Validation of Pan-Arctic Soil Temperatures in Modern Reanalysis and Data Assimilation Systems

Herrington, Tyler C.; Fletcher, Christopher G.; Kropp, Heather

doi:https://doi.org/10.5194/tc-2022-5

Preprints

https://doi.org/10.5194/tc-2022-5

Preprints

25 Jan 2022

Status: this preprint is currently under review for the journal TC.

Validation of Pan-Arctic Soil Temperatures in Modern Reanalysis and Data Assimilation Systems

Tyler C. Herrington¹, Christopher G. Fletcher¹, and Heather Kropp² Tyler C. Herrington et al.

Show author details

¹Department of Geography and Environmental Management, University of Waterloo, 200 University Ave., Waterloo, Ontario, Canada, N2L 3G1
²Environmental Studies Program, Hamilton College, 198 College Hill Road, Clinton, 13323, New York, U.S.A.

Received: 04 Jan 2022 – Discussion started: 25 Jan 2022

Abstract. Reanalysis products provide spatially homogeneous coverage for a variety of climate variables in regions where observational data are limited. However, very little validation of reanalysis soil temperatures in the Arctic has been performed to date, because widespread in situ reference observations have historically been unavailable there. Here we validate pan-Arctic soil temperatures from eight reanalysis and Land Data Assimilation System (LDAS) products, using a newly-assembled database of in situ data from diverse measurement networks across Eurasia and North America. We find that most products have soil temperatures that are biased cold by 2–7 K across the Arctic, and that biases and Root Mean Square Error (RMSE) are generally largest in the cold season. Monthly mean values from most products correlate well with in situ data (R > 0.9) in the warm season, but show lower correlations (r = 0.6–0.8), in many cases, over the cold season. Similarly, the magnitude of monthly variability in soil temperatures is well captured in summer, but overestimated by 20 % to 50 % for several products in winter. The suggestion is that soil temperatures in reanalysis products are subject to much higher uncertainty when the soil is frozen and/or when the ground is snow-covered. We also validate the ensemble mean of all products, and find that when all seasons, and metrics are considered, the ensemble mean generally outperforms any individual product in terms of its correlation and variability, while maintaining relatively low biases. As such, we recommend the ensemble mean soil temperature product for a wide range of applications – such as the validation of soil temperatures in climate models, and to inform models that require soil temperature inputs, such as hydrological models, or for permafrost simulations.

Download & links

Preprint (PDF, 22872 KB)

Supplement (413 KB)

Tyler C. Herrington et al.

Status: final response (author comments only)

RC1:
'Comment on tc-2022-5', Anonymous Referee #1, 11 Feb 2022
General comments

The authors present a comprehensive evaluation of soil temperature from eight reanalyses and LDAS products based on in-situ observations. Authors reported that the soil temperature bias is generally larger in the cold season, indicating the presence of soil freeze/thaw transition and snow layer introduced additional difficulties for soil temperature simulations. Authors also found that the ensemble generally outperforms any individual product. These findings improved the understanding of the current body of knowledge on soil temperature simulations in land surface models. However, a large part of the manuscript remained unfocused and an in-depth discussion is missing. I hence strongly suggest authors reformulate and shorten the manuscript (maybe as a brief communication) with a very specific focus on soil temperature validation.

Manuscript structure and discussion

In Sec. 4 & 5, the authors present the evaluation results together with a large part of the discussion, and additional discussions are given in Sec. 6. This makes the manuscript very unclear and difficult to follow.

The discussion in Sec 6 is very general and superficial. Most of the part has been extensively discussed in other studies, and they are not tightly connected with the previous parts. For instance, the gab of site-scale observation and model grid (about 10–100 km), or so-called scale effects, is widely reported. P23, L392–398, this part is very confusing. Does the misclassification of permafrost affect the results? Please make sure only to present the most relevant parts here to avoid diluting your real contributions.

Soil temperature standard deviation

The authors presented and discussed the reanalysis soil temperature deviation. I am wondering why this is important here and how this could be used for validation purposes? The strong variation of soil temperature in the cold season could be expected due to the presence of a snow layer, see Figure 6 from Burke et al., (2020).

Climatology

Again, the climatology based on the ensemble results is somehow unfocused. The purpose of this study is "validation of pan-Arctic (and Boreal) soil temperatures from eight reanalyses and land data assimilation system (LDAS) products." (see P2, L53–54), rather than analyzing the climatology. To be more focused, authors could compare and evaluate the trend of ensemble results with site-scale observations. Otherwise, I do not see the necessity of this part.

Specific comments:

P2, L24: Permafrost carbon and climate warming loop are complex, and thus ...could act as a "possibly/potentially" positive...

P2, L31: Qinghai-Tibetan Plateau.

P2, L45–49: Ensemble simulation has also been used for permafrost simulation, for instance, Cao et al., (2019), although these studies do not directly use the soil temperature.

P4, L122: The variation of soil temperature is complex and typically depends on surface condition (i.e., snow layer, vegetation), soil properties (i.e., soil organic content), and soil depth. It could vary very large at the hourly and daily scales.

P6, L135: How much the difference could be? Could you please write it down?

P6, L141: ...2 to 12...

P6, L142: The so-called "scale effects" has been widely reported, see Gubler el al., (2011) for the Alps and Cao et al., (2019) for high latitudes. Please cite relevant references.

P8, L172: you have two "also" here

P8, L180: Then why not directly use the IPA map? You could also find the global permafrost zonation index map from Gruber et al., (2012).

P9, L192: "more" → greater/larger

P14, L250: Qinghai-Tibetan Plateau

P16, L256: Zero curtain period is heavily dependent on the soil moisture rather than the active layer thickness.

P22, L357: Remove the redundant ')'.

Tables & Figures

Table 1: Could you please also add the soil discretization information here, such as depth for each layer and the total soil column depth? Please double-check the spatial resolution of all the reanalyses, ERA5 should be 0.25°, ERA-Interim is 0.75°, and MERRA-2 is 0.5°×0.625°. Depending on the datasets you used, JRA-55 is 1.25° for the reanalysis level and 0.56° for the model level.

