The QTP, extending across 26°00^′–39°47^′ N and 73°19^′–104°47^′ E, represents the world's largest and highest physiographic unit, covering an area of approximately 2.5 million km² with an average elevation exceeding 4000 m a.s.l. (Fig. 1a). The region is characterized by a pronounced continental climate marked by significant spatial gradients; mean annual air temperature (MAT) decreases from approximately 10 °C in the humid southeast to -8 °C in the arid northwest (Yu et al., 2024). Mean annual precipitation (MAP) follows a similar trend, declining from over 1000 mm to less than 100 mm (Yang et al., 2010). Alpine grassland is the dominant land cover type (Wang et al., 2022), reflecting these severe climatic constraints. Permafrost is widespread, underlying approximately 67 % of the plateau's total area and acting as a primary regulator of cryogenic soil processes (Cao et al., 2023). In the core permafrost zones, intense freeze-thaw cycles, limited vegetation cover, and sparse observational data create particularly challenging conditions for both soil formation and large-scale modeling.

Soils on the QTP are typically pedogenically young and fragile, shaped by the combined effects of intense glacial erosion, pervasive freeze-thaw cycles, and wind erosion (Wei et al., 2025; Montgomery, 2002; Li et al., 2012). These active geomorphic processes drive rapid surface renewal, resulting in a complex soil-forming environment where diverse soil types exhibit high spatial heterogeneity (Fig. 1b).

Multiple datasets were compiled to drive the revised mass balance model, calibrate its parameters, and validate the outputs. A total of 552 soil profile observations were compiled across the QTP to serve as the ground truth for model calibration and performance assessment (Fig. 1a). These profiles were integrated from several primary sources, including the Second National Soil Survey (Pan and Shi, 2015), the Chinese Soil Series Chronicle (Central and Western Volumes) (Zhao and Li, 2020; Li et al., 2020; Yuan, 2020; Huang and Lu, 2020; Wu et al., 2020; Yang and Zhang, 2020), and a study of parent material age in the Qinghai Lake Basin (Zhang et al., 2024b). A critical methodological distinction is that these observations represent the vertical distance from the ground surface to the base of the pedogenic horizons, effectively measuring solum thickness. Because excavations in these national surveys typically terminate at the C horizon interface rather than the lithic contact (Shangguan et al., 2013), these data provide a consistent reference for the solum thickness simulated in this study.

Climate forcing was derived from high-resolution datasets for the period 2000–2015 to represent the modern steady-state climate. MAT data were sourced from the Dataset of 0.01° Surface Air Temperature over Tibetan Plateau (1979–2018) (Ding et al., 2022), while MAP data were extracted from the 1 km Monthly Precipitation Dataset for China (1901–2024) (Peng, 2025). Vegetation conditions, which regulate surface erosion, were parameterized using a 1 km MODIS(MOD13A2)-derived Normalized Difference Vegetation Index (NDVI) product (2001–2020) (Zhu, 2022) over the 2000–2015 interval. Topographic attributes, including slope, aspect, and curvature, were extracted from the 30 m NASADEM digital elevation model and subsequently resampled to a 1 km spatial grid using bilinear interpolation. Land use data for 2015 at a 300 m resolution (Xu, 2019) and soil parent material from the 1:2.5 million scale Quaternary Geological Map of the QTP (Li, 2020) were similarly standardized to a 1 km resolution grid.

The model's erosion framework is driven by spatially explicit estimates of hydraulic and aeolian mass transfer. The hydraulic erosion rate (Ewater) was adopted from a 1 km resolution dataset (2002–2016) (Teng et al., 2018) generated using the Revised Universal Soil Loss Equation (RUSLE), while the aeolian erosion rate (Ewind) was sourced from a 1 km resolution map (1980–2015) (Teng et al., 2021) created via the Revised Wind Erosion Equation (RWEQ) model. To ensure geomorphic realism, we integrated the global depth-to-bedrock (DTB) dataset developed by Shangguan et al. (2017) as a critical structural constraint. This dataset, generated at a 1 km resolution using an ensemble of machine learning algorithms, defines the total available weathered mantle. However, it should be noted that uncertainties remain in the interior regions of the QTP due to highly sparse field observations and a heavy reliance on model-based extrapolation.

