County-level evaluation of large-scale gridded datasets of irrigated area over China

Authors: TIAN, X - Tianjin University; DONG, J - Tianjin University; CHEN, X - Tianjin University; ZHOU, J - Tsinghua University; GAO, M - Tianjin University; WEI, L - Nanjing University; KANG, X - Tianjin University; ZHAO, D - Tianjin University; ZHANG, H - Tianjin University; Crow, Wade

Submitted to: Journal of Geophysical Research Atmospheres
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/15/2024
Publication Date: 2/28/2024
Citation: Tian, X., Dong, J., Chen, X., Zhou, J., Gao, M., Wei, L., Kang, X., Zhao, D., Zhang, H., Crow, W.T. 2024. County-level evaluation of large-scale gridded datasets of irrigated area over China. Journal of Geophysical Research Atmospheres. Article e2024GL108326. https://doi.org/10.1029/2024GL108326.
DOI: https://doi.org/10.1029/2024GL108326

Interpretive Summary: Given the large, and growing, importance of irrigation agriculture for global food production, it is important to consider the role of irrigation in the large-scale tracking of agricultural drought and water-resource availability. At a minimum, such tracking requires accurate maps of irrigated areas. A number of remote sensing methods have recently been proposed for obtaining such maps at a global scale; however, little is known about the accuracy of such methods – since ground-based data sets required for validation are difficult to collect and process. This study describes the, extremely laborious, processing of a ground-based irrigation area dataset using county-level maps and the application of this new dataset to improve the evaluation of global irrigation area datasets based on remote sensing. Results highlight key limitations in these irrigated-area global datasets and identify new strategies for addressing these limitations. The dataset and evaluation results presented in this paper will eventually be applied to improve our ability to globally track and model irrigation within agricultural regions.

Technical Abstract: Accurately tracking irrigation occurrence and amount is crucial for global food security, regional hydrology, and the study of landatmosphere interactions. The accuracy of irrigation tracking models is largely determined by the quality of their ancillary irrigated area (IA) inputs, which can be mapped using either remote sensing or government censored data (GCD). However, the reliability of IA datasets remains unknown. Here, county-level GCD during 2000 to 2021 are collected, cross-validated and employed to evaluate commonly applied gridded IA datasets. Results show that IA datasets based on the direct interpolation of Food and Agriculture Organization (FAO) agricultural census are more robust than other datasets in capturing the spatial distribution of IA. However, FAO statistics are updated only every 5 to 10 years and can easily become outdated. Therefore, FAO-based IA datasets may lead to significant inter-annual irrigation errors when applied to long-term irrigation modeling. On the other hand, IA products entirely based on remote sensing (i.e., not constrained by GCD) tend to contain strong positive IA biases over humid regions. These biases are likely related to the lack of contrast between vegetation dynamics in irrigated and rainfed croplands over humid regions. Machine learning approaches, which train remote sensing data to capture IA, are relatively more accurate. However, these datasets still contain substantial temporal errors that may be related to the reliability of training IA data samples. Although GCD data is generally reliable, it cannot provide pixel-wise IA information and updating GCD information is laborious. Therefore, new algorithms that can combine multi-source information and effectively capture the spatiotemporal variability of IA are necessary for large-scale irrigation modeling.