GPTCast: a weather language model for precipitation nowcasting

TL;DR

GPTCast is a novel deep-learning model inspired by large language models, designed for ensemble precipitation nowcasting using radar data, providing realistic probabilistic forecasts with uncertainty estimation.

Contribution

This paper introduces GPTCast, a new approach that applies GPT-based models to spatiotemporal radar data for precipitation forecasting, with a specialized tokenizer and no reliance on randomness during training.

Findings

Outperforms state-of-the-art ensemble extrapolation methods

Provides realistic probabilistic precipitation forecasts

Accurately estimates uncertainty in predictions

Abstract

This work introduces GPTCast, a generative deep-learning method for ensemble nowcast of radar-based precipitation, inspired by advancements in large language models (LLMs). We employ a GPT model as a forecaster to learn spatiotemporal precipitation dynamics using tokenized radar images. The tokenizer is based on a Quantized Variational Autoencoder featuring a novel reconstruction loss tailored for the skewed distribution of precipitation that promotes faithful reconstruction of high rainfall rates. The approach produces realistic ensemble forecasts and provides probabilistic outputs with accurate uncertainty estimation. The model is trained without resorting to randomness, all variability is learned solely from the data and exposed by model at inference for ensemble generation. We train and test GPTCast using a 6-year radar dataset over the Emilia-Romagna region in Northern Italy, showing superior results compared to state-of-the-art ensemble extrapolation methods.