT interactions were
assumed to be random effects and therefore appeared in the SAS random statement, in
the order presented above. Growth period (PER), LEAF, DAY, and their interactions
were considered to be fixed effects. Since measurements were not taken every week or
even on the same day of the week, DAY was treated as a continuous variable and entered
as a covariate. Because there was only a single value for Asat, canopy Pl500, and
canopy respiration for each plot on each sampling day, LEAF was not included in the
analysis of these variables and a reduced version of the model was used for these
Yijk = CI + Ai + Bj + AiBj + C + AiC + BJC + eij
Least squares means were calculated for PER. Means separation for PER was directly
from the ANOVA. The oc=0.10 level was selected as the threshold for determining the
significance of all effects and contrasts.
Results and Discussion
Statistically, total plant growth (A Total Plant Mass) was greater for PER 2 than
PER 1 (Table 3-3). The dramatic loss of total plant mass in PER 1 masked a net increase
in leaf mass, overshadowing the fact that canopy growth was greater in PER 1 than PER
2. The loss of total plant mass was driven almost entirely by a decrease in root mass in
PER 1. In PER 2 root mass stabilized (Table 3-3) and a net gain of total plant mass was
realized (Figure 3-3).
Initial root mass for PER 1 was high at 11 500 kg DM hal but dropped to less than
1/3 of the original mass by WEEK 7 of PER 1 (Figure 3-4). The consistent decline
during PER 1 and the relatively constant root mass from the end of PER 1 through PER 2
tends to dismiss random sampling error as the cause of this loss. Alternatively, the
Spartan maintenance of the site in past years may have played a role in this behavior.
There had been no fertilizer or irrigation applications over several years and the extensive
root system may have developed to more thoroughly mine the soil for water and
nutrients. At the time we started measuring plant growth, irrigation had been available
for three months and a second fertilizer application had just been applied. As PER 1
progressed, nutrients may have become available in sufficient concentrations that plant
needs could be met with a less extensive root system. Excess root mass may have been
mobilized for new shoot growth or merely allowed to die, resulting in the pattern of loss
observed. In contrast to this loss in PER 1, root mass remained relatively constant in
PER 2. There was a slight increase in root mass in WEEK 6 (Figure 3-4) which
coincided with a period of increasing stolon mass and, thus, could be related to a
dormancy-induced change in priority of assimilate partitioning.
A more obvious signal of approaching dormancy may be seen in the pattern of
stolon growth. Throughout PER 1 stolon mass remained unchanged at approximately
4700 kg DM ha-l (Figure 3-5). Stolen mass increased in PER2, peaking at 8980 kg DM
ha-l on WEEK 7 (Figure 3-5). This change in growth pattern, as evidenced by the
significant WEEK and PER x WEEK interaction effects, resulted in greater A stolon
mass in PER 2 (Table 3-3). This late-season shift in partitioning of growth towards
storage tissue may be part of a dormancy response to shorter daylengths. Increased
allocation of growth to stolons may have contributed to the lower increment of leaf and
stem mass observed in PER 2 (Table 3-3). The combination of increasing stolon mass
and stable-to-increasing root mass in PER 2 is in contrast to the observations of Sinclair
et al. (2003) who reported steady or decreasing below-ground plant mass for Pensacola
bahiagrass between 22 Sep. 1999 and 1 Dec. 1999 sampling dates at Ona, FL. Our
longer, 8-wk, harvest interval may have allowed greater accumulation of stolon mass
between harvests than the 4 to 5-week interval of Sinclair et al. (2003). The management
history of the two sites could also have had an influence, as judged by the exceptional
pattern of root growth observed at our site.
Changes not only in leaf and stem mass (Table 3-3), but also changes in the
"character" of the canopy were observed. Shoot growth during the fall season (PER 2)
was slower than in summer (PER 1) resulting in less stem and leaf dry weight
accumulation (Table 3-3, Figures 3-6 & 3-7) as well as fewer new mature leaves (Figure
3-8) at the end of PER 2. Stem weights were lower for all weeks in PER 2 compared to
PER 1 but the decrease in stem mass was not different between periods (Table 3-3). The
linear decline in stem mass in both periods and even the "bump" in stem mass in PER 1
(Figure 3-6) was, at least in part, an artifact of the partitioning scheme employed. Our
partitioning strategy grouped developing leaves, still encased in the sheath, with stems.
Once the leaves began elongating and emerged from the sheath, the leaves became part of
the leaf mass and the fraction of their mass that had previously been developing in the
sheath was lost from the stem mass. Very little stem elongation was observed except for
the few tillers that developed seedheads, leaving little opportunity to increase stem mass
as the plant matured.
Development of leaves in the fall regrowth period (PER 2) was quite different from
that in PER 1. Leaf mass increased in a quadratic manner (Table 3-3) to peak on WEEK
6 of both periods (Figure 3-7), with average leaf mass and A leaf mass slightly lower in
PER 2. In contrast, the A V-stage in PER 2 was only 3.41 leaves, less than half of the
7.65 leaves added in PER 1 (Table 3-3, Figure 3-8). Average LAI followed leaf mass
more closely than V-stage, and a quadratic progression in LAI development was
observed in both periods. The LAI in PER 1 was higher than that in PER 2 (Table 3-3).
Despite the slower development, the initial and final LAI were the same for both periods
(Figure 3-9). Thus, the same final LAI was achieved in PER 2 with only half as many
mature leaves as were observed in PER 1. It should be noted that only leaves with ligules
were included in the V-stage count but all leaf blade material extending from the leaf
sheath was included in the LAI measurements.
Accordingly, the SLA (leaf area per g of leaf mass) was slightly larger, indicating
thinner leaves, in PER 2, although a quadratic decrease in SLA was observed in both
regrowth periods (Table 3-3, Figure 3-10). Ghannoum et al. (2001a) also observed
seasonal effects on SLA in a variety of perennial C4 graSses grown in summer and winter;
however, the response was species-dependent, positive in some cases and negative in
others. Although peak daily solar radiation was not recorded, total daily solar radiation
was lower in PER 2 (Table 3-2). Our SLA values are in accord with the range of values
reported by Boote et al. (1999) (88 to 108 cm2 -1l) for greenhouse-grown bahiagrass at
this location. One might expect these levels to be lower (thicker leaves) as high SLAs are
generally associated with shaded leaves and leaves grown under reduced light levels such
as in greenhouses while our plants were grown outdoors in full sun. Our values are quite
low compared to SLAs reported for other perennial C4 graSses (Ghannoum et al., 2001a,
2001b) and even C3 graSses (Ryser and Wahl, 2001).
The mean SLA reported for 11 NADP-1VE-type C4 graSses grown inside a
glasshouse in summer, with midday PAR levels averaging 860 Cpmol m-2 S-1, was 314 cm2
g- almost five times the level observed in our study (Ghannoum et al., 2001a).
However, their plants were harvested only 46 d (approx 6.5 wk) after planting, much
younger than the average age of the leaves on our plants. In contrast, the harvest interval
employed by Boote et al. (1999) was longer than the current 8 wk. The fine-leaf structure
of Pensacola bahiagrass may predispose this cultivar to have a lower than average SLA;
however, the low SLA values merit further measurements on this species and its cultivars
grown in other locations.
Similarly, the LAI values that we are reporting are lower than those observed by
others. At the extreme are LAI values above 8.0 reported for bahiagrass by Agata
(1985a; 1985b), which are considerably higher than our values of 1.75 and 1.67 for
8-week re growth in PER 1 and PER 2, respectively (Figure 3-9), or even our GrAI values
which include stem area as well as leaf area (Figure 3-1 1) Unfortunately, the
methodology used by Agata (1985a, 1985b) to determine LAI was not clear, hindering
any further comparisons. Other reports give considerably lower LAI values. Pedreira
and Brown (1996b) reported LAI for stubble and 13-d regrowth for three populations of
bahiagrass grown in the field near Athens, GA. Reported values were averages for two
cutting heights (3.5 and 10 cm). This stubble LAI (comparable to our WEEK 0 values)
ranged from 0.42 for selection T14 in August to 1.68 for Pensacola bahiagrass in July
while re growth LAI values ranged from 1.67 for selection T14 in August to 2.30 for
Tifton 9 in July. These values compare very favorably to our values of 0. 19 for WEEK 0
(stubble) and 1.16 for WEEK 2 in PER 1 (Figure 3-9). Methodology for the Athens
study was similar to ours in that leaf area measurements were based on leaf lamina only;
however, their sample size was much smaller (20 leaves vs. 100-500 leaves per
replication in our study).
