FEATURES BASED ASSESSMENTS OF WARM SEASON CONVECTIVE PRECIPITATION FORECASTS FROM THE HIGH RESOLUTION RAPID REFRESH MODEL
February 24, 2017
Janice Bytheway
Hosted by Chris Kummerow (advisor), Dave Randall, Russ Schumacher, Curtis Alexander (NOAA), Chandra Venkatachalam (Electrical and Computer Engineering)
Abstract
Forecast models have seen vast improvements in recent years, via increased spatial and temporal resolution, rapid updating, assimilation of more observational data, and continued development and improvement of the representation of the atmosphere. One such model is the High Resolution Rapid Refresh (HRRR) model, a 3 km, hourly-updated, convection-allowing model that has been in development since 2010 and running operationally over the contiguous US since 2014. In 2013, the HRRR became the only US model to assimilate radar reflectivity via diabatic assimilation, a process in which the observed reflectivity is used to induce a latent heating perturbation in the model initial state in order to produce precipitation in those areas where it is indicated by the radar.In order to support the continued development and improvement of the HRRR model with regard to forecasts of convective precipitation, the concept of an assessment is introduced. The assessment process aims to connect model output with observations by first validating model performance then attempting to connect that performance to model assumptions, parameterizations and processes to identify areas for improvement. Observations from remote sensing platforms such as radar and satellite can provide valuable information about three-dimensional storm structure and microphysical properties for use in the assessment, including estimates of surface rainfall, hydrometeor types and size distributions, and column moisture content.
A features-based methodology is used to identify warm season convective precipitating objects in the 2013, 2014, and 2015 versions of HRRR precipitation forecasts, Stage IV multisensor precipitation products, and Global Precipitation Measurement (GPM) core satellite observations. Quantitative precipitation forecasts (QPFs) are evaluated for biases in hourly rainfall intensity, total rainfall, and areal coverage in both the US Central Plains (29-49N, 85-105W) and US Mountain West (29-49N, 105-125W). Features identified in the model and Stage IV were tracked through time in order to evaluate forecasts through several hours of the forecast period. The 2013 version of the model was found to produce significantly stronger convective storms than observed, with a slight southerly displacement from the observed storms during the peak hours of convective activity (17-00 UTC). This version of the model also displayed a strong relationship between atmospheric water vapor content and cloud thickness over the central plains. In the 2014 and 2015 versions of the model, storms in the western US were found to be smaller and weaker than the observed, and satellite products (brightness temperatures and reflectivities) simulated using model output indicated that many of the forecast storms contained too much ice above the freezing level.
Model upgrades intended to decrease the biases seen in early versions include changes to the reflectivity assimilation, the addition of sub-grid scale cloud parameterizations, changes to the representation of surface processes and the addition of aerosol processes to the microphysics. The effects of these changes are evident in each successive version of the model, with reduced biases in intensity, elimination of the southerly bias, and improved representation of the onset of convection.