Early prediction of corn yields from daily weather data and single predetermined seasonal constants

Abstract

Since a large portion of year-to-year fluctuation in crop yields is due to weather variations, there have been many efforts to model crop—environment relationships and develop operational yield-prediction systems. However, this effort has been complicated by the fact that the end result (yield) is the summation of many diverse daily responses of plants to environmental factors. Further, most plant responses are not linear, therefore effects of extreme conditions are lost if daily weather variables are averaged over periods of days or weeks and then correlated with plant growth and/or development. The objective of this study was to quantify daily growth responses of maize ( Zea mays L.), determine generalized growth—environment relationships, and develop a yield-prediction system adapted to daily input of commonly published weather variables for large areas of production. Daily growth rates obtained in Illinois, Iowa and Minnesota (1978–1979), from measurements of leaf extension, were regressed on selected environmental variables to obtain a growth-prediction equation applicable to average temperature and rainfall data for areas such as crop reporting divisions. Daily values of the growth-prediction equation applied to recorded weather data from Iowa (by crop reporting divisions) for the period 1950–1973 were summed each year for 20, 40 and 60 days from the date 50% of the crop was planted (beginning date), to obtain growth indices for subsequent use in the analysis of growth—yield relationships. Recorded yields 1950–1973 (USDA) were regressed on: growth index, latitude, beginning date, running 15-day sum of maximum temperature, running 15-day sum of percent estimated soil moisture, sum of percent estimated soil moisture from beginning date to prediction, fertilizer applied (nitrogen); and various transformations of these seven basic variables. The prediction equation (specific for Iowa) obtained from this regression (along with the growth-prediction equation) applied to independent data for 1974-177, resulted in yield predictions a few days to a month earlier, and equal or better than the first CRB-SRS-USDA forecasts each year. To extend the system to other states the same procedure is applied for development of appropriate coefficients.