A Deep Learning Approach to Mapping Irrigation: IrrMapper-U-Net
Accurate maps of irrigation are essential for understanding and managing water resources. We present a new method of mapping irrigation and demonstrate its accuracy for the state of Montana from years 2000-2019. The method is based off of an ensemble of convolutional neural networks that use reflectance information from Landsat imagery to classify irrigated pixels, that we call IrrMapper-U-Net. The methodology does not rely on extensive feature engineering and does not condition the classification with land use information from existing geospatial datasets. The ensemble does not need exhaustive hyperparameter tuning and the analysis pipeline is lightweight enough to be implemented on a personal computer. Furthermore, the proposed methodology provides an estimate of the uncertainty associated with classification. We evaluated our methodology and the resulting irrigation maps using a highly accurate novel spatially-explicit ground truth data set, using county-scale USDA surveys of irrigation extent, and using cadastral surveys. We found that that our method outperforms other methods of mapping irrigation in Montana in terms of overall accuracy and precision. We found that our method agrees better statewide with the USDA National Agricultural Statistics Survey estimates of irrigated area compared to other methods, and has far fewer errors of commission in rainfed agriculture areas. The method learns to mask clouds and ignore Landsat 7 scan-line failures without supervision, reducing the need for preprocessing data. This methodology has the potential to be applied across the entire United States and for the complete Landsat record.