Figure 4: Do you really need so many sub-plots? The inter-comparisons among different reanalyses are shown here but not discussed in the main text. Did I miss something important? Please also add the 1:1 line, so that readers could clearly see the cold/warm bias.

Figure S3: Could you please improve the resolution of Figure S3?

References

Burke, E. J., Zhang, Y., and Krinner, G.: Evaluating permafrost physics in the Coupled Model Intercomparison Project 6 (CMIP6) models and their sensitivity to climate change, The Cryosphere, 14, 3155–3174, https://doi.org/10.5194/tc-14-3155-2020, 2020.

Cao, B., Quan, X., Brown, N., Stewart-Jones, E., and Gruber, S.: GlobSim (v1.0): deriving meteorological time series for point locations from multiple global reanalyses, Geosci. Model Dev., 12, 4661–4679, https://doi.org/10.5194/gmd-12-4661-2019, 2019.

Gruber, S.: Derivation and analysis of a high-resolution estimate of global permafrost zonation, The Cryosphere, 6, 221–233, https://doi.org/10.5194/tc-6-221-2012, 2012.

Gubler, S., Fiddes, J., Keller, M., and Gruber, S.: Scale-dependent measurement and analysis of ground surface temperature variability in alpine terrain, The Cryosphere, 5, 431–443, https://doi.org/10.5194/tc-5-431-2011, 2011.
Citation: https://doi.org/10.5194/tc-2022-5-RC1
- AC1:
  'Reply on RC1', Tyler Herrington, 29 Aug 2022
  The authors would like to thank Referee 1 for their helpful comments. We are planning a substantial restructuring of the Results (Sections 4 and 5) and the Discussion/Conclusion (Section 6) to shorten its length and remove superficial discussion material. We are also planning to rework Section 5 to instead investigate trends in the ensemble mean soil temperature against a subset of stations with longer temperature records.
  
  Major Comments
  Reformulate and shorten the manuscript (maybe as a brief communication) with a very specific focus on soil temperature validation
  
  The authors are not aware of an option for a brief communication style manuscript in The Cryosphere. Instead, the revised manuscript will be substantially rewritten in response to the reviewer comments. We are shortening Section 4 (Results) by removing discussion material and moving it to Section 6 (Discussion/Conclusion). Section 5 will be revised, with focus on comparing soil temperature trends in the ensemble mean soil temperature product against a subset of stations with longer soil temperature records. Section 6 will be shortened, by including only the most relevant details that describe uncertainties in the analysis related to instrumental uncertainty, modeling uncertainty, and sampling uncertainty.
  In Sec. 4 & 5, the authors present the evaluation results together with a large part of the discussion, and additional discussions are given in Sec. 6. This makes the manuscript very unclear and difficult to follow.
  
  We will separate the discussion material in Section 4 and 5 and move it instead to Section 6. Within Section 6 we plan to more concisely describe how model uncertainty, instrumental uncertainty and sampling uncertainty could impact our results (see our response above and for Major Comment #3 for further information).
  The discussion in Sec 6 is very general and superficial, and is not tightly connected to previous sections. For instance, the gap of site-scale observation and model grid (about 10–100 km), or so-called scale effects, is widely reported. P23, L392–398, this part is very confusing. Does the misclassification of permafrost affect the results? Please make sure only to present the most relevant parts here to avoid diluting your real contributions.
  
  We thank the reviewer for their helpful suggestions on Section 6. Our plan is to move pertinent discussion material from Sections 4 and 5 to Section 6. By doing so, we believe that the discussion will more directly tie-in with the results. In addition, we are planning to show how our estimate of instrumental uncertainty is qualitatively similar to the findings of studies exploring scale effects in frozen soils (e.g. Gubler et al., 2011; Morse et al., 2012; Gisnås 2014; Cao et al., 2019) (see our response to Minor Comment #5 for more details).
  As the number of validation grid cells in the discontinuous permafrost zone is substantially smaller than those in the continuous permafrost zone (particularly so when we use the Obu et al. (2019) permafrost zonation), the authors do not believe that any potential misclassification of permafrost in this region would fundamentally alter the conclusions of the study. We will be sure to shorten this section, and point out that the results appear not to be impacted by any potential misclassifications of permafrost in the discontinuous permafrost region.
  
  The authors presented and discussed the reanalysis soil temperature deviation. I am wondering why this is important here and how this could be used for validation purposes? The strong variation of soil temperature in the cold season could be expected due to the presence of a snow layer, see Figure 6 from Burke et al., (2020).
  
  The authors thank the reviewer for bringing this figure to our attention. We use the normalized standard deviation as a measure of the temporal variance of soil temperatures across the grid cells. Doing allows us to assess the range of simulated soil temperatures at each grid cell for a particular product, to see if it can capture a similar seasonal cycle of temperatures to that of the observations.
  When we describe soil temperature variability in the cold season we are referring to two main features – first that the individual products themselves show a larger variance in soil temperatures than they do during the warm season. Second, we are also describing the spread in soil temperature variance between products. Figure 6 in Burke et al. (2020) clearly shows how differences in snow insulation between products could contribute for the spread in soil temperatures over the winter, and we will include a reference to Burke et al. (2020) in our discussion section describing this.
  While snow cover could very well account for the larger variance in soil temperatures within a product over the cold season – Figure 6 in the Burke et al. (2020) paper is unable to fully explain why an individual product’s cold season soil temperature variance is often substantially larger than its warm season variance. We are planning to explore the impact of snow cover on soil temperatures in reanalysis products in a follow up paper.
  The climatology based on the ensemble results is somehow unfocused. The purpose of this study is "validation of pan-Arctic (and Boreal) soil temperatures from eight reanalyses and land data assimilation system (LDAS) products." (see P2, L53–54), rather than analyzing the climatology. To be more focused, authors could compare and evaluate the trend of ensemble results with site-scale observations.
  
