the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 Oct 2024
Diagnosing Aerosol-Meteorological Interactions on Snow within the Earth System: A Proof-of-Concept Study over High Mountain Asia
Abstract. Snowmelt in the Third Pole, particularly in High Mountain Asia (HMA), is strongly influenced by interactions between aerosols and meteorology. However, understanding these interactions remains uncertain due to their complexity and limitations in existing approaches using model sensitivity and process-denial experiments. In addition, these interactions are insufficiently represented in current climate models. Equally ambiguous is the impact of these interactions on snow processes in the context of climate change. Here we use network theory to identify key variables that influence non-linear processes within snowmelt using daily data for the late snowmelt season (May–July). We combine statistical and machine learning methods using observational and model data, to highlight the underappreciated relevance of coupled processes between aerosols and meteorology on snow, as well as the inconsistent representation of aerosol-meteorology interactions on snow within major reanalyses, reflective of differences in model design. In particular, dust interactions with near-surface temperature and large-scale circulation are underrepresented, as well as gaps in cloud cover interactions especially in the least coupled reanalysis. Carbonaceous aerosols and large-scale circulation emerge as main drivers of aerosol-meteorology onto snow interactions, highlighting their relevance in Earth system models (ESMs) for the accurate assessment of water availability in developing economies. These diagnoses point to the degree of complexity of these interactions and their relative strength of representation across ESMs. The proposed framework can thus be extended to diagnose other complex Earth system processes, providing a pathway for improving Earth system predictability and reducing climate change uncertainties.
- Preprint
(4062 KB) - Metadata XML
-
Supplement
(1631 KB) - BibTeX
- EndNote
Status: open (until 10 Jan 2025)
-
RC1: 'Comment on egusphere-2024-2298', Anonymous Referee #1, 19 Nov 2024
reply
General impressions
In this work, the authors use statistical and machine learning methods to assess how well reanalysis-derived aerosol, surface and meteorological predictor variables can predict MODIS-derived snow cover fraction in high mountain Asia. They focus on the relative importance of different non-linear interaction terms between the predictors and snowmelt during May-June. From this information, they make suggestions on which aerosol/meteorological interaction processes models can improve on to better model snow-related processes in this region.This seems like a very interesting approach that really could be useful to other applications, as the authors suggest. Their method can provide clues on what processes to prioritize for future study, either in models or with field studies. I found the scientific results to be very interesting, assuming they didn't misattribute something due to noise. However, while they discuss uncertainties broadly in section 4.3, I do believe that they need a more complete discussion of the uncertainties in the paper when they discuss individual findings. Also, while my impression of the quality of the work is generally favorable, I am not an expert in machine learning methods and so it would be helpful to have the complementary opinions of another reviewer with that expertise.
Otherwise, the figures are attractive and fairly easy to understand. It is well-written overall, but the introduction seemed too theoretical and could be made more succinct to encourage readers to really grasp what is new and important in this study. The abstract also doesn't quite mirror section 4 in terms of the implications of the study. Also, while I was impressed with the paper's thoroughness in some places, the methods are incompletely described in some places (see more details below), such that I believe that the study would not be easy to replicate as currently written.
Specific comments
More discussion of the uncertainties is merited:
- Please provide information on how good the predictors (e.g., dust) are in the reanalyses. Has there been any validation for these predictors in this region? It may be worth noting in the text that the mountainous terrain can make these values less certain. It is also important to more clearly note that a low predictability could either be due to low importance of the factor or poor reanalysis skill of the factor.
- Relatedly, in Table 2: Please discuss the role of data noise in the results from this table and implications for the interpretation of it.
- Fig. 4: In the text, it might be helpful to discuss the implications on the results of missing/not including any key variables in the analysis.
- Figure 6: I'd like to see a more detailed discussion of possible uncertainties in this figure.
- section 3.3.4. Very interesting scientific results, assuming the authors didn't misattribute something due to noise in the data. Authors, please address this point in the text.Clarification on the methods is also requested:
- Section 2: Please explain where were these reanalysis output data were taken from? Near the site? Over the site?
- Discussion on Equation 1: The code for the project is not publicly available yet, and so more detail needs to be provided here for others to replicate this work. For example, what software and method/packages were used in this part of the analysis? Did the authors take steps against over-fitting the data?
- L680: "The codes for MATCHA’s model framework, the regression algorithms, and the analysis will be made available in a public repository after publication." It would be better if these were available for the publication. Without that, or a much more detailed description of the methods, I don't think people will be able to replicate these results, as is the standard for most journal articles.
