PCSDiff: Diffusion-Based Bias Correction and Super Resolution Toward Practical Operational Medium-Term Precipitation Forecast
Abstract
Medium-range precipitation forecasts are impaired by persistent systematic biases, lead-time-dependent error accumulation, and coarse spatial resolution, restricting their reliability for flood-drought risk assessment. Existing AI correction techniques lack dedicated modeling for multi-day dynamic bias evolution and proper meteorological constraints, often generating over-smoothed rainfall structures, and cannot meet operational deployment demands. This work introduces PCSDiff, a cascaded task-decoupled diffusion framework targeting 10-day precipitation bias correction and downscaling. To jointly counteract temporal error drifts and reconstruct physically plausible local precipitation details, PCSDiff integrates the Precipitation Intensity-aware Multi-branch Decoder (PIMD) module for dynamic multi-day error mitigation using synoptic-temporal features, followed by a two-phase conditional diffusion super-resolution module to restore fine-scale precipitation patterns. Evaluated against CMA-CRA observations over China after global-data training, PCSDiff cuts RMSE by 16.1% and lifts ACC by 13.9% relative to raw ECMWF forecasts at 3-10-day lead times, and consistently outperforms mainstream deep-learning baselines on both general and extreme-precipitation metrics. Benefiting from a streaming inference pipeline, our method achieves low-latency rolling forecasting for practical meteorological operations.
Related
- SwAIther-Precip: Lead-Time-Aware Bias Correction Enables Kilometer-Scale Downscaling of Global AI Precipitation Forecasts over Switzerland
- Spatio-temporal precipitation downscaling at hourly resolution based on km-scale regional simulations
- Diffusion-Based Refinement for Kilometer-Scale Probabilistic Precipitation Nowcasting
- Skillful and realistic precipitation forecasting with reinforced diffusion
- High-resolution Calibrated Probabilistic Hourly Precipitation from a Deterministic Forecast
BibTeX
@misc{sun2026pcsdiff,
title = {{PCSDiff: Diffusion-Based Bias Correction and Super Resolution Toward Practical Operational Medium-Term Precipitation Forecast}},
author = {Yuze Sun and Shiyi Wang and Jiancheng Pan and Die Wang and Andreas F. Prein and Wentao Luo and Linhan Jiang and Jie Wu and Quan Zhang and Xiaomeng Huang},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.06942},
eprint = {2609.06942},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.06942},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - PCSDiff: Diffusion-Based Bias Correction and Super Resolution Toward Practical Operational Medium-Term Precipitation Forecast
AU - Yuze Sun
AU - Shiyi Wang
AU - Jiancheng Pan
AU - Die Wang
AU - Andreas F. Prein
AU - Wentao Luo
AU - Linhan Jiang
AU - Jie Wu
AU - Quan Zhang
AU - Xiaomeng Huang
PY - 2026
DA - 2026/09/07
PB - arXiv
DO - 10.48550/arxiv.2609.06942
UR - https://arxiv.org/abs/2609.06942
AB - Medium-range precipitation forecasts are impaired by persistent systematic biases, lead-time-dependent error accumulation, and coarse spatial resolution, restricting their reliability for flood-drought risk assessment. Existing AI correction techniques lack dedicated modeling for multi-day dynamic bias evolution and proper meteorological constraints, often generating over-smoothed rainfall structures, and cannot meet operational deployment demands. This work introduces PCSDiff, a cascaded task-decoupled diffusion framework targeting 10-day precipitation bias correction and downscaling. To jointly counteract temporal error drifts and reconstruct physically plausible local precipitation details, PCSDiff integrates the Precipitation Intensity-aware Multi-branch Decoder (PIMD) module for dynamic multi-day error mitigation using synoptic-temporal features, followed by a two-phase conditional diffusion super-resolution module to restore fine-scale precipitation patterns. Evaluated against CMA-CRA observations over China after global-data training, PCSDiff cuts RMSE by 16.1% and lifts ACC by 13.9% relative to raw ECMWF forecasts at 3-10-day lead times, and consistently outperforms mainstream deep-learning baselines on both general and extreme-precipitation metrics. Benefiting from a streaming inference pipeline, our method achieves low-latency rolling forecasting for practical meteorological operations.
ER -