  We agree that a focused evaluation of the trends against a subset of the stations with longer timeseries would be of value, and will restructure Section 5 to instead validate the trends in the ensemble mean soil temperature against station observations.
  P8, L180: Then why not directly use the IPA map? You could also find the global permafrost zonation index map from Gruber et al., (2012).
  
  The authors thank the reviewer for this suggestion. We have incorporated the Obu et al. (2019) permafrost zonation index map into our analysis. Figure R1 compares permafrost zonation using the Smith and Riseborough (2002) air temperature method and the Obu et al. (2019) index. The Smith and Riseborough (2002) method appears to overestimate the number of permafrost grid cells relative to Obu et al. (2019), however this difference does not fundamentally alter the conclusions of our study.
  
  Minor Comments
  P2, L24: Permafrost carbon and climate warming loop are complex, and thus ...could act as a "possibly/potentially" positive…
  
  We will revise this statement to read as: “Continued warming, and thawing of permafrost soils, and related decomposition of carbon could act as a potential positive feedback on warming.”
  P2, L31: Qinghai-Tibetan Plateau
  
  Corrected.
  P2, L45–49: Ensemble simulation has also been used for permafrost simulation, for instance, Cao et al., (2019), although these studies do not directly use the soil temperature
  
  We have added a reference to Cao et al. (2019) in our introduction where we describe the use of ensemble mean datasets in the literature.
  P4, L122: The variation of soil temperature is complex and typically depends on surface condition (i.e., snow layer, vegetation), soil properties (i.e., soil organic content), and soil depth. It could vary very large at the hourly and daily scales.
  
  The authors thank the reviewer for this helpful comment. As we are using soil temperatures averaged between 0cm and 30cm in the near surface, and between 30cm and 300cm at depth, we had presumed that the soil temperatures should show reduced variation on daily and hourly timescales. We will revise this sentence to read “Many of the in situ (station) sites reported measurements at hourly or daily frequency, however we chose to perform the analysis at monthly time scales, in order to focus on processes controlling the seasonal cycle of soil temperatures.”
  P6, L135: How much the difference could be? Could you please write it down?
  
  The authors presume that the reviewer is asking by how much the soil temperatures may vary between stations within a grid cell. Panel B of Figure 1 gives an estimate of the variability of soil temperatures within a grid cell – most stations have a median spatial standard deviation of ~2^oC, however soil temperatures may vary by as much as 7^oC in some cases. The variability is of a similar magnitude to previous studies exploring sub-grid scale variability in cryospheric soil temperatures (e.g. Gubler et al., 2011; Morse et al., 2012; Gisnås 2014; Cao et al., 2019).
  P6, L141: ...2 to 12..
  
  We chose to write 2 as “two”, since style conventions in The Cryosphere specify that all single digit numbers (unless they are followed by units) should be written as a word.
  P6, L142: The so-called "scale effects" has been widely reported, see Gubler el al., (2011) for the Alps and Cao et al., (2019) for high latitudes. Please cite relevant references.
  
  We thank the reviewer for pointing out several relevant references on scale effects. In our revisions, we plan to link our estimates of scale effects with those in the literature, and show that our results qualitatively agree with those exploring scale effects in seasonally frozen and permafrost soils (e.g. Gubler et al., 2011; Morse et al., 2012; Gisnås 2014; Cao et al., 2019) – see our response to Minor Comment # 5 for more details.
  P8, L172: you have two "also" here
  
  Corrected.
  P9, L192: "more" → greater/larger
  
  Revised.
  P14, L250: Qinghai-Tibetan Plateau
  
  Corrected.
  P16, L256: Zero curtain period is heavily dependent on the soil moisture rather than the active layer thickness
  
  The authors thank the reviewer for this helpful comment. We will remove the following sentence here: “In regions where the active layer is deep (such as over the discontinuous permafrost zone), the zero curtain period is often longer and more pronounced (Chen et al., 2021) than it is over the continuous permafrost zone or regions with seasonally frozen soil.”
  The section will now read: “Thirdly, latent heat interactions in the active layer during spring can lead to long periods of time where the soil remains at or close to freezing - the zero-curtain period. Many of the processes that control the zero-curtain effect, such as freeze-thaw parameterizations are relatively simplistic in many land models (Cao et al., 2020; Chen et al., 2015), and their coarse resolution would fail to capture local scale variations in the zero-curtain period.”
  P22, L357: Remove the redundant ')'
  
  We will remove the extra bracket here.
  Table 1: Could you please also add the soil discretization information here, such as depth for each layer and the total soil column depth? Please double-check the spatial resolution of all the reanalyses, ERA5 should be 0.25°, ERA-Interim is 0.75°, and MERRA-2 is 0.5°×0.625°. Depending on the datasets you used, JRA-55 is 1.25° for the reanalysis level and 0.56° for the model level
  
  We will fix the spatial resolutions in the table for ERA5, ERA-Interim and MERRA2. For JRA55, we used data at the model resolution of 0.56^o. We will also add another column to the table to include information about the soil depths included.
  
  Figure 4: Do you really need so many sub-plots? The inter-comparisons among different reanalyses are shown here but not discussed in the main text. Did I miss something important? Please also add the 1:1 line, so that readers could clearly see the cold/warm bias
  
  The most important comparisons to be made here are the performances of the individual products against the station – the outer margins of Figure 4 in the paper – which is the focus of the text. We also used the histograms of the warm/cold season to look at the variability of soil temperatures in the warm and cold season. We will include a paired down figure in the revisions - similar to the one below (Figure R2).
  
  Figure S3: Could you please improve the resolution of Figure S3?
  
