Mapping Crop Cycles in China Using MODIS-EVI Time Series

Wednesday, 17 December 2014
Le Li1,2, Mark A Friedl3, Qinchuan Xin4, Josh M Gray5, Yaozhong Pan1 and Steve E Frolking6, (1)Beijing Normal University, Beijing, China, (2)IB Institute of Botany, Chinese Academy of Sciences, State Key Laboratory of Vegetation and Environmental Change, Beijing, China, (3)Boston University, Boston, MA, United States, (4)Tsinghua University, Ministry of Education Key Laboratory for Earth System Modeling, Beijing, China, (5)Boston University, Earth and Environment, Boston, MA, United States, (6)Univ New Hampshire, Durham, NH, United States
As the Earth’s population continues to grow and demand for food increases, multiple cropping is an effective way to increase crop production. Cropping intensity, which we define here as the number of cropping cycles per year, is an important dimension of land use that is strongly influences water demand and agricultural production. Satellite data provide global land cover maps with indispensable information regarding areal extent of global croplands and its distribution. However, the land use information such as cropping intensity is not routinely provided by global land cover products from instruments such as MODIS, because mapping this information from remote sensing is challenging. We present a straight forward but efficient algorithm for automated mapping of agricultural intensity over large geographic regions using 8-day MODIS EVI time series data derived from Terra and Aqua MODIS surface reflectance products. The proposed algorithm first applies an adaptive Savitzky-Golay filter to smooth Enhanced Vegetation Index (EVI) time series derived from MODIS surface reflectance data, and then uses an iterative moving-window methodology to identify cropping cycles from the smoothed EVI time series. Comparison of results from our algorithm with national survey data at both the provincial and prefectural level in China show that the algorithm provides estimates of gross sown area that agree well with inventory data. Accuracy assessment using data generated by expert classification of randomly selected pixel samples indicates an overall accuracy of 91.0% for three agricultural intensity classes. More generally, the algorithm shows significant potential to automatically estimate reliable cropping intensity information in support of large-scale studies of agricultural land use and land cover dynamics.