MODIFIED STABILITY ANALYSIS AND ON-FARM RESEARCH
TO BREED SPECIFIC ADAPTABILITY FOR ECOLOGICAL DIVERSITY Peter E. Hildebrand
In this paper, I argue that the cumulative effect of four factors has led to at least a quarter century of rejecting genetic material which would have demonstrated superior yield capability in the best or the poorest farm environments,
and has not necessarily led to varieties or hybrids with superior yield capability in average environments.
These four factors are: one, statistical dependence on ANOVA that leads to the concern with reducing genotype-by-environment interaction.
This in turn leads to the nearly universal practice of evaluating material on experiment stations and farms with real or artificially created superior environments.
Adding to the effect has been the capability of many farmers in the developed world to use their resources to modify unfavorable environments and thus adjust their conditions to those required by the varieties and hybrids developed under superior conditions.
The final factor has been the widespread use of a regression coefficient of one as a measure of stability when evaluating environmental effect on genotype.
8 (Wheat trial)
For more than a half century, dating, to my knowledge, from the Yates and Cochran article in 1938, agricultural scientists, and in particular plant breeders, have been using some form of stability analysis to evaluate germplasm over different locations and environments. Recently, persons in developing countries and working with farming systems research and extension methodologies have been using a modified form of the procedure to develop practices and germplasm for the widely varying environments found in many of these countries. While the activities cited above use a common methodology to find superior technology, the reasons for its use, the philosophy behind it and the results differ.
9 Irrigated Tobacco
Over the last few decades, and in the developed world where resources were relatively abundant, widely adaptable technology, achieved by manipulation of the crop growing environment on farms, has been feasible. Mechanization, irrigation and heavy use of chemicals allowed farmers to provide the right environment to achieve maximum production with widely adapted crop varieties. However, in the developing world where resources are not sufficient to allow farmers to modify adverse environments, farmers and crop scientists have had to tailor technology to specific environments. The 'Green Revolution' had a large and critical impact on food production in this part of the world, but it affected mostly those farms blessed with the better environments.
It has been only recently that organized development efforts have been able to reach the majority of farmers -- those with few resources and living in relatively hostile environments -- by adapting technology to their environments.
11 Perennial peanut
Now in the developed world, with increasing concern for an agriculture that is environmentally and economically more sustainable, crop adaptation to environment, rather than adapting environment to the crop, is also becoming more important.
Breeding for specific adaptability requires a different interpretation and approach to the stability analysis procedure than breeding for broad adaptability.
This paper discusses the use of a modified approach to stability analysis as an important research tool which can provide the basis for an effective and efficient plant breeding program for which adaptability to specific environments is desired as a critical factor in enhancing the sustainability of agriculture throughout the world.
First, in this paper, the philosophical history of stability analysis is discussed.
Then the effect of this history on plant technology development is argued.
Finally, the farming systems philosophy is discussed and the use of stability analysis in this approach is presented.
Yates and Cochran, apparently the first to conceptualize stability analysis, understood the potential to differentially recommend varieties based on their performance in different environments: "The fact that the experimental sites are a random sample of this nature does not preclude different recommendations being made for different categories included in this random sample."
In fact, they argue that the deliberate inclusion of sites representing extreme conditions may be of value.
However, their primary concern was the statistical problem of how to evaluate mean differences among or between treatments which were tested over a wide range of environments and where genotype-by-environment interaction was present.
Finlay and Wilkinson refined and enhanced the Yates and Cochran procedure. Perhaps their most important contribution was to describe the characteristics of varieties with high (> 1.0) and low (< 1.0) stability (regression) parameters and relate these to variety mean yield over environments. They defined varieties with general adaptability as those with average stability (b = 1.0) when associated with high mean yield over all tested environments.
However, they also recognize that above average stability (b approaching zero) is associated with increasing adaptability to low-yielding environments,
and below average stability with increasing adaptability to high-yielding environment. They argue in favor of testing over a wide range of environments and not discarding low yielding environments because this ". . will bias the selection towards types specifically adapted to high-yielding environments, and will pass over those with general adaptability".
As this statement emphasizes, however, their interest ultimately was on generally adapted varieties. In speaking of the varieties included in the tests they were reporting, they said:
"The varieties with the high phenotypic stability all have low mean yields.
They are so stable, in fact, that they are unable to exploit high-yielding
environments. On the other hand, varieties can be too sensitive to
environmental change, as is shown by the low mean yields of the varieties with high regression coefficients. The generally adapted varieties balance
between these extremes, the actual point of balance depending on the
particular genotype and range of environments" (p. 752).
At the same time, Finlay and Wilkinson were cognizant of the hazard resulting from discarding potentially valuable materials when searching only for varieties with high mean yields over all environments.
