the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 30 Apr 2026
Modeling ice rich permafrost landscapes with CLM5 using dynamically coupled tiles
Abstract. Thawing of extended amount of ground ice in permafrost regions can lead to rapid, large-scale landscape changes known as thermokarst, which significantly alter the thermal, hydrological and biogeochemistry state of the soil and the land surface. These thermokarst processes are driven by excess ground ice and permafrost microtopography. However, large-scale land surface models, used in coupled earth system models for climate predictions, do not represent such small-scale processes, and may therefore miss important mechanism that could contribute to underestimation of current greenhouses emission predictions from the permafrost regions. In this study we implement a new tiling approach in the Community Land Model, version 5.0, which is used in several Earth System Models, which already includes representation of excess ground ice, to resolve permafrost. The approach divides the vegetated land unit of a grid cell in two interacting tiles, that exchange snow, heat, and water, enabling simulation of rapid thaw processes under permafrost degradation. We evaluate this model configurations at two contrasting sites: a palsa mire landscape in northern Norway, and an ice-wedge polygon landscape in northeastern Siberia. The new implementation significantly improves the representation of soil temperature and soil moisture dynamics. It successfully captures the coexistence of two contrasting landscapes, a cold dry elevated higher tile and a warm, saturated lower tile. At the palsa sites, the tiling approach proves to be necessary to maintain stable Palsa conditions until 2014. These results demonstrate that explicitly representing excess ice and landscape dynamics in land surface models improves simulation of permafrost dynamics and may help reduce uncertainty in projections of permafrost-carbon feedbacks.
- Preprint
(2352 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-1039', Anonymous Referee #1, 28 May 2026
-
RC2: 'Comment on egusphere-2026-1039', Anonymous Referee #2, 04 Jul 2026
The manuscript is well written and provides a valuable proof of concept for representing ice-rich permafrost landscapes in CLM5. The comments below are intended to strengthen the manuscript and clarify several questions regarding the model formulation, interpretation, and potential broader applicability.
This manuscript presents a novel two-tile, dynamically coupled representation of ice-rich permafrost landscapes in CLM5. Rather than representing the vegetated portion of a grid cell as a single, uniform soil column, the authors divide it into elevated and lower landscape tiles that exchange snow, lateral heat, and lateral water. The two tiles can differ in soil stratigraphy, excess-ice content, snow cover, moisture regime, and elevation, while excess-ice melt allows subsidence and changes in relative topography over time.
The approach is a useful and promising step toward representing thermokarst-related heterogeneity in an Earth system modeling framework. However, the manuscript should clarify how tile-resolved state variables and fluxes are aggregated back to the land-unit or grid-cell scale for coupling with the atmosphere, river-routing scheme, and, ultimately, carbon-cycle calculations. It is not yet clear whether aggregation is simply area-weighted, how conservation is ensured across tiles, or how variables with different vertical structures and surface elevations are reconciled.
At present, the paper demonstrates improved representation of soil physical processes rather than improved projections of greenhouse-gas emissions. Carbon, methane, nitrogen, vegetation, and dynamic ecosystem feedbacks are intentionally left for future work. Given the broader motivation of reducing uncertainty in permafrost-carbon feedbacks, the discussion would benefit from a clearer outline of how this framework could be extended to biogeochemistry and eventually scaled to circumpolar applications.
Several key inputs remain site-specific and partly prescribed, including tile area, contact length, inter-tile distance, initial elevation difference, soil profiles, and the partitioning of excess ice between tiles. This is entirely reasonable for a site-scale proof of concept, but the manuscript should discuss how these parameters could be derived from observations, remote sensing, landform maps, or statistical parameterizations at regional and global scales.
Forcing uncertainty also remains an important limitation. At Samoylov, simulated snow depth is generally too high, and winter soil temperatures are correspondingly too warm. At Iškoras, the Mire retains permafrost despite being permafrost-free in reality, possibly because of cold biases in the forcing and/or overly strong thermal insulation in the prescribed peat profile. These mismatches do not undermine the value of the tiled approach, but they should be more explicitly considered when interpreting the apparent success of the model.
