Crop Yield Forecast Models that Maximize Explanatory Power of Climate Based Inputs from Satellite Observations

The most essential factors influencing variability in crop yields at a particular location are inter-annual variations of moisture and temperature. Therefore these two primary factors are often used to predict the yields for corn, soybean and wheat in the U.S.A. Another important factor is the trend increase due to the improvements in seed stock and agricultural practices. Yield predictions based on the trend and weather factors have proven valuable. Nonetheless, additional predictors exist for the yield predictions, including the greenness index (measured by NDVI) for monitoring the plant growth situation. Unfortunately many of these predictors are correlated, which weakens the accuracy and stability of regression-based models. Moreover, the spatial resolution of the historical yield data and the satellite observations of the weather, soil, and plant growth conditions can have spatial autocorrelations. To coupe with the collinearity in both time and space, we have chosen to use Canonical Correlation Analysis (CCA) to load the satellite-based predictors and to make predictions in the spectral space. At the meeting, we will present the procedures and results, and demonstrate the value of this work and its applications.