. A data assimilation framework to estimate irrigation: merging soil moisture retrievals with land surface models

Abstract
Water withdrawals for irrigated agriculture represent the single largest consumptive use for much of the world, bearing a large footprint on the water and energy cycles. Despite efforts to document these withdrawals, practical challenges exist in quantifying irrigation magnitude over large areas of the Earth surface. Here we describe a framework that uses data assimilation to estimate irrigation magnitude, designed to be used with soil moisture retrievals from NASA's SMAP satellite. The results from a synthetic experiment are presented that evaluate the accuracy of irrigation estimates over an agricultural region of the central USA. Variations of the experiment are conducted to quantify the impact of key uncertainties, using land surface model outputs in the place of remote sensing. The presented method performs with high skill in idealized scenarios but is highly sensitive to the amount of irrigation signal captured by soil moisture retrievals and model versus retrieval algorithm parameterizations. The seasonal percent bias and correlation between estimated and truth irrigation in idealized scenarios (e.g. no observational noise and optimal irrigation timing) were 0.66% and 0.95, respectively; scenarios with observational noise representing SMAP's unbiased RMSE and overpass timing resulted in seasonal percent biases and correlations ranging from -5.7-11.3% and 0.64-0.88, respectively. The findings from this evaluation will guide a continental-scale application towards an irrigation magnitude product, useful for model development as well as water management efforts.