A New Method to Map Croplands in Pakistan based on Dynamic Time Warping and Density-based Spatial Clustering of Applications with Noise using Landsat Time Series on GEE Platform

Abstract:

National-scale cropland mapping is an important approach for ensuring food security, particularly in developing countries like Pakistan. Agriculture is one of the most important industries in Pakistan, and within its territory, regional differences in climatic conditions led to a wide variety of crops with complex growth patterns. For the reason that mapping of croplands in Pakistan at the national scale is a challenge. This study proposed an automated cropland extraction method based on dynamic time warping (DTW) and density-based spatial clustering of applications with noise (DBSCAN) (ACE-DTW). The combined DTW algorithm and DBSCAN clustering were capable of automatically generating typical time series for land cover types in a complete and accurate manner. Frist, the proposed method selected 422 frames of multispectral remote sensing images on Google Earth Engine and constructed a normalized difference vegetation index (NDVI) time series. Next, 2409 training samples were obtained through visual interpretation. Then, typical time series of land cover types in Pakistan were automatically generated, pixels whose DTW distance from the cropland was smaller than that from other land cover types were classified as cropland pixels, thereby cropland mapping was finished. Research results revealed that the overall accuracy of ACE-DTW was 91.0%, which exceeded existing cropland maps and results derived supervised classifiers. In addition, the accuracies achieved by ACE-DTW in the northern mountain region and southeastern plains of Pakistan were higher than existing cropland maps, which demonstrates the advantages of ACE-DTW for obtaining detailed descriptions of croplands in diverse geological regions. ACE-DTW can automatically generate typical time series of various cropland types in large-scale, without omitting crops cultivated in smaller areas. Therefore, it can produce relatively complete descriptions of crop spatial differentiation, intra-class variation, and crop rotation patterns, and shows promise for improving the completeness and accuracy of cropland maps.