The two-tile framework is an elegant simplification, but real thermokarst landscapes often comprise more than two interacting elements, including rims, polygon centers, troughs, ponds, drainage pathways, peat plateaus, collapse scars, and lakes. Future work should evaluate whether two tiles are sufficient across different degradation stages and landscape types, or whether selected environments require additional tile classes or an adaptive tiling structure.
The three research questions posed in the Introduction are addressed through the results and sensitivity experiments. Nevertheless, the manuscript would be strengthened by a brief, explicitly structured synthesis in the Discussion or Conclusion that addresses each question in sequence and distinguishes robust general findings from site-specific or forcing- and parameter-choice-sensitive conclusions.
The lateral water-exchange scheme also deserves further discussion. In the current formulation, lateral water transfer is driven by differences in perched water-table elevation between the two tiles. This is a reasonable first-order approximation, but it conceptualizes the system as two adjacent vertical columns connected through a relatively simple conductance term.
Previous work by Harp et al. and Zlotnik et al. suggests that drainage from inundated low-centered polygons is spatially heterogeneous rather than uniformly distributed across the polygon center. Much of the lateral subsurface drainage may occur through a relatively narrow annular zone near the polygon rim. Their analyses further indicate that drainage depends on polygon geometry, seasonal thaw depth, hydraulic anisotropy, and hydraulic resistance at the polygon center–trough interface. Incorporating these controls into the lateral-flow parameterization could improve the physical realism of the Samoylov application and provide a useful pathway toward representing transitions between ponded low-centered polygons and more efficiently drained polygon states.Harp, D. R., Zlotnik, V., Abolt, C. J., Busey, B., Avendaño, S. T., Newman, B. D., Atchley, A. L., Jafarov, E., Wilson, C. J., and Bennett, K. E.: New insights into the drainage of inundated ice-wedge polygons using fundamental hydrologic principles, The Cryosphere, 15, 4005–4029, https://doi.org/10.5194/tc-15-4005-2021, 2021.
Zlotnik VA, Harp DR, Jafarov EE, Abolt CJ. A Model of Ice Wedge Polygon Drainage in Changing Arctic Terrain. Water. 2020; 12(12):3376. https://doi.org/10.3390/w12123376
Citation: https://doi.org/10.5194/egusphere-2026-1039-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 276 | 69 | 20 | 365 | 22 | 25 |
- HTML: 276
- PDF: 69
- XML: 20
- Total: 365
- BibTeX: 22
- EndNote: 25
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
line 46 You generalize that all current ESMs omit thermokarst, but coauthors have published papers using CESM/CTSM to do so. Reword to be more precise.
line 70 site CLM5, CESM, and NorESM
line 71 CLM already has microtopography and subsidence due to excess ice melt; be more specific of how this adds to the existing scheme
line 77 It is not clear that the sentence beginning "The approach..." is talking about the modified CLM, not the standard CLM that you were discussing in the previous sentences. Perhaps have a different paragraph for the two
Figure 1 is confusing. For a)-d) it is not clear what the correspondence between the reality and conceptual parts is or what the colors represent. In f) I see A1 and A2, but they are not described. Perhaps these should be two figures (a-d) and (e-f) so that you have space to more fully describe them. Similarly, the figure caption should be more descriptive.
line 90 what does 'version 2.0' refer to?
line 100 There is no maximum number of soil layers, and 54 soil layers is not the default; perhaps list the default.
line 104
The description of the CLM spatial heirarchy could be improved. Define landunits, columns, and patches. 'Landscape types' are not columns, they are landunits. CLM has a concept of microtopography; explain what it is and how you are using the term here differently.
line 110 is there a new spatial structure (tile), or just two columns? Explain your model structure in terms of figure 1 in more detail.
line 156 why is thermal conductivity called 'ksat'?
line 160 are there any soil layers that are not thermally active?
