Incorporating phenology into yield models

Abstract:

Because the yields of many crops are sensitive to meteorological forcing during specific growth stages, phenological information has potential utility in yield mapping and forecasting exercises. However, most attempts to explain the spatiotemporal variability in crop yields with weather data have relied on growth stage definitions that do not change from year-to-year, even though planting, maturity, and harvesting dates show significant interannual variability. We tested the hypothesis that quantifying temperature exposures over dynamically determined growth stages would better explain observed spatiotemporal variability in crop yields than statically defined time periods. Specifically, we used National Agricultural and Statistics Service (NASS) crop progress data to identify the timing of the start of the maize reproductive growth stage ("silking"), and examined the correlation between county-scale yield anomalies and temperature exposures during either the annual or long-term average silking period. Consistent with our hypothesis and physical understanding, yield anomalies were more correlated with temperature exposures during the actual, rather than the long-term average, silking period. Nevertheless, temperature exposures alone explained a relatively low proportion of the yield variability, indicating that other factors and/or time periods are also important. We next investigated the potential of using remotely sensed land surface phenology instead of NASS progress data to retrieve crop growth stages, but encountered challenges related to crop type mapping and subpixel crop heterogeneity. Here, we discuss the potential of overcoming these challenges and the general utility of remotely sensed land surface phenology in crop yield mapping.