| 02 Sep 2026
Precipitation phase-partitioning schemes introduce substantial uncertainty into simulations of the permafrost thermal regime over the Tibetan Plateau
Abstract. Precipitation phase-partitioning schemes determine how precipitation forcing is divided into snowfall and rainfall at each time step in land surface modelling. However, their influence on simulated permafrost thermal regimes remains poorly quantified. In this study, we use the one-dimensional transient permafrost model CryoGridLite to assess the uncertainty in permafrost simulations induced by different schemes over the Tibetan Plateau during 1951–2024. By keeping the meteorological forcing dataset, model configurations, and input parameters identical, we implemented eight commonly used schemes, including binary threshold schemes, temperature-ramp schemes, piecewise schemes, and wet-bulb/humidity-dependent schemes, allowing the effect of precipitation phase assignment to be isolated. Our simulations indicate that precipitation phase-partitioning schemes produce substantial differences in snowfall and rainfall forcing, which propagate into simulated ground thermal regimes. Across the ensemble-defined permafrost area, 83.4 % of the area shows an inter-scheme range in mean annual ground temperature at 10m depth exceeding 0.5 °C, and 48.9 % exceeds 1.0 °C. Uncertainty in simulated thaw depth is more spatially localized than that in mean annual ground temperature, but 47.2 % and 18.3 % of the ensemble-defined permafrost area still show inter-scheme thaw depth ranges exceeding 0.25 and 0.5 m, respectively. These thermal differences further affect threshold-based permafrost diagnosis. Differences among schemes lead to measurable changes in diagnosed permafrost area, with the largest diagnostic uncertainty occurring in marginal permafrost regions where simulated ground temperatures are close to 0 °C. Although precipitation phase assignment is commonly treated as a preprocessing step in meteorological forcing preparation, our controlled experiments show that the choice of scheme can generate pronounced differences in simulated ground thermal states and permafrost area diagnostics. Phase partitioning should therefore be considered a previously underexamined forcing-related source of uncertainty in high-elevation permafrost thermal simulations.
This manuscript presents an investigation into the uncertainty introduced by precipitation phase-partitioning (PPP) schemes into numerical simulations of the permafrost thermal regime across the Tibetan Plateau (TP). Using the one-dimensional numerical permafrost model CryoGridLite driven by the China Meteorological Forcing Dataset (CMFD v2.0) over the period 1951–2024, the authors evaluate eight widely used PPP formulations (binary threshold, linear ramp, piecewise, and wet-bulb/humidity-dependent schemes). The paper is well structured, timely, and addresses an overlooked aspect of permafrost modeling. While PPP uncertainty has been explored extensively in catchment hydrology and snowpack modeling, its downstream propagation into ground thermal regimes and threshold-based permafrost diagnostics has rarely been isolated. The authors highlight a notable regional physical mechanism: over the TP, schemes that partition more precipitation into snowfall lead to colder ground thermal regimes, primarily due to warm-season snow persistence and the associated albedo cooling effect, which contrasts with the classic winter snow-insulation warming observed across much of the Arctic. Overall, I consider the manuscript suitable for publication after revision, provided that the authors address the comments below.
1. The Methods state that TD is reported only for grid cells where MAGT10m is below 0 °C, whereas Section 3.3 states that TD is evaluated using the same fixed ensemble-defined permafrost mask used for the MAGT10m analysis. Please clarify how TD is treated when a grid cell belongs to the ensemble-defined permafrost mask but is diagnosed as non-permafrost in one or more individual PPP experiments. Are these values treated as missing values, excluded from the range calculation, or handled in another way? This choice could directly affect the estimated TD uncertainty, particularly in marginal permafrost regions where the PPP schemes most frequently differ in permafrost diagnosis. The authors should carefully check the calculation procedure and ensure that the descriptions in the Methods, Results, and Fig. 4 caption are fully consistent.
2. Figure 7 shows that the high-snowfall schemes produce greater warm-season snow depth, higher surface albedo, greater upward shortwave radiation, and lower ground surface temperature than the low-snowfall schemes. These results support an important role of warm-season snow persistence and associated shortwave radiative effects in linking PPP differences to simulated ground temperatures. However, the manuscript further states that the differences are “primarily attributable” to this pathway and describes it as the “dominant scheme-induced signal”. Because the complete surface energy balance is not shown, the relative contributions from longwave radiation, sensible and latent heat fluxes, ground heat flux, rainfall infiltration, soil-moisture changes, and other related processes cannot be quantitatively separated. I therefore suggest that the authors either (i) provide additional energy budget diagnostics to support the relative importance of this pathway, or (ii) moderate the corresponding statements and describe the results as suggesting an important role of warm-season snow persistence and shortwave radiative effects.
3.Figure 1: The map legend refers to the “Third Pole region boundary”, whereas the manuscript consistently describes the study domain as the Tibetan Plateau, and Table 1 refers to a “Tibetan Plateau Boundary Dataset”. Please use consistent terminology.
4.Lines 178-180: Please provide brief explanations of ΔT, ΔS, and T0 in Eqs. (6)-(8). In addition, please clarify how relative humidity (RH) was calculated, since CMFD2.0 is described as providing specific humidity rather than relative humidity, and specifies the source of the surface elevation used in the Ding scheme.
5.Line 413: “Appendix D1” appears to be an incorrect cross-reference. Please change this to “Fig. D1” or “Appendix D”, as appropriate.
6.Equation (A1): Please check whether the contribution of air to volumetric heat capacity has been intentionally neglected. The model description includes an air volume fraction, and the volumetric heat capacity of air (ca), is listed in Table C3, whereas the term θaca is absent from Eq. (A1).
7.Around Eq. (A10): The text states that both km,sat and ko,sat “are given by”, whereas Eq. (A10) only defines km,sat. Since the ko,sat has been provided in Table C3, please revise the text accordingly.
6.Table C3: Please verify the listed ice density, ρi=1000 kg m−3. If using the same density for liquid water and ice is an intentional simplification in CryoGridLite, please state this explicitly or provide the corresponding model reference.
8.Figure D1: CMFD2.0 shows a pronounced change in precipitation during the early part of the record, particularly for total precipitation and precipitation occurring under near-freezing conditions. Please verify that this feature is indeed present in the original CMFD2.0 dataset.
There are some minor grammatical issues in the manuscript, like in Line 480, “a new soil thermal conductivity and heat capacity scheme ... that account for the effect of gravel” should be changed to “... that accounts for the effect of gravel”. Please check the grammatical again throughout the manuscript.