As the stem tissue is also green, leaf + stem or "green" area index (GrAI) may
present a more accurate measure of photosynthetically active plant area than LAI alone.
As with LAI, GrAI was higher for PER 1 than PER 2 at 2.49 and 1.85, respectively
(Table 3-3). Orthogonal contrasts showed the relationship between GrAI and WEEK to
be cubic (Table 3-3), this is likely an artifact of the variation in stem mass resulting from
our partitioning scheme and may not be a biologically relevant pattern. The time-series
change in GrAI (Figure 3-11) clearly illustrates, more so than LAI, a considerably slower
increase and overall lower photosynthetic area throughout most of PER 2. Like LAI,
though, initial GrAI was the same for both periods and final GrAI were also much closer
than for the middle of the regrowth periods.
The slower leaf growth rate cannot be attributed to differences in initial leaf mass
and initial leaf area as neither differed between periods, although the slower growth rate
would reinforce itself through lower leaf mass and LAI once regrowth began. The slower
development of LAI and GrAI may have decreased potential photosynthesis during much
of PER 2; however, final LAI values were similar or identical for both periods (Figure
3-9). Cooler temperatures and lower solar radiation levels (Table 3-2) likely were maj or
factors reducing fall growth rates. The increased partitioning of growth to stolon tissue
could also have reduced leaf growth in the second half of PER 2.
Some caveats apply to the photosynthesis results. First, due to a combination of
equipment repair and availability issues, photosynthesis measurements for PER 1 were
made with a different LI-COR LI-6200 than was used in PER 2. However, both
instruments were calibrated using the same reference gas and procedures. Second, the
complete set of leaf photosynthetic measurements (two 15-s values for 3 leaves per plot)
included a few values that were not physiologically realistic. The leaf data were analyzed
to identify and remove outlying data points. Of the 197 photosynthesis measurements,
one was more than 1.5 interquartile ranges from the 75th percentile and one was more
than 1.5 interquartile ranges below the 25th percentile; both data points were removed
from the analysis These were the only data points removed from the analysis. The
values removed were quite extreme (gross photosynthesis levels of 70.35 and -20.47
Cpmol CO2 m-2 S-1) and may have been the result of air leaking into the leaf chamber
during measurement, beginning measurements before CO2 COncentration in the chamber
had begun to drop, or simply due to high sensitivity to the very small leaf area in the leaf
chamber (approximately 1 cm2) and a correspondingly small drop in [CO2] Over the
measurement period causing excessive variation on measurements. Under the reduced
dataset (n=195), the covariate, DAY, was significant (Table 3-3) with leaf photosynthetic
rate being highest during early regrowth (Figure 3-12). Despite lower temperatures in
PER 2 and a positive leaf temperature to leaf photosynthesis correlation coefficient of
r=0.45, PER did not have a significant effect on leaf photosynthesis. Our measured leaf
photosynthesis values of 31.0 and 26.6 Cpmol CO2 m-2 S-1 for PER 1 and PER 2,
respectively, fall well within the range reported by Boote et al. (1999) (24.8 35.2 Cpmol
CO2 m-2 S-1) and Fritschi et al. (1999) (19.0 35.4 Cpmol CO2 m-2 S-1) for greenhouse-
grown bahiagrass at 350 CLL L1 CO2 COncentration at this site.
As expected, predicted leaf Asat was considerably higher than the measured leaf
photosynthesis values (Table 3-3). Like measured leaf photosynthesis, the predicted Asat
values were not different between periods, however, unlike the measured data, DAY did
not affect Asat. This would indicate that the maximum potential leaf photosynthetic rate
remained the same over the temperature range experienced in this study. There is
precedent for this. Asat has been shown frequently to decrease rapidly below 200C
(reviewed by Long ); however, the lowest temperature recorded in the canopy
chamber during photosynthesis measurements was 270C, considerably higher, where the
impact may be slight and difficult to discern. This might also help explain why
temperature accounted for such a small proportion of the variation in leaf photosynthetic
rate. Although there is no way to verify the accuracy of our predicted Asat using the data
we collected, the values seem within reason as a handful of measured values were at or
above the predicted rates.
To allow us to compare photosynthetic performance across sampling dates and
periods despite the varying light conditions, the canopy photosynthesis data were fit to an
asymptotic exponential function and the results used to predict canopy photosynthesis at
1500 Cpmol PAR m-2 S-1 (Pl500). Model fit was good with an r2> 0.99 for most plots and
sample days and the lowest r2 for a plot/day was 0.89. Analysis of the adjusted canopy
data showed a higher canopy gross photosynthetic rate in PER 1 than PER 2 (Table 3-3,
Figure 3-13). This is consistent with the greater plant mass (particularly leaf and stem)
observed in PER 1. Interestingly, when we fitted a regression model using daily
maximum temperature (Tmax), leaf + stem mass (Greenkg), GrAI, and SLA to the Pl500
data, Tmax had the best fit (r2= 0.48) and the addition of either Greenkg or GrAI did not
significantly improve the fit of the model. The PER 2 average of 43.0 Cpmol CO2 m-2 land
s^l was within the range of 31.6 to 47. 1 Cpmol CO2 m-2 S-1 reported by Boote et al. (1999)
for Pl500 in greenhouse-grown bahiagrass canopy gross photosynthesis at 350 CLL CO2
L CO2, whereas our PER 1 rate of 55.9 Cpmol CO2 m-2 S-1 was outside this range but
below the highest rate of 60.7 Cpmol CO2 m-2 S-1 reported by Fritschi et al. (1999) for
Pl500 of greenhouse-grown bahiagrass in the establishment year. As our plants were
grown under full sun, a higher photosynthetic rate than for greenhouse-grown plants
would be expected. That Fritschi observed higher rates may be related to the differences
in the age of the stands, if the rates are different at all.
Concurrent with the higher photosynthetic rate in PER 1 were higher canopy + root
+ soil respiration levels. This measurement was based on the CO2 exchange rate
measured in total darkness. Respiration rate is dependent on both the amount of tissue
respiring as well as the temperature. In regression analysis using the "Backwards" option
in PROC REG of SAS (SAS Institute Inc., 1987), both Tair (air temperature) and
Greenkg contributed significantly to the model and, combined, they could account for
87% of the variation observed in respiration. Canopy + root + soil respiration rates were
24.0 Cpmol CO2 m-2 land s-l in PER 1 and 13.9 Cpmol CO2 m-2 S-1 in PER 2 (Table 3-3),
slightly higher than the levels reported by Boote et al. (1999) but within the range of
values reported by Fritschi et al. (1999) for bahiagrass grown at 350 CIL L1 CO2.
Respiration rates also varied by day of regrowth (Table 3-3) but since DAY was a
covariate, orthogonal contrasts could not be used to discern a pattern of response.
