Integrating remote sensing, irrigation suitability and statistical data for irrigated cropland mapping over mainland China

Knowledge of irrigation location and extent is essential for irrigation-water use estimation and water resource management. However, it remains a great challenge to map irrigated areas at large spatial scales due to the great variation in climate, geography, and agricultural practices, as well as the lack of sufficient ground truth data. This study proposed a novel approach to develop the first 250-m irrigated cropland map in mainland China (CIrrMap250) by integrating remote sensing, irrigation suitability, and irrigated area statistics. We assessed the performance of CIrrMap250 and compared it with three irrigation maps (i.e., EVI-map, NDVI-map, GI-map) generated using the threshold-based classification method and four other existing maps, including GMIA2005 (Siebert et al., 2005), GIAM2000 (Thenkabail et al., 2009), Zhu-map (Zhu et al., 2014), and Meier-map (Meier et al., 2018). Results indicate that CIrrMap250 and all other maps capture well the intensively irrigated areas such as the North China Plain and Northwest China, as well as many large-scale irrigation districts. However, all maps except CIrrMap250 tend to underestimate irrigated cropland in river valleys while overestimating irrigated cropland in the mountainous areas, as illustrated by the field-surveyed irrigation districts, due to the neglect of the mixed grid effects. Compared to other irrigation maps, CIrrMap250 exhibits a better agreement with the reference points, achieving improvements in Kappa coefficient and overall accuracy by 8% up to about 2 times. The irrigated area estimates of CIrrMap250 are very close to the statistical data due to their usage in generating the training pool. Further analysis indicates CIrrMap250 has a greater proportion of irrigated cropland at lower elevations, on smaller slopes, and near water bodies than the other maps. There is large uncertainty in irrigation ratio estimates due to the varying cropland area from multiple sources. This study demonstrates the effectiveness of the new irrigation mapping method and highlights the great potential of combining irrigation suitability with remote sensing and statistical data to improve the accuracy of large-scale irrigated cropland mapping.