Seth McGinnis

An AI‐based Limited‐Area Model (LAM) is developed for dynamical downscaling over the Southern Great Plains and the southeastern United States, with strong generalization abilities under diverse boundary conditions. The model is trained using 0.25, 3‐hourly ERA5 as forcings and CONUS404 as targets in 1980–2019, producing 4‐km, hourly dynamical downscaling outputs; it is also connected to a post‐processing model to derive additional diagnostic variables. The model is evaluated across multiple forcing data sets, time periods, and climate regimes. For present‐day downscaling in water years 2021–2024, the model produces stable multiyear simulations with no unrealistic drift; its deterministic verification scores are comparable to other weather‐forecasting‐oriented AI models. The model also generalizes robustly to a 1.0, 6‐hourly non‐ERA5 forcing data set, yielding only minor performance changes. Frontal cyclone and hurricane case studies further demonstrate that the model reconstructs realistic, interpretable weather‐scale dynamical and thermodynamic structure from coarse boundary information. The AI‐based LAM is further tested by downscaling 30‐year global climate model runs in 1980–2010 and 2070–2100, and climate model ensembles in 2025–2027. In this application, the model remains stable at hourly downscaling frequencies for all 30 years and effectively captures climate‐change signals, indicating meaningful generalization across different climate regimes. When downscaling ensembles, the model produces well‐posed member distributions without collapsing the ensemble spread. Overall, the AI‐based LAM of this study offers good downscaling performance and generalization abilities. It provides a practical example of adapting AI weather prediction models for regional climate applications, with generalization across boundary forcing data sets and climate regimes.

This study presents a process-informed framework for assessing the differential credibility of diverse downscaling methodologies, including both statistical (simple and complex) and dynamical approaches. The methods evaluated include a convolutional neural network (CNN), the Locally Constructed Analog Method (LOCA), the Statistical DownScaling Model (SDSM), quantile delta mapping (QDM), simple interpolation with bias correction, and two regional climate models. As proof of concept, we apply the framework to evaluate the physical consistency of processes associated with wet-day occurrence at a site in the southern USA Great Plains. Additionally, we introduce a relative credibility metric that quantifies cross-method performance and outlines how this framework can be extended to other variables, regions, and downscaling applications. Results show that all downscaling methods perform credibly when the parent global climate model (GCM) performs credibly. However, complex statistical methods (CNN, LOCA, SDSM) tend to exacerbate GCM errors, while simpler methods (QDM, interpolation + bias correction) generally preserve GCM credibility. Dynamical downscaling, by contrast, can mitigate inherited biases and improve overall process-level credibility. These findings underscore the importance of process-based evaluation in downscaling assessments and reveal how downscaling model complexity interacts with GCM quality.

The coastal low-level jet, or coastal low-level jet (CLLJ), is a synoptically-forced meteorological feature frequently present offshore the western United States (U.S.). Characterized by a wind speed maximum that resides at the top of the marine boundary layer, the CLLJ is largely controlled by the location and strength of the North Pacific High (NPH) as well as the coastal geometry. Considering the rich wind resource available in this offshore region, the Bureau of Ocean Energy Management identified wind energy lease areas offshore California and supported the deployment of two U.S. Department of Energy wind lidar buoys near Morro Bay and Humboldt. Despite our relatively good understanding of the fundamental mechanisms responsible for large-scale CLLJ properties offshore the western U.S., future changes in CLLJ characteristics are less clear. To address this research challenge, and ultimately to better inform future wind turbine deployments, we use simulations driven by three global climate models (GCMs). We apply self-organizing maps to the model outputs for a historical and two future climate periods to show the range of NPH regimes that support CLLJ conditions during the warm seasons, as well as the subtle contribution from land-falling cyclones approaching the mainland during the cold seasons. Compared to the historical period, the three GCM-driven simulations agree that CLLJ conditions will become more (less) prevalent from central California northward (southward). They agree less with respect to future changes in maximum CLLJ wind speeds and CLLJ heights. However, after considering model biases present during the historical period, wind speeds between the models are actually more similar during the 2070–2095 period than during the historical period. The potential combination of more frequent CLLJ conditions characterized by relatively consistent wind speeds occurring at lower heights across northern California suggests that the Humboldt lease area may be ideal for a long-term wind turbine deployment.