Winter dormancy, the seasonal depression of canopy growth, in bahiagrass often
has been attributed to a decrease in temperature. More recently, daylength has been
identified as having a role in triggering dormancy (Mislevy, 1998; Gates et al., 2001;
Sinclair et al., 2003). The obj ective of our study was not to identify the cause of
dormancy but rather to quantify growth and photosynthesis during the late summer and
fall in more detail than previous studies. Our study points out several changes in the
pattern of plant growth and photosynthetic rate that may illuminate parts of the
underlying mechanism of dormancy. Two key observations were the sudden increase in
stolon growth half-way through PER 2 and that the leaf photosynthetic rate was not
different between PER 1 and PER 2. The effect of air temperature could explain most of
the variation observed in canopy root soil respiration, but could explain only 25% of
daily variation in leaf photosynthesis or 50% of the variation in canopy photosynthesis
(data not shown). The rate of V-stage progression was dramatically lower in the fall
(PER 2), much more so than the reduction in leaf mass. Despite having fewer mature
leaves per tiller, initial and final LAI were nearly the same for both 8-wk regrowth
periods. Consistent with the lower leaf mass and similar LAI, SLA was higher in PER 2.
In the case of shaded leaves, the resulting higher SLA is associated with lower
concentrations of photosynthetic apparatus and lower potential photosynthetic rates,
however, our predicted Asat was not different between periods.
Likely, growth reduction during winter dormancy is the culmination of a number of
factors; reduction in growth rate due to lower temperatures, change in partitioning of
assimilate favoring storage tissue over leaf growth, and changes in leaf characteristics
related to lower light levels. Use of the results of this study to develop parameters for
modeling bahiagrass growth testing would allow exploration of "what-if scenarios and
possibly help us better understand how these factors interact to reduce forage production
during winter dormancy.
Table 3-1. Schedule of sampling and harvest activities.
Per 1 Date
Per 2 Date
Mow to 10-cm stubble height
Sample growth Final
Table 3-2. Weekly averages of daily temperatures and daily solar radiation and total
weekly rainfall + irrigation water applied to bahiagrass grown at the Irrigation
Park, Gainesville, FL -2001
Daily Temperature Rainfall Solar Radiation
(MJm-2 day- )
Table 3-3. Results of statistical comparison of treatment effects on plant growth and
photosynthesis. Period means are least squares means. Significance
determined by ANOVA for Period and orthogonal contrast for Week and Per
X Week interaction.
Period Mean Statistics
Growth Variable 1 2 Period Week Per X Week
Leaf Mass (kg DM ha-' ) 2150 1700 ** **/qdr ns
A Leaf Mass (kg DM ha-l )a 3015 251 1*----
Stem Mass (kg DM ha-l ) 3690 1960 *** ***/lin ns
A Stem Mass (kg DM hal ) a -456 -1396 ns----
Stolon Mass (kg DM ha-l ) 4740 6870 *** */lin */lin
A Stolen Mass (kg DM ha-l ) a 612 2350 **----
Root Mass (kg DM ha-l ) 6270 3800 ***"/lin ***/lin
A Root Mass (kg DM ha-l ) a -8009 -344 ***----
Total Plant Mass (kg DM ha-l ) 16845 14325 ns ***/lin
A Total Plant Mass (kg DM ha-l ) a -4839 3121 **----
Canopy Height (cm) 29.8 22.5 *** ***/cub ***/lin
V-stage (number of fully emerged
leaves tiller- ) 3.52 1.70 *** ***/qdr ***/qdr
A V-stage (number of fully emerged
leaves tiller ) a 7.65 3.41 ***----
SLA (cm2 -1f ~ leaf) 64.8 73.3 **/qdr ns
LAI (m2 leaf m-2 land) 1.28 1.10 **/qdr ns
A LAI (m2 leaf m-2 land) 1.56 1.52 Ns----
GrAI (m2 leaf + stem m-2 land) 2.49 1.85 *** **/cub ns
A GrAI (m2 leaf + stem m-2 land) 1.84 1.12*----
Photosynthesis Variable Period Day Per X Da
Measured Leaf Photosynthesi s
(ymol CO2 m-2 leaf si) 31.0 26.6 ns *** ns
Predicted Max Leaf Photosynthesis
(Cpmol CO2 m-2 leaf s^l) 44.2 39.9 ns ns ns
Predicted Canopy Gross
Photosynthesis at 1500 Cpmol
photons (Cpmol CO2 m-2 land s^l) 55.9 43.0 ** ns
Canopy Root Soil Respiration
(pol CO2 m-2 land si) 24.0 13.9 ** *ns
*P<0.10, **P<0.05, ***P<0.01, ns=not significant, --- does not apply to this variable.
lin, qdr, cub = linear, quadratic or cubic orthogonal contrasts, respectively, are significant
" A values are net change over period = Week 8 values Week 0 values.
Figure 3-1. Sod core as removed from the soil.
Figure 3-2. Example of a separated subsample of bahiagrass after removing roots.
0 1 2 3 4 5 6 7 8 9
Weeks of Regrowth
Figure 3-3. Total plant mass for established bahiagrass grown at Gainesville, FL from 18
July to 12 Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001.
0 1 2 3 4 5 6 7 8
Weeks of Reg rowth
Figure 3-4. Root mass for established bahiagrass grown at Gainesville, FL from 18 July
to 12 Sept. (--) and 12 Sept. to 7 Nov. (- -), 2001.
0 1 2 3 4 5 6 7 8
Weeks of Reg rowth
Figure 3-5. Stolen mass for established bahiagrass grown at Gainesville, FL from 18 July
to 12 Sept. (--) and 12 Sept. to 7 Nov. (- -), 2001.
U 1000 -- i -- -
0 1 2 3 4 5 6 7 8
Weeks of Reg rowth
Figure 3-6. Stem mass for established bahiagrass grown at Gainesville, FL from 18 July
to 12 Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001.
0 1 2 3 4 5 6 7 8
Weeks of Reg rowth
Figure 3-7. Leaf mass for established bahiagrass grown at Gainesville, FL from 18 July
to 12 Sept. (--) and 12 Sept. to 7 Nov. (- -), 2001.
0 1 2 3 4 5 6 7 8
Weeks of Reg rowth
Figure 3-8. V-stage for established bahiagrass grown at Gainesville, FL from 18 July to
12 Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001.
0 1 2 3 4 5 6 7 8
Weeks of Reg rowth
Figure 3-9. Leaf area index (LAI) for established bahiagrass grown at Gainesville, FL
from 18 July to 12 Sept. (-) and 12 Sept. to 7 Nov. (- -), 2001.
Weeks of Regrowth
Figure 3-10. Specific leaf area (SLA) for established bahiagrass grown at Gainesville,
FL from 18 July to 12 Sept. (--) and 12 Sept. to 7 Nov. (- -), 2001.
0 1 2 3 4 5 6 7 8
Weeks of Reg rowth
Figure 3-11. Leaf + Stem (green) area index (GrAI) for established bahiagrass grown at
Gainesville, FL froml8 July to 12 Sept. (-) and 12 Sept. to 7 Nov. (- -),
Days of Reg rowth
Figure 3-12. Leaf photosynthetic rate for established bahiagrass grown at Gainesville, FL
from 18 July to 12 Sept. (m) and 12 Sept. to 7 Nov. (0), 2001.
Days of Regrowth
Figure 3-13. Canopy gross photosynthetic rate adjusted to 1500 Cpmol Par m-2 S-1 (Pl500)
for established bahiagrass grown at Gainesville, FL from 18 July to 12 Sept.
(m) and 12 Sept. to 7 Nov. (0), 2001.
DEVELOPMENT OF CROPGRO SPECIES FILE PARAMETERS FOR
CROPGRO is a mechanistic model that predicts yield and composition of crops
based on plant, soil, management, and weather inputs. As such, it appears well suited to
the task of modeling forage growth and nutrient concentration. Additionally, the ability
to simulate soil water and N balances, soil organic matter residue dynamics, and
pest/disease damage increase CROPGRO's utility as a tool for evaluating potential
environmental consequences of management changes. Its generic, process-oriented
design has allowed it to be adapted to model a variety of different species including
soybean (Glycine max L.), peanut (Arachis hypogaea L.), dry bean (Pha~seolus vulgaris
L.), faba bean (Vicia faba L.), and tomato (Lycopersicon esculentum Mill.) (Scholberg et
al., 1997; Boote et al., 1998a, 1998b, 2002). Adaptation is accomplished by changing a
set of parameters and relationships describing the species' response to environmental
variables. The procedure is described in Boote et al. (2002).
