Predicting yield of major crops at a state scale through NIRv

Abstract:

In the agricultural industry, it has been substantial and challengeable task to develop a reliable way for predicting crop yield of extensive fields prior to harvest. Although existing methods utilizing surface reflectance derived from satellites could readily quantify crop yield, it is difficult to apply them at global scale and they largely remain uncertain and dependent on crop types and agrarian practices. Here, we developed a novel approach, employing near‐infrared reflectance of vegetation (NIRv), which could accurately and consistently anticipate yield of major crops at state scale ahead of harvests, in comparison with normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI). We calculated daily NIRv, NDVI and EVI of primary countries croplands at 1 km spatial resolution and filled data gaps of each pixel of DOY using temporal and spatial interpolation. We extracted representative values for each crop and state through a method based on maximum and sum of values within growing seasons of the year, and compared them with crops yield data obtained from the Food and Agricultural Organization Statistical Database (FAOSAT). Our approach displayed that the NIRv values were highly and consistently correlated with crop yield over crop types and countries, while those of NDVI and EVI remarkably varied. It is observed that disparities between NIRv and the other vegetation indices appeared particularly in heterogeneous croplands and double cropping regions. The proposed approach is also capable of forecasting the crop yield more easily in a timely manner than extant methods based on meteorological data and processing models. This study suggests that NIRv has a potential for robust estimation of crop yield across assorted crops and countries.