Date of Award

6-8-2026

Document Type

Thesis - SCU Access Only

Publisher

Santa Clara : Santa Clara University, 2026

First Advisor

David Anastasiu

Abstract

Rainfall prediction is important for local decision-making because precipitation can a!ect transportation and public safety. However, broad regional weather models may not always capture local precipitation patterns accurately enough for users in specific areas such as Santa Clara or the San Francisco Bay Area. This project developed a machine learning pipeline that uses historical forecast data and high-resolution observational data to improve precipitation prediction. The system also integrates the trained model with a web application that presents forecast outputs through an interactive, user-friendly map interface. The project was designed as a local correction system that uses existing weather forecasts as a baseline and refines them using observed precipitation data. The completed work included data preprocessing, model development, and front-end visualization. Experimental evaluation showed that the machine learning correction framework improved precipitation prediction relative to baseline High-Resolution Rapid Refresh(HRRR) forecasts at most forecast hours. The convolutional neural network (CNN)-based approach reduced the overall precipitation error while improving precipitation occurrence detection and spatial precipitation realism. To make these forecasts accessible to end users, the project also included a web-based visualization platform. The web application organizes prepared forecast outputs through a back-end service and displays them in a format that is accessible to nontechnical users. Internal testing evaluated data handling, API behavior, back-end reliability, interface clarity, and missing-data handling. In general, the project demonstrated that a local-first prediction and visualization system can make rainfall information more relevant and accessible for users who need location-specific precision.

Download

Share

COinS