  We will make sure to improve the resolution of Figure S3 in the revisions.
  References:
  Cao, B., Quan, X., Brown, N., Stewart-Jones, E., and Gruber, S.: GlobSim (v1.0): deriving meteorological time series for point locations from multiple global reanalyses, Geosci. Model Dev., 12, 4661–4679, https://doi.org/10.5194/gmd-12-4661-2019, 2019.
  Gisnås, K., Westermann, S., Schuler, T. V., Litherland, T., Isaksen, K., Boike, J., and Etzelmüller, B.: A statistical approach to represent small-scale variability of permafrost temperatures due to snow cover, The Cryosphere, 8, 2063–2074, https://doi.org/10.5194/tc-8-2063-2014, 2014.
  Gubler, S., Fiddes, J., Keller, M., and Gruber, S.: Scale-dependent measurement and analysis of ground surface temperature variability in alpine terrain, The Cryosphere, 5, 431–443, https://doi.org/10.5194/tc-5-431-2011, 2011.
  Morse, P. D., Burn, C. R., and Kokelj, S. V.: Influence of snow on near-surface ground temperatures in upland and alluvial environments of the outer Mackenzie Delta, Northwest Territories., Can. J. Earth Sci., 49, 895–913, https://doi.org/10.1139/e2012-012, 2012.
  
  Citation: https://doi.org/10.5194/tc-2022-5-AC1
RC2:
'Comment on tc-2022-5', Anonymous Referee #2, 30 Jun 2022

General comments

The authors perform a comprehensive validation of current reanalysis and LDAS model products using a set of in situ soil temperature measurements across the northern hemisphere. Validation of soil temperatures in large-scale reanalysis data products is an important task given their frequent use in global ecological and hydrology models and permafrost simulations, for example. It is crucial that the underlying biases are clearly acknowledged by the users of such products. A novelty of this study is that it validates multiple reanalysis soil temperature products at the same time and covers the entire pan-Arctic area.

The authors find that most reanalysis products are considerably cold-biased and that the biases and inter-model variability are larger during the cold season compared to the warm season. The authors also compile an ensemble from the assessed reanalysis data and show how it interestingly overperforms most/all individual models. The authors list potential future applications of the ensemble mean product, but I would wish to see a bit more discussion on its current usability, given that the recorded biases remain quite high and display some regional patterns. The underlying reasons for these are addressed in the manuscript but not how the biases would affect, e.g., permafrost simulations where a bias or RMSE of above 2° C can have notable implications.

The article is very well written and figures are great. At places the text is hard ro follow (especially Section 4.3, see detailed comments below) owing to the multiple simultaneous comparisons: near surface vs. at depth soil temperatures, cold season vs. warm season, permafrost vs. no to little permafrost, North America vs. Eurasia, and DJF vs. JJA. I suggest the authors to make sure all sections are clearly defined.

I recommend the publication of this manuscript after the authors have considered my minor suggestions and comments.

Specific comments

L104: The authors suggest that their study is "To the authors’ knowledge, this one of the first studies to compile a comprehensive set of in situ soil temperature measurements across the Eurasian and North American Arctic, from multiple diverse sparse networks". While it may be true that this is true for the “one of the first” part, it should be noted that the compilation is not totally novel, given that similar in situ temperature datasets have been compiled not only by Cao et al. (2020, in the references) but also, e.g., by Karjalainen et al. (2019) and Ran et al. (2022) who used mostly the same data sources, albeit computing temperatures averages for a much larger depth (several meters deep in permafrost but also in non-permafrost soils). Moreover, Lembrechts et al. (2020) have published a global soil temperature compilation of soil and near-surface temperatures. I suggest the authors to consider if their statement needs some elaboration, e.g., does the compiled dataset differ from previous datasets in some ways.

The authors recognize the notably different sampling size for North America but retain from explaining why no more data were collected, apart from mentioning the overall data scarcity in northern Canada, to correct the imbalance between North America and Eurasia. Based on the previous data compilations (see above), there should be suitable measurement time series available from North America. Notwithstanding, the authors satisfactorily show how the sampling imbalance does not affect the fundamental conclusions (LL378-391).

LL140-141: "Panel B of Figure 1 shows the spatial standard deviation of monthly surface soil temperatures for grid cells with more than two stations included.” However, in Figure 1b, grid cells with two stations are also shown. Also, I remain unsure whether there are any grid cells with >1 stations in Eurasia?

L236: Reference should be to Fig. S1, right?

LL 236-238: "The mean bias and RMSE are typically 1â¦ C to 3â¦ C smaller over North America, relative to the permafrost zone in Eurasia (see Figure S3); however with fewer grid cells over North America, the uncertainty is also larger - as evidenced by the larger error bars.” I cannot see the mentioned difference between North America and Eurasia in Figure S3 (biases in ERA5-Land) or in the associated bar graph (ERA5-Land) in Figure S2. However, all products considered the said difference between regions is visible. Consider checking and editing the text so that it corresponds to the results shown in Figure S3.

L239: What correlations, the ones between measurements and reanalysis temperatures? A slight elaboration would help the reader to see that what are compared in the sentence.

LL240-241: I also struggled with this sentence. What is the opposite situation here? It is hard to follow the comparisons between permafrost and little to no permafrost, as well as near-surface and at depth temperatures at the same time, especially since the results are not shown.

LL243-246: Are these results related to the permafrost binning? It’s fine if they are not, but overall Section 4.3 is at times hard to follow because it deals with both permafrost binning and regional comparisons.

L405: Instead of the cold season standard deviations, should you not refer here to cold stations/observations? That is, figure 6 does not distinguish between warm and cold season.

L261: The ensemble mean product is not properly addressed until deep into the results (validation) section. I suggest presenting the ensemble mean product and its calculation procedure already in the early stages (possibly inside section 2.1.).

L303: I find “coastal regions” not the ideal term here because the regions with the highest variability are far more than that. In winter, greatest variation associates with the coldest regions, yet not exclusively either. Could the variation here be related to snow cover duration / snow thickness as mentioned elsewhere in the text?

Technical corrections

L61: Please, open the abbreviation GLDAS-CLSM already here.

LL80-83: Check grammar of the sentence. Maybe delete the word "that" at line 81?

L191: Figure 2 does not have panels C and D.