Kelly (1995) previously attempted to adapt CROPGRO to model the growth of
bahiagrass with the obj ective of using the model as a component of a system for
simulating peanut cropping systems. Simulation results were incorporated into an
economic model to predict the sustainability and profitability of the cropping systems.
The species, cultivar, and ecotype files developed were later released as a "pasture"
model in DSSAT v 3.5 (ICASA, 1998). Our application of this model to simulate data
sets of bahiagrass hay production revealed consistent overprediction of DM yields,
particularly in the cooler months of the year. More rigorous applications and objectives
for the use of the model impose different standards of accuracy and our proposed use as a
practical planning and teaching tool requires a more accurate prediction capability and a
more faithful representation of the seasonal patterns of growth of bahiagrass. The
obj ective of this work was to develop parameters, from searching the literature,
experiments, and calibration, to model bahiagrass growth and composition with the
CROPGRO CSM model.
Materials and Methods
In deriving model parameters to describe bahiagrass growth and composition, we
followed the general adaptation procedure described by Boote et al. (2002). Where
possible, parameters describing the basic processes of photosynthesis, respiration, N
assimilation, and plant development in bahiagrass were derived from the literature.
Parameters describing basic biochemical processes assumed to be conserved, or similar
(e.g.. growth respiration cost per unit of protein), among species are universal throughout
all CROPGRO species Eiles. For some less conserved processes and traits where data
were lacking, parameters from the CROPGRO soybean species Eile were used. Soybean
was selected as it is one of the original models used to develop CROPGRO and is
arguably the most tested and robust version of the model. Examples of parameters
incorporated from the soybean Eiles are lipid, lignin, organic acid, and mineral
composition, as well as carbon cost to mobilize N from senesced proteins (Penning de
Vries et al., 1974).
Where processes or parameters were believed to be divergent from soybean or
thought to be unique to perennial forage species, parameter estimates were interpolated
from literature data from other tropical perennial grass species or selected through
sensitivity analysis of the bahiagrass model.
We also developed an "optimized" set of parameters using a custom built
optimization program utilizing a "brute-force" optimization strategy. In the program, the
user specifies a minimum and maximum acceptable value for each parameter and the
desired number of "steps" between those limits for up to five parameters. Simulations
were run using all possible combinations of the specified parameters. Results from each
set of parameters were statistically analyzed for mean of simulated results, slope and
intercept of a fitted regression line of predicted and observed data, r2, d-index value
(index of agreement [Willmott, 1981]), and root mean square error (RMSE). Results
from all runs were saved to an output file and the combination with the lowest RMSE
was listed at the end. The output file was then exported to a spreadsheet, parsed, and
sorted from highest (best fit) to lowest d-index rating. The optimized parameter values
were selected using a combination of high d-index ranking, low RMSE value, and most
logical fit. The optimized parameter set was subsequently tested for fit, using the same
procedure as used for testing the literature-based species file.
Description of Data Sets Used to Fit Parameters
Two data sets were selected for use in fitting parameters and testing the optimized
model. Only a brief listing of growing conditions will be given here. A more complete
description of each data set may be found in the cited articles.
The study at Ona, FL was part of a three-species study of forage protein response to
N fertilization and cutting date (Johnson et al., 2001). The experiment was conducted at
the Range Cattle Research and Education Center (REC) at Ona, FL (270 25'N, 810 55'W;
elevation 27.4 m) on a Pomona fine sand (sandy siliceous, hyperthermic Ultic
Haplaquod) soil. Pensacola bahiagrass received five fertilizer treatments (0, 39, 78, 118,
and 157 kg N hal cutting- ), equivalent to annual applications of 0, 234, 468, 708, and
942 kg N ha-l supplied as ammonium nitrate. Fertilizer was applied on 5 May and on the
day after each cutting except for the October (last) harvests. Staging harvests marking
the beginning of each growing season were made on 5 May 1997 and 4 May 1998 with
successive harvests every 28 d until October. Forage yield and crude protein
concentration were measured for all but the staging harvests. Daily weather data were
acquired from the REC's weather station. Temperatures rarely dropped below 00C in the
winter. Rainfall totaled 1142 mm for October 1996 through September 1997 and 2110
mm from October 1997 through September 1998.
The Eagle Lake, TX (290 35'N, 960 20'W, elevation 46 m) experiment was part of
a larger study of N contributions of arrowleaf (Trifolium vesiculosum Savi) and
subterranean (Trifolium subterraneum L.) clovers overseeded on bahiagrass and
bermudagrass [Cynodon dactylon (L.) Pers.] conducted over the 1979-1981 growing
seasons (Evers, 1985). The study was located in southeastern Texas at the Texas
Agricultural Experiment Station at Eagle Lake on a Crowley fine sandy loam (fine
montmorillonitic, thermic, Typic Albaqualfs) soil. The Pensacola bahiagrass treatments
included fertilizer applied at annual rates of 0, 84, 168, 252, or 336 kg N ha- The
fertilizer was split into three equal applications made on or about 1 April, 1 June, and 1
August of each year. All plots were harvested monthly from May through October.
Forage yield and crude protein data were reported. Daily weather data were acquired
from the experiment station's weather station. Freezing temperatures were not
uncommon in the winter with minimum temperatures as low as -90C. Average rainfall
was less than for Ona, with annual precipitation of 1354 mm, 765 mm, and 1223 mm for
1979, 1980, and 1981 seasons, respectively.
The two data sets were split for optimization and testing. The two lowest N
treatments and the two highest N treatments from each site were used in the optimization
process. The middle N treatment from both data sets was reserved for testing the
literature-based and optimized species files. The rationale behind the splitting scheme
was to maximize the range of N fertilization and number of observed data pairs (108)
available for the optimization process. At the same time the test data sets would be most
indicative of how the model will perform under "normal" or the most frequently
encountered conditions. The primary objective for the optimization process was to
minimize RMSE for the prediction of herbage mass (leaf + stem weight) or herbage N
concentration (leaf + stem N concentration), depending on the variables being optimized.
Preparation of Datasets
There were no data available regarding initial plant mass or soil conditions for
either experiment, so actual initial crop condition could not be input into the model.
Instead, we estimated the initial conditions by running each simulation for one full
growing season/winter cycle prior to the measured seasons. Actual weather data were
used for the prior year. The season began with an established plant stand cut to the same
stubble height used in the measured years. Fertilization during the prior year was
consistent with the medium N fertilization treatment for each site (468 kg N ha-l yr- at
Ona, FL, and 168 kg N ha-l yr- at Eagle Lake, TX).
To compare simulated and observed growth, the two sets of results had to be
expressed on a common basis. The field studies reported yield as herbage (leaf + stem)
mass harvested above a base cutting or stubble height while simulation results reported
yield as the total amount of leaf and stem. The difference between the two is the amount
of leaf and stem mass in the stubble left after each harvest. Using the results of other
studies (Beaty et al., 1968; Pedreira and Brown, 1996b; Rymph and Boote, 2002), we
developed estimates of post-harvest stubble mass for the different cutting heights used in
the Ona and Eagle Lake experiments. These estimated stubble masses were added to the
reported harvest yields to approximate total herbage mass observed for these
experiments. Estimates for stubble mass left under 3.5-cm, 5-cm, 7.5-cm, and 10-cm
cutting heights were 1500, 1800, 2400, and 3000 kg DM ha l. These values may apply
only to Pensacola bahiagrass. Newer bahiagrass varieties with more upright growth
habits may have considerably less stubble mass (Pedreira and Brown, 1996b). This tactic
basically creates a consistent season-long offset while actual stubble mass may vary with
season and N fertilization. This approach represents a compromise between accuracy and
simplicity of implementation/utility.
