• TABLE OF CONTENTS
HIDE
 Introduction
 Methods and materials
 Results
 Discussion
 References






Title: Testing modified stability analysis with biophysical process models
CITATION THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00082699/00001
 Material Information
Title: Testing modified stability analysis with biophysical process models
Physical Description: Book
Creator: Hildebrand, Peter E.
Bowen, W. T.
Kelly, T. C.
Publisher: Unknown
 Record Information
Bibliographic ID: UF00082699
Volume ID: VID00001
Source Institution: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: oclc - 212627747

Table of Contents
    Introduction
        Page 1
    Methods and materials
        Page 2
    Results
        Page 3
        Page 4
        Page 5
        Page 6
        Page 7
        Page 8
        Page 9
        Page 10
    Discussion
        Page 11
        Page 12
        Page 13
        Page 14
        Page 15
        Page 16
        Page 17
    References
        Page 18
Full Text

/16, &69

q%, 001-
TESTING MODIFIED STABILITY ANALYSIS
WITH BIOPHYSICAL PROCESS MODELS 1

P.E. Hildebrand, W.T. Bowen and T.C. Kelly 2

Crop growth computer simulation models have been developed over the
past decade and are increasingly being used in agricultural
research and management. Their ability to accurately predict crop
growth in a wide range of environments has improved markedly in the
past few years. These models use crop physiological
characteristics, related to the genetic make-up of specific lines,
to predict responses of these lines to various environmental
situations. The models can predict the response of crops to
numerous combinations of planting dates, soils, climatic
conditions, fertilizations, and plant populations.

Selected maize cultivars, for example, could be evaluated for many
different conditions, representing various soils, weather years and
management practices, in a matter of minutes or hours. To do the
same under real field conditions would require several months for
each year's data, and two or more years to match the output of the
computer models. While, theoretically, recommendations for farmers
could be made directly on the basis of simulated experiments, the
prospects for this application are limited, at least until greater
experience with this approach has been gained, and the range of
environments in which the crop models are valid is known (Harrison
et al., 1990).

Meanwhile, these models still have many limitations. Their data
requirements are high, and their application, particularly in
environments for which they have not been validated, requires a
high degree of familiarity with the internal workings of the
models. Perhaps the most severely limiting characteristic at the
present time is the incapability of current models to account for
stresses due to pests, diseases, and inter-species competition,
although these deficiencies are currently being addressed.
Computer solutions closely simulate observed field results under
near optimum production conditions (such as found on experiment
stations or on high technology farms) but overestimate production
rather severely when pest or disease problems negatively affect
field yield (Gilbert, 1992).

Under the auspices of the International Benchmark Sites Network for
Agrotechnology Transfer (IBSNAT), scientists at the University of
Florida, and elsewhere, have been involved with crop growth model
development and applications (IBSNAT, 1989). Recent work has begun
at the University of Florida and elsewhere to analyze the utility

1 For presentation at the 12th annual Farming Systems
Symposium, Michigan State University, September 13-18, 1992.
2 University of Florida, Gainesville, FL 32611-0240










of these models to complement on-farm research. Validated crop
models have the ability to screen agricultural technology options
so that only the most promising go to on-farm trials. Technology
packages can be designed and assessed efficiently using computer
simulation. Management practices and genetic characteristics can
be pretested in other locations by changing model parameters and
inputs. Finally, by using historical or generated weather
sequences, variability associated with particular practices can be
assessed through space and time (Dent and Thornton, 1988; Thornton,
1991).

One potentially useful aspect of crop models is to generate data
which simulate an on-farm research program. Results could then be
used to help understand on-farm research analysis and design and
could provide an efficient source of data for training purposes.
This paper explores this use. Specifically, the CERES-maize model
(Jones and Kiniry, 1986) and the Strategy Evaluation in the DSSAT
(IBSNAT, 1989) were used to generate maize response to a number of
soils conditions representative of those in the southeastern United
States and recorded climatic conditions over a four year period for
10 locations in this area. The data thus generated are analyzed by
Modified Stability Analysis (Hildebrand, 1984).


