Model calibration

Improving the accuracy of land surface models (LSMs) is crucial for reducing uncertainties in climate change projections. Parameter data assimilation (DA), which fine-tunes model parameters to better match observed data, is key to enhancing LSM performance. However, the complexity of LSMs poses challenges for global optimization. Advances in computational power, novel data sets, and machine learning (ML) offer promising solutions to improve LSMs. ML can streamline the DA process, handling large data sets and reducing computational demands. This article discusses the progress made in LSM parameter estimation and the challenges faced by the community. We then discuss how ML can help address these challenges and outline future priorities. International collaboration, fostered by initiatives like the Analysis, Integration and Modeling of the Earth System Land DA Working Group and the International Land Model Forum, is essential for accelerating progress, facilitating knowledge exchange, and developing standardized methods for more accurate climate modeling.