Published results of sensitivity experiments on agricultural models by international authors are analyzed with a simple univariate sensitivity index. Values for system parameter and response‐effects at ratios greater than 2.0 or 3.0 are presented for different crop plants, ecosystem components and processes. Responses are ranked according to their level of sensitivity to direct and indirect changes that conceivably could be imposed by manufactured chemicals. Of several hundred agricultural modeling publications, only 8.5% was found to contain results from sensitivity analysis experiments performed by their original authors. This low proportion means that environmental interactions and biological species other than those for which such results are published might be of equal or greater sensitivity and importance if comprehensive sensitivity analyses were available across all crops and ecological processes.
Results imply that hydrological flows, soil nutrient losses, and pesticide losses might all be quite vulnerable to disruption if indirect effects could be transferred through ecosystems by changes in weather and climate. Below‐ground primary production processes in agroecosystems such as cotton, semiarid range pastureland, perennial ryegrass and sugar yield of sugar beets might be vulnerable to potential, direct, sublethal effects of chemical substances. If climatic processes were changed, indirect effects could be expected in cotton and pasture root growth, and potato yields. Results also indicate that vegetative growth of corn, alfalfa, soybean and perennial ryegrass might all be extremely vulnerable to any indirect effects from chemically induced climatic changes. Reproductive parts that could be especially vulnerable to possible potential direct effects from chemical substances include seed yield of annual semiarid pastures, fruit quality of hardwood trees and, specifically, apple fruit production. If any indirect effects are possible, it appears that grain sorghum and grain corn yields, and apple fruit yields might be very vulnerable. All these results are limited to the particular designs of the models used and the response variables on which data are available from simulations.
