Evaluation of AI-models for convective parameters
Benchmark datasets:
- Weatherbench (ERA5 and TIGGE for 2020) -> limited levels
https://weatherbench2.readthedocs.io/en/latest/data-guide.html
- Observations -> soundings from University of Wyoming
Convective parameters:
- Instability -> CAPE
- Shear -> surface and 500 hPa
wrf-python
Target:
Severe environments as a 0-1 binary -> fractional skill score (FSS)
Values of instability and shear -> RMSE, BIAS, SAL-score
Investigation of bias -> role of moisture
Evaluation of convective season NH&SH -> North America, Europe, Australia, Argentina
Year 2020
Notes:
No surface humidity, but pressure levels up to 1000 hPa;
FCN-V2 outside of 0-100% RH bounds by ~20%
FCN-V2 issues with humidity overall