Using Machine Learning and Distributed Hydrologic Modeling to Predict Soil Texture, Surface Soil Moisture and Evapotranspiration in Jornada Experimental Range, Southwestern U.S.
In water-limited ecosystems, detailed knowledge of the soil, vegetation, and atmosphere interactions is critical to understand the processes that control the partitioning of energy, water fluxes, and biogeochemical cycles within the critical zone. This Master’s thesis is divided into two main contributing sections. The first, is on the use of machine learning to reconstruct missing soil type information, and the second, on the calibration and validation of a physically-based distributed hydrological model to estimate soil moisture and evapotranspiration within the Jornada Experimental Range of the U.S. in southern New Mexico. For the first contribution, three explainable, shallow machine-learning techniques are used to predict an observed gap in the SSURGO’s soil textural types within the study watershed with accuracy beyond 95% for the best-performing method (Gradient Boosting). For the second contribution, the information re-constructed about soils is used within the TIN-based Real-time Integrated Basin Simulator (tRIBS). The model ingests terrain, soils, atmospheric forcing, vegetation cover, and activity. A small-size watershed (69 km2 ) was delineated and selected as the study zone encompassing two eddy covariance stations with soil moisture probes, a sap flow network, and a uniformly distributed summer sampling network (5x5) of soil water content profiles, using a 30 m resolution DEM to derive the Triangular Irregular Network that the model uses to run. tRIBS is used to simulate three variables (1) evaporation, (2) transpiration, and (3) surface soil water content at the watershed scale during the North American Monsoon (NAM) season of 2019. tRIBS simulations obtain Kling-Gupta Efficiency Coefficients above 0.4 for soil moisture and evapotranspiration on a daily scale. The calibrated model could be used to understand the effects of the terrain spatial heterogeneity on the distribution of such processes, including the effect of vegetation activity on the partitioning of evapotranspiration and the components of the surface energy budget.
Ecology|Hydrologic sciences|Environmental science
Mayo-Rios, Jorge Andres, "Using Machine Learning and Distributed Hydrologic Modeling to Predict Soil Texture, Surface Soil Moisture and Evapotranspiration in Jornada Experimental Range, Southwestern U.S." (2023). ETD Collection for University of Texas, El Paso. AAI30635218.