Global seasonal forecast skill data for heat-impact related indicators using the German Climate Forecast System (GCFS) v2.1

Description of Forecast Skill Dataset

The skill estimates support a climate risk assessment exercise conducted under the funding of the FPCUP Action U-CLIMADAPT (link). The open-source and open-access risk assessment tool CLIMADA (link) and the elaborated use cases for which the skill data serve as input can be found via climada_petals (link) using the Copernicus_forecast module. The use cases showcase the application of seasonal forecast information provided by the Copernicus Climate Change Service for climate risk assessments. As the related skill data are evaluated based on past model data (hindcasts) and are rather demanding to calculate, they are included via this repository and not determined dynamically within the CLIMADA model.

 This dataset provides forecast skill estimates for climate model predictions based on the German Climate Forecast System (GCFS v2.1) developed by Deutscher Wetterdienst (DWD) Fröhlich et al. (2020). The data focuses specifically on heat-impact-related indicators at a global scale for the seasonal prediction time scale. It is designed to evaluate the quality (or "skill") of seasonal climate predictions by comparing past model forecasts (hindcasts) with actual observed climate data from ERA5 reanalysis.

The hindcast data cover the period from 1990 to 2019, enabling a robust long-term evaluation of seasonal forecast skills. The dataset provides a global spatial resolution of 1° x 1° (approximately 110 km by 110 km), offering sufficient detail for regional-scale analysis.

The dataset includes two primary datasets:

Maximum daily temperature (tasmax): This is a key variable for assessing heat-related risks such as heatwaves and their potential impacts on human health, agriculture, and infrastructure. Tasmax serves as a direct indicator of extreme heat conditions over specific regions.

Labour productivity (lpr): Derived from tasmax, this variable offers valuable insights into how heat exposure affects labour productivity. The derivation is based on empirical modelling from Dasgupta et al. (2021), highlighting the combined effects of climate-induced heat stress on workforce efficiency and supply.

Key Features of the Dataset

The dataset is comprehensive and supports climate risk assessment with the following features:

Skill Metrics:

tasmax_fc_mse (Forecast Model Error): This metric quantifies the forecast error of the model based on the Mean Squared Error (MSE). It serves as an indicator of how much the forecast deviates from observed values.

tasmax_ref_mse (Reference Model Error): This metric provides the error of a reference to a climatological baseline forecast model, offering a benchmark for comparison.

tasmax_msess (Mean Squared Error Skill Score - MSESS): This is a normalised score that evaluates the relative performance of the forecast compared to the reference. Higher values indicate better performance, with values greater than 0 suggesting skill beyond the reference.

tasmax_msessSig (Statistical Significance of MSESS): This metric assesses the statistical significance of the MSESS values. It helps determine whether the skill score is meaningful or a result of random chance, ensuring reliable interpretation.

Time Frame and Spatial Coverage:

The dataset includes hindcast data aggregated over a forecast period of six months, with starting points for each of the 12 months of the year (e.g., shc01 corresponds to January, shc05 to May, etc.).

The data is grid-based with global spatial coverage, providing resolution across longitude (384 grid points) and latitude (192 grid points).

Data Aggregation Over Six Months

The forecast skill data is aggregated over a six-month prediction period, meaning that for each forecast starting month (e.g., January or May), the skill metrics represent the model's performance in predicting the subsequent six months. For example:

A forecast initiated in May (shc05) evaluates model performance for predictions covering May to October.

This six-month aggregation provides insight into the model's reliability for extended seasonal predictions, helping to assess skill over a critical window for decision-making.

Structure of the Dataset:

Temporal Dimension: Each dataset file corresponds to a single starting month and aggregates forecast skill metrics over the six-month prediction horizon.

Spatial Resolution: Longitude and latitude values define the grid, covering the globe at a sufficient resolution for regional analysis.

Variables: The dataset includes multiple variables (tasmax_fc_mse, tasmax_ref_mse, tasmax_msess, and tasmax_msessSig) for assessing forecast skill across different skill metrics.

References

Dasgupta et al. (2021): Explores the combined effects of heat on labour productivity. Read more.

Fröhlich et al. (2020): Details on the German Climate Forecast System (GCFS). Read more.

Licensing and Acknowledgments

The dataset is provided under the terms of use by DWD and the FPCUP Action U-CLIMADAPT. For more information, visit:

DWD Climate Forecast Information

FPCUP U-CLIMADAPT

CLIMADA SEASONAL FORESCAST MODUL

Important Notes

Skill estimates represent past performance and should be used cautiously for decision-making.

Always consider the statistical significance (tasmax_msessSig) before interpreting tasmax_msess.