To evaluate the performance of our model, two existing national-scale datasets were utilized for comparison. The first dataset, the Shangguan map (Shangguan et al., 2013), was developed at a spatial resolution of 30 arcsec (∼1 km) using a polygon linkage method that integrated 8979 soil profiles from the Second National Soil Survey (1979–1985) with the 1:1 million Soil Map of China. Notably, the Shangguan map explicitly delineates non-soil map units (e.g., glaciers, rock debris, and water bodies), which appear as void spaces (null values) in the dataset. The second dataset, the Liu map (Liu et al., 2021), is a high-resolution (90 m) gridded dataset generated through a predictive mapping paradigm using a quantile regression forest algorithm. The model was trained on approximately 5000 representative soil profiles from the more recent National Soil Series Survey (2009–2019). While both datasets utilize extensive national survey data, their performance on the QTP is potentially limited by the lower density of profiles in high-altitude regions compared to eastern China. The observational dataset compiled for our study integrates the primary data sources used in both the Shangguan map and the Liu map, supplemented by additional regional survey data to maximize representativeness.

Finally, to constrain the simulation physically, non-soil map units were excluded by masking glaciers and water bodies using the Second Chinese Glacier Inventory (Guo et al., 2015b) and the China Lake Dataset (1960s–2020) (Zhang et al., 2019), with the aquatic mask specifically based on the 2020 lake distribution.

To simulate the spatial heterogeneity of solum thickness across the QTP, we revised the classical mass balance framework to account for the region's specific geomorphic processes. Unlike traditional landscape evolution models that treat the subsurface as a single homogeneous layer, our framework employs a coupled two-interface system. This system distinguishes between the solum (hs; m), defined as the genetically developed A and B horizons, and the underlying saprolite or C horizon (hc; m). This distinction is critical for the QTP, where the physical and chemical coupling between surface erosion and the deep weathering front is mediated by the thickness of the solum.

Applying the mass balance model (Eq. 9) requires the specification of distinct kinetic and transport coefficients that govern soil evolution. Given the vast area and extreme environmental gradients of the QTP, assuming a single set of global parameters would oversimplify regional geomorphic distinctiveness and introduce systematic bias. To address this, we implemented a zonally adaptive parameterization strategy that partitions the plateau into homogeneous environmental clusters, allowing the model to explicitly account for the spatial non-stationarity of weathering and erosion processes. The process involved two main steps.

We first evaluated the predictive importance of 16 potential covariates derived from topographic, climatic, and ecological datasets (Table 1) (Yamashita et al., 2024; Wang et al., 2021). A Random Forest algorithm was employed to rank the predictive importance of these variables for solum thickness using the Mean Decrease Accuracy (MDA) metric. Variables with non-positive MDA values were deemed non-contributory and were excluded to reduce noise.

Second, based on the remaining key predictors, we partitioned the study area into eight environmentally distinct clusters using the k-prototypes (Huang, 1998) unsupervised clustering algorithm (k=8). This algorithm was specifically selected for its ability to handle mixed data types, effectively integrating both continuous and categorical variables. Consistent with the data processing in Sect. 2.2, non-soil areas (glaciers and water bodies) were excluded from the clustering process. This resulting zoning system provided the framework for calibrating the revised mass balance model independently for each cluster, thereby better capturing the region's complex environmental gradients.

Obtaining direct field measurement of the model parameters (Table 2) required by the mass balance model is practically infeasible across the vast and extreme environment of the QTP. We therefore implemented a parameter optimization approach using inverse modeling against the available soil profile datasets. The model parameters (Table 2) were calibrated independently for each of the eight environmental clusters using the Shuffled Complex Evolution (SCE-UA) algorithm (Duan et al., 1993). This method was selected for its robust performance in calibrating physical land surface models (Guo et al., 2013; Vrugt et al., 2003) as its complex shuffling mechanism is highly effective at avoiding convergence on local minima. To prevent equifinality and maintain physical consistency, the parameters were constrained within physically permissible ranges derived from literature (Table 2).

To rigorously evaluate the reliability of the simulated solum thickness, we implemented a two-tiered assessment strategy comprising internal cross-validation and external benchmarking against existing national datasets. Given the practical difficulty of collecting independent validation samples in the extreme environment of the QTP, we employed a k-fold cross-validation scheme (with k=4) to maximize the utility of the 552 available soil profiles. Within each environmental cluster, the observational data were randomly partitioned into four equal subsets. The model was then iteratively trained using three of the subsets, with the remaining subset reserved for validation. This process was repeated four times to ensure every subset served as a validation set exactly once.