Initial testing of the model revealed some characteristics of the CROPGRO
program code that were not compatible with a perennial forage. FREEZ l and FREEZ2
are parameters describing temperatures where all leaves fall off of the crop or the entire
crop dies (respectively) due to cold. We found that after a FREEZ l event occurred, there
was no regrowth of new leaves, resulting in the plants exhausting all reserves on
maintenance respiration and dying. The problem was related to the strategy used to end
photosynthesis of grain legumes after a foliage-killing freeze event. Since we could not
alter the code, both FREEZ l and FREEZ2 were set to -250C, essentially disabling the
FREEZ l function but allowing the simulation to continue through the winter.
Additionally, we simulated frost damage of leaves by partially defoliating the crop each
January using the PEST routine.
Results and Discussion
This discussion will be confined to parameters that were unique to perennial
tropical grasses or required redefinition or alteration in concept. A complete list of
parameter values is provided in Table 4-1.
CROPGRO has two options for predicting daily assimilate production: a daily
canopy option and an hourly leaf-level option. The daily canopy option is the more
simplistic approach, predicting photosynthate production as an asymptotic light response
to total daily solar radiation levels. The leaf-level photosynthesis option predicts hourly
photosynthetic rates for sunlit and shaded leaf area by simulating the dynamics of
Rubisco activity and electron transport and integrates them within the hourly hedgerow
approach to yield a daily canopy rate. Both options include adjustments for current
temperature, CO2 COncentration, and leaf N concentration conditions.
All previous efforts to adapt CROPGRO involved crops using the C3
photosynthesis pathway/mechanism. In contrast, bahiagrass expresses the C4
photosynthetic pathway, more specifically, it is an NADP-1VE type species (Hattersley
and Watson, 1976), the same pathway that is expressed in maize (Zea mays L.).
Concentration of CO2 in the bundle sheath chloroplasts through the "CO2 COncentrating
shuttle" contributes several advantages to C4 plants. Since CO2 is concentrated around
the CO2-fixing Rubisco enzyme in the bundle sheath chloroplasts, the relatively lower
solubility of CO2 at high temperatures is of little consequence, allowing higher rates of
carboxylation and suppressing oxygenation activity of Rubisco at higher temperatures
and light levels than is generally possible in C3 plants. Likewise, the quantum efficiency
(QE) is not temperature sensitive, not decreasing at high temperatures as in C3 plants. As
photorespiration is reduced, less Rubisco is required to maintain high carbon exchange
rates (CERs), and leaf protein levels are generally lower for C4 plants. These differences
must be reflected in our predicted patterns of photosynthetic response to light intensity,
CO2 COncentration, leaf N concentration, and temperature in the model.
The asymptotic light response curve used to predict daily canopy photosynthesis is
defined by two parameters; PARMAX the level of photosynthetically active radiation
(PAR) at which photosynthetic rate is 63% of maximum (moles [quanta PAR]m-2d- ), and
PHTMAX, the asymptote (maximum) of daily assimilation rate (g CH20 m-2 d- )
occurring at very high light (at least three times as high as PARMAX). These values are
not generally presented in the literature so preliminary values (Table 4-1) were estimated
as 150% of the corresponding parameter values used for soybean.
The lower rate of photorespiration observed in C4 Species results in a lower CO2
compensation point (CCMP) (atmospheric CO2 COncentration where the rate of CO2
uptake by photosynthesis equals the rate of CO2 CVOlution by respiration) in C4 plants.
Values of 0 to 14 CIL L-1 have been reported as the CO2 COmpensation point for various
C4 Species (Bolton and Brown, 1980; Raj endrudu and Das, 1981), most on the order of 0
to 5 CIL L^(~Rajendrudu and Das, 1981). Based on these results we selected a value of 5
CIL L^1 for CCMP which is used for the daily canopy photosynthesis option.
The leaf-level photosynthesis option is a more complicated system requiring
several more parameters than the daily canopy option, but the model at the leaf and
chloroplast level incorporates several conserved processes for which parameters may be
directly measured. Leaf quantum efficiency (QE) is typical of these conserved
parameters/processes. Quantum efficiency (parameter name PGEFF) or quantum yield is
broadly defined as the initial slope of the leaf CO2 aSSimilation:absorbed PAR response.
A value of 0.0541 Clmol CO2 ClmOl-1 absorbed photons (Ehleringer and Bjoirkman, 1977)
is typically used in CROPGRO for all C3 Species, including soybean. While the same
biochemical processes are used in both C3 and C4 photosynthesis, the CO2 COncentrating
effect of the C4 Systems increases their QE. Differences in efficiency exist between the
three variations of the C4 photosynthetic pathway (NAD-ME, NADP-ME, and PCK-type)
with NADP-ME species exhibiting the highest QE with an average QE of 0.065 Clmol
CO2 ClmOl-1 absorbed photons (Ehleringer and Pearcy, 1983). We selected this value for
the bahiagrass species file (Table 4-1). This value appears to be quite robust as it falls
well within the range of QE values predicted from bahiagrass canopy photosynthetic light
response data (0.054 0.081 Clmol-l absorbed photons) (Rymph and Boote, 2002), and
reported QE values (0.062 to 0.075 Clmol CO2 ClmOl-1 absorbed photons) for another
NADP-ME species, sugarcane (Saccharum spp.) (Meinzer and Zhu, 1998).
One inconsistency that remains is the relationship between temperature and QE. In
C3 plants, as temperature increases, the solubility of CO2 decreases relative to the
solubility of Oz, lOwering QE of C3 Species at high temperatures. Because of the high
CO2 COncentration surrounding Rubisco in bundle sheath cells of C4 plants, the effect of
temperature on QE is negligible. However, the temperature effect on QE is hard-coded
into CROPGRO and currently, there are no user parameters to modify that response.
The other parameter required is light-saturated leaf assimilation (LFMAX) for
leaves at high N concentration, 300 C, and a given specific leaf weight. We based our
estimate of LFMAX (and PGREF) on a predicted maximum leaf photosynthetic rate
developed from bahiagrass canopy light response data (Rymph and Boote, 2002) of
approximately 40.0 Cpmol CO2 m-2 leaf s^l. Relative differences among cultivars are
modeled by changing the ratio of LFMAX (maximum leaf photosynthetic rate for the
cultivar) to PGREF (maximum leaf photosynthetic rate for the species). As Pensacola
was the "reference" cultivar on which the species parameters are based and was the
cultivar measured, PGREF=LFMAX=1.760 mg CO2 m-2 S-1
The amount of photosynthetic enzymes in the leaf affect photosynthetic rate as
well. Generally, higher N concentrations in the leaves are correlated with higher levels of
these enzymes and higher photosynthetic capacity. Bahiagrass and other C4 graSses are
generally considered to have low concentrations of N in the leaves, yet maintain high
photosynthetic rates. Thus, optimal N concentrations for photosynthesis of bahiagrass
are likely to be lower than for soybean. We could find no reports of the minimum N
concentration required for photosynthesis [FNPGN(1)], so we defined this lower
threshold of the N response function [FNPGN(1)] from the lowest reported leaf N
concentration 7.6 g N kg-l leaf (Beaty and Tan, 1972). Sugimoto and Nikki (1979)
observed a curvilinear increase in bahiagrass leaf photosynthetic rate as leaf N
concentration increased from approximately 20 up to 30 g N kg- The rate remained
constant from 30 g N kg-l to almost 40 g N kg- Hence, we chose a curvilinear
(quadratic) shape to define the response of bahiagrass to leaf N concentration, with
optimum photosynthetic rates beginning at 30 g N kg-l [FNPGN(2)] and no decline in
rate at higher N concentration. The 30 g N kg-l optimum was also used for LNREF, the
N concentration at which PGREF is defined for the species.
The high concentration of CO2 arOund Rubisco in the bundle sheath chloroplasts
permits high photosynthetic rates at higher temperatures than typically observed in C3
plants. Although the mechanism is not well understood, C4 Species generally also have a
greater sensitivity threshold for low temperature reduction of photosynthetic rate than C3
species (Long, 1983; 1999)). Thus bahiagrass should have a base temperature required
for photosynthesis that is higher than soybean and it should have higher optimum and
maximum (highest temperature at which photosynthesis occurs) temperatures as well.