Methods and materials

Crop growth simulation model

The DSSAT suite of crop models takes historical or generated
weather data and user-supplied soils and management practices to
simulate crop growth. For this research, historical weather data
were available for four years, 1984-87, at ten different locations,
three in southern Alabama, four in southern Georgia, and three in
northern Florida. At each site, a representative soil type was
selected such that all ten soils were different. Soil types ranged
from silty clay to silt loam to sandy loam to deep fine sands.

From a wide array of possibilities, the DSSAT model was used to
generate simulated trial data over four years and for 10 soils for
two maize cultivars (PIO 3382 and CESDA-28), one planting date
(March 15), two plant populations (30,000 and 60,000 plants ha'1),
and two fertilizer levels (50 and 150 kg N ha-1). Fertilizer was
applied in two equal applications at planting and 30 days after
planting. Ultimately, because the two cultivars were very similar,
except that one consistently out yielded the other, only one
cultivar (PIO 3382) is used in the analysis. Hence, the simulated
on-farm trial represents a 2x2 factorial with two plant populations
and two fertilizer levels. Only one complete block was planted for
each environment each year.










Modified Stability Analysis

Modified Stability Analysis (MSA) has been suggested as an
efficient and effective method for designing and analyzing on-farm
research data when the purpose is to make recommendations for
specific environmental conditions and for farmers' varying
evaluation criteria (Stroup, et al., 1991). The procedure uses an
environmental index (EI) as a continuous, quantifiable measure of
the quality of the environment for producing the crop or crops
being evaluated (Hildebrand, 1984). Responses of the individual
treatments are regressed across environments and the treatments are
compared in order to select those which excel in different kinds of
environments and for different evaluation criteria.

It has been observed that if the data generated by on-farm research
meet three criteria, relationships among treatments and
environments found in one year will be stable over years (Stroup,
et al., 1991). These three criteria are:

1) The range of environments, as measured by the difference
between the highest and lowest EI, is as least as large as the
overall mean yield for the set of environments included in the
on-farm trial;

2) The range of yields obtained represents approximately what
would be expected over a period of years for the environments
included in the trial; and

3) The Els are reasonably well distributed.

If this observation proves sound, it can materially shorten the
amount of time between technology evaluation and diffusion. Rather
than the wait of several years suggested by conventional wisdom,
recommendations confidently can be made to farmers after only one
year of on-farm trials.


Results

Mq ha-1 as a criterion

Data for all four years are shown in Table 1. The question to be
examined is whether the data from the first year (1984), or for
that matter, any year that might have been first, represent the
same relationships among treatments and between treatments and
environment that would occur after several years of trials. Linear
regression results for 1984 are in Figure 1. Do the data satisfy
the three criteria above? For 1984, the range of Els was very











4

Table 1. Simulated annual on-farm maize research results,
southeastern United States, 1984 1987.
---------TREATMEMTS---------
YEAR "SITE" P1N1 P2N1 P1N2 P2N2 El
====S== B=fBCSB == *=== === = =S = === if========


1984 UFU
1984 GAHO
1984 GATI
1984 AUHE
1984 UFQU
1984 GABL
1984 AUGE
1984 AUDO
1984 UFGA
1984 GAAB

AVERAGE, 1984

1985 GAAB
1985 UFU
1985 GAHO
1985 GATI
1985 AUHE
1985 AUGE
1985 UFQU
1985 UFGA
1985 AUDO
1985 GABL

AVERAGE, 1985

1988 GAAB
1986 GAHO
1986 UFU
1986 GATI
1986 UFQU
1988 AUGE
1986 GABL
1986 AUHE
1986 UFGA
1986 AUDO

AVERAGE, 1986

1987 AUGE
1987 AUHE
1987 GABL
1987 GATI
1987 GAHO
1987 GAAB
1987 AUDO
1987 UFU
1987 UFQU
1987 UFGA

AVERAGE, 1987


2.09 2.50 1.94 2.33 2.22
1.52 1.35 3.29 3.53 2.42
3.88 2.55 5.68 7.35 4.87
4.75 3.64 5.95 9.47 5.95
5.38 4.09 6.25 8.88 6.15
5.50 7.22 5.48 7.47 6.42
3.28 5.66 6.20 10.79 6.48
5.78 8.49 5.78 8.52 7.14
6.29 8.65 6.29 9.24 7.62
5.95 9.44 5.95 11.33 8.17

