long_term_precipitation_forecast

Long-term precipitation forecast

The long-term precipitation forecast downstream task consists of forecasting the global distribution of daily accumulated precipitation from satellite observations up to 4 weeks into the future. The input data for this task is derived from gridded satellite observations from geostationary and polar-orbiting platforms and aggregated into daily files. The output data are corresponding daily precipitation accumulations derived from gauge-calibrated satellite-based precipitation data records.

Refer to the OVERVIEW.md for results from the baseline model.

Creating training and test datasets

Installing dependencies

Most depencies for the data extraction can be installed using conda.

conda create --name precip --file conda-requirements.txt --channel conda-forge
conda activate precip

Three packages, however, are still under active development and must be installed from github using pip. The commit hashed unique identify the development snapshots with which the data can be extracted.

pip install git+ssh://[email protected]/see-geo/pansat@55c7b991ab4f19223664ad896caa9cb575221aee
pip install  git+ssh://[email protected]/simonpf/chimp@e1701da4d7d1a1db5b3701852cd92c48aea3b872

Setting-up the environment

Downloading IMERG data requires a NASA. After creating the account, the username and password must be added to the pansat environment:

NOTE: pansat stores the password in an encrypted file on the machine. Despite the file being encrypted, it is recommended to use a throwaway password

pansat account add ges_disc <user_name>

Downloading the data

The following scripts downloaded the data from the four sources:

Name	Data
extract_precip_data.sh	Daily precipitation
extract_gridsat_data.sh	Geostationary observations
extract_ssmi_data.sh	PMW observations from polar orbiting satellites
extract_patmosx.sh	VIS/IR observations from polar orbiting satellites

Evaluation and forecast baselines

The aim of the long-term precipitation forecast task is to improve subseasonal-to-seasonal (S2S) precipitation forecasts. The principal baselines for the long-term precipitation forecast task are derived two state-of-the-art, conventional numerical weather prediction models (NWP). In addition to this, a machine-learning-based baseline model is provided with this repository to prove that the feasibility of the proposed task.

The conventional NWP forecasts are derived from the ECMWF S2S database (Vitart, 2017) and are part of the task's test data. The code implementing the machine-learning baseline model is located in the baseline folder.

References

Vitart, F., and Coauthors, 2017: The Subseasonal to Seasonal (S2S) Prediction Project Database. Bull. Amer. Meteor. Soc., 98, 163–173, https://doi.org/10.1175/BAMS-D-16-0017.1.