Researchers at Colorado State University have developed deep learning models to improve rainfall prediction using radar data. These models outperform traditional methods by accurately estimating precipitation rates with high spatial and temporal resolution, crucial for severe weather forecasts and water resource management.
Accurate rainfall estimation is essential for decision-making in weather forecasting and water management. Conventional radar-based methods have limitations in representing complex precipitation dynamics, leading to errors in rainfall predictions. Deep learning offers a promising solution by leveraging radar observations to enhance precipitation estimation accuracy.
This study introduces deep learning models designed to utilize polarimetric radar observations for quantitative precipitation estimation (QPE). These models are trained using radar data and surface gauge measurements, achieving superior performance compared to traditional radar-rainfall relations. Dense blocks-based models demonstrate the best performance, capturing spatial correlations and improving rainfall estimation accuracy. Of the models studied, RQPENet D1 had the best performance for QPE with a mean absolute error (MAE) of 1.58 mm, a root mean squared error (RMSE) of 2.68 mm, a normalized standard error (NSE) of 26%, and a correlation of 0.92 for hourly rainfall estimates.
Available for Exclusive Licensing
TRL: 5
Copyright
Haonan Chen
Aly Hoeher
Aly.Hoeher@colostate.edu
970-491-7100