4.44 5.36 5.28 7.89 5.74

0.00 0.00 0.00 0.00 0.00
1.94 1.88 1.84 2.07 1.93
3.42 2.46 3.36 3.96 3.30
3.93 2.55 4.87 8.21 4.89
5.68 4.32 5.94 9.44 6.35
5.88 5.15 6.22 11.09 7.09
5.95 6.03 5.95 10.75 7.17
5.91 8.42 5.91 8.97 7.30
5.98 10.39 5.98 10.62 8.24
6.18 10.39 6.18 11.14 8.47

4.49 5.16 4.63 7.63 5.47

0.00 0.00 0.00 0.00 0.00
2.47 2.48 2.51 2.61 2.52
2.65 2.93 2.67 2.93 2.80
2.00 1.32 5.12 4.01 3.11
3.30 2.30 3.99 4.49 3.52
3.91 2.89 4.72 5.72 4.31
4.38 4.32 4.38 4.71 4.45
4.89 3.11 5.59 7.35 5.24
5.81 4.99 5.81 7.72 6.08
5.57 9.62 5.57 9.78 7.64

3.50 3.40 4.04 4.93 3.97

0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
2.94 2.23 3.85 3.94 3.24
5.33 4.07 5.94 9.04 6.10
5.78 8.19 5.78 10.00 7.44

1.41 1.45 1.56 2.30 1.68





&c) 62/
/ ,, j.'., ,


V 1ic1


SIMULATED MAIZE TRIAL
1984


ENVIRONMENTAL INDEX, El


Figure 1.
measured by


Response of four maize treatments to environment as
the environmental index, EI, 1984.


-p


P1 N1

P2N1
-8f-
P1 N2

P2N2










near the overall mean (range/overall mean = 1.0), so the first
criterion is met. Because they are simulated data, it is somewhat
difficult to decide the second. However, except for years when
frost kills early-planted maize, particularly in the more northern
part of this geographic area as happened in 1987, these yield
levels are fairly representative. Therefore, the second criterion
can be accepted for 1984 as well. The distribution of observations
for 1984 was quite acceptable, given the nature of on-farm
research, so this criterion was also fulfilled, even if one or two
additional low observations would be helpful. Because all three
criteria were at least marginally satisfied, we are hypothesizing
that the relationships among the treatments (population and
fertilizer) and between the treatments and environment obtained the
first year (1984) will be stable over time.

The 1985 data included frost loss in the most northern site, so the
range of Els was greater than for 1984, and greater than the
overall mean (range/overall mean = 1.5), Table 1 and Figure 2. The
appearance of the figure is somewhat different from Figure 1, but
the relationships are quite similar. In 1985, the P2N2 treatment
was superior overall, as it was in 1984. In both 1984 and 1985,
the higher population combined with the lower fertilizer rate
(P2N1) was superior to either of the lower population treatments in
the better environments. However, the relationships were reversed
in the poorer environments which were unable to support the higher
plant populations. In 1984, UFLI, GAHO and GATI were in the poorer
environments. In 1985, UFLI and GAHO, and possibly GATI were there
as well as GAAB where there was frost kill. In 1984, the response
of the lower population to fertilizer was slightly stronger than it
was in 1985, but this response was considerably stronger with the
higher population in either year.

Had either 1984 or 1985 been the first year, the conclusions would
have been essentially the same. The combined 1984-85 regression
relationships (Figure 3), had they been combined, showed a great
deal of similarity to the relationships from the individual years.
The 1986 data were also very similar to the previous two years,
Figure 4.

Only 1987 data resulted in somewhat different relationships, Figure
5, resulting from a widespread freeze which affected seven of the
10 environments, Table 1. In 1987, treatment P2N2 was still
superior overall, but there was no difference among the other three
treatments, contrary to results in the other years. What if 1987
had been the first year rather than the last? The first two
criteria would have been met, but the third definitely was
violated. Hence, there would have been limited confidence in the
resulting relationships from 1987 data if it had been the first
year of the trial, so the criteria would have provided sufficient
warning.







Vi^7







SIMULATED MAIZE TRIAL
1985
12-
-U-
P1 N1
10- .......... ......-.. -.. ................---------
P2N1








2-- ---------------- ------------- --- --- -
0111P1 N2









0 1 2 3 4 5 6 7 8 9
ENVIRONMENTAL INDEX, El






Figure 2. Response of four maize treatments to environment as
measured by the environmental index, EI, 1985.















