Identifying Physical Drivers of Climate and Land-surface Biases in a Regional Climate Model Using Perturbed Parameter Ensembles

Regional climate modeling is an important tool for providing high-resolution climate projections at scales appropriate for management and adaptation actions. One of the key advantages of regional dynamic downscaling is the ability to resolve interactions between components of the climate system, such as the biosphere and the atmosphere, at finer spatial scales. However, regional models inherit biases from the global climate models used to provide lateral boundary conditions. Additionally, models are often parameterized at the global scale and must be adapted for the specific region of interest. Through exploration of parameter sensitivities and interactions, the physical drivers of biases can be better understood and, in some cases, improved.

Using a perturbed parameter ensemble (PPE) generated with the regional climate model HadRM3P-MOSES2, we performed a sensitivity experiment to assess the influence of model parameter adjustments on land-surface and atmospheric processes. We utilized statistical emulation to identify processes and parameters with high sensitivity. Parameters related to hydrology and photosynthesis had the largest influence on temperature biases, highlighting the importance of land-atmosphere interactions. Parameter adjustments reduced the biases in primary productivity by 26% and evapotranspiration by 10%. Western North America contains diverse ecosystems and domain wide model improvement demonstrates that alternate model parameterizations can improve model performance for a wide range of ecosystems simultaneously.

Our sensitivity experiment used a PPE to identify key model structural limitations which hinder the ability to capture plant response to extreme drought and temperature stress, a process essential to capturing carbon and water cycling under future climate conditions. This study demonstrates the value of using PPEs to identify processes that may require improved mathematical representations, and parameters that may be constrained through targeted observational studies.