The accuracy of the solum thickness simulation was quantitatively characterized using three statistical metrics: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Relative Error (MRE). These metrics characterize the deviation between simulated values (m^(xi)) against measured values (m(xi)): 10RMSE=1n∑i=1nm^xi-mxi2,11MAE=1n∑i=1nm^xi-mxi,12MRE=1n∑i=1nm^xi-mximxi. Beyond internal consistency, we conducted a comparative performance analysis against two state-of-the-art regional products: the Shangguan map and the Liu map. Error metrics for all three datasets were calculated against the compiled 552 soil profiles. It is important to acknowledge that these 552 observations partially overlap with the source data used to generate the reference maps. Therefore, the performance comparison must be interpreted as a relative assessment under shared observational constraints, rather than a fully independent external validation. This approach still provides a rigorous benchmark of how well the process-based framework reproduces geomorphic reality compared to traditional polygon-linkage and machine-learning paradigms using the best available data for the region.

The Random Forest-based importance analysis revealed significant disparities in the predictive power of the 16 potential environmental variables. As illustrated in Fig. 2a, elevation and NDVI emerged as the dominant factors controlling the spatial heterogeneity of the QTP environment. This reflects the fundamental thermodynamic constraints on pedogenesis across the QTP: elevation governs the thermal energy available for chemical weathering, while precipitation and vegetation regulate both the moisture necessary for mineral transformation and the surface stability against erosion. Conversely, local morphometric variables such as general curvature (GeC), the Sediment Transport Index (STI), land use type (LUT), the Terrain Ruggedness Index (TRI) and the Topographic Wetness Index (TWI) exhibited non-positive MDA values, indicating negligible contribution to the environmental zoning. Consequently, these five variables were excluded, retaining 11 key environmental variables for the clustering process.

Based on these variables, the QTP was partitioned into eight spatially distinct clusters (Fig. 2b). These clusters exhibit high internal homogeneity but significant inter-cluster heterogeneity, effectively capturing the plateau's diverse environmental regimes (Table 3). Cluster 1, located in the southeastern and eastern margins, represents the zone with the most favorable moisture and thermal regimes; it features the lowest mean elevation (2136 m a.s.l.) and the highest precipitation (832.41 mm) and air temperature (MAT: 13.09 °C), supporting the densest vegetation on the plateau (mean NDVI: 0.86) and rapid soil development. In sharp contrast, Cluster 2 is concentrated in the Qaidam Basin (northeastern QTP) and represents the plateau's arid interior, characterized by flat terrain, minimal precipitation (∼222 mm), and desert vegetation. Clusters 3 and 4 serve as transitional zones in the eastern and southeastern QTP; they exhibit similar temperature and precipitation patterns and vegetation conditions to one another but are differentiated by elevation (3597 and 4354 m, respectively).

The remaining groups, Clusters 5 through 8, constitute the high-altitude cold zones. Cluster 5 is predominantly located in the northeastern QTP, while Cluster 6 extends across the central and eastern-central portions of the plateau, including the Three Rivers Source Region. Both are defined by high elevations (>4200 m a.s.l.) and reduced climatic variability. Cluster 7 covers the central and western interior, the core permafrost zone, exhibiting low temperatures (MAT: -3.32 °C) and sparse vegetation. Finally, Cluster 8 occupies the most extreme high-elevation zone, such as the northern Himalayas where soil formation is kinetically limited by low temperatures and sparse biomass production.

The calibrated model parameters (Table 4) provide physical insights into these zonal differences. The lithology adjustment coefficient (a), which scales the maximum potential weathering rate, reveals a fundamental divergence between the plateau's interior core and its margins. In the central permafrost and high-mountain zones (Clusters 7 and 8), a values are extremely low (<1 m a⁻¹), reflecting severe lithological constraints that inhibit soil production regardless of climatic inputs. In contrast, the surrounding regions (Clusters 1–4, 6) exhibit high potential weathering rates (a>80 m a⁻¹). This suggests that in these active soil-forming zones, the calibrated model requires relatively high effective production rates to reproduce the observed solum thickness under these environmental conditions.

The erosion weighting coefficients (α, β, γ) highlight spatial disparities in sediment transport. In the arid interior (Cluster 7), coefficients are near unity (≥0.96), indicating that empirical erosion estimates closely match the low weathering rates. Conversely, in the humid southeast (Cluster 1) and the source regions (Cluster 6), the hydraulic erosion weights are notably low (α<0.06), suggesting that vegetation cover significantly dampens actual sediment export relative to the potential erosion energy predicted by RUSLE.