Several studies have been conducted to quantify the cardinal temperatures for tropical C4
grasses (Ludlow and Wilson, 1971; Wilson, 1975; Unruh et al., 1996); unfortunately
none included bahiagrass. Our interpretation of these results are that the optimum range
for leaf photosynthesis for a tropical grass species should be between 35 and 450C, with a
base temperature around 70C and a maximum critical temperature for zero rate near
The daily canopy calculations use a daily, rather than hourly, time step. To find
daytime temperature thresholds for the daily canopy option, we compared the threshold
temperatures for the leaf-level option to daily minimum and/or maximum temperatures
for Gainesville, FL and calculated a corresponding average daytime temperature. The
corresponding temperatures were: base temperature [FNPGT(1)], 120C, optimum range
[FNPGT(2), FNPGT(3)] from 250C to 380C, and maximum temperature of 500C
Low temperatures may also have a prolonged effect on photosynthesis, affecting
photosynthetic rate after temperatures have returned to the optimal range. CROPGRO
uses another set of temperature parameters, FNPGL(1-4) and TYPPGL, to describe the
effect of minimum night temperature on the subsequent day's light saturated leaf
photosynthetic rate. West (1973) observed that Digitaria decumbens grown at 300C and
subj ected to just one night at 100C and returned to 300C, showed a 40% decrease in
photosynthetic rate compared to plants held continuously at 300C. Based on this, we set
the minimum temperature [no photosynthesis on the day after experiencing this
temperature FNPGL(1)] to 70C, optimum night temperature [no effect on subsequent
days photosynthesis -FNPGL(2)] to 180C, with a quadratic (curvilinear) response
between these points (Table 4-1).
Bahiagrass poses an additional challenge to modeling its growth using CROPGRO
because a significant proportion of total plant mass is represented by stolon mass and
CROPGRO does not include a stolon organ in its structure. To include stolons in the
stem fraction would have confounded the computation of protein/N removed at harvest
and further complicated the estimation of stubble mass. Thus we redefined "roots" in the
model to include both stolons and roots. This "redefinition" without a code change
required considerable modification of the growth and senescence parameters relative to
those used for other species modeled by CROPGRO. The largest adjustment was for the
root length density (RFAC1) parameter (cm of root length per g of root). Stolons are
much thicker than roots and may represent more plant mass than the roots. Additionally,
N uptake per length of stolon (if any) is likely to be much lower than for roots, further
decreasing their "effective" length as a root. Based on the relative proportions of stolons
and roots reported by Rymph and Boote (2002), RFAC1 was reduced to 5000 cm g l,
33% lower than the value used for soybean roots.
As stolon mass is routinely mobilized to support new growth, the maximum
senescence rate (RTSDF) of the combined organs was increased from 0.01 to 0.02 or 2%
per day. In preliminary simulations this yielded a maximum predicted root mass of
approximately 10 000 kg root dry matter (DM) ha- in the range of the combined stolon
and root mass observed by Boote et al. (1999) (10 660 to 15 370 kg ha- ) and Rymph and
Boote (2002) (7155 to 15 740 kg ha- ).
Carbon and Nitrogen Mobilization Parameters
Another area where modeling perennial forages and annual grains differs is N
mobilization. The basic concept is the same, but the timing and purpose differ. Nitrogen
reserves in annual grain crops are generally mobilized for filling seed. Although many
perennial forages such as Pensacola bahiagrass may set seed, they are generally harvested
at a younger stage of maturity and reserves are used primarily to speed vegetative
regrowth after a harvest or in the spring. Since perennial forages must be able to do this
repeatedly over several growing seasons, the rate and extent of N mobilization may be
quite different than that observed in annual grain crops. Reports from Skinner et al.
(1999), estimating N and total nonstructural carbohydrate (TNC) mobilization in blue
grama grass [Bouteloua gracilis (H.B.K.) Lag ex Steud] during regrowth after cutting,
showed quite high rates of N mobilization. On average, 36% of the available N was
mobilized within 7 to 10 d of cutting. This translates into approximately 5% d-l or a
maximum available N mobilization fraction (NMOBMX) of 0.05. Total nonstructural
carbohydrate mobilization was also reported but the measured TNC concentrations were
2.5 to 3 times the levels found in bahiagrass, and we could not adapt them to our purpose
with any confidence. With no TNC data to support changes, the soybean values were
used for maximum available CH20 mobilization fraction (CMOBMX) (Table 4-1).
Vegetative Partitioning Parameters
During vegetative growth, partitioning of new growth among leaf, stem, and roots
is a function of the vegetative stage of the crop (V-stage). This is another area where the
concept of V-stage is different between annual crops and perennial forage crops. Annual
crops, as well as seedling forages, progress through the sequential increase in leaf
numbers in a relatively orderly fashion. Established perennial forages, however, are
periodically "re-staged" by harvests and frosts, interrupting the orderly pattern. As a
seedling, bahiagrass could reach a V-stage of 4 (four fully-expanded leaves) with a
relatively small root mass and few, if any, stolons. An established stand of bahiagrass,
with a relatively large root and stolon system capable of mobilizing significant amounts
of N, could also have the same V-stage rating of 4 after a harvest. In the CROPGRO
model, partitioning of subsequent growth is handled identically in both scenarios. A
unifying assumption is that, in both cases, if V-stage is low, the priority for partitioning is
towards growing leaf mass/area to establish photosynthetic capacity. As V-stage
increases, more DM may be partitioned to stolon and root. Additionally, since stolon and
root mass were combined, partitioning of new growth between organs required
modification from the proportions used for soybean.
While partitioning in seedlings may be measured by changes in leaf, stem, and root
mass over time, the presence of older, senescing material in established plants prevents
such a simple determination. Assuming that the model would be used most often to
predict growth of established stands, we developed the partitioning parameters around
observed patterns of regrowth, rather than purely on seedling growth. Parameter values
were estimated prior to optimization using growth patterns reported by Rymph (2002)
and Boote et al. (1999) and then refined by running simulations and manually adjusting
the parameters to match growth patterns and relative magnitudes of each organ (leaf,
stem, or root) (Table 4-1).
Leaf Growth and Senescence Parameters
Complications caused by repeated re-setting of the V-stage of the crop within a
growing season mandated some other modifications in addressing leaf growth-related
parameters. As the V-stage of the plants is reset after each harvest, there is potential for
V-stage to be quite low for a mature plant with numerous growing points stolonss). Use
of the VS SINTK function which allows photosynthesis and leaf expansion to be limited by
sink strength rather than assimilate supply, while potentially appropriate for a small
seedling, is not likely to fit the conditions of the older plant. To prevent potential
limitations to growth in older plants, the VS SINK function was disabled by assigning a
value of 0.0 to the VS SINK parameter.
Senescence parameters were modified very little from Kelly's (1995) pasture
model. The time constant for senescence (TCMP) was set to 25 thermal days based on
the weekly counts of dead leaves and weather from the raw data of Rymph and Boote
(2002). The light compensation point trigger for leaf senescence (ICMP), which triggers
leaf senescence due to shading of lower leaves was set to 0.8 moles photons m-2 d- the
same as soybean. In a similar vein, the V-stage trigger for senescence (when 12% of the
plant' s leaf number is assumed to have been senesced) (XSTAGE) was lowered from 14
leaves for soybean to 9. This was necessary because of the relatively low number of
leaves on a bahiagrass plant compared to a soybean plant.
