High-resolution precipitation is critical for hydrologic and climate-risk applications, but climate models are too coarse to resolve storm-scale extremes. We compare a deterministic U-NET with two generative models, a Wasserstein Generative Adversarial Network (WGAN) and a diffusion model, for 8× and 16× precipitation downscaling using ERA5-Land. U-NET is stable but smooths extremes, whereas generative models better capture variability and heavy tails while introducing greater uncertainty.

Seawater composition affects carbon cycling in the ocean and has changed over Earth history, requiring corrections when reconstructing past marine carbonate systems. We present a new correction scheme for the intermediate complexity Earth system model cGENIE based on the ion interaction model MyAMI. We validate the new scheme, find significant improvements over the default scheme, and discuss the relevance of accurate and consistent major ion correction in carbon cycle reconstructions.