Date of Graduation
7-2026
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Geosciences (PhD)
Degree Level
Graduate
Department
Geosciences
Advisor/Mentor
Cheng, Linyin
Committee Member
Peter, Brad
Second Committee Member
Feng, Song
Third Committee Member
He, Yaqian
Keywords
deep learning applications, weather, extreme events forecasting, Standardized Temperature Index (STI), artificial intelligence
Abstract
Weather forecasting and extreme events prediction poses a significant challenge in modern day computing. Accuracy in weather forecasts is dependent on how well the terrestrial, oceanic and atmospheric processes are represented by a physical model. The Numerical Weather Prediction (NWP) models which are the primary tools in weather forecasting have limitations in providing accurate and long-range weather forecasts due to imperfect representations of large-scale earth system processes. The NWP models require extensive computational resources to capture complex atmospheric mechanisms on smaller grids. The finer the grid resolution, the greater the computational resources and time required to generate forecasts. NWP models use parametrization schemes which approximate the fine scale physical processes while missing important details and therefore the resulting forecasts are not accurate. Recently due to advancements of Artificial Intelligence (AI) in every field of science, there is a growing popularity of using AI techniques in climate and weather forecasting. This research focuses on data driven deep learning techniques for weather and climate extremes forecasting. It develops on a concept of utilizing land-atmosphere variables and trains them using deep learning approaches which can predict the future state of atmosphere. The idea is to simulate the atmosphere like physical model with less computational cost while keeping the prediction accuracy. The deep learning methodology is applied to daily summer and winter temperature forecasts with lead times of up to 30 days for the prediction of temperature extremes (hot and cold events) using the Standardized Temperature Index (STI), as well as compound climate extremes using the Standardized Dry and Hot Index (SDHI). A particular emphasis is given on the potential for exploring sub-seasonal to seasonal weather forecasts. This study employs a purely data-driven deep learning approach, offering a faster and more cost-effective method that can be further extended to operational weather forecasting.
Citation
Ali, J. (2026). Deep Learning Applications in Weather and Extreme Events Forecasting. Graduate Theses and Dissertations Retrieved from https://scholarworks.uark.edu/etd/6409