To meet US goals of deploying additional wind energy as part of the decarbonization strategy, wind plants are being planned for the deep water offshore the western US. The wind flow in that region is complex due to the proximity to the coast, cold water upwelling, and persistent stratiform clouds that interact with radiation in ways that have the potential to destabilize the atmosphere. That flow has the potential to change with a changing climate. To address these issues, we assess the flow and the clouds in that region using downscaled climate model data, under both historic climate (1975–2005) and projected future (2025–2055) conditions. We note that the climate simulations agree fairly well with the cloud patterns observed by satellite data in the nearshore and offshore regions. We then assess the projected changes in clouds, wind speed, and other important variables, noting that our simulations project that the predominant north/northwesterly low-level jet is expected to strengthen and clouds are likely to be commensurately enhanced, although projected changes are within about 10% of current conditions. Our examination of the dynamics associated with the changes in the climate simulations provides confidence in the dynamical consistency of these projected changes.

This paper introduces a novel framework for a process-informed, differential assessment of credibility across various spatial and point-based downscaling methodologies, including both complex and simple statistical approaches, as well as dynamical downscaling. The methods assessed include a convolutional neural network (CNN), the locally constructed analog method (LOCA), the statistical downscaling model (SDSM), quantile delta mapping (QDM), simple spatial interpolation plus bias correction, the Regional Climate Model (RegCM) and the Weather Research and Forecasting Model (WRF). For proof of concept of our framework, our study focuses specifically on the physical consistency of wet days in a location in the southern US Great Plains. We find that all downscaling methods perform credibly when the parent global climate model (GCM) performs credibly. However, complex statistical methods like CNN, LOCA, and SDSM exacerbate inaccuracies when the GCM outputs are unreliable, performing worse as the GCM’s credibility decreases. On the other hand, simpler methods like QDM and bias-correction generally retain the GCM’s credibility. Notably, dynamical models can mitigate issues inherited from GCMs, enhancing the overall credibility of the data. These results highlight the need for careful evaluation of complex statistical downscaling techniques and suggest that further scrutiny is warranted. Finally, we summarize our process-informed analysis of credibility into a relative credibility metric, offering a quantifiable way to compare different downscaling approaches, and we provide guidance on the application and expansion of our framework for future research.

The 1997 New Year's flood event was the most costly in California's history. This compound extreme event was driven by a category 5 atmospheric river that led to widespread snowmelt. Extreme precipitation, snowmelt, and saturated soils produced heavy runoff causing widespread inundation in the Sacramento Valley. This study recreates the 1997 flood using the Regionally Refined Mesh capabilities of the Energy Exascale Earth System Model (RRM‐E3SM) under prescribed ocean conditions. Understanding the processes causing extreme events informs practical efforts to anticipate and prepare for such events in the future, and also provides a rich context to evaluate model skill in representing extremes. Three California‐focused RRM grids, with horizontal resolution refinement of 14 km down to 3.5 km, and six forecast lead times, 28 December 1996 at 00Z through 30 December 1996 at 12Z, are assessed for their ability to recreate the 1997 flood. Planetary to synoptic scale atmospheric circulations and integrated vapor transport are weakly influenced by horizontal resolution refinement over California. Topography and mesoscale circulations, such as the Sierra barrier jet, are better represented at finer horizontal resolutions resulting in better estimates of storm total precipitation and storm duration snowpack changes. Traditional time‐series and causal analysis frameworks are used to examine runoff sensitivities state‐wide and above major reservoirs. These frameworks show that horizontal resolution plays a more prominent role in shaping reservoir inflows, namely the magnitude and time‐series shape, than forecast lead time, 2‐to‐4 days prior to the 1997 flood onset.