Figure 3: This is a nice figure with lots of information in it. The letters in “Correlation coefficient” are clumped together and could be corrected.

Figure 4: Stratification of the values in histograms is not explained. Please add it to the caption.

Figure 5: Y-axis is a bit messy. Consider adjusting the interval at which temperatures are denoted.

Figure 8: DJF missing from Panel A label.

L286: NH à northern hemisphere

L290: Why are ensemble mean at depth temperatures not shown? Could be part of the supplement. Figure 9 also shows at depth results, so it would be interesting to see how the models reconstruct frozen ground in JJA, although it is acknowledged that this is not explicitly representative of permafrost.

L366: Please put Gruber et al. 2018 inside parentheses.

L369-370: "Moreover, the impact of snow cover on soil temperature is generally more pronounced over permafrost regions (regions of seasonal frost).“ Is something missing here? Should it be "compared to regions of seasonal frost" or what is the idea?

LL418-419: Could you elaborate, what does it mean "is being explored"?

L428: Please provide a url for the ensemble mean dataset on the ADC.

L583: Database title and url missing.

References

Karjalainen et al. 2019 Circumpolar permafrost maps and geohazard indices for near-future infrastructure risk assessments. Scientific Data https://doi.org/10.1038/sdata.2019.37

Lembrechts et al. 2020 SoilTemp: A global database of near-surface temperature. Global Change Biology https://doi.org/10.1111/gcb.15123.

Ran et al. 2022 New high-resolution estimates of the permafrost thermal state and hydrothermal conditions over the Northern Hemisphere. Earth System Science Data https://doi.org/10.5194/essd-14-865-2022

Citation: https://doi.org/10.5194/tc-2022-5-RC2
- AC2:
  'Reply on RC2', Tyler Herrington, 29 Aug 2022
  The authors would like to thank Referee 2 for their helpful comments. As a part of our revisions, we have gathered substantially more data for North America, and have recalculated all metrics. In our updated database, we now have 135 validation grid cells over North America; 30 of which are located over the permafrost region. By utilizing soil temperature data from a variety of hydrometeorological and agricultural monitoring networks, our dataset now provides the most comprehensive analysis to date of soil temperatures across northern and southern Canada and the Great Lakes basin.
  
  Major Comments
  The authors list potential future applications of the ensemble mean product, but I would wish to see a bit more discussion on its current usability, given that the recorded biases remain quite high and display some regional patterns. The underlying reasons for these are addressed in the manuscript but not how the biases would affect, e.g., permafrost simulations where a bias or RMSE of above 2° C can have notable implications.
  
  We thank the reviewer for this comment. Several of the products have an RMSE of ≤4^oC – particularly over permafrost regions (as shown in S1). In most products, this is expressed as a cold bias, which would suggest that reanalysis products may overestimate permafrost extent and underestimate active layer thickness. The ensemble mean biases and RMSE are generally better than (or similar to) the best performing product, especially when all seasons and depths are considered. In addition, the ensemble mean soil temperatures show a more realistic pattern of soil temperature variability in the permafrost zone compared to the individual products themselves.
  The ensemble mean product provides gridded, monthly-averaged soil temperature estimates of near surface, and deeper soil temperatures at a 1^o resolution. Therefore, it is most suitable to regional or hemispheric-scale analyses of soil temperature climatologies, or their seasonal cycle, or to explore recent trends in soil temperatures (since 1980). The product could also be used to provide boundary conditions for hydrological models. In fact, a higher resolution version of this product (see our response to Question 12 in Minor Comments) is being used for such a purpose and will be described in a follow-up study.
  The authors acknowledge that the ensemble mean soil temperature product would most likely yield an overestimation of permafrost extent, given that it is biased cold by 3-5^oC, on average, at high latitudes. That being said, over permafrost regions, the RMSE of the ensemble mean product outperforms the RMSE the best performing product by ~2^oC, on average, and hence it may still provide some added value for estimation of high latitude soil temperatures relative to the individual products.
  
  At places the text is hard to follow (especially Section 4.3, see detailed comments below) owing to the multiple simultaneous comparisons: near surface vs. at depth soil temperatures, cold season vs. warm season, permafrost vs. no to little permafrost, North America vs. Eurasia, and DJF vs. JJA. I suggest the authors to make sure all sections are clearly defined.
  
  We will restructure the results section to make things clearer. The general delineation will remain the same, with a pan-Arctic assessment (section 4.1 – Warm Season and section 4.2 – Cold Season), and a permafrost-focused section (section 4.3), as we believe that this is a natural way to delineate the results, but we will try to more logically separate the near-surface results from the results at depth. We will also separate out the regional results, especially now that we have a greater number of validation grid cells over North America (making the comparisons between Eurasia and North America more meaningful).
  
  L104: The authors suggest that their study is"To the authors’ knowledge, this one of the first studies to compile a comprehensive set of in situ soil temperature measurements across the Eurasian and North American Arctic, from multiple diverse sparse networks". While it may be true that this is true for the “one of the first” part, it should be noted that the compilation is not totally novel, given that similar in situ temperature datasets have been compiled not only by Cao et al. (2020, in the references) but also, e.g., by Karjalainen et al. (2019) and Ran et al. (2022) who used mostly the same data sources, albeit computing temperatures averages for a much larger depth (several meters deep in permafrost but also in non-permafrost soils). Moreover, Lembrechts et al. (2020) have published a global soil temperature compilation of soil and near-surface temperatures. I suggest the authors to consider if their statement needs some elaboration, e.g., does the compiled dataset differ from previous datasets in some ways.
  
