Spatially and Physically Consistent Weather Estimation using Bias-Corrected Climate Simulations and Reanalysis Analogs

Global Climate Models (GCMs) are essential for projecting future climate; however, despite the rapid advance in their ability to simulate the climate system at increasing spatial resolution, GCMs cannot capture the local and regional features necessary for climate impacts (e.g., hydrologic, agricultural, forest, health) assessments well. Temperature and precipitation, for which dense observational records are available, can be bias corrected and downscaled, but many climate impacts models require a larger set of variables, such as relative humidity, cloud cover, wind speed and direction, and solar radiation. While this larger set of variables can be created by a variety of methods ranging from simply using the raw GCM output to deploying a stochastic weather generator to running a regional climate model, these approaches introduce errors and often require significant effort.

We developed an efficient weather estimator based on sampling analogs from reanalysis to produce daily weather that is physically, spatially, and temporally consistent. Specifically, we use the Bias Correction with Constructed Analogs (BCCA) temperature and precipitation data topographically downscaled (~1 km) to identify a corresponding day in the North American Regional Reanalysis (NARR) dataset based on a close match of temperature, precipitation, and season for our region of interest, the Lake Champlain Basin in Vermont. We then use the full suite of atmospheric and surface variables from NARR for that day to create a realistic and consistent set of weather inputs for climate impacts modeling.

We find that daily distributions of variables with strong seasonal cycles, such as evapotranspiration and solar radiation, are estimated best relative to NARR, while variables such as wind are difficult to estimate through temperature and precipitation alone. We explore the sensitivity of our method to day selection parameters and across four BCCA ensemble members. We also analyze the relative importance of selection based on temperature and precipitation in reproducing the NARR daily distributions of estimated variables.