Spatial Alignment of CAM Ensemble Members to Improve Ensemble Consensus Precipitation

Several methods have been devised to find consensus among ensemble forecast members. The ensemble mean, a simple point-wise arithmetic average of ensemble members, is one of the simple methods. However, due to the difference in spatial distribution and intensity of precipitation features in each ensemble member, the ensemble mean tends to reduce the magnitude of precipitation maxima while increasing the areal coverage of light values. The probability-matched (PM) ensemble mean and localized PM (LPM) mean methods have been introduced to overcome these problems. The PM and LPM use ensemble members' probability distribution function (PDF) to preserve the ensemble forecast's maxima. PM and LPM methods redistribute the values of each grid point of the ensemble mean but may not preserve the spatial structures of the features themselves, which can be blurred if there are offsets among feature locations.

This study aims to find a way to improve ensemble consensus precipitation by considering the spatial offsets among ensemble members. This study uses the phase-correcting method to align the fields of each ensemble member to a common location. Each ensemble member is re-aligned with respect to other members in pairs, a vector mean offset is calculated, and the Spatially Aligned Mean (SAM) is obtained by averaging the re-aligned members. SAM is applied to an operational high-resolution (3km) Convection-Allowing Model (CAM) ensemble, the US High-Resolution Ensemble Forecast (HREF), which has ten members. Also, to preserve the ensemble forecast's maxima, LPM is applied to the SAM results.

This experiment evaluates ensemble mean, SAM, and spatially aligned LPM of 3-hour accumulated precipitation over the contiguous United States (CONUS) using Stage IV precipitation data as verification for lead times of 15 to 36 hour. The verification is done for four weeks in the summer of 2022 during the Hydrometeorology Testbed (HMT) Flash Flood and Intense Rainfall (FFaIR) experiment. The point-wise verification (Bias, POD, FAR, ETS) and spatial feature verification are performed with several thresholds using the Meteorology Evaluation Tools’ (MET) Method for Object-based Diagnostic Evaluation (MODE) program.

The verification results will be presented and discussed.

Figure Caption: Preliminary results: Frequency Bias and ETS for 20 mm/3h rainfall threshold over the first 3 weeks of 2022 FFaIR.