Search

Excessive farming has transformed large proportions of Xinjiang's terrestrial surface, which leads to widespread loss and degradation of ecosystems and biodiversity. Mapping croplands at regional scale and in a rapid and less costly approach is becoming increasingly needed. This study explored an integrated approach based on a combined use of multiple remotely sensed data to map croplands in the oasis of northwest China. The Terra MODIS land surface temperature, Enhanced vegetation index, TRMM precipitation and SRTM digital elevation model data products were examined and combined. The Random Forest models were established to estimate fractional croplands at the regional scale. The influences of basin, images and percentage of croplands were all examined on the accuracy of cropland estimation. Moreover, time series analysis of croplands from 2000 to 2017 was conducted across the basin scale. The results are: ① the integrated approach for cropland mapping at regional scale is promising with the explained variances of fifteen trial models ranging from 58%–93% and two basin scale models ranging from 77.21%–86.10%. ② The prediction accuracy of croplands in Tarim Basin was much higher than that in Junggar Basin indicated by explained variances. ③ Prediction errors increase and then decrease with the increase of percentage of croplands. ④ EVI and DEM were identified to be the most important variable in Junggar and Tarim basin. ⑤ During the past 18 years, the area of croplands has been expanding at approximately 641.3km2/year in Junggar Basin and 271.3 km2/year in Tarim Basin respectively with no sign of cease. The fastest expansion period and the most dramatic changing area were identified. This study is especially valuable for time series analysis of croplands and its correlation with land degradation in arid region.