Eberhart and Russell, authors of probably the most cited article regarding the use of stability parameters, were concerned that large genotype-by-environment interactions reduced 'progress' from selection.
One means of reducing this interaction, they reported, is to stratify environments to make them more similar. However, they thought the resulting interaction still "frequently remains too large".
Their solution was a method to select stable genotypes that interact less with the environments in which they are grown and then use only the more stable genotypes for the final stages of testing. They then set their task in the article as finding the criteria necessary to rank varieties by stability. Ultimately they define a stable variety as one with an average response to environment, or what Finlay and Wilkinson called average stability, and a minimum of deviations from regression.
Based on this interpretation of the philosophical history of stability analysis, the hypothesis to be explored here is that the cumulative effect of four factors has led to at least a quarter century of rejecting genetic material which would have demonstrated superior yield capability in the best or the poorest farm environments and has not
necessarily led to varieties or hybrids with superior yield capability in average environments.
The four factors are, again, statistical dependence on ANOVA that leads to the concern with reducing genotype by environment interaction, which in turn leads to
the nearly universal practice of evaluating material on experiment stations and farms with real or artificially created superior environments, in order to control this interaction or to permit the material to manifest its 'yield potential'.
Third, the capability of many farmers in the developed world, over the last few decades, to use their resources to modify unfavorable environments; and
fourth, the widespread use of a regression coefficient of one as a measure of stability.
If progress from selection has been slow, it was the result of these factors, not the presence of genotype-by-environment interaction which Eberhart and Russell considered to have a negative impact.
If the present argument is convincing, then it could be concluded that genotype-byenvironment interaction should be viewed as positive.
Furthermore, the philosophy toward the use of stability analysis, and the reasons to use on-farm testing, should change if we are to effectively and efficiently move toward a more sustainable agriculture.
Analysis of variance is an indispensable tool to evaluate replicated research conducted at only one location. It can be used by a skilled statistician to evaluate multilocation research, but it is easy to make errors (in part depending upon whether the locations are considered fixed or random) if the analyst is not skilled. Furthermore, in the field, but seldom published, it is not uncommon to see an analysis showing significant location-bytreatment interaction which masks mean treatment differences (the concern of Yates and Cochran). The researcher concludes it is not possible to detect treatment differences and recommends either repeating the experiment or controlling the sources of interaction in a new experiment.
This latter recommendation leads many researchers to try to minimize differences among locations by stratification; by environmental control techniques such as irrigation, fertilization and rigorous pest control; by excluding low-yielding environments, and/or by rejecting on-farm testing in favor of the experiment station. As a consequence, the results apply only to farms and farmers with similar, usually superior environments. The only reason hybrids (or other material) bred and tested under these conditions have had any successful adoption is that many farmers in the developed world, and a select few in developing countries, have had the resources and capability to duplicate favorable environments similar to those created for purposes of the testing program.
Because testing is carried out only under a very narrow range of environments, extrapolation to poorer environments (frequently those found on the majority of farms)
may well convey erroneous conclusions. This is not to argue that experiment stations and farms with superior environments should not be used in a testing program.
To the contrary, when used in conjunction with average and poor environments, data from superior environments can improve the regression estimates.
Informal observation indicates that if the range of environments in one year approximates the range to be expected in the same region over a period of years, and/r the magnitude of the range is at least as large as the overall mean yield, then the regression estimates stabilize with only one year of testing.
Adherence to the concept that a regression coefficient of unity is favorable has a separate set of implications. This figure shows hypothetical results of three varietie that have been tested over an appropriately wide range of environments. In this hypothetical case, all three have the same overall mean yield and s 2di = 0. The regression coefficients are 1.5, 1.0 and 0.5 for varieties A, B and C, respectively. In the absence of other disqualifying characteristics, variety B (the most generally adaptable according to Finlay and Wilkinson, or the most stable according to Eberhart and Russell) would be selected based on the value of the regression coefficient. The argument against variety A is that because it has a coefficient much higher than unity, it is too sensitive to environmental change and does poorly in poor environments. Variety C, because it has a coefficient much lower than unity, is unable to exploit high-yielding environments. Therefore, variety B, which is not superior in any environment, is chosen as the best of the three.
Notice that the argument against variety A with a high coefficient, moves from right to left or toward low environments (it does poorly in poor environments). The opposite is true of the argument against variety C with a low coefficient, which moves from left to
right or toward high environments (it is unable to exploit good environments). These are negative interpretations which lead to the rejection of these varieties for -all environments and the selection of variety B for all environments even though it excels in none.
