Assessing Machine Learning Utility in Predicting Hydrologic and Nitrate Dynamics in Karst Agroecosystems
Author ORCID Identifier
Year of Publication
Master of Arts (MA)
Biosystems and Agricultural Engineering
Dr. William Ford
Seasonal hypoxia in the Gulf of Mexico and harmful algal blooms experienced in many inland freshwater bodies is partially driven due to excessive nitrogen loading seen from agricultural watersheds. Within the Mississippi/Atchafalaya River Basin, many areas are underlain with karst features, and efforts to reduce nitrogen contributions from these areas have had varying success, due to lacking a complete understanding of nutrient dynamics in karst agricultural systems. To improve the understanding of nitrogen cycling in these systems, 35 months of high resolution in situ water quality and atmospheric data were collected and fed into a two-hidden layer extreme learning machine (TELM) to predict discharge and nitrate exports from a karst agroecosystem in the Inner Bluegrass region, to improve the understanding of nitrate dynamics in karst and determine the variables of influence driving nitrate loading in karst systems. Including atmospheric and soil moisture and temperature records to 100 cm in modeling resulted in the TELM providing accurate estimates of both nitrate concentration and flowrate (NSE=0.9328 and NSE=0.9363 respectively) and represented short term storm event hysteresis and diurnal signals in model predictions. The TELM also showed the variables most influential in training were the soil moisture and temperature parameters levels, pointing to the importance of focusing future work on understanding how temperature influences matrix-macropore interactions in the temperate karst environment. Finally, the ELM showed the fertilizer application data was not influential in model training, indicating that, at this study site, the fertilizer application has little control over nitrate loading. This should be studied further in other landscapes with higher rates of fertilizer application where alternative hysteretic patterns have been observed.
Digital Object Identifier (DOI)
McGill, Timothy, "Assessing Machine Learning Utility in Predicting Hydrologic and Nitrate Dynamics in Karst Agroecosystems" (2022). Theses and Dissertations--Biosystems and Agricultural Engineering. 90.
Bioresource and Agricultural Engineering Commons, Environmental Engineering Commons, Environmental Monitoring Commons, Hydrology Commons, Soil Science Commons, Water Resource Management Commons