The actual resetting of V-stage after a harvest is done in the PEST routine, using
either the MOW function (remove herbage to a designated residual mass) or a
combination of the HARV and HRVS (remove a designated proportion of existing
herbage mass and number of leaves) functions, and no modification of the species file
was required. To implement the MOW function, the user supplies the harvest date(s) and
the amounts) of stubble mass to remain after harvest. On the harvest date, CROPGRO
then calculates proportion of canopy mass removed and leaf mass, stem mass, and
V-stage are each reduced by that proportion. The HARV and HRVS functions work
similarly except that the user sets the proportion of herbage mass to be removed (HARV)
separately from the proportion of V-stage lost (HRVS).
The influence of temperature on the rate of phenological development of bahiagrass
is not well documented in the literature. Therefore, we set the cardinal temperatures for
base (no new leaves or seeds), optimum (maximum rate of leaf or seed addition), and
maximum (upper failure temperature) points (Table 4-1) based on our experiences
growing bahiagrass (K.J. Boote, personal communication).
Testing of Literature-Based Parameters
Testing of the preliminary, literature-based, species file was encouraging with
d-index values of 0.843, 0.605, and 0.925 for accumulated herbage mass, herbage N
concentration and accumulated herbage N mass, respectively, using the leaf-level
photosynthesis option and 0.851, 0.531, and 0.907 for the daily canopy photosynthesis
option (Table 4-2). Despite the moderate d-index values for predicted herbage N
concentration, the r2 ValUeS were quite low at 0. 18 and 0. 19 for the leaf and canopy
models, indicating that the N response was a weak point. The higher d-index value for
predicted herbage N mass than for either herbage mass or herbage N concentration
represents the effect of offsetting errors (underestimation of yield coincident with an
overprediction of N concentration).
Reviewing the predicted pattern of growth, however, showed excessive rates of
winter and spring growth (Figure 4-la). Water and N demand associated with this
excessive growth caused elevated water and N stress throughout the spring and early
summer (Figure 4-la), reducing predicted growth rates in May and June. The principle
cause of this discrepancy was our failure to simulate winter dormancy. Compounding
this was the lack of a working freeze damage routine promoting an artificially high LAI
and photosynthetic capacity through much of the winter. CROPGRO has no provisions
for modeling dormancy so we attempted to duplicate the dormancy effect through other
To reduce winter growth rate we used the PEST routine in CROPGRO to reduce
daily photosynthesis production by 70% from 23 October through 30 March. This
approach reduces photosynthetic rate but does not concurrently reduce transpiration.
Coincident with this change, the periods of water and N stress were shortened
considerably (Figure 4-1b). Statistically, however, there were minimal changes in the fit
of either leaf or canopy models after this modification (Table 4-2). This approach
resulted in slight improvements in d-index values for leaf and canopy model predicted
herbage mass, leaf model predicted herbage N mass, and slightly lower (worse) d-index
values for canopy model predicted herbage N concentration and herbage N mass (Table
4-2). Late spring regrowth was still considerably reduced compared to the observed
growth (Figure 4-1b), possibly because the PEST option reducing photosynthesis does
not reduce transpiration except indirectly through the lower LAI resulting from slowed
Further investigation revealed that two mechanisms may have been responsible for
the early season water and N stress. In reducing photosynthesis in the PEST routine, the
normal photosynthetic rate and transpiration rate were calculated and then the
photosynthetic rate was reduced by the designated percentage. Transpiration, however,
was not reduced so water uptake continued at the normal (now excessive) rate, depleting
available soil water. The only "reduction" in transpiration was due to the lowered LAI
that resulted from the slowed growth. Also, root N was mobilized throughout the winter
and into the spring to compensate for the reduced assimilate production, reducing root
mass considerably by the end of the winter/early spring period (data not shown). While
more water and N may have been available in the soil, the diminished root system had a
reduced capacity to exploit them, suppressing early season growth rates. Despite the
failure to statistically improve the fit, we used this strategy in all optimization and testing
runs as the patterns of N stress were more realistic than before.
Performance of the model using the daily canopy option and winter photosynthetic
reduction was quite similar to the leaf-level option performance for predicting herbage
mass (Figure 4-2). The d-index values for the fit of the predicted data to herbage mass
were identical to those for the leaf-level option. The predicted pattern of growth was also
quite similar with slightly reduced winter growth rates but nearly identical summer
growth (Figure 4-2). Fit of predicted herbage N concentration was slightly poorer for the
canopy option, but fit of predicted herbage N mass was similar for both options (Table
Fit of predicted herbage N concentration was not as good as was herbage mass for
either option (Table 4-2). Leaf + stem N concentration was consistently overpredicted
for the Ona, FL site (Figure 4-3). During winter regrowth, after the "simulated frost"
defoliation, herbage N concentration exceeded 40 g N kg- equivalent to 250 g CP (crude
protein) kg- higher than the "maximum" leaf N concentration set by PROLFI. This is
related to the N allocation problem in the code cited earlier that prohibited leaf regrowth
after a FREEZ l event. As the goal of the present exercise was to calibrate the parameters
without changing any source code, this problem could not be addressed.
For the Eagle Lake, TX data, predicted herbage N concentration appeared to follow
a more accurate pattern despite greater variation in the observed values (Figure 4-4).
Prediction of herbage N concentration was more balanced, being both over- and under-
predicted. The improved prediction pattern may be related to the lower fertilizer levels
used at Eagle Lake and the lower yields for that site. Values from the daily canopy
photosynthesis option were generally higher than for the leaf-level option (Figures 4-3
Since we were unable to accurately predict the spring growth pattern, some of the
early season yield data were excluded from the optimization process. The rationale for
"culling" these two data points was that the model was consistently predicting early
season N and water stress when there was none; thus keeping those data points in the
optimization would influence the final parameter values in order to compensate for the
predicted stresses. This left 52 data pairs for calibration. No data were excluded from
the datasets used to test the performance of the model. All testing runs used all of the
observed data available for the site/fertility combinations used (27 data pairs). The
distinction being made here is that we wanted to develop the most accurate parameters
for the model through optimization (hence leaving out the early season data) while
presenting a fair evaluation of the performance of the model through testing (by including
Our strategy was to first optimize the leaf-level option temperature parameters to
establish proper general patterns of growth, then refine the prediction by optimizing
parameters that affect the growth response to N. After optimizing the leaf-level option,
the temperature parameters for the daily canopy photosynthesis option were optimized
followed by simultaneous optimization of PARMAX and PHTMAX.
All temperature parameters (leaf-level and daily canopy options) and
PARMAX/PHTMAX optimizations used only the two highest N fertility treatments from
each experiment, assuming that N would not be limiting growth for those treatments.
The two highest and two lowest N fertilization treatments from each study were used in
optimizing the N parameters as this presented the broadest range of conditions.
Testing of Optimized Parameters
Optimization improved the predicted winter growth pattern (Figures 4-5a and 4-5b)
but fit of predicted herbage mass during the growing season was generally unaffected
(Table 4-2) with similar d-index values for both optimized and literature-based species
Eiles. Winter growth pattern was improved most for the daily canopy option where there
was almost no regrowth through the winter (Figure 4-5b). A wider range of potential
parameter values were offered in the daily canopy option optimization as there was less
data available to define the ranges. Winter regrowth was curtailed by increasing
FNPGT(1), only allowing growth on days with average temperatures greater than 200C
(Table 4-1). Normally, this would affect growth rates well into the spring and fall
growing season, but PARMAX and PHTMAX were also boosted in the optimization
(Table 4-1). This combination allowed growth rates to be nearly identical during the
normal growing season (Figures 4-5a and 4-5b). Nitrogen stress was reduced in the
optimized simulations (Figures 4-5a and 4-5b); however, water stress was still extensive
in the spring, even for the daily canopy option, resulting in a continued poor prediction of
first cutting regrowth at Ona, particularly in the second growing season. To compensate
for the failure of the model to properly simulate winter dormancy, the optimization
process promoted combinations of extreme parameter values to improve the fit of the
Overall, both the leaf-level and daily canopy options tended to overpredict herbage
mass at lower yields and underpredict at higher yields with this phenomena most evident
in the daily canopy option results (Figures 4-6a and 4-6b). Despite the improved fit, the
optimized values for these three parameters, FNPGT(1), PARMAX, and PHTMAX, are
not realistic and reflect the emphasis on compensating for the excessive winter growth
pattern, not improving the relevancy of the parameter value.
