NCAR Bias Calibration and Downscaling Statistical Analysis
Observation Datasets
ASOS (https://mesonet.agron.iastate.edu/request/download.phtml)
Mesonet (Index of /madisPublic1/data/archive/2023/08/29/LDAD/mesonet/ (noaa.gov))
Analysis
GFS Medium Range - 96 hour forecast
HRRR/RAP Short Range - 18 hour forecast
Timeline of Analysis - Start time: 2023-08-30 12:00:00, End time: 2023-09-04-01
- Performed analysis during Hurricane Idalia landfall in North Florida and movement into the Southeastern US to highlight a range of meteorological summertime conditions (dry western US, wet eastern US) that would give us a broad understanding of the impact of bias calibration and downscaling methods over a range of domain configurations (NoahMP gridded, NextGen hydrofabric, coastal unstructured mesh)
- Only linked observation data with nearby model grid cells or elements when the centroid of the model data is within 10 kilometers of the observation station location
- Mesonet data quality filters - Utilized data quality flags to only extract hourly observations when data passed through 3 out of 4 or more data quality filtering methods (MADIS-mesonet data and quality memo - Google Docs)
- ASOS data quality filters - Developed similar data quality filters analogous to Mesonet methods that essentially filter out signal-noise ratios using a Locally Weighted Least Squares (LOWESS) filter method over a 2-month period of ASOS data.
GFS Medium Range Operational Configuration
Results and Implications
- Bias correction alone may have a slight benefit for air temperature and wind vector fields across all domain configurations although there’s no strong correlation to currently conclude this for community users. Since bias correction is a function of the time of day and the day of the year, statistical improvements could very well vary across the seasons. Further research is likely needed for a clear-cut consensus on this general topic, but users may turn on this feature in the NextGen Forcings Engine for some potential benefit during the medium range operational configuration.
- Downscaling methods are very sensitive to “truth” estimates for the domain configuration (NextGen hydrofabric, coastal mesh) element/grid feature heights. If the domain element/grid feature heights are subject to large errors (i.e. NextGen hydrofabric catchment feature heights), then the downscaling methods will greatly impact and degrade the quality of the estimates for surface pressure, air temperature, specific humidity, and incoming short-wave radiation estimates. In our analysis with GFS downscaling here, the errors associated between catchment/element height estimates compared to surface observations were very large and attributed to the degradation of the downscaled meteorological forcings for coastal meshes and the NextGen hydrofabric. Statistical results were much stronger overall if a user neglects the downscaling option for unstructured meshes (hydrofabric, coastal mesh). We did not observe the same impact for the WRF-hydro gridded domain as much with better overall height estimates were observed. Users are still cautioned for downscaling methods to be applied to gridded domains contingent on the quality of the surface height estimates of each grid cell.
- Errors from the coastal mesh analysis are slightly higher due to the smaller sample set of points along the eastern US coastline, making those statistical results rather trivial and more comparable overall to the statistical correlation and root mean square errors of the NextGen hydrofabric and NWM gridded domain.
Scatter Plots
Histogram, Correlation, and Spatial Plots
F1: No_Bias_Correction_or_Downscaling
F2: With_Bias_Correction_and_Downsling
F3: With_Bias_Correction_Only
F4: With_Downscaling_Only
- NextGen Hydrofabric digital elevation map (DEM) in the background overlaid with ASOS/Mesonet station locations where differences in elevations are greater than +/- 100 meters. This highlights the vast errors in the NextGen Hydrofabric DEM compared to surface observations, which cascades into significantly larger errors in downscaling meteorological forcings within the CONUS regions highlighted with the blue dots in the figure below. Caution should be taken when utilizing the NCAR downscaling methods in the NextGen Forcings Engine, particularly for the NextGen hydrofabric for catchments within the regions shown below.
HRRR/RAP Short Range Operational Configuration
Results and Implications
- Statistical trends remain nearly identical for bias calibration and downscaling impacts on various domain configurations (NextGen hydrofabric, coastal unstructured mesh, NWM gridded domain) between the NWM Short Range operational configuration results shown below and the NWM Medium Range operational configuration results earlier in this document.
- Overall statistical correlations across bias calibration and downscaling configurations are slightly higher overall for the NWM Short Range operational configuration 18-hour cycle compared to the NWM Medium Range 4-day analysis previously. This is likely an artifact in itself from the analysis time period from real-time between the two operational configurations.
- User advice for utilizing bias calibration and downscaling configurations based on the domain configurations of interest in this analysis is consistent between the Medium and Short-Range operational configurations. Bias correction methods may have a slight benefit across domain configurations, but downscaling methods are highly sensitive to the height estimates within the domain configuration of interest (NextGen hydrofabric, coastal meshes) and subject to large errors in surface meteorological estimates when there are moderate to large errors within the height estimates.
Scatter Plots
Histogram, Correlation, and Spatial Plots
F1: No_Bias_Correction_or_Downscaling
F2: With_Bias_Correction_and_Downscaling
F3: With_Bias_Correction_Only
F4: With_Downscaling_Only