SIMULATED MAIZE TRIAL
1984-85
12

10 ----------------------------------------------------------- -- ........---.
P1 N1

P2N1
co 8 .................... .---- -----------.-.- --.-
8-
I P1 N2

0 ,-. ,P2N2
w
5 4 ..--- ---- - - -................. ......... .

2-.... ..............----- .- ------------------------ ----------

011
0 1 2 3 4 5 6 7 8 9
ENVIRONMENTAL INDEX, El






Figure 3. Response of four maize treatments to environment as
measured by the environmental index, EI, 1984 1985.
















SIMULATED MAIZE TRIAL
1986
12
-----
P1N1
1 0 -0 -- ---------- ------- ---- ---- --- -- ----- --- -............-.......... .............................................
P2N1
cz 8 .-.-----------------. .... ...........------- ... ....... -.....
-c P1 N2

o P2N2
w






0
0 1 2 3 4 5 6 7 8 9
ENVIRONMENTAL INDEX, El







Figure 4. Response of four maize treatments to environment as
measured by the environmental index, EI, 1986.














SIMULATED MAIZE TRIAL
1987


P1N1

P2N1

P1 N2

P2N2


ENVIRONMENTAL INDEX, El


Figure 5. Response of four maize treatments to environment as
measured by the environmental index, EI, 1987.










Stability of the individual treatment regressions over years

Figures 6 9 show the linear regression relationships for each
year and for the four year period for the individual treatments.
At least for this set of data, these relationships are remarkably
stable, even though the range of Els in the individual years is
different from the range over the four year period.

If the four year period, Figure 10, is considered to provide the
real range of possibilities from which to make a decision regarding
recommendations, these would have been little different from any of
the individual years, even possibly 1987 when a widespread freeze
killed 70 percent of the trial. The poorer environments over the
four year period, Table 2, were UFLI, GAHO and GATI, the same as
appeared in both 1984 and 1985.


Discussion

As the data for these simulated on-farm trials were being
generated, the crop growth simulation model used demonstrated one
of the still-existing weaknesses of these models -- overestimation
because of the incapability of the model to account for stresses
due to pests, diseases and weeds. When usual maize planting dates
for the region were used in the models, yields inevitably were
uncharacteristically high with few low yields included. The
results violated both the first and second confidence criteria.
Only by moving to earlier and earlier planting dates could a
realistic range of yields be achieved.

Having achieved realistic results from the crop simulation model,
the analysis demonstrated that relationships among treatments and
between treatments and environment as measured by Modified
Stability Analysis, do stabilize in one year if the three
confidence criteria are met.

Finally, it would appear that there is potential for using crop
simulation models for training purposes, and to simulate on-farm
research results. However, these models need to be more realistic
in their responses to insect pests, diseases and weeds before they
can begin to provide the same kind of information available from
well designed and conducted on-farm research.










12

Table 2. Simulated four year on-farm maize research results,
southeastern United States, 1984 1987.
........--------TREATMEMTS --------.
YEAR "SITE" P1N1 P2N1 P1N2 P2N2 El
:===== =wwww M==== ==== ==-== = == = = =-


1987 GATI
1987 AUGE
1985 GAAB
1987 GABL
1986 GAAB
1987 GAAB
1987 AUDO
1987 AUHE
1987 GAHO
1985 UFL
1984 UFU
1984 GAHO
1986 GAHO
1986 UFU
1986 GATI
1987 UFU
1985 GAHO
1986 UFQU
1986 AUGE
1986 GABL
1984 GATI
1985 GATI
1986 AUHE
1984 AUHE
1986 UFGA
1987 UFQU
1984 UFQU
1985 AUHE
1984 GABL
1984 AUGE
1985 AUGE
1984 AUDO
1985 UFQU
1985 UFGA
1987 UFGA
1984 UFGA
1986 AUDO
1984 GAAB
1985 AUDO
1985 GABL


