OneScience-Group/RISE-UNet

<p align="center"><strong><span style="font-size: 30px;">RISE-UNet</span></strong></p>

Model Introduction

RISE-UNet combines deep learning and dynamical forecasts for subseasonal root-zone soil-moisture prediction. It recursively predicts five weekly anomalies and evaluates drought probabilities through ensembles.

Paper: Skillful subseasonal soil moisture drought forecasts with deep learning-dynamic models https://doi.org/10.1038/s41467-025-62761-3

Model Description

The model was proposed by researchers at Auburn University. It was trained with GLEAM root-zone soil moisture, ERA5 reanalysis, and GEFSv12 and ECMWF S2S reforecasts. By combining residual, inception, squeeze-and-excitation, and UNet++ operations with recursive predictions, it supports weekly root-zone soil-moisture and flash-drought forecasting over the contiguous United States, China, and Australia.

Use Cases

Usage Instructions

Download

hf download OneScience-Group/RISE-UNet --local-dir ./RISE-UNet
cd RISE-UNet

Environment Dependencies

Hardware Requirements

• —A GPU or DCU is recommended.
• —A CPU can run the default small-sample connectivity configuration.
• —DCU users should install DTK 25.04.2 or a compatible OneScience-recommended version first.

DCU Environment

# Activate DTK and Conda first
conda create -n onescience311 python=3.11 -y
conda activate onescience311
pip install onescience[earth-dcu] -i http://mirrors.onescience.ai:3141/pypi/simple/ --trusted-host mirrors.onescience.ai

GPU Environment

# Activate Conda first
conda create -n onescience311 python=3.11 -y libstdcxx-ng=12 libgcc-ng=12 gcc_linux-64=12 gxx_linux-64=12
conda activate onescience311
pip install onescience[earth-gpu] -i http://mirrors.onescience.ai:3141/pypi/simple/ --trusted-host mirrors.onescience.ai

Synthetic Data

The paper uses a 48x96 0.5-degree regional grid, 11 ensemble members, weekly historical and forecast variables, and GLEAM 0–100 cm root-zone soil-moisture anomalies as targets. Synthetic data preserve the grid, member count, recursive five-week protocol, and RISE operators while reducing initialization count, width, and epochs. Results verify the workflow only and do not represent paper performance.

python scripts/fake_data.py

Training

python scripts/train.py
torchrun --nproc_per_node=2 --nnodes=1 --master_addr="localhost" --master_port=29500 scripts/train.py

The default synthetic run completes five-week recursive optimization, deep-supervision losses, and the ensemble-spread constraint, and both single-process and two-process DDP training have been verified. It produces one recoverable checkpoint and records the CRPSexp training result. Training results are saved to:

result/checkpoints/rise_unet.pt
result/training/metrics.json

Weights

The paper's code is available at https://osf.io/6y4kh/, but an independently licensed official pretrained checkpoint was not confirmed.

Inference

python scripts/inference.py

Inference restores the checkpoint, retains stochastic dropout, and recursively generates weeks 1–5 for 11 members. The output shape is [11,5,48,96] and has passed finite-value checks. Inference results are saved to:

result/output/predictions.npz

Evaluation

python scripts/result.py

Evaluation computes weekly ACC, CRPS, and drought GSS and creates a week-3 spatial error figure. All metrics are finite, and the PNG has passed format and non-empty-pixel checks; synthetic results do not represent paper performance. Evaluation results are saved to:

result/evaluation/metrics.json
result/evaluation/comparison.png

Official OneScience Information

Citation and License

This repository is an independent engineering reproduction of the public RISE-UNet specifications, with code licensed under the Apache License 2.0.

The original paper is licensed under CC BY-NC-ND 4.0; the paper and GLEAM, ERA5, GEFSv12, and ECMWF S2S data remain subject to their respective licenses and terms.