Towards an Optimal Representation of Sub‐Grid Heterogeneity in Land Surface Models

Abstract

One of the persistent challenges of Land Surface Models (LSMs) is to determine a realistic yet efficient sub‐grid representation of heterogeneous landscapes. This is particularly important in emulating the fine‐scale and nonlinear interactions between water, energy, and biogeochemical fluxes at the land surface. In LSMs, landscape heterogeneity can be represented using sub‐grid tiling techniques, which partition macroscale grid cells (e.g., 1°) into smaller units or “tiles.” However, there is currently no formal procedure to define the number of tiles required to adequately represent the heterogeneity of hydrologic processes within a macroscale grid cell, and across spatial scales. To address these challenges, a new approach is presented to diagnose sub‐grid process heterogeneity formally and to infer an optimal number of tiles per macroscale grid cell. The approach is demonstrated using the HydroBlocks modeling framework coupled to Noah‐MP LSM implemented over a 1.0‐degree domain in Western Colorado in the United States. Our results show that (a) a surrogate model can accurately infer the spatial structure of the LSM's time‐averaged hydrological fields, with over 95% overall R2 performance in the validation stage; (b) the optimal configurations for a target level of complexity can be determined using a multi‐objective Pareto efficiency analysis, which includes the simultaneous representation of the multi‐scale heterogeneity of several processes; (c) the use of ∼100 tiles effectively reproduces a quasi‐fully distributed LSM setup (i.e., 83,000 tiles) with approximately 1% of the computational expense. This method provides a path forward to efficiently determine the optimal tile configurations for LSMs while simultaneously considering the spatial heterogeneity and spatial accuracy of hydrologic processes.

A Two‐Column Model Parameterization for Subgrid Surface Heterogeneity Driven Circulations

“…The method used is described in greater detail in Torres‐Rojas et al. (2022), with the decay cutoff in our case as 5% of the variance. This particular cutoff is chosen to produce a wide range of values for different surfaces without saturating at either end of the range.…”

The observed spatio-temporal patterns of Land surface temperature over the Contiguous United States

“…Despite the significant advances regarding tiling schemes over the last decade, many issues persist, including the fact that over large-scale domains, LSM sub-grid outputs are mostly only summarized and evaluated via macroscale grid statistics: spatial mean and variance. Although informative, these statistics are insensitive to the tiles' large-scale spatial patterns (i.e., patternagnostic metrics) (Jupp & Twiss, 2006;Torres-Rojas et al, 2022). This issue is critical as emerging work shows the importance of correctly representing the sub-grid spatio-temporal patterns of surface states to explain the role of sub-grid heterogeneity on atmospheric response (Simon et al, 2021).…”

“…More specifically, the multi-scale spatial heterogeneity of the physical environment (e.g., vegetation, soils, elevation, and land use) has been acknowledged manuscript submitted to JGR: Atmospheres to influence the spatial and temporal distribution of the fluxes in a nonlinear manner (Dickinson, 1995;R. A. Fisher & Koven, 2020;Koch et al, 2017;Nicholson, 1988;Simon et al, 2021;Tesfa et al, 2014;Torres-Rojas et al, 2022;Vergopolan et al, 2022). Besides, it is widely recognized that certain fluxes and associated processes and variables tend to recur and appear consistently across different scales in space, time, or both.…”

Detecting vegetation stress in mixed forest ecosystems through the joint use of tree-water monitoring and land surface modeling

“…This issue is further complicated in highly segregated landscapes where multiple PFTs, located within the same model grid cell, interact through the shared resources (e.g., soil water). While previous studies have tackled the challenge of representing sub-grid heterogeneity to understand the impacts of topographical gradients (Tesfa & Leung, 2017), hydrological processes (Torres-Rojas et al, 2022), or vegetation types (Ke et al, 2013) on the estimation of energy and water fluxes, no study has yet evaluated the impact of sub-grid heterogeneity on the simulated plant water stress in LSMs. In particular, the effect of homogenizing PHTs within single PFTs and/or across the landscape on the simulated plant water stress is still poorly understood.…”