0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00
1.94 1.88 1.84 2.07 1.93
2.09 2.50 1.94 2.33 2.22
1.52 1.35 3.29 3.53 2.42
2.47 2.48 2.51 2.61 2.52
2.65 2.93 2.67 2.93 2.80
2.00 1.32 5.12 4.01 3.11
2.94 2.23 3.85 3.94 3.24
3.42 2.46 3.36 3.96 3.30
3.30 2.30 3.99 4.49 3.52
3.91 2.89 4.72 5.72 4.31
4.38 4.32 4.38 4.71 4.45
3.88 2.55 5.68 7.35 4.87
3.93 2.55 4.87 8.21 4.89
4.89 3.11 5.59 7.35 5.24
4.75 3.64 5.95 9.47 5.95
5.81 4.99 5.81 7.72 6.08
5.33 4.07 5.94 9.04 6.10
5.38 4.09 6.25 8.88 6.15
5.68 4.32 5.94 9.44 6.35
5.50 7.22 5.48 7.47 6.42
3.28 5.66 6.20 10.79 6.48
5.88 5.15 6.22 11.09 7.09
5.78 8.49 5.78 8.52 7.14
5.95 6.03 5.95 10.75 7.17
5.91 8.42 5.91 8.97 7.30
5.78 8.19 5.78 10.00 7.44
6.29 8.65 6.29 9.24 7.62
5.57 9.62 5.57 9.78 7.64
5.95 9.44 5.95 11.33 8.17
5.98 10.39 5.98 10.62 8.24
6.18 10.39 6.18 11.14 8.47





((pe pb /l
(/4 /0/q ( ~)~


SIMULATED MAIZE TRIAL
P1 N1, INDIVIDUAL YEARS & ALL YEARS


1984

1985


1987

ALL


ENVIRONMENTAL INDEX, El


Figure 6. Response of P1N1 to environment as measured by the
environmental index, EI, individual years (1984-87) and overall.





c uP 4


9
~& 1<


SIMULATED MAIZE TRIAL
P2N1, INDIVIDUAL YEARS & ALL YEARS


ENVIRONMENTAL INDEX, El


Figure 7. Response of P2N1 to environment as measured by the
environmental index, EI, individual years (1984-87) and overall.


1984

1985

1986

1987

ALL














SIMULATED MAIZE TRIAL
P1N2, INDIVIDUAL YEARS & ALL YEARS


ENVIRONMENTAL INDEX, El


Figure 8. Response of P1N2 to environment as measured by the
environmental index, EI, individual years (1984-87) and overall.


1984

1985

1986

1987

ALL






16


SIMULATED MAIZE TRIAL
P2N2, INDIVIDUAL YEARS & ALL YEARS


3 4 5 6
ENVIRONMENTAL INDEX, El


1984

1985

1986

1987

ALL


Figure 9. Response of P2N2 to environment as measured by the
environmental index, EI, individual years (1984-87) and overall.






17


SIMULATED MAIZE TRIAL
1984 -1987


P1N1

P2N1

P1N2

P2N2


ENVIRONMENTAL INDEX, El


Figure 10. Response of four maize treatments to environment as
measured by the environmental index, EI, 1984 1987.










REFERENCES

Dent, J.B. and P.K. Thornton. 1988. The role of biological
simulation models in farming systems research. Agricultural
Administration and Extension, 29:111-122.

Gilbert, R.A. 1992. On-farm testing of the Pnutgro crop model in
Florida. Unpublished M.S. thesis, Department of Agronomy,
University of Florida.

Harrison, S.R., P.K. Thornton and J.B. Dent. 1990. The IBSNAT
project and agricultural experimentation in developing
countries. Experimental Agriculture, 26:369-380.

Hildebrand, P.E. 1984. Modified stability analysis of farmer-
managed on-farm trials. Agronomy Journal, 76:271-274.

IBSNAT. 1989. Decision Support System for Agrotechnology Transfer
(DSSAT) Version 2.1. Dept. of Agronomy and Soils, University
of Hawaii, Honolulu, HI 96822.

Jones, C.A., and J.R. Kiniry. 1986. CERES-Maize: A simulation model
of growth and development. College Station, TX: Texas A&M
University Press.

Stroup, W.W., P.E. Hildebrand and C.A. Francis. 1991. Farmer
participation for more effective research in sustainable
agriculture. Staff Paper SP91-32. Food and Resource
Economics Department, University of Florida.

Thornton, Phillip. 1991. Application of crop simulation models in
agricultural research and development in the tropics and
subtropics. International Fertilizer Development Center,
Muscle Shoals, Alabama.




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs