Correlation and regression approaches to modeling rainfall-crop yield relationships in a tropical environment

Abstract

Simple correlation and multiple regression of total rainfall and actual water availability on maize yield for 3 different phenological stages of maize (Zea mays L.) during 1970-98 growing seasons were done. Both rainfall and actual water
availability were each significantly correlated with yield, but rainfall at every stage showed a comparatively lower
coefficient of determination of yield than actual water availability, However, the simple correlation model appeared to overestimate the contribution due to rainfall, consequently the sum of the contribution due to both rainfall and actual
water availability (values ranging from 124-173% during the 3 phenological stages) looked unrealistic. However the
multiple regression model produced a more realistic result. It showed that the combination of rainfall and actual water
availability during each phenological stage accounted for less than 100% variation in yield and this implies that some other
variables other than rainfall and A W A might also be important in explaining the variability in yield. Generally the need to be cautious in using rainfall data for yield prediction has been stressed while the advantage of a carefully selected relevant independent agroclimatological indices and adoption of holistic rather than parametric approach to modeling climate-crop yield has been highlighted.