How much land-surface complexity is needed to simulate drought processes in boreal forests? A calibration-free CLASS assessment across an energy-water gradient

Land surface models are often made increasingly complex to represent heterogeneity in soils and vegetation. However, the level of horizontal and vertical complexity actually required to capture drought processes in boreal forests remains unclear. This issue is critical for large-sample hydrological and climate applications, where detailed site-by-site calibration is rarely feasible. In this study, the Canadian Land Surface Scheme (CLASS v3.6) is applied in point mode and driven by ERA5-Land to quantify trade-offs between structural complexity and hydrological realism under a strictly calibration-free, rule-based parameterization.

The analysis is conducted at Forêt Montmorency, a humid, snow-dominated boreal catchment in Québec, Canada, instrumented with eddy-covariance towers, soil water content and temperature profiles, and long-term hydrometeorological observations. Vegetation and soil parameters are constrained by field data, LiDAR-based canopy metrics and CLASS defaults, without tuning to match fluxes. Two experiment sets are considered. In the first, the impact of progressively increasing the number of grouped response units (GRUs) and soil layers (from reduced 3–4 layer profiles to an 8-layer column) on model skill is assessed under identical ERA5-Land forcing. Second, the multi-decadal ERA5-Land record is used to isolate and evaluate model behavior during independently defined drought windows. Therefore, performance metrics specifically target water-limited conditions rather than aggregates over mixed wet and wet–dry periods.

Model behavior is evaluated for total evaporation and its components, soil moisture and temperature, diagnostically derived soil water potential (psi), and simple runoff and low-flow indicators at seasonal to annual scales, using Kling–Gupta efficiency, bias and error metrics. Drought periods are defined independently of the model from standardized climatic and soil-based indices (SPEI, SSMI, REW and psi thresholds), ensuring that the assessment targets genuinely water-limited conditions rather than artifacts of model structure.

Results indicate clear diminishing returns from added structural complexity. Increasing the number of GRUs and extending the soil column beyond a limited number of layers does not systematically improve evaporation skill and can degrade the coherence of shallow soil water content dynamics. However, a parsimonious configuration with a small set of dominant GRUs and a moderately deep soil profile is sufficient to reproduce seasonal energy partitioning and to capture the timing and relative severity of drought events within forcing-related uncertainty. These findings provide quantitative evidence that robust drought diagnostics in snow-affected boreal forests do not require highly complex land-surface setups and that carefully designed, rule-based configurations offer a pragmatic benchmark for regional hydrological and climate modelling studies.