COMMITTEE CHAIR: Dr. Ali Fares
CO-COMMITTEE CHAIR: Dr. Ripendra Awal

TITLE: ASSESSING THE FUTURE IMPACTS OF DROUGHT ON COTTON YIELD IN TEXAS USING MACHINE LEARNING AND CLIMATE INDICES

ABSTRACT: Climate change and increasing drought frequency pose significant challenges to agricultural sustainability in Texas. It particularly impacts cotton production, which is highly sensitive to temperature variability and water availability. Despite numerous studies evaluating climate impacts on agriculture, limited research integrates climate projections to predict cotton yield. There is also few research on drought indices, machine learning models, and crop simulation models to assess future cotton yield vulnerability across Texas climate divisions. Therefore, this study evaluates historical climate trends, compares predictive modeling approaches, and projects future cotton yield under climate change scenarios. Historical climate variables including maximum temperature (Tmax), minimum temperature (Tmin), precipitation (PR), potential evapotranspiration (PET), soil water storage (STOR), and drought indices (SPI-3, SPI-6, SPEI-3, and SPEI-6) were analyzed with cotton yield data from 1968 – 2024. Multiple Linear Regression (MLR) and Random Forest (RF) models were developed and evaluated using correlation coefficient (R) and Root Mean Square Error (RMSE). Future climate projections from CMIP6 LOCA datasets under SSP245 and SSP585 scenarios were used to project cotton yield for 2030–2050. Additionally, the FAO AquaCrop model was used to simulate crop response to water availability and validate machine learning projections. Results indicate that temperature and PET were the most influential predictors of cotton yield across Texas climate divisions. RF models showed improved predictive performance with R values ranging from approximately 0.62 to 0.89, compared to MLR values ranging from 0.48 to 0.81. AquaCrop validation showed low to moderate correlation with observed yields (R ˜ -0.05 to 0.48). Future projections indicate yield increases of approximately 9–21% in North Central and High Plains divisions, while declines of approximately 6–11% are projected in South Central and Upper Coast regions. Irrigated cotton showed greater resilience compared to non-irrigated systems. Overall, this study demonstrates that integrating machine learning and crop simulation models improves climate impact assessment and supports climate-resilient agricultural management strategies in Texas.

Keywords: Cotton yield, machine learning, drought indices (Standardized Precipitation Index (SPI), Standardized Precipitation Evapotranspiration Index (SPEI)), AquaCrop model, Random Forest (RF), Texas agriculture

Room Location: Jesse and Mary Gibbs Jones CAFNR Research Building, Seminar Room 015