The simulation results from the revised mass balance model indicate that solum thickness across the QTP ranges from 0.39 to 2.04 m, with a spatially averaged mean of 0.89 m. Spatially, solum thickness generally decreases from the southeast to the northwest (Fig. 3). Regions with the greatest solum thickness are primarily concentrated in the southeastern and eastern margins of the plateau, particularly within the Yellow River Basin, where the average thickness reaches 1.16 m. This area corresponds mainly to Cluster 1, which features the most favorable moisture and thermal regimes on the plateau, fostering dense vegetation cover (NDVI > 0.6) and rapid soil development. Crucially, the high-resolution simulation resolves distinct geomorphic signatures that are often obscured in coarser empirical maps. The eastern drainage systems (Yangtze, Mekong, and Salween basins) display distinct dendritic patterns of solum thickness (Fig. 3). Thicker colluvial deposits (indicated by yellow to orange bands in Fig. 3) extend along the river valleys, contrasting sharply with the thinner solum on the adjacent bedrock-dominated mountain ridges. This ridge-valley differentiation provides geomorphic support for the model's gravitational erosion module, which simulates the transport of material from steep slopes to valley floors.

A similar depositional signature is evident in the Qaidam Basin (Cluster 2), which appears as a coherent “island” of moderately thick solum amidst the rugged terrain of the surrounding Qilian and Altun Mountains. This reflects the basin's nature as a stable depositional zone. Similarly, in the south, a continuous linear band of high solum thickness marks the Yarlung Zangbo River (Brahmaputra) valley. This depositional corridor stands in stark contrast to immediately adjacent high-altitude zones, the Himalayas to the south and the Gangdise Mountains to the north, which exhibit some of the lowest thickness values on the plateau due to intense glacial and periglacial erosion.

Conversely, the vast interior of the plateau (Clusters 7 and 8) is dominated by uniformly low solum thickness (<0.60 m). This broad expanse reflects the core permafrost zone, where extreme cold and aridity kinetically inhibit pedogenesis, and the lack of vegetation leaves the surface vulnerable to deflation by strong westerly winds.

The frequency distribution of simulated solum thickness (Fig. 3b) shows that the model predominantly produces moderately developed soils (0.6–1.2 m), with thin and thick extremes largely partitioned by geomorphic regimes. Basin-level analysis (Fig. 3c and Table B1) further reveals clear differences among major catchments. The Yarlung Zangbo Basin in the southeastern QTP exhibits the greatest mean solum thickness and the largest internal variability (standard deviation), consistent with its complex terrain and strong topographic gradients. In contrast, the permafrost-dominated basins in the inner QTP (e.g., Inner Plateau) show the smallest mean solum thickness and the lowest variability, reflecting kinetically limited pedogenesis under cold and arid conditions.

To verify that the model's underlying thermodynamic parameterizations were preserved in the final steady-state solutions, we conducted a Pearson correlation analysis between simulated solum thickness and key environmental variables (Fig. 4). Because the initial soil production function (P0) is mathematically driven by the EEMT framework, the final results are structurally expected to correlate with both primary climatic and ecological inputs.

As anticipated, MAT (r=0.66) and elevation (r=-0.66) emerge as the primary correlates of the regional solum thickness gradient. This strong coupling serves as an internal validation of our modeling framework. It confirms that the broad, regional-scale thermodynamic forcing is preserved through the steady-state mass balance equations. Vegetation exerts an equally important control. NDVI (r=0.45) shows a clear positive correlation with simulated solum thickness, confirming that areas with denser vegetation cover experience reduced erosion rates and therefore support greater solum accumulation. This relationship validates the multi-process erosion framework, particularly the role of the hydraulic and aeolian weighting coefficients in dampening erosion under higher vegetation conditions.

Conversely, the near-zero plateau-scale correlation with general curvature (r=-0.01) should not be interpreted as evidence of negligible topographic influence. Rather, it indicates that curvature-related mass redistribution fluctuates rapidly around zero at the hillslope scale, and this high-frequency signal is statistically masked by the immense, low-frequency climatic gradients when aggregated across the entire plateau.

