This study evaluates the grid-length dependency of the Weather Research and Forecasting (WRF) Model precipitation performance for two cases in the Southern Great Plains of the United States. The aim is to investigate the ability of different cumulus and microphysics parameterization schemes to represent precipitation processes throughout the transition between parameterized and resolved convective scales (e.g., the gray zone). The cases include the following: 1) a mesoscale convective system causing intense local precipitation, and 2) a frontal passage with light but continuous rainfall. The choice of cumulus parameterization appears to be a crucial differentiator in convective development and resulting precipitation patterns in the WRF simulations. Different microphysics schemes produce very similar outcomes, yet some of the more sophisticated schemes have substantially longer run times. This suggests that this additional computational expense does not necessarily provide meaningful forecast improvements, and those looking to run such schemes should perform their own evaluation to determine if this expense is warranted for their application. The best performing cumulus scheme overall for the two cases studies here was the scale-aware Grell–Freitas cumulus scheme. It was able to reproduce a smooth transition from subgrid- (cumulus) to resolved-scale (microphysics) precipitation with increasing resolution. It also produced the smallest errors for the convective event, outperforming the other cumulus schemes in predicting the timing and intensity of the precipitation.

This study evaluates the sensitivity of wind turbine hub-height wind speed forecasts to the planetary boundary layer (PBL) scheme, grid length, and initial condition selection in the Weather Research and Forecasting (WRF) Model over complex terrain. Eight PBL schemes available for the WRF-ARW dynamical core were tested with initial conditions sources from the North American Mesoscale (NAM) model and Global Forecast System (GFS) to produce short-term wind speed forecasts. The largest improvements in forecast accuracy primarily depended on the grid length or PBL scheme choice, although the most important factor varied by location, season, time of day, and bias-correction application. Aggregated over all locations, the Asymmetric Convective Model, version 2 (ACM2) PBL scheme provided the best forecast accuracy, particularly for the 12-km grid length. Other PBL schemes and grid lengths, however, did perform better than the ACM2 scheme for individual seasons or locations.

Many statistical downscaling (SD) methods require observational inputs and expert knowledge, and thus cannot be generalized well across different regions. Convolutional Neural Networks (CNNs) are deep learning models that have generalization abilities for various applications. In this research, we modify UNet, a semantic-segmentation CNN, and apply it to the downscaling of daily maximum/minimum 2-m temperature (TMAX/TMIN) over the western continental US from 0.25-degree to 4-km grid spacings. We select high resolution (HR) elevation, low resolution (LR) elevation and LR TMAX/TMIN as inputs, train UNet using Parameter-Elevation Regressions on Independent Slopes Model (PRISM) data over the south- and centralwestern US from 2015 to 2018, and test it independently over both the training domains and the northwestern US from 2018 to 2019. We found the original UNet cannot generate enough fine-grained spatial details when transferred to the new northwestern US domain. In response, we modified the original UNet by assigning an extra HR elevation output branch/loss function and training the modified UNet to reproduce both the supervised HR TMAX/TMIN and the unsupervised HR elevation. This improvement is named “UNet-AE”. UNet-AE supports semi-supervised model fine-tuning for unseen domains and showed better grid-point-level performance with more than 10% mean absolute error (MAE) reduction compared to the original UNet. Based on its performance relative to the 4-km PRISM, UNet-AE is a good option to provide generalizable downscaling for regions that are under-represented by observations.

Statistical downscaling (SD) derives localized information from larger scale numerical models. Convolutional Neural Networks (CNNs) have learning and generalization abilities that can enhance the downscaling of gridded data (Part I of this study experimented with 2-m temperature). In this research, we adapt a semantic-segmentation CNN, called UNet, to the downscaling of daily precipitation in western North America, from the low resolution (LR) of 0.25-degree to the high resolution (HR) of 4-km grid spacings. We select LR precipitation, HR precipitation climatology and elevation as inputs, train UNet over the subsetted south- and central-western US using Parameter-Elevation Regressions on Independent Slopes Model (PRISM) data from 2015 to 2018, and test it independently in all available domains from 2018 to 2019. We proposed an improved version of UNet, that we call “Nest-UNet”, by adding deep-layer aggregation and nested skip connections. Both the original UNet and Nest-UNet show generalization ability across different regions, and outperform the SD baseline (bias-correction spatial disaggregation) with lower downscaling error and more accurate fine-grained textures. Nest-UNet also shares the highest amount of information with station observations and PRISM, indicating a good ability to reduce the uncertainty of HR downscaling targets.

Spurious grid-scale precipitation (SGSP) occurs in many mesoscale numerical weather prediction models when the simulated atmosphere becomes convectively unstable and the convective parameterization fails to relieve the instability. Case studies presented in this paper illustrate that SGSP events are also found in the North American Regional Reanalysis (NARR) and are accompanied by excessive maxima in grid-scale precipitation, vertical velocity, moisture variables (e.g., relative humidity and precipitable water), mid- and upper-level equivalent potential temperature, and mid- and upper-level absolute vorticity. SGSP events in environments favorable for high-based convection can also feature low-level cold pools and sea level pressure maxima. Prior to 2003, retrospectively generated NARR analyses feature an average of approximately 370 SGSP events annually. Beginning in 2003, however, NARR analyses are generated in near–real time by the Regional Climate Data Assimilation System (R-CDAS), which is identical to the retrospective NARR analysis system except for the input precipitation and ice cover datasets. Analyses produced by the R-CDAS feature a substantially larger number of SGSP events with more than 4000 occurring in the original 2003 analyses. An oceanic precipitation data processing error, which resulted in a reprocessing of NARR analyses from 2003 to 2005, only partially explains this increase since the reprocessed analyses still produce approximately 2000 SGSP events annually. These results suggest that many NARR SGSP events are not produced by shortcomings in the underlying Eta Model, but by the specification of anomalous latent heating when there is a strong mismatch between modeled and assimilated precipitation. NARR users should ensure that they are using the reprocessed NARR analyses from 2003 to 2005 and consider the possible influence of SGSP on their findings, particularly after the transition to the R-CDAS.