  The authors recognize the notably different sampling size for North America but retain from explaining why no more data were collected, apart from mentioning the overall data scarcity in northern Canada, to correct the imbalance between North America and Eurasia. Based on the previous data compilations (see above), there should be suitable measurement time series available from North America.
  The authors thank the reviewer for making us aware of these studies. As a result, the biggest change in the revised manuscript is the inclusion of a large amount of new soil temperature data from North America. Figure R1 compares the previous and updated distribution of validation grid cells, which now contains 135 validation grid cells over North America near the surface; 30 of which are located over the permafrost region. This means that our sample of sites for North America is now more comparable to the 247 grid cells in Eurasia (45 of which span the permafrost region).
  The new data are drawn from multiple sources, and we reiterate our claim from the original manuscript that this collection of pan-Northern Hemisphere soil temperature data constitutes a novel and important contribution to the permafrost research community. Over the permafrost region, we’ve assembled data from the Yukon (Yukon Geological Survey, 2021) and the NWT (Cameron et al., 2019; Ensom et al., 2019; Gruber et al., 2019; GTN-P, 2018; Spence and Hedstrom, 2018a; Spence and Hedstrom, 2018b; Street, 2018).
  In addition, we have incorporated data from several soil monitoring and hydrometeorological networks across Southern Canada and the Great Lakes basin of the United States, that, to our knowledge, are not included in any of the above papers. These include 85 stations from the Manitoba Mesonet network (RoTimi Ojo and Manaigre, 2021), 83 stations in Michigan and western Wisconsin (MAWN, 2022), 31 stations from the Alberta Climate Information Service network (Alberta Agriculture, Forestry and Rural Economic Development, 2022), and 150 stations from North Dakota (NDAWN, 2022). We are also including data from a peatland ecosystem in Metro Vancouver (Lee et al., 2017; Lee et al., 2021), as well as data from 11 stations in central and Northern BC (Déry, 2017; Hernández-Henríquez et al., 2018; Morris et al. 2021), and 2 stations in southern Quebec (Arsenault, 2018; Fortier, 2020).
  We have also been in contact with the data providers from the Real Time In-Situ Monitoring Network (RISMA), however the data was not available to include at the time this response was submitted. We hope to include the RISMA dataset (which includes 13 stations in southern Manitoba, 6 stations in southeastern Ontario, and 4 stations in southern Saskatchewan) in follow-up studies.
  While the Ran et al. (2022) study included borehole measurements from southern Canada, the data did not include information about the seasonal cycle of soil temperatures. Thus, our work presents the most comprehensive analysis to date of soil temperatures across northern and southern Canada and the Great Lakes basin.
  
  Minor Comments
  LL140-141:"Panel B of Figure 1 shows the spatial standard deviation of monthly surface soil temperatures for grid cells with more than two stations included.” However, in Figure 1b, grid cells with two stations are also shown. Also, I remain unsure whether there are any grid cells with >1 stations in Eurasia?
  
  This was a typo. It should read “Panel B of Figure 1 shows ... grid cells with two or more stations included.”
  Based on the grid cells that met our criteria for validation, there were no grid cells in Eurasia with two or more stations included. A clarification has been added to the text.
  
  L236: Reference should be to Fig. S1, right?
  
  L236 mentions that “several factors may explain the increased variability in soil temperatures over permafrost regions.” We presume that you may have meant L226, which describes the difference in the mean bias/RMSE over North America versus Eurasia?
  Figure S1 displays the mean bias and RMSE over the combined Pan-Arctic permafrost zone. Here we meant to refer to Figure S2, which shows the difference in bias between Eurasia and North America. We will make this correction in the revisions.
  
  L239: What correlations, the ones between measurements and reanalysis temperatures? A slight elaboration would help the reader to see that what are compared in the sentence.
  
  Yes, these are referring to correlations between the observed soil temperatures and the reanalysis temperatures. We will rephrase this sentence to “Correlations between observed soil temperatures and soil temperatures in the reanalysis products are generally quite similar in both the permafrost region and zone with little to no permafrost.”
  
  LL240-241: I also struggled with this sentence. What is the opposite situation here? It is hard to follow the comparisons between permafrost and little to no permafrost, as well as near-surface and at depth temperatures at the same time, especially since the results are not shown.
  
  Here we were referring to the fact that the reanalysis products are more likely to overestimate the observed variance over the permafrost region at depth. We will split L240-241 into two sentences. “Individual models are more likely to overestimate the near surface soil temperature variance over the zone with little to no permafrost. At depth, however, reanalysis products are more likely to overestimate the variance over the permafrost region.”
  
  LL243-246: Are these results related to the permafrost binning? It’s fine if they are not, but overall Section 4.3 is at times hard to follow because it deals with both permafrost binning and regional comparisons.
  
  We have revised Section 4.3 to focus entirely on permafrost regions so that this, and other, ambiguities in this section are corrected.
  
  L405: Instead of the cold season standard deviations, should you not refer here to cold stations/observations? That is, figure 6 does not distinguish between warm and cold season.
  
  What we were describing here is that when soil temperatures are frozen (and particularly for soil temperatures below –20^oC), soil temperature standard deviations increase to near 10^oC in several products. We will make this clear that we are explicitly referring to frozen soil conditions.
  
  L261: The ensemble mean product is not properly addressed until deep into the results (validation) section. I suggest presenting the ensemble mean product and its calculation procedure already in the early stages (possibly inside section 2.1.).
  
  As suggested, we will add some further details about the methodology used to create the ensemble mean soil temperature product in Section 3 (Methods).
  
  L303: I find “coastal regions” not the ideal term here because the regions with the highest variability are far more than that. In winter, greatest variation associates with the coldest regions, yet not exclusively either. Could the variation here be related to snow cover duration / snow thickness as mentioned elsewhere in the text?
  
  We agree that “coastal regions” does not adequately describe the spatial pattern of variability – particularly in winter – and we have changed the description to “colder regions” in the revised text. Figure R2 shows a scatterplot of the relationship between mean annual air temperature (MAAT) and soil temperature standard deviation, when soil temperature variance is largest. The figure shows that soil temperature standard deviation and MAAT have a moderately strong negative correlation of -0.69. Moreover, it appears that regions with extreme continentality (such as eastern Siberia) show the largest standard deviations. While it is possible that snow cover characteristics may be important in certain regions, a detailed snow cover analysis is beyond the scope of this paper – and will be the focus of a follow-up paper.
  
