A comprehensive assessment of the correlations between field crop yields and commonly used MODIS products

Abstract

An exploratory assessment was undertaken to determine the correlation strength and optimal timing of several commonly used Moderate Resolution Imaging Spectroradiometer (MODIS) composited imagery products against crop yields for 10 globally significant agricultural commodities. The crops analyzed included barley, canola, corn, cotton, potatoes, rice, sorghum, soybeans, sugarbeets, and wheat. The MODIS data investigated included the Normalized Difference Vegetation Index (NDVI), Fraction of Photosynthetically Active Radiation (FPAR), Leaf Area Index (LAI), and Gross Primary Production (GPP), in addition to daytime Land Surface Temperature (DLST) and nighttime LST (NLST). The imagery utilized all had 8-day time intervals, but NDVI had a 250m spatial resolution while the other products were 1000m. These MODIS datasets were also assessed from both the Terra and Aqua satellites, with their differing overpass times, to document any differences. A follow-on analysis, using the Terra 250m NDVI data as a benchmark, looked at the yield prediction utility of NDVI at two spatial scales (250m vs. 1000m), two time precisions (8-day vs. 16-day), and also assessed the Enhanced Vegetation Index (EVI, at 250m, 16-day). The analyses spanned the major farming areas of the United States (US) from the summers of 2008–2013 and used annual county-level average crop yield data from the US Department of Agriculture as a basis. All crops, except rice, showed at least some positive correlations to each of the vegetation related indices in the middle of the growing season, with NDVI performing slightly better than FPAR. LAI was somewhat less strongly correlated and GPP weak overall. Conversely, some of the crops, particularly canola, corn, and soybeans, also showed negative correlations to DLST mid-summer. NLST, however, was never correlated to crop yield, regardless of the crop or seasonal timing. Differences between the Terra and Aqua results were found to be minimal. The 1000m resolution NDVI showed somewhat poorer performance than the 250m and suggests spatial resolution is helpful but not a necessity. The 8-day versus 16-day NDVI relationships to yields were very similar other than for the temporal precision. Finally, the EVI often showed the very best performance of all the variables, all things considered.