Advancing the Representation of Land Surface Heterogeneity in Land Surface Models

Abstract

Due to its profound influence on various environmental processes and phenomena, the correctrepresentation of landscape physical heterogeneity in models is vital for applications spanning a wide range of scales, from global climate prediction to field-scale hydrological forecasting. Land Surface Models (LSM), Earth System Models (ESMs), and satellite remote sensing provide spatially distributed fields of surface fluxes and states, making them critical scientific tools for understanding the impact of physical heterogeneity. Enhanced understanding of heterogeneity's spatial and temporal effect can significantly improve our comprehension of hydrological, energy, and biogeochemical cycles at multiple scales. Under this framework, the dissertation focuses on optimizing, evaluating, and improving heterogeneity representations for LSM and ESM applications. Chapter 2 introduces a novel multi-objective optimization approach to efficiently determine optimal heterogeneity representation configuration for LSMs while considering the spatial structure of the generated fields, the accuracy of the representation of hydrological processes, and the computational trackability of the resulting structure. Chapter 3 builds upon the spatial nature of this approach and presents the Empirical Spatio-Temporal Covariance Function (ESTCF), a tool based on geostatistics that allows to efficiently and effectively characterize the spatio-temporal patterns observed in remotely sensed fields and relate them to physical characteristics of the environment. Intending to use remote sensing elevation data to its maximum, Chapter 4 proposes strategies to improve the coupling between river networks and heterogeneity representations in LSMs. Experiments demonstrate the sensitivity of spatiotemporal patterns in the land surface to the heterogeneity representation. Finally, the tool developed in Chapter 2 and the heterogeneity representation proposed in Chapter 4 are combined in Chapter 5, where the spacetime covariance is used to evaluate LSM simulated spatio-temporal patterns of land surface temperature. The proposed method efficiently summarizes complex patterns and offers valuable insights into model strengths and weaknesses. Overall, this dissertation contributes to a stricter description and assessment of the landscape heterogeneity representation in LSMs and ESMs, providing a foundation for a more comprehensive model development.