ClimaLand: A Land Surface Model Designed to Enable Data-Driven Parameterizations

Land surface models (LSMs) are essential tools for simulating the coupled climate sys27 tem, representing the dynamics of water, energy, and carbon fluxes on land and their in28 teraction with the atmosphere. However, parameterizing sub-grid processes at the scales 29 relevant to climate models (∼ 10–100 km) remains a considerable challenge. The pa30 rameterizations typically have a large number of unknown and often correlated param31 eters, making calibration and uncertainty quantification difficult. Moreover, many ex32 isting LSMs, written in legacy languages, are not readily adaptable to the incorporation 33 of modern machine learning parameterizations trained with in-situ and satellite data. 34 This article presents the first version of ClimaLand, a new LSM designed for overcom35 ing these limitations, including a description of the core equations underlying the model, 36 the results of an extensive set of validation exercises, and an assessment of the compu37 tational performance of the model. We show that ClimaLand can leverage graphics pro38 cessing units (GPUs) for computational efficiency, and that it’s modular architecture and 39 high-level programming language, Julia, allow for integration with machine learning li40 braries. In the future, this will enable efficient simulation, calibration, and uncertainty 41 quantification with ClimaLand.