Where do you source the data?
To create a consistent product, we need to combine multiple data sources because each source has some of these limitations, or multiple limitations:
- 99% of raw data sources contain only historical data (no forward-looking view).
- oher datasets began in the 1980s and stopped 5, 10 or 20 years ago: never updated.
- many sources cover only one country (typically when it’s a national service).
- the majority of open data sources cover just one hazard, for example, only storms, rainfall, or just the wildfire, and nothing else.
- most sources have a coarse time step, for example, certain satellite data are available only once per week, so we combine complementary observations from several satellites.
- almost all official public flood maps aren’t truly digital, they’re just PNG images, not geolocated data in the proper sense.
The primary data sources for the historical period include ERA5-land reanalysis and observations.
ERA5-land weather reanalysis provides reconstructed hourly fields of all weather parameters. It assimilates and integrates weather and satellite observations. This global dataset is at 0.1-degree resolution (~10km) and it is updated several times per month with new weather observations.
For the forward-looking analysis we use three dynamic climate models :
- CNRM-CM6-1-HR
- MPI-ESM1-2-LR, and
- EC-Earth3
These are numerical High Resolution global climate models from the most recent CMIP6 run [Seferian et al. 2019].
The CMIP6 abbreviation stands for the Coupled Model Intercomparison Project Phase 6.
A multi-model approach offers more objective and more stable and reliable results while reducing uncertainty.
The raw CMIP6 climate model data can be downloaded from several nodes from the Earth System Grid Federation:
- https://esgf-index1.ceda.ac.uk/
- https://esgf-node.ipsl.upmc.fr/
Many alternative raw data sources are not used as ingridients for our risk scores, but appear to be useful for the validation of our products, and also for the bias correction. Logically, we can validate our products only against independent data sources.
High resolution Risk Assessment provided by Weather Trade Net data is a reprocessed in-house commercial product.
We don't re-distribute hourly temperatures or wind speed data.
Our Risk Assessment data is quality checked, bias corrected and continuously updated.
Quantile mapping method is used for the bias correction: individually for each calendar month, for each weather parameter and for each geolocation. The reference period for the data bias correction is 1991-2020. Same bias correction is applied to the historical and forward-looking climate model data.
Multi-year and season-specific tendencies are preserved. Trends are unique for each geolocation, each calendar month and each weather parameter.
WTN's blog is another great source of information about the data collection and post-processing.