Research on the Optimal Thresholds for Crop Start and End of Season Retrieval from Remotely Sensed Time-Series Data Based on Ground Observations

Abstract

Crop phenology is of great importance to crop growth monitoring, crop classification and yield prediction. Obtaining accurate phenological information is still difficult due to the complexity of cropping systems. A variety of methods have been utilized in retrieving crop phenology from remotely sensed time-series vegetation index data, among which the so-called threshold method is the mostly used. However, currently the thresholds are set empirically and a validation of the obtained phenological information is often lacking. This paper attempted to investigate the optimal thresholds for retrieving the Start of Season (SOS) and the End of Season (EOS) of different crops from MODIS EVI time-series data by using the dynamic threshold method. The crop growth and development dataset from National Meteorological Information Center of China were used to select analysis samples. The recorded green-up dates or three-leaf dates were used as reference SOS, and the recorded maturity dates were used as reference EOS. Four indicators, Bias, Biassign, Root Mean Square Error(RMSE) and Correlation Coefficient(R) were used to assess the accuracy of the retrieved phenology. Results show that: (1) the often used 20% threshold is not optimum in retrieving crop SOS and EOS. (2) Optimum thresholds are 24% and 48% for retrieving SOS and EOS, respectively. (3) Accuracy assessment results indicate that it is better to set different thresholds in retrieving crop SOS and EOS.