Fit of herbage N concentration predictions improved considerably for both the
optimized leaf-level and optimized daily canopy photosynthesis options (Table 4-2,
Figures 4-7 and 4-8). The d-index rating for both options improved considerably after
optimization; however, the r2 ValUeS remained quite low (Table 4-2). Predicted herbage
N concentrations were still consistently overpredicted at Ona (Figure 4-7) despite
optimization toward lower PROLFI, PROLFG, PROSTI, and PROSTG parameters
(Table 4-1). The pattern of predicted herbage N concentration for Eagle Lake remained
realistic after optimization (Figure 4-8), however, the optimized parameter values were
generally lower than the literature-based parameters (Figure 4-4). The difference
between predicted and observed values was also reduced for both the leaf-level (Figure
4-9a) and daily canopy options (Figure 4-9b) indicating a more consistent prediction.
Both leaf-level and daily options tended to overpredict herbage N concentration, but this
was more pronounced in the daily canopy option predictions (Figures 4-9a and 4-9b).
Despite the improvement in herbage N concentration prediction, fit of predicted herbage
N mass showed little improvement (Table 4-2) but had been quite high to begin with.
Performance of the literature-based parameters was quite good, especially related to
predicting herbage mass and herbage N mass. The prediction of herbage N concentration
needed improvement. On review of the results, there appeared to be some features of
CROPGRO that may have made significant contributions to the errors in predicting both
herbage N concentration and herbage mass. The absence of a dormancy routine to
control vegetative growth during the winter and spring months had a profound effect on
early season N and water availability, contributing to low predicted herbage mass
throughout the season. The absence of a storage organ such as a rhizome or stolon
contributed to this problem by confounding effects of changing proportions of stolon and
root mass. Quirks related to modeling of freeze damage and patterns of "refi11ing" of N
in old tissues complicated matters even more. Imposing a 70% reduction in potential
daily photosynthesis during the winter months compensated for some of the problems,
albeit in an artificial way.
Optimization did improve the fit of both the leaf and canopy models compared to
simulations using the literature-based parameters. Winter growth was slow, and
excessive levels and variation in leaf N concentration were controlled using the optimized
parameters. However, some of the optimized parameters are at the edges of their
biological range or beyond as a result of compensating for the missing/problem
components in the model code. The optimization was more an exercise in compensating
for the model than in divining more accurate parameter values.
In order to better mimic the biology of perennial, tropical grasses, modifications
must be made to the model code itself. Primary among these changes is the addition of a
dormancy routine. Evidence for this and the mechanism required is available in the
literature. Sinclair et al. (2003), Mislevy et al. (2000), and Gates et al. (2001) clearly
demonstrate the role of daylength in controlling dormancy. Hints for the mechanism
involved in reducing leaf and stem growth during dormancy can be found in Rymph and
Boote (2002) and Boote et al. (1999) where significant shifts in allocation of new growth
from shoots to stolons were observed in the fall. Adding a mechanism controlled by
daylength to reduce partitioning and mobilization to the shoot while increasing the same
to the stolon would complement the maturity, temperature, and stress mechanisms
already present in CROPGRO.
Addition of the storage organ would also allow more realistic prediction of the
patterns of accumulation and depletion of roots, avoiding confounding root mass and N
uptake parameters to compensate for the presence of stolons in the root mass. Providing
a storage organ not only provides a sink to store the excess assimilate that is currently
allocated to leaves and stems in the winter, it would also supply a source of CH20 and N
for regrowth after frosts, in the spring, and after harvests. This would prevent the current
situation of the plants dying after a frost and allow for more rapid regrowth in the spring.
Other elements of the model, such as the freeze damage scheme and the
partitioning of N to replenish old leaves, likely stem from past approaches to modeling an
annual grain crop compared to a perennial forage. Situations such as low leaf mass after
a harvest coupled with large amounts of available N from the roots are not generally
encountered in the life cycle of maize or soybean but are dominant features of the pattern
of growth of a perennial grass. These differences are better addressed through adapting
the model code than by adjusting species parameters.
Consideration of these differences notwithstanding, the overall performance of both
the literature-based and optimized parameters was good. If used carefully, the optimized
leaf and canopy models should perform well. More testing would be in order if these
models were to be used extensively. As mentioned earlier, further optimization will only
improve our ability to compensate for the model code, not improve the quality of the
parameters. Taking steps such as running the simulation for a year prior to the measured
growing seasons to establish initial conditions, addition of defoliation events to simulate
frosts, and addition of photosynthesis reduction schemes to reduce winter growth will be
as critical as changing parameter values in establishing a good fit of model predictions to
observed data. The bulk of future efforts should be directed at changing the model code
to more accurately reflect the life cycle of perennial grasses.
Table 4-1. Bahiagrass parameter values for the CROPGRO species file. Preliminary
values were derived from the literature. Optimized values were derived from
optimization runs made based on the preliminary values.
Parameter Name Preliminar value Opiized value
PARMAX 60.0 140.0
PHTMAX 90.0 180.0
FNPGN(1-4) 0.75, 3.0, 10.0, 10.0 1.0, 3.0, 10.0, 10.0
FNPGT(1-4) 12.0, 25.0, 38.0, 50.0 20.0, 25.0, 30.0, 50.0
XLMAXT -5.0, 7.0, 35.0, 45.0, 55.0, 60.0 -5.0, 10.0, 26.0, 45.0, 57.0, 60.0
YLMAXT 0.0, 0.0, 1.0, 1.0, 0.0, 0.0 0.0, 0.0, 1.0, 1.0, 0.0, 0.0
FNPGL(1-4) 7.0, 18.0, 45.0, 57.0 7.0, 18.0, 45.0, 57.0
PROLF I, G, and F 0.22, 0. 11, 0.05 0.15, 0.05, 0.04
PROST I, G, and F 0. 11, 0.07, 0.033 0.125, 0.04, 0.022
PRORT I, G, and F 0.101, 0.040, 0.022
PLIP LF, ST, RT 0.025, 0.020, 0.020
PLIG LF, ST, RT 0.04, 0.06, 0.07
PCAR LF, ST, RT 0.602, 0.697, 0.702 0.672, 0.682, 0.702
XLEAF 0.0, 1.5, 2.0, 3.0, 5.0, 7.0, 30.0
YLEAF 0.45, 0.5, 0.6, 0.4, 0.25, 0.2,
YSTEM 0.05, 0.05, 0.1, 0.1, 0.05, 0.05,
XVGROW 0.0, 5.0, 10.0, 15.0, 20.0, 25.0
YVREF 0.0, 10.0, 20.0, 30.0, 40.0, 50.0
XSLATM -50.0, 00.0, 10.0, 30.0, 60.0
Table 4-1. Continued
Parameter Name Preliminar value Opiized value
YSLATM 0.25, 0.25, 0.25, 1.00, 1.00
FREEZ l, FREEZ2 -25.0, -25.0
XSTAGE 0.0, 5.0, 9.0, 50.0
XSENMX 3.0, 5.0, 10.0, 50.0
TB, T1, T2, TMax 9.0, 32.0, 40.0, 45.0
TB, T1, T2, TMax 10.0, 28.0, 32.0, 45.0
TB, T1, T2, TMax 10.0, 28.0, 32.0, 45.0
XVSHT (1-10) 0.0, 1.0, 4.0, 6.0, 8.0, 10.0,
14.0, 16.0, 20.0, 40.0
YVSHT (1-10) 0.0150, 0.0265, 0.0315,
0.0330, 0.0345, 0.0330,
0.0310, 0.0255, 0.0170, 0.0030
YVSWH (1-10) 0.0150, 0.0255, 0.0310,
0.0320, 0.0330, 0.0315,
0.0295, 0.0230, 0.0125, 0.0005