Submitted to: Crop Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/8/2007
Publication Date: 7/1/2008
Citation: Williams, C.L., Liebman, M., Edwards, J.W., James, D.E., Singer, J.W., Arritt, R.W., Herzmann, D. 2008. Predicting spatial variation of crop yield across a landscape using aggregated environmental data. Crop Science. 48:1545-1559. Interpretive Summary: Geographic Information Systems (GIS) provide powerful new tools for deriving geographic, soil, and climatic data that can be used for modeling agricultural productivity over wide areas in order to improve policy and research decision making. This study was undertaken to determine if available information could be used to predict county-average crop yields for four major crop species in Iowa. Using county averages over a 20 year period for corn, soybean, alfalfa, and oat, between 65 and 79 percent of the variability in county-average crop yields was predicted using a multiple regression model using GIS-derived predictors. Between 51 and 80 percent of the inter-annual variability in county average crop yields was predicted. These results demonstrate that crop yields can be predicted across large geographic regions using coarse data on soil, climate, and geographical factors. The ability to make such predictions will benefit agricultural researchers and policy makers by providing methods for predicting changes in crop productivity with environmental changes and for predicting crop productivity for agricultural species in areas where they are not commonly grown. The ability to predict productivity of a crop where it is not grown will help researchers and policy makers identify new crops in particular regions in order to improve productivity and stability of agricultural production.
Technical Abstract: Crop yield variability is an important attribute of agroecological systems. For decision-makers to make informed choices, it is necessary to understand spatial distribution of yield variability across landscapes. Spatial patterns of yield variability are associated with underlying environmental variability. Application of locally-derived process-based crop models to very large areas has had limited success. Thus, landscape-scale spatial analysis is emerging as an important direction in agroecological research. Use of spatially-referenced environmental data in a geographic information system (GIS), combined with regression techniques, provide particularly powerful tools in such analyses. The aim of this study was to predict spatial distribution of crop yields and their interannual variability among counties in Iowa, USA, using a limited set of environmental predictors derived through GIS in mixed step-wise regression models. Predictors included measures of variability of climatic, edaphic and topographic parameters. Significant (p < 0.001) models of mean yield were obtained for corn (Zea mays L.; R2 = 0.74), soybean [Glycine max (L.) Merr.]; R2 = 0.65), alfalfa (Medicago sativa L.); R2 = 0.75), and oat (Avena sativa L.; R2 = 0.79). Significant (p < 0.001) models were also obtained for interannual yield variability of corn (R2 = 0.80), soybean (R2 = 0.65), alfalfa (R2 = 0.51), and oat (R2 = 0.62). The approach may be used to predict suitability of currently grown crops under expected future environmental conditions, and for prediction of alternative crop yields under current and expected future environmental conditions.