Improving irrigation planning and early prediction for agricultural drought in Wisconsin

Conference: ‘American Geophysical Union, Fall Meeting 2020, H165-000’

Abstract: In our ability to sustain food production, it is important to know how much water our crops use. The risks of climate extremes such as heat waves and droughts are increasing and have already threatened the North American agricultural system in the form of increased drought duration, intensity, and reduced crop yield. Current drought prediction models cover large regions and are not specific to individual farms. The goal of this study is 1) Forecast field-scale short-term evapotranspiration (ET) for different crops in the U.S. Midwest and 2) improve a commonly used satellite-driven regional crop evapotranspiration model (WISP). The first goal is achieved by building a prediction and forecasting empirical model based on machine learning. We compare two approaches, neural networks with Long short-term memory (LSTM) and random forest (RF) applied to nineteen crop sites in a range of crop types, soils, and irrigation usage that have continuous ET measurements made by eddy covariance. In general, performance of RF exceeded LSTM in most cases and forecasts were found to be reliable out to three days. Unlike rainfed crops, we found irrigated crops ET prediction requires including additional information related to physical properties of sites (soil types, crop coefficient, cumulative growing degree days). For the second goal, we evaluated ET predictions from WISP against some of these towers. Overestimation of ET in WISP arises from faulty parameterization of longwave radiation clear-sky emissivity. We applied an inverse approach to estimate corrected clear sky emissivity parameters. ET calculated based on corrected parameters reduced most of the model’s ET bias and lowered RMSE. New field-scale actual ET measurements provide critical insights to improve ET models and hence, enhance drought monitoring and forecasting necessary to be able to use limited water resources efficiently to maintain productivity.