MPR-enabled hydro-thermal soil physics in mHM: scaling and transferability tests

Obtaining accurate large-scale estimates of top-soil water content is a grand challenge in land-surface modelling [1]. Soil moisture (SM) is a key climate variable for understanding changes in the terrestrial water cycle [2], monitoring drought evolution [3], predicting drought severity [4], and improving flood forecasting by constraining antecedent wetness [5].  Predicting SM in extreme climates (paramo or permafrost) is further complicated by the coupling of soil water flow and heat transport. Decades of research have been invested in this subject, yet a scalable and transferable solution has not emerged.

Evidence from controlled multi-model experiments with harmonised forcings, geodata, and initial conditions (e.g., ESA https://4dhydro.eu/) suggests that epistemic uncertainty in simulated SM is dominated by model structure, soil parameterisations, and the scaling of soil properties.  The spread can be substantial; Wang et al. noted that "differences in model-predicted soil moisture can be quite large" [6]. A common narrative is that Richards-equation-based (RE) land-surface models are impractical at the kilometre scale: their effective parameters are difficult to infer, transfer across scales is often unsuccessful [7], and calibration against control variables such as streamflow is considered computationally intractable.  Simpler infiltration-capacity (IC) schemes or conceptual models, while readily calibrated against streamflow, are often assumed to yield poorer SM dynamics.

We revisit these assumptions by embedding a fast RE solver—the SLI module as implemented in CABLE [8,13]—into the mesoscale Hydrologic Model (mHM) and parameterising it with Multiscale Parameter Regionalization (MPR) [9].  MPR uses pedo-transfer relationships, high resolution physiographic datasets, and upscaling operators to derive effective, scale-consistent soil hydraulic parameters, while mHM provides the distributed water-balance and streamflow (Q)-based calibration framework. This design targets transferability across basins and resolutions. Using the results of Kholis et al. [10], we show that, when implemented in mHM, RE and IC yield similar streamflow performance under consistent calibration, while their SM states diverge. RE- and IC-based simulations agree on SM anomalies, but differ in volumetric water content, with discrepancies increasing with soil depth. The SLI module adds a thermal diffusion equation to mHM-RE, enabling joint tests of SM and soil temperature (Ts). We evaluate across German sites using station-based soil moisture and soil temperature observations and report mean daily performance of KGE(Q) = 0.89, KGE(SM) = 0.40 and KGE(Ts) = 0.90. In addition, we will present a first cold-region application using the 20-year Bayelva permafrost record (1998–2017) from Spitsbergen [11].

We conclude that (1) MPR enables practical parameterisation and scale transfer of RE across locations, (2) an RE+MPR SM module can be optimised without sacrificing streamflow skill, and (3) the mHM-RE infrastructure enables consistent multi-variable evaluation (SM and Ts), including EO-based benchmarking where available. Next steps include benchmarking against CryoGrid[12] and CABLE[13], extending evaluation to alpine and paramo observatories to probe combined hydro-thermal realism, and developing a long-term global SM reconstruction to advance state-of-the-art drought monitoring [2].

References:

[1] 10.1029/2010WR010090 
[2] 10.1126/science.adw5851
[3] 10.1088/1748-9326/11/7/074002
[4] 10.1029/2021EF002394
[5] 10.1038/s41467-024-48065-y
[6] 10.1175/2008JCLI2586.1
[7] 10.22541/essoar.174982768.80043676/v1
[8] 10.1016/j.jhydrol.2010.05.029
[9] 10.1029/2008WR007327
[10] 10.1029/2024WR039625
[11] 10.5194/essd-10-355-2018
[12] 10.5194/gmd-16-2607-2023
[13] 10.1002/2017MS001100