Phenology-Based Classification Of Crop Species And Rotation Types Using Fused Modis And Landsat Data: The Comparison Of A Random-Forest-Based Model And A Decision-Rule-Based Model

The spatial and temporal variations of crop species and rotation types play a key role in crop yield estimation, natural resources management and climate change research. Remote sensing is an effective approach for monitoring agricultural management at the regional scale. However, the lack of remote sensing data with high spatiotemporal resolutions results in a generally limited crop mapping accuracy. To overcome this limitation, we employed an improved flexible spatiotemporal data fusion (IFSDAF) model to conduct data fusion using MODIS and Landsat imagery and extract NDVI time series with both high spatial and temporal resolution. Following this, we proposed a Random-Forest-based model and a decision-rule-based model for mapping crop species and rotation types. According to the accuracy assessment, the Random-Forest-based model with an overall accuracy of 90 % was more suitable for mapping crops with two or more growing seasons, such as double-cropping rice. Meanwhile, the decision-rule-based model with an overall accuracy of 89.7 % was more suitable for monitoring crops with only one growing season compared to the Random-Forest-based model. In comparison with traditional ways such as field survey, the classification models proposed in this study provide useful information for better monitoring spatiotemporal variations of crops species and rotation types, and guiding agricultural management accordingly.