Background: Wildfire simultaneity affects the availability and distribution of resources for fire management: multiple small fires require more resources to fight than one large fire does. Aims: The aim of this study was to project the effects of climate change on simultaneous large wildfires in the Western USA, regionalised by administrative divisions used for wildfire management. Methods: We modelled historical wildfire simultaneity as a function of selected fire indexes using generalised linear models trained on observed climate and fire data from 1984 to 2016. We then applied these models to regional climate model simulations of the 21st century from the NA-CORDEX data archive. Key results: The results project increases in the number of simultaneous 1000+ acre (4+ km2) fires in every part of the Western USA at multiple return periods. These increases are more pronounced at higher levels of simultaneity, especially in the Northern Rockies region, which shows dramatic increases in the recurrence of high return levels. Conclusions: In all regions, the models project a longer season of high simultaneity, with a slightly earlier start and notably later end. These changes would negatively impact the effectiveness of fire response. Implications: Because firefighting decisions about resource distribution, pre-positioning, and suppression strategies consider simultaneity as a factor, these results underscore the importance of potential changes in simultaneity for fire management decision-making.

The 1997 New Year’s flood event was the most costly in California’s history. This compound extreme event was driven by a category 5 atmospheric river that led to widespread snowmelt. Extreme precipitation, snowmelt, and saturated soils produced heavy runoff causing widespread inundation in the Sacramento Valley. This study recreates the 1997 flood using the Regionally Refined Mesh capabilities of the Energy Exascale Earth System Model (RRM-E3SM) under prescribed ocean conditions. Understanding the processes causing extreme events inform practical efforts to anticipate and prepare for such events in the future, and also provides a rich context to evaluate model skill in representing extremes. Three California-focused RRM grids, with horizontal resolution refinement of 14km down to 3.5km, and six forecast lead times, 28 December 1996 at 00Z through 30 December 1996 at 12Z, are assessed for their ability to recreate the 1997 flood. Planetary to synoptic scale atmospheric circulations and integrated vapor transport are weakly influenced by horizontal resolution refinement over California. Topography and mesoscale circulations, such as the Sierra barrier jet, are prominently influenced by horizontal resolution. The finest resolution RRM-E3SM simulation best represents storm total precipitation and storm duration snowpack changes. Traditional time-series and causal analysis frameworks are used to examine runoff sensitivities state-wide and above major reservoirs. These frameworks show that horizontal resolution plays a more prominent role in shaping reservoir inflows, namely the magnitude and time-series shape, than forecast lead time, 2-to-4 days prior to the 1997 flood onset.

The US National Center for Atmospheric Research (NCAR) has published several large datasets in the Amazon Web Services (AWS) cloud, thanks to support from the NCAR “Science at Scale” project, the AWS Open Data Sponsorship program, and the Amazon Sustainability Data Initiative. In each case we selected a subset comprising the most useful variables from the original data, and converted that subset from NetCDF to Zarr before publication. The Zarr format supports the same data model as netCDF and is well suited to object storage and distributed computing in the cloud using the Pangeo libraries in Python. Each dataset has an accompanying Intake-ESM catalog to facilitate data discovery and reading via Xarray, and each also has a sample Jupyter Notebook to illustrate how to access and analyze the data. Egress for these data are free, but users are encouraged to bring their compute to the data. The datasets currently published are: Community Earth System Model Large Ensemble (CESM LENS): https://doi.org/10.26024/wt24-5j82 North American Coordinated Regional Downscaling Experiment (NA-CORDEX): https://doi.org/10.26024/9xkm-fp8 CESM version 2 Large Ensemble (CESM2-LE): https://doi.org/10.26024/y48t-q717 Data Assimilation Research Testbed (DART) Reanalysis: https://doi.org/10.26024/sprq-2d04 This paper will provide information about the datasets and summarize lessons learned from the data conversion and publication.