[A-CG37] Satellite Earth Environment Observation

[ACG37-P20] Performance Assessment of Irrigation Projects in Nepal

*Adarsha Neupane¹, Yohei Sawada¹ (1.University of Tokyo)

Keywords:Landsat, random forest classification, classification, conjunctive use

An accurate and efficient method to monitor irrigation projects is important, especially in developing world, where the performance of irrigation is often suboptimal. In Nepal, one of the developing countries, the irrigated area has not been objectively recorded, although the assessment of the irrigated area has substantial implications on national policy, projects’ annual budgets and donor funding. Here we present the application of Landsat images to measure irrigated areas in Nepal for past 20 years toward the robust assessment of the irrigation performance. In this study, Landsat 7 (from 2003 to 2012) and Landsat 8 (from 2013 to 2022) images were used to develop a machine learning model which classifies irrigated and non-irrigated areas in the Sunsari Morang Irrigation Project, Nepal. The overall accuracy of our proposed random forest-based classification was 88.6% and the producer’s and consumer’s accuracy for the agriculture class was 86.8% and 81.5%, respectively. The classified image shows that a large area is irrigated in the upstream portion of the canals while the irrigation performance is low in the tail-end portion of the canals. In addition, we find that the tail-end portion of the command area significantly increased the irrigated area in 2017-2018 and 2021-2022 compared to 2013-2014 and 2015-2016, which implies that the use of groundwater for irrigation has flourished in recent years based on the country’s new policy of the conjunctive use of groundwater along with surface irrigation. It should be noted that an existing global cropland map cannot accurately reproduce the small-scale distribution of irrigated land in this region, which reveals the advantage of regionally trained machine learning models.