  Technical Corrections
  L61: Please, open the abbreviation GLDAS-CLSM already here.
  
  We will expand GLDAS-CLSM to read “Global Land Data Assimilation System – Catchment Land Surface Model” here.
  
  LL80-83: Check grammar of the sentence. Maybe delete the word "that" at line 81?
  
  This sentence should say “In ERA5, a weak coupling exists between the land surface and atmosphere. It includes an advanced LDAS 80 that incorporates information regarding the near-surface air temperature, relative humidity, as well as snow cover (de Rosnay et al., 2014), along with satellite estimates of soil moisture and soil temperature from the top 1m of soil (de Rosnay et al., 2013).”
  
  L191: Figure 2 does not have panels C and D.
  
  This sentence should read “Warm season biases tend to be slightly larger at depth (Figure 2, Panel B) for most products (by 1^oC – 2^oC).”
  
  Figure 3: This is a nice figure with lots of information in it. The letters in “Correlation coefficient” are clumped together and could be corrected.
  
  This will be improved in the revised version.
  
  Figure 4: Stratification of the values in histograms is not explained. Please add it to the caption.
  
  The data in the histogram are stratified based on the Berkeley Earth Surface Temperature (BEST) 2m air temperature. The cold season (blue) refers to soil temperatures occurring when the air temperature is ≤ -2^oC, while the warm season (red) refers to soil temperatures when the air temperature is > -2^oC. We will ensure that this is included in our revisions.
  It is also now apparent that there is a typo in this caption. It should have read “Seasons are stratified by the BEST air temperature, with the cold season (≤ -2◦C) in green and the warm season (>-2◦C) in red.”
  
  Figure 5: Y-axis is a bit messy. Consider adjusting the interval at which temperatures are denoted.
  
  We will change the Y-Axis to include major ticks at every 2◦C, which should hopefully declutter the Y-Axis.
  
  Figure 8: DJF missing from Panel A label.
  
  Figure has changed in response to other revisions: captions/labels will be corrected as appropriate.
  
  L286: NH -> northern hemisphere
  
  We will expand NH to read “northern hemisphere” here.
  
  L290: Why are ensemble mean at depth temperatures not shown? Could be part of the supplement. Figure 9 also shows at depth results, so it would be interesting to see how the models reconstruct frozen ground in JJA, although it is acknowledged that this is not explicitly representative of permafrost.
  
  Our decision to not include the results at depth was because the pattern correlations were quite similar to those near the surface (with a pattern correlation of ~0.95 over the study area). The overall features were generally quite similar, however showing a smaller annual range of temperatures. We have added a sentence explaining our reasoning behind the decision to focus on the near surface, but will include the depth results as a supplemental figure.
  
  L366: Please put Gruber et al. 2018 inside parentheses.
  
  We will change “Gruber et al. (2018) to “(Gruber et al., 2018)”.
  
  L369-370:"Moreover, the impact of snow cover on soil temperature is generally more pronounced over permafrost regions (regions of seasonal frost).“ Is something missing here? Should it be "compared to regions of seasonal frost" or what is the idea?
  
  We agree that this sentence is confusing. It should read “Moreover, the impact of snow cover on soil temperature is generally more pronounced over permafrost regions relative to regions of seasonal frost.”
  
  LL418-419: Could you elaborate, what does it mean "is being explored"?
  
  Using a similar blending methodology, we have been investigating the performance of a 0.31-degree product (using a smaller subset of products that provide data at higher spatial resolution). We have also performed similar analyses with a 0.05-degree soil temperature product, using interpolated soil temperatures from the Arctic System Reanalysis version 2 (ASR), ERA5-Land, and the Famine Early Warning Systems Network (FLDAS). The goal has been to assess the impact of spatial resolution on performance of the ensemble mean product. We are hoping to include these results in a follow-up paper.
  
  L428: Please provide a url for the ensemble mean dataset on the ADC.
  
  The original version we submitted had all URLs as hyperlinks. We see that the hyperlinks are not present in the version available online, so we will be sure to include a URL for the dataset in our revisions.
  
  L583: Database title and url missing.
  
  We will be sure to correct this and include a proper link to the Arctic Data Center repository for the Kropp dataset.
  