At the hillslope scale, however, topography-driven differentiation remains clearly visible. This is well illustrated in the transect analysis across the Hengduan Mountains (Fig. 5), a region of extreme relief and vertical zonation. Here, solum thickness varies systematically with terrain position: steep ridges act as sediment sources where active erosion restricts accumulation, whereas gentler slopes and valley bottoms function as depositional zones. Consequently, a distinct inverse relationship is observed (Fig. 5c), where solum thickness peaks in the warm, vegetated valley bottoms and diminishes rapidly towards the cold, barren ridges. These spatial patterns confirm that terrain attributes successfully drive local differentiation within the model, even though their signal is averaged out at the plateau scale.

The accuracy of spatially explicit soil models on the QTP has long been constrained by the scarcity and bias of observational data. As evidenced by the distribution of the 552 soil profiles used in this study (Fig. 1a), sampling is predominantly clustered in accessible, lower-elevation regions, while high-altitude zones, which are characterized by cryogenic conditions, low atmospheric pressure, and rugged terrain, remain significantly undersampled. This spatial imbalance poses a fundamental challenge for empirical and machine learning approaches (e.g., the Shangguan and Liu maps), which operate on the assumption that the training data statistically represent the entire domain. When trained on biased datasets where hillslopes are underrepresented, these models tend to overfit the characteristics of flat terrain, leading to poor generalization in complex topography. This bias explains the artifacts observed in the reference datasets, such as the noisy fragmentation in the Qaidam Basin (Fig. 7e) or the artificial layered patterns in the Qiangtang Plateau (Fig. 8d).

Despite utilizing the same biased observational dataset for calibration, the revised mass balance model offers an important advantage that stems from its reliance on physical mechanisms of pedogenesis rather than statistical correlation. By grounding the simulation in the mass balance equations (Eqs. 1 and 2), the model reduces dependence on dense field observations, and the structural constraint forces solum thickness to respond to topographic curvature and slope-dependent erosion, even under uneven training data distributions. Specifically, even if the model parameters are calibrated using data primarily from flat areas, the inclusion of the gravitational erosion term ensures that the model systematically removes solum from convex ridges and accumulates it in concave valleys, even in areas with no calibration data. This physically constrained behavior allows the model to infer plausible spatial patterns in undersampled regions and reproduce realistic geomorphic patterns (e.g., sharp basin-ridge boundaries in the Qaidam Basin and elevation-dependent gradients in the Qiangtang Plateau).

While the mass balance framework ensures physical consistency, the accuracy of the simulation remains partially contingent on the quality of the environmental forcing data and internal model parameters. A primary source of uncertainty stems from the empirical RUSLE and RWEQ erosion rates. These models were originally developed for low-altitude temperate zones and may have reduced applicability on the QTP due to its extreme altitude, widespread permafrost, frequent freeze-thaw cycles, and distinct surface conditions.

To quantify how uncertainty in these inputs propagates into simulated solum thickness, we independently perturbed Ewater and Ewind within ±40 % of their baseline values using Latin Hypercube Sampling (LHS). This perturbation range was selected because previous studies have reported typical uncertainties of 30 %–60 % when these empirical models are applied outside their original calibration domains, particularly in cold, high-altitude, or permafrost-affected regions (Wei et al., 2025; Schürz et al., 2020). A total of 150 ensemble simulations were generated.

As shown in Fig. 9, the coefficient of variation (Fig. 9a), 5th–95th percentile uncertainty range (Fig. 9b), and standard deviation (Fig. 9c) consistently indicate higher uncertainty in the central and western permafrost core. The mean relative change between the ensemble mean and the baseline simulation (Fig. 9d) remains small over most of the plateau, suggesting that the baseline result is generally representative. Sobol total-order sensitivity indices (Fig. 9e and f), which measure the total contribution of an input (including its interactions with other inputs) to the variance of the simulated solum thickness, reveal that aeolian erosion exerts a stronger influence on output uncertainty across large areas, especially in arid and permafrost zones, while hydraulic erosion contributes more in the humid southeastern margins.

Despite these limitations, these RUSLE and RWEQ datasets remain the best available spatially explicit erosion products for the QTP. Previous validations have demonstrated their reliability, with reported coefficients of determination of 0.93 for aeolian erosion and 0.81 for hydraulic erosion against ¹³⁷Cs measurements (Teng et al., 2018, 2021). Moreover, our model structure mitigates these input uncertainties through cluster-based calibration of the weighting coefficients (α, β, γ). By allowing these coefficients to adjust during calibration, the inverse modeling process functions as a physically constrained bias correction. Consequently, while the model relies on the spatial pattern of the erosion inputs, their magnitudes are rigorously scaled to better match local observations.

