Abstract

Irrigated agriculture is the largest consumer of freshwater globally. Effective water management is crucial to support ongoing agricultural intensification to meet increasing demand for food, fuel, and fiber production. Knowledge of where and when irrigation occurs is critical for effective management and hydrological modeling, yet data on patterns of irrigation through time are surprisingly rare. Existing regional datasets in the United States tend to be either aspatial county-level estimates or static, single-year remotely sensed products with relatively low spatial resolution (~250 m or coarser). Spatially explicit, dynamic maps are needed to understand water use trends, create accurate hydrological models, and inform forecasts of future water availability under projected climate change. In the High Plains Aquifer (HPA), repeat mapping efforts in 2002 and 2007 indicated only 60% of irrigated lands were static between these periods. To better understand annual irrigation dynamics, we used remote sensing to produce annual maps of irrigated cropland across the HPA region from a data fusion of Landsat satellites, annual time series of vegetation indices, and ancillary data such as precipitation, soil properties, and terrain slope. We performed machine learning classification using Google Earth Engine, allowing efficient image processing over a large region for multiple years. We then analyzed maps for water use trends and found that although total irrigated area has increased only slightly, there was substantial variability in the spatial pattern of irrigated lands over time. This dataset will support efforts towards groundwater sustainability by providing consistent, spatially explicit tracking of irrigation dynamics over time.