line 163 I assume the term in parentheses is a transmissivity? If so, describe that in the text and/or add an equation T=... Why is an absolute value needed (if that is what the vertical lines indicate)
line 167 the heat gradient appears to just be a height difference?
line 174 what is 'excess ice standing surface water', and why do you need to prevent it?
line 175 conceptually, this pair of tiles is embedded in a landscape of similar tiles, so where would this overflow go?
line 206 this is not the correct reference for the meteorological data. What is the spatial resolution, and what fields are used to force the model?
line 234 what are the functions of the 'maximum saturated area fraction' and the mean topographic slope in the model?
line 242 what is the 'snow accumulation factor'?
line 248 why was the period 1930-1950 chosen to spin up the model, given that earlier forcing data is available?
line 258 I did not understand the reason why one simulation started with initial subsidence, can you explain more clearly? Similarly, why would the other simulation start with an elevation of 0.38?
line 274 why does the tile with the deeper snow depth melt-out earlier?
line 277 does the phrase ' snow remains deeper at the Center than at the Rim' refer to observations or the simulation?
line 280 is there a reference for this: 'the absence of wind-driven snow removal, an important process in the Lena River delta.'? Why would one disable wind compaction?
line 293 why do you discuss Figure 3 before Figure 2? Similarly, you discuss figures from the appendix, then return to discussing figure 2 (line 316). I recommend trying to discuss each figure in turn fully before switching to another figure.
line 301 you note that the temperature differences in the model are due to differences in snow depth, but the observed palsa snow depth is similar or even smaller than the model. Why do the observations not show the same temperature difference as the model?
Figure 2a snow exists through summer at Samoylov? And snow height decreases from october to january?
Figure 2b why is there a large temperature difference for the model between rim and center in the autumn (starting around october), but it is gone by april?
Figure 2d obs show no difference in ALT, but model shows center being colder. Does this give you information about the site soil properties, i.e. perhaps the model soil properties are more different than observed?
line 333 you say REF lacks ice melting, why is that? It also has excess ice (?).
line 341 I did not understand i) and ii). For i) you are not comparing maximum values, you are comparing to the model at the same time. For ii), I only see a rim observation, not an ice-wedge observation.
line 362 you say the Rim becomes wetter but the water table is deeper. Usually a deeper water table indicates drier conditions; can you clarify?
line 376 you say 'REF fails to capture the observed soil moisture regime', but how could it capture both? It is only one column. At best it might capture one or the other, but saying it failed to capture two moisture states at the same time is not really correct.
line 380 snowmelt appears to occur in may and june. Why does surface water peak in july or august? I would think evaporation would reduce surface water after snowmelt is over. Is the increase due to summer precipitation?
Figure 5 the model Center shows surface water above a dry layer near the surface. How is that possible? Wouldn't the surface water infiltrate into the soil? REF shows similar behavior.
line 407 Can you further explain why there is greater energy input during snowmelt for the Mire?
line 417 is this statement also true for REF (not just the tiled simulations)?
line 424 where does surface runoff go? In the model, runoff can simply be removed from the system, but is surface runoff from the sites actually observed?
line 428 in satellite phenology mode (no active biogeochemistry), plants of the same type should have identical leaf area index. If vegetation ET is different, it may be due to differences in snow cover; please confirm.
line 435 is there outflow from the Center in reality? If not, can the model be parameterized to reproduce such behavior?
line 496 perhaps this is a fine point, but is the soil column split, or are you simply adding a second soil column? The latter implies no change to the model spatial heirarchy, but the former seems to add a new layer ('tile'). I would recommend using the model terminology ('column') rather than 'tile' to be clear if the number of columns is simply increased. (Similarly for line 510.) If a new spatial unit is added to the model, explain how the tile fits within the previous structure (perhaps in section 2.1.1), and why adding a tile was done rather than adding another column.
Perhaps the authors could add some discussion of how this site-level study relates to global ESM simulations. Can this implementation be used more broadly, or is it primarily useful for site-level research? For example, how could the snow redistribution scheme be used in a larger domain?