Journal of Geographical Sciences ›› 2011, Vol. 21 ›› Issue (4): 705-718.doi: 10.1007/s11442-011-0874-1

• Applications of GIS • Previous Articles     Next Articles

Integrating geographical data and phenological characteristics derived from MODIS data for improving land cover mapping

CAI Hongyan1,2, ZHANG Shuwen1, BU Kun1, YANG Jiuchun1, CHANG Liping1   

  1. 1. Northeast Institute of Geography and Agroecology, CAS, Changchun 130012, China;
    2. Graduate University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2011-01-18 Revised:2011-02-27 Online:2011-08-15 Published:2011-08-05
  • Supported by:

    The National 973 Program, No.2010CB950901-2-1; The program of Ministry of Science and Technology, No.SB2007FY110300-1-2


The study developed a feasible method for large-area land cover mapping with combination of geographical data and phenological characteristics, taking Northeast China (NEC) as the study area. First, with the monthly average of precipitation and temperature datasets, the spatial clustering method was used to divide the NEC into four ecoclimate regions. For each ecoclimate region, geographical variables (annual mean precipitation and temperature, elevation, slope and aspect) were combined with phenological variables derived from the moderate resolution imaging spectroradiometer (MODIS) data (enhanced vegetation index (EVI) and land surface water index (LSWI)), which were taken as input variables of land cover classification. Decision Tree (DT) classifiers were then performed to produce land cover maps for each region. Finally, four resultant land cover maps were mosaicked for the entire NEC (NEC_MODIS), and the land use and land cover data of NEC (NEC_LULC) interpreted from Landsat-TM images was used to evaluate the NEC_MODIS and MODIS land cover product (MODIS_IGBP) in terms of areal and spatial agreement. The results showed that the phenological information derived from EVI and LSWI time series well discriminated land cover classes in NEC, and the overall accuracy was significantly improved by 5.29% with addition of geographical variables. Compared with NEC_LULC for seven aggregation classes, the area errors of NEC_MODIS were much smaller and more stable than that of MODIS_IGBP for most of classes, and the wall-to-wall spatial comparisons at pixel level indicated that NEC_MODIS agreed with NEC_LULC for 71.26% of the NEC, whereas only 62.16% for MODIS_IGBP. The good performance of NEC_MODIS demonstrates that the methodology developed in the study has great potential for timely and detailed land cover mapping in temperate and boreal regions.

Key words: geographical data, vegetation phenology, MODIS, land cover, Northeast China