We further evaluated the sensitivity of simulated solum thickness to the erosion weighting coefficients through a One-At-A-Time (OAT) analysis by applying ±20 % perturbations to α, β, and γ based on the calibrated baseline map. The overall response is small across most of the plateau (Fig. 10). Detailed cluster-level elasticity coefficients and summary statistics are provided in Fig. C1 and Table C1. Plateau-wide elasticities, defined as the ratio of the relative change in simulated solum thickness to the relative change in the input parameter, remained consistently low (α≈-0.01, β≈-0.07 to -0.08, γ≈-0.01 to -0.02), indicating that the model is robust to reasonable variations in these parameters. A value close to zero indicates low sensitivity, while negative values indicate that solum thickness decreases as erosion increases. However, sensitivity exhibits spatial variation: gravitational weighting (γ) shows relatively higher influence in high-relief areas (e.g., Hengduan Mountains), while aeolian weighting (β) is more sensitive in the arid interior and permafrost zones (Clusters 7 and 8).

Overall, the combined uncertainty and sensitivity analyses demonstrate that the revised mass balance model produces stable solum thickness estimates under reasonable perturbations of both empirical inputs and internal parameters. The elevated uncertainty observed in the permafrost core primarily reflects limitations in the empirical erosion forcing rather than instability in the model itself.

Although the revised mass balance model provides a physically consistent approach to mapping solum thickness across the QTP, its application relies on several important assumptions that warrant critical discussion.

The first assumption is the use of a steady-state framework. Our solution for solum thickness (Eq. 9) assumes that soil production is balanced by surface erosion over long timescales (Dietrich et al., 1995). On the tectonically active QTP (Ding et al., 2019; Pan et al., 2022), this assumption is an idealization. Soil systems are continuously adjusting in response to ongoing uplift (Zhao et al., 2023; Wei et al., 2025), climatic fluctuations, and permafrost degradation (Li et al., 2012). Localized non-linear processes, such as landslides, debris flows, and retrogressive thaw slumps (Luo et al., 2022), can also cause rapid changes in solum thickness that are not captured by the current linear diffusion-based formulation (Dixon et al., 2009; Dietrich et al., 1995; Roering et al., 1999).

To partially address the limitations of the steady-state assumption, we introduced adjustable weighting coefficients (α, β, γ). These coefficients should be interpreted as effective scaling factors rather than direct proxies for deviations from geomorphic equilibrium. As demonstrated in our uncertainty and sensitivity analyses (Sect. 4.2), they absorb multiple sources of uncertainty, including biases in the empirical erosion estimates (Ewater and Ewind), unresolved transient dynamics, and scale mismatches between processes.

A second important limitation arises from the implicit representation of cryogenic processes. Potential soil production is estimated using the EEMT framework driven by MAT and MAP. While this provides a robust thermodynamics basis for regional patterns, the formulation does not explicitly account for freeze-thaw cycles and active layer dynamics, which are critical in permafrost regions. Similarly, the empirical RUSLE and RWEQ erosion inputs do not fully account for the complex interactions between seasonal freeze-thaw cycles, ice-rich permafrost, and sediment mobility. As quantified by our LHS analysis, these combined omissions introduce additional uncertainty that is most pronounced in the permafrost core. Further research should therefore prioritize the development of QTP-specific cryogenic parameterizations for both weathering and erosion.

The partitioning of the regolith into solum and saprolite (hc=DTB-hs) is structurally constrained by the global depth-to-bedrock product of Shangguan et al. (2017). Although this dataset provides the best available estimate of the total weathered mantle, it is limited by the scarcity of deep borehole data across much of the plateau interior. To evaluate the influence of potential DTB uncertainty on our results, we applied a ±40 % perturbation to the DTB values. Even under a -40 % perturbation scenario, solum thickness is truncated by the DTB constraint in only 0.001 % of the total simulated area. This demonstrates that uncertainties in the DTB dataset have negligible impact on the reported solum thickness, which is derived independently through the erosion-production balance (Eq. 9).

Looking forward, coupling the current mass balance framework with landscape evolution models and time-dependent permafrost simulations would enable a more explicit representation of transient responses. Such integration would allow the simulation of how solum thickness may evolve under future climate scenarios, including potential deepening of the active layer and changes in erosion regimes associated with permafrost thaw.