Paddy Fields Extraction and Spatiotemporal Evolution Analysis Based on Remote Sensing Cloud Platform-A case study of Hubei Province

Paddy fields are the material basis for human survival and development, especially in China. With the acceleration of urbanization, cultivated land is transforming and decreasing continuously. Therefore, it is of great significance to monitor the changes of paddy fields area and analyze their evolution of spatial-temporal distribution for food supply, land resources management and sustainable development. Taking Hubei Province as the study area, Google Earth Engine (GEE) was utilized to extract the paddy fields and identify their types between 1990 and 2020, based on Landsat series remote sensing images and Random Forest algorithm in this study. Firstly, preliminary cropland samples in 2020 were generated from the union of three public Land Use and Land Cover (LULC) datasets, then combining with the phenological information, different kinds of paddy fields reference samples in 2020 were extracted from the union. Secondly, the samples migration method was applied to obtain paddy fields samples in other target years, by comparing Spectral Angle Distance and Euclidean Distance. Then, six spectral bands, two normalized difference indices and the time series of monthly mean Normalized Difference Vegetation Index (NDVI) were used to classify single-cropping and double-cropping paddy fields. Meanwhile, the accuracy was evaluated based on the confusion matrix, overall accuracy and Kappa coefficient. Finally, area changes and spatio-temporal distribution of paddy fields were analyzed to discover the evolution characteristics and trend. The results showed that:(1) From 1990 to 2020, the overall area of paddy fields maintained stable, and the area of single-cropping was five times greater than double-cropping, and continued to increase. (2) Almost 90% of paddy fields were distributed in the plain area of central Hubei Province, while only 0.5% were distributed in the mountainous area of western Hubei Province. (3) The method of union of public LULC datasets and samples migration method used in this study can effectively generate reference sample datasets for past target years to execute land use classification in long-term time series.