  References:
  Agriculture and Agrifoods Canada. Real Time In-Situ Monitoring for Agriculture. (Retrieved: July 2022 from: https://agriculture.canada.ca/SoilMonitoringStations/historical-data-en.html)
  Alberta Agriculture, Forestry and Rural Economic Development, Alberta Climate Information Service (ACIS). Current and Historical Alberta Weather Station Data. (Retrieved: July 2022 from: https://acis.alberta.ca).
  Cameron, E. A., Lantz, T. C., O’Neill, H. B., Gill, H. K., Kokelj, S. V., and Burn, C. R.: Permafrost Ground Temperature Report: Ground temperature variability among terrain types in the Peel Plateau region of the Northwest Territories (2011-2015), Northwest Territories Geological Survey, Northwest Territories, Canada, 2019.
  Déry, S.: Cariboo Alpine Mesonet (CAMnet) Database,. Zenodo. https://doi.org/10.5281/zenodo.1195043 , 2017.
  Ensom, T., Kokelj, S. V., and McHugh, K. K.: Permafrost Ground Temperature Report: Inuvik to Tuktoyaktuk Highway stream crossing and alignment sites, Northwest Territories, Northwest Territories Geological Survey, Northwest Territories, Canada, 2020.
  Gruber, S. Derivation and analysis of a high-resolution estimate of global permafrost zonation. The Cryosphere, 6(1), 221–233. https://doi.org/10.5194/tc-6-221-2012. 2019.
  Gruber, S., Brown, N., Stewart-Jones, E., Karunaratne, K., Riddick, J., Peart, C., Subedi, R., and Kokelj, S. V.: Permafrost Ground Temperature Report: Ground temperature and site characterisation data from the Canadian Shield tundra near Lac de Gras, Northwest Territories, Canada, Northwest Territories Geological Survey, Northwest Territories, Canada, 2019.
  Lee, S., Black, T. A., Nyberg, M., Merkens, M., Nesic, Z., Ng, D., and Knox, S. H.: Biophysical Impacts of Historical Disturbances, Restoration Strategies, and Vegetation Types in a Peatland Ecosystem, J Geophys Res Biogeosci, 126, https://doi.org/10.1029/2021JG006532, 2021.
  Lee, S.-C., Christen, A., Black, A. T., Johnson, M. S., Jassal, R. S., Ketler, R., Nesic, Z., and Merkens, M.: Annual greenhouse gas budget for a bog ecosystem undergoing restoration by rewetting, Biogeosciences, 14, 2799–2814, https://doi.org/10.5194/bg-14-2799-2017, 2017.
  Hernández-Henríquez, M. A., Sharma, A. R., Taylor, M., Thompson, H. D., and Déry, S. J.: The Cariboo Alpine Mesonet: sub-hourly hydrometeorological observations of British Columbia’s Cariboo Mountains and surrounding area since 2006, 18, 2018.
  MAWN.: Michigan Automated Weather Network. (Retrieved: July 2022 from: https://mawn.geo.msu.edu/)
  Morris, J., Hernández-Henríquez, M., and Déry, S.: Cariboo Alpine Mesonet meteorological data, 2017-2021. Zenodo. https://doi.org/10.5281/zenodo.6518969, 2021.
  NDAWN.: North Dakota Mesonet Network. (Retrieved: July 2022 from: https://ndawn.ndsu.nodak.edu/).
  Obu, J., Westermann, S., Bartsch, A., Berdnikov, N., Christiansen, H. H., Dashtseren, A., et al. Northern Hemisphere permafrost map based on TTOP modelling for 2000–2016 at 1 km2 scale. Earth-Science Reviews, 193, 299–316. https://doi.org/10.1016/j.earscirev.2019.04.023. 2019.
  RISMA.: Real-Time In-Situ Soil Monitoring for Agriculture. (Retrieved: July 2022 from: https://agriculture.canada.ca/SoilMonitoringStations/)
  RoTimi Ojo, E. and Manaigre, L.: The Manitoba Agriculture Mesonet: Technical Overview, Bulletin of the American Meteorological Society, 102, E1786–E1804, https://doi.org/10.1175/BAMS-D-20-0306.1, 2021.
  Spence, C. and Hedstrom, N.: Baker Creek Research Catchment Hydrometeorological and Hydrological Data, 2018a.
  Spence, C. and Hedstrom, N.: Hydrometeorological data from Baker Creek Research Watershed, Northwest Territories, Canada, Earth Syst. Sci. Data, 10, 1753–1767, 2018b.
  Street, L. E., Mielke, N., and Woodin, S. J.: Phosphorus Availability Determines the Response of Tundra Ecosystem Carbon Stocks to Nitrogen Enrichment, Ecosystems, 21, 1155–1167, https://doi.org/10.1007/s10021-017-0209-x, 2018.
  Streletskiy, D. A., Shiklomanov, N. I., & Nelson, F. E. Spatial variability of permafrost active-layer thickness under contemporary and projected climate in Northern Alaska. Polar Geography, 35(2), 95–116. https://doi.org/10.1080/1088937X.2012.680204. 2019.
  Tao, J., Koster, R. D., Reichle, R. H., Forman, B. A., Xue, Y., Chen, R. H., and Moghaddam, M.: Permafrost variability over the Northern Hemisphere based on the MERRA-2 reanalysis, The Cryosphere, 13, 2087–2110, https://doi.org/10.5194/tc-13-2087-2019, 2019.
  Yukon Geological Survey: Yukon permafrost reports data. In: Yukon Permafrost Database. Government of Yukon, 2021.
  
  Citation: https://doi.org/10.5194/tc-2022-5-AC2

Tyler C. Herrington et al.

Supplement

https://doi.org/10.5194/tc-2022-5-supplement

Data sets

Ensemble Mean Reanalysis Soil Temperature Dataset (1981–2018) Tyler Herrington and Christopher G. Fletcher https://doi.org/10.18739/A2RN3085P

Tyler C. Herrington et al.

Viewed

Total article views: 561 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
398	139	24	561	48	16	15

HTML: 398
PDF: 139
XML: 24
Total: 561
Supplement: 48
BibTeX: 16
EndNote: 15

Views and downloads (calculated since 25 Jan 2022)

Month	HTML	PDF	XML	Total
Jan 2022	72	21	3	96
Feb 2022	84	27	6	117
Mar 2022	48	17	4	69
Apr 2022	37	9	0	46
May 2022	43	9	1	53
Jun 2022	26	12	1	39
Jul 2022	21	9	2	32
Aug 2022	28	17	4	49
Sep 2022	12	9	1	22
Oct 2022	18	4	2	24
Nov 2022	9	5	0	14
Dec 2022	0

Cumulative views and downloads (calculated since 25 Jan 2022)

Month	HTML	PDF	XML	Total
Jan 2022	72	21	3	96
Feb 2022	84	27	6	117
Mar 2022	48	17	4	69
Apr 2022	37	9	0	46
May 2022	43	9	1	53
Jun 2022	26	12	1	39
Jul 2022	21	9	2	32
Aug 2022	28	17	4	49
Sep 2022	12	9	1	22
Oct 2022	18	4	2	24
Nov 2022	9	5	0	14
Dec 2022	0

Viewed (geographical distribution)

Total article views: 531 (including HTML, PDF, and XML) Thereof 531 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 02 Dec 2022

Short summary

Here we validate soil temperatures from eight reanalysis products across the pan-Arctic, and compare their performance to a newly calculated ensemble mean soil temperature product. We find that most products are biased cold by 2–7 K, especially in the cold season, and that the ensemble mean product outperforms individual reanalysis products. As such, we recommend the product for a wide range of applications – including validation of climate models, or as input to permafrost models.


Total:	0
HTML:	0
PDF:	0
XML:	0