Enhancing MODIS land cover product with a spatial–temporal modeling algorithm

MODIS Collection 5 land cover product (MCD12Q1) provides annually updated global land cover maps since 2001. This time series product has become an essential data source for the generation of many other land surface products and for a variety of regional and global studies. However, classification errors are inherent in the land cover product, which can misrepresent land cover transitions. In particular, land cover transitions are illogical if they contradict ecological rules and are unlikely to be observed. In this study, we evaluated the MODIS land cover product by analyzing the nature and magnitude of its illogical land cover transitions using annual MCD12Q1 land cover maps from 2001 to 2010. Our analysis revealed that illogical transitions exist in the product for all consecutive years, and are distributed most commonly in several regions over the world. To enhance the MODIS land cover product, we applied a spatial–temporal modeling algorithm that incorporates expert knowledge to reduce illogical transitions on five such “hotspot” tiles. The results showed substantial improvements in both accuracy and consistency of the land cover product using the spatial–temporal modeling algorithm. The percentage of illogical transitions in each of the five tiles was significantly reduced among consecutive years and across the entire time series. This study demonstrates the effectiveness of the spatial–temporal modeling algorithm for producing high quality time series of land cover maps, and also highlights the importance of temporal consistency in land cover mapping.