Crop Identification by Using Seasonal Parameters Extracted from Time Series Landsat Images in a Mountainous Agricultural County of Eastern Qinghai Province, China

Time series vegetable indexes (Vis) have been evidenced a useful data to extract vegetable phenology and identify crop types. This paper conducted such a research in Qinghai Province by using Landsat TM images, via four steps, i) sampling single-crop plots and extracting crop spectrums based on pure signle-crop pixels; ii) building time-series vegetable indexes by using Landsat 8 TM images (2013-2014); iii) extracting seasonal parameters according to algorithms defined in TIMESAT program; vi) generating a decision tree for identifying crop types and validate classification accuracy via ground investigation. The results indicate that crops planted in a larger continuous range, such as spring wheat, potato and rapeseed, achieved an acceptable accuracy of above 70%, while crops planted too dispersedly (like broad bean, which is often inter-planted with other crops) or with a too smaller planting range (like barley), remained a poor recognition rates (below 50%). The value of this work lies in it displayed not only the classification accuracy of crop types in this region by using such methodology, but also the feasibility of integrating VIs calculation, seasonal parameter extracting and decision tree generation into one computer program, which is highly desired in this region.