Mapping interannual variability of maize cover in a large irrigation district using a vegetation index – phenological index classifier

Abstract

Accurate mapping interannual variability of crop cover is a pre-request for modern agricultural management, while most published algorithms require re-calibration when crop cover is mapped over multiple years, and hence greatly hinder their applicability. In addition, these algorithms are often not applicable for areas with complex planting patterns. Here we propose a vegetation index – phenological index (VI-PI) classifier to map interannual variability of crop cover (using maize, which is one of the major crops in the study area as a demonstration case) in the Hetao Irrigation District of North China from 2003 to 2012 using the MODIS data at 250m spatial resolution. Representative MODIS Normalized Difference Vegetation Index (NDVI) time series of maize is obtained during a field survey in late August, 2012, which is fitted with an asymmetric logistic curve to obtain the phenological indices. The maize classifier (an ellipse on the VI-PI space) is shaped based on the in situ data and adjusted by the official statistics in 2010–2012. The performance of the developed classifier is then tested with the official data from 2003 to 2009. Results show that the asymmetric logistic curve performs excellent in describing the NDVI time series of maize, and the estimated distribution of maize agrees reasonably well with the independent official data. The relative errors are lower than 7% in the training years, and lower than 30% during the testing years which is considered acceptable for crop mapping in an area with complex planting patterns. And the kappa coefficient was as high as 0.86. These results indicate that the proposed VI-PI classifier can be used effectively for crop mapping over multiple planting years and in areas with a complex planting structure.