If the emphasis regarding varieties with a high regression coefficient were toward, rather than away from the best environments (which variety can exploit the best environments?), variety A would be selected. Likewise, if for varieties with a low coefficient, emphasis were toward (rather than away from) the poorer environments (which variety can maintain yield even in poor environments?), variety C would be selected, as in this figure.
'Me difference is not one of analytical procedure, but of philosophy, goal and/or attitude toward selection. The impact on a breeding program could be dramatic.
The farming systems research-extension approach to technology development
was a response in the early 1970s to a worldwide concern with the plight of small-scale, resource poor farmers, particularly in the third world, who were not benefitting from the international public and private investment in agricultural research, extension and technology development efforts.
The methods developed to reach these clients revolved around a multidisciplinary approach based on the philosophy not that these farmers resisted new technology and change, rather that the technology being offered was not appropriate to their conditions. Early thoughts that social scientists (anthropologists, sociologists and economists) were
necessary to help "sell" modern technology to these farmers (that is, "convince" them to use it) were soon changed.
It became apparent that the production environments on their farms were as much a product of the socioeconomic factors facing these families as of the biophysical conditions associated with the often unfavorable locations of their farms. This meant that technology had to be tailored to the socioeconomic (date of planting, pest control practices, fertilizer use, soil preparation) as well as the biophysical (soils, climate, altitude) factors affecting the environments of these farmers. It also became apparent that technology testing had to be carried out under the wide range of environments represented and that recommendations varied depending on the nature of the environment.
That is, "scale neutral" does not apply just to field size. It also applies to the whole package of resources available to the farmer. An important characteristic of farming systems methodology is extensive use of on-farm evaluation under the biophysical and socioeconomic conditions faced by the farmers who are expected to adopt the varieties or hybrids and practices being developed. As early in the technology development process as feasible, the new material or practices are evaluated under the "good" and the "bad" that farmers are able to provide. This reduces the possibility of eliminating technology that would be superior in poor as well as in good environments.
How early in the development process new genetic material is placed in on-farm evaluation depends on the breeders who must supply it to the persons who conduct the on-farm testing. In the most successful programs the field testing teams are considered as essential elements of the breeding program and the breeders participate actively in the on-farm activities. Without these field teams, it would be difficult for the breeders to have the opportunity for widespread evaluation of their material under the conditions of
the small farmers who are the principal clients of the institute. Responses in the field help orient the direction of the breeding program. In the least Successful rograms, field testing teams have no official connection with the breeding program.-4he have access only to material that has already been released by the breeding program. There is little or no feedback of the results of the field testing program to the breeding program, and there is a low rate of adoption of the tested materials which seldom excel in the conditions of the farmers. Even worse, there is a continual loss of genetic material which might have excelled under the farmers' conditions.
The number and range of environments under which on-farm tests are conducted depend on the testing budget (personnel, transportation and materials); the philosophy of the breeding and/or testing program leaders; and crop production conditions. The range and distribution of environments is probably more important than number so long as some minimum number (probably around 10) is achieved.
As mentioned above, the magnitude of the range of environments divided by the overall mean yield is a useful measure as a basis for evaluating adequacy. A quick search resulted in locating the following ratios:
A potato study on agronomic practices from 20 farms by the International Potato Center in the Mantaro Valley of Peru: 1.42/1.
A maize variety study conducted by the farming systems program in Paraguay on 24 farms: 1.21/1.
A maize variety and fertilizer use study conducted on 14 farms by the farming systems team in Malawi: 1.57/1.
These compare with a ratio of 0.55/1 in the Yates and Cochran study when years are combined (as the authors did) or 0.83/1 when years and locations are separate,
and ratios of 0.39/1 up to 0.78/1 in a study on peanut production in stress and non-stress environments over four years reported by Knauft and Gorbet. Other cited articles do not show data. However, Eberhart and Russell's recommendation of minimizing genotype-by-environment interaction and then selecting the highest yielding materials would have the tendency to reduce this ratio by either reducing the environmental range used for testing or increasing the overall mean of the materials tested, or both.
The conclusions of this paper are based on the premises that:
1) in the developing world, there is an urgent need for a diversity of genetic
materials, as well as other agricultural technologies, that excel in the poorest as
well as the best farm environments; and
2) in the developed as well as the developing world, the concern for achieving a
more sustainable agriculture, also creates a need for some agricultural
technologies that excel in inferior environments and not just those which function
well only in favorable environments.
These needs can be satisfied most efficiently by
1) enhancing and taking advantage of genotype-by-environment interaction, rather
than discouraging, depressing or rejecting it;
2) subjecting new materials as early in the development process as possible to
poor as well as good farm conditions; and
3) searching for those materials which have the capability to maintain productivity
in poor environments or which excel in superior environments (low and high regression coefficients, respectively) rather than suppressing them in favor of
materials with a regression coefficient of unity.