Ecosystem Insights through Extreme Values: A Fresh Look at Meteorological Drivers

Understanding the influence of meteorological drivers on the ecosystem is a cen-tral problem in Earth’s system science. Deriving these influences directly fromobservations is crucial, especially when models struggle due to scale mismatchesbetween the leaf or plant level to the ecosystem level. However, natural systemsoften have complex interactions, where multiple drivers influence each other andthe target variable simultaneously. These interactions complicate our understandingof individual variable effects. For instance, the relative importance of soil moistureand temperature on net biome production remains unclear, with in-situ measure-ments and remote sensing products yielding contradicting results. In this work, wepropose a novel approach: training on extreme values only. By focusing only onthese values, we can better approximate the influences of meteorological drivers.We demonstrate the potential of this approach through a toy example, validating itanalytically and empirically.