Dissertation

Prediction and Forecasting of Evapotranspiration and Groundwater Anomalies, Along With Improved Parameterization of ET in Agricultural Lands

Talib, Ammara

2023

Overview

Predicting and forecasting evapotranspiration (ET) and groundwater (GW) variations are essential for sustainable water use in agriculturally intensive areas. Despite its importance in linking energy cycles and water, ET is challenging to measure. Further, to accurately estimate ET and GW dynamics, input uncertainty and deficiencies in hydrologic models pose fundamental challenges. Moreover, for land surface model-based ET, process models GW reanalysis, and remote sensing products, performance varies with the spatiotemporal scale due to the complex nonlinear relationships among meteorological and biophysical predictors of ET and GW dynamics in managed landscapes. Because of complicated boundary conditions, heterogeneous hydrogeological characteristics, groundwater extraction, and nonlinear interactions between these factors, it is proved difficult to predict and forecast ET and GW anomalies over the long term in agricultural areas. Nevertheless, data-driven methods and deep learning have shown promising results when identifying variables' dependencies.Here, this dissertation addresses this gap by 1) evaluating sources of bias in the regional Wisconsin Irrigation and Scheduling Program (WISP) models and developing a correction based on eddy covariance (EC) observations. 2) developing and evaluating the performance of data-driven models such as random forests (RF) and long short-term memory (LSTM) to predict and forecast daily ET on diverse agricultural farms in the Midwest, USA. and 3) utilizing recurrent neural network of LSTM as a method for forecasting GW anomalies two months in advance and for analysis of drivers that affect GW dynamics.To accomplish the first objective of this dissertation, ET, observations were made for five years in agricultural fields in the Wisconsin Central Sands (WCS) region, one of the most productive agricultural regions of the country, using EC systems. WISP model ET bias was traced to the underestimation of net longwave radiation (LWnet) owing to a biased specification of effective clear sky atmospheric emissivity ( , ). Correcting the , reduced the WISP model's bias and error for both LWnet and ET. The second objective was to apply the ET modeling framework developed for Midwest for nineteen fields where eddy covariance ET and meteorological observations are available during the growing season (April-October). In terms of daily ET prediction, a 16 parameter random forest approach outperformed a process-based land surface model. In irrigated crops, vapor pressure and crop coefficients were the most important predictors, while in non-irrigated crops, short wave radiation and enhanced vegetation index were the most important predictors. Finally, for the third objective, a modeling approach to forecast GW anomalies was developed and evaluated in the WCS region in the U.S. Groundwater anomalies showed high spatiotemporal variability, and their responses differed across locations depending on boundary conditions, catchment geology, climate, and topography. Land use change and irrigation pumping have interactive effects on GW anomalies forecasting. By understanding the critical processes underlying hydrologic and climatic variability and change over land, these findings may enable improved and more accurate hydrologic and climatic simulations. Using our framework, we can model water cycle components and dynamics in areas with unknown or uncertain subsurface properties.

• Publisher: ProQuest Dissertations & Theses
• Subject: Agriculture / Environmental engineering / Meteorology / Water Resources Management
• ISBN: 9798380106368