367452 A Python-Based Quantitative Precipitation Estimate Over Alaska Using Rain Gauge Kriging and the HRRR-AK Precipitation Forecast

Wednesday, 15 January 2020
Hall B1 (Boston Convention and Exhibition Center)
Brett T. Hoover, CIMSS, Madison, WI; and J. A. Otkin, E. Petrescu, and E. Niebuhr

An accurate quantitative precipitation estimate (QPE) is essential for hydrological forecasting and proper calibration of quantitative precipitation forecasts. However, the Alaska region suffers from poor observational coverage; the rain gauge and radar networks are sparse beyond the Gulf of Alaska coastline and most of Alaska is too far north to make use of geostationary satellite data. As a result, the NCEP Stage IV precipitation analysis currently does not extend into the Alaska region, and no accepted standard QPE exists there.

CIMSS has partnered with NWS Anchorage, the NWS Alaska Region Headquarters, and the Alaska River Forecast Center to develop a quantitative precipitation estimate that utilizes in situ precipitation observations and precipitation forecasts from the High-Resolution Rapid Refresh forecast of the Alaska region (HRRR-AK), which became operational in November of 2018. The HRRR-AK provides forecasts of precipitation on a 3 km resolution grid initialized every three hours, which serve as a first-guess for the precipitation analysis. Available rain gauge observations from MesoWest are incorporated into the first-guess through kriging, by taking point-measurements of the difference between observations and the first-guess at gauge locations, constructing a 1-dimensional model of the spatial autocorrelation as a function of distance, and dispersing the point-observation corrections to the entire HRRR-AK grid via interpolation to provide the Best Linear Unbiased Estimate of the state. This technique has previously been successfully applied to snow analyses at the National Operational Hydrologic Remote Sensing Center, and as such serves as a good basis for this project. The QPE system being developed for Alaska is Python-based, utilizing the PyNiO GRIB reading module from the National Center for Atmospheric Research and R’s GSTAT kriging module within Python through rpy2.

The QPE from the version-1 product demonstrates an improved fit to available rain gauge observations over the HRRR-AK forecast alone, both with respect to rain gauge observations assimilated through kriging as well as rain gauge observations not utilized for kriging but retained for independent cross validation. The primary limiting factor to additional improvement of QPE is believed to be inaccuracy of the HRRR-AK forecast, which produces many outlier observation-minus-forecast innovations that cannot be used by the kriging algorithm. Efforts are underway to perform statistical bias correction of the HRRR-AK precipitation forecasts to further improve QPE.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner