MODELING IRRIGATED CROP PRODUCTION TO BACKFILL MISSING WATER DATA

While United States crop production data extend over 150 years, few records are categorized by irrigated or dryland conditions. This is a particularly notable data gap as irrigation covers about one third of total cropland and is beginning to extend beyond the semi-arid western states. Furthermore, most irrigation information is summarized at the state-level making small-scale land use patterns more difficult to identify. The purpose of this study was to parse out irrigated and dryland production data from recorded cumulative averages to develop a temporally complete dataset at the county-level for major row crops across the country. Specifically, this study uses machine learning through a boosted regression model to backfill missing data. As part of the boosted regression, we include several physical drivers to agricultural production including climatic and underlying geologic conditions. Initial results contain dryland and irrigated production data from 1945-2018 for corn, cotton, hay, soybeans, and wheat, and methods can be replicated for other timespans and commodities. Results from this study can be used to better identify farmer decision-making, analyze agricultural responses to policy, and describe future water demands as farmers use irrigation to adapt to climate change, mitigate annual production risks, and boost annual profits.