- L224. "Both metrics are calculated to add up to 100%, making it easier to interpret these sensitivities." This is actually not very easy to interpret, at least not for me. For example, in Fig. 3, it looks like carbonaceous aerosols predict ~25% of the variability in circulation, but that is not actually correct, as I understand it. I think more explanation here and in the text discussing Fig. 3 is necessary for readers to not accidentally be misled by this approach.
- L241: "non-linear interaction terms defined as product terms between these predictors (253 in total)" Please provide more information here. As written, it is not clear what the 253 predictors are, how they were derived, and what they are meant to represent. This was somewhat further explained in section 2.7, but details are still lacking. For example, it would be helpful to explain where the 231 value comes from (Line 274). Please also define a "pairwise interaction contribution."Other specific comments:
L583: "remain the primary driver of SCF ( 60% contribution)" I don't think the authors can say that 60% of the driving factors are MMI. What they can say is that 60% of the predictable variability in SCF comes from MMI in their analysis. It would be helpful to clarify this throughout section 4.
Section 4: The confidence in their methods in this section seems a bit overblown. It is not that I disagree too much with the main points, but I feel that greater emphasis on the uncertainties would be appropriate. For example, on L608: the authors say, "Although both MERRA-2 and MATCHA incorporate some degree of coupling within their models, interactions of dust with circulation variables need more attention within the two." I would instead have said, "Although both MERRA-2 and MATCHA incorporate some degree of coupling within their models, our study suggests that interactions of dust with circulation variables need more attention within the two."
L201. It would be helpful to refer the reader to Supplemental Table 2 when mentioning the predictors.
Supplementary Fig. 1a: These figures are confusing, because it is hard for a reader to orient in terms of location and area. I recommend adding information on latitude, longitude and a length scale bar. Fig. 1b is very good.
Intro: might be good to mention the fraction of the planet that is covered by the Third Pole for context
It would be helpful to define "network theory" in the abstract when it is first discussed since this is not a term many in the field will be familiar with.
L240: what is C2? please define.
L 448: "in MERRA-2, second-order interactions between DU and AOD with circulation and temperature variables are absent" Please describe some examples of such potentially important second-order interactions
L. 501: "We see that interactions between DU and PRECIP in MERRA-2 and MATCHA are given unnecessary importance, while for ERA5/CAMS4, it is the interactions between DU and SKT that are overemphasized." Is there any supporting evidence for this hypothesis in the literature?
L580: "We substantiated the importance of AMI on snow in driving SCF variability..." The wording here seems a bit strong. To substantiate the importance of something to me means that you are proving something, but what I think the authors meant (and what I think it more accurate) is that they provided additional information in support of this importance.
Section 4.2: In addition, in principle, if the method is as accurate as they think it is, could be used to help focus field campaigns to get the most bang-for-the-buck. But it would be good to get supporting literature evidence for their approach working first, as they mention in Section 4.3.
Great that they have supplementary table 1. That's a good resource and important to have available.
Section 4.3 is really great, and I am glad they added that in.
Sentence starting on L. 360: this is a really important point, and I am glad the authors included it.
Technical comments
I really like that they provided an appendix with acronyms and I used it a lot when I was reviewing the paper. But there were a few things missing.
- Figure 1. Since the authors have an appendix for acronyms, I suggest they add the definitions of the abbreviations for the individual predictors there as well, and reference that in the figure caption.
- R22 is not defined in the text or in the appendix. Presumably it is Roychoudhury et al., 2022?
- Please add RGI to the abbreviations list.L44: Did the authors mean, "and potentially leading to the misattribution of the relevant drivers to snowmelt"?
L. 63: Two sentences right after each other with lists seems overwhelming. Consider re-writing.
The intro seemed to ramble a bit. There is the opportunity to improve the paper by getting to the point more succinctly in this section.
Equation 1: why is YSCF^s,t used instead of Y^s,t? on Line 213, Y is defined as SCF. And just Y is used in equations 2 and 3.
Equation 1: Please indicate what the double O in term 3 means/represents.L240: also reference supplemental Table 2.
Table 1: please define mu and sigma in the Table caption, and not just in the text.
Figure 3. Please describe what "Others" means in the caption (or just replace "Others" with "AOD550+Sea salt")
Figure 3. In the figure, the word "contribution" is too vague. Please clarify what this means in the caption or use more specific wording in the figure.
Citation: https://doi.org/10.5194/egusphere-2024-2298-RC1
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 116 | 43 | 11 | 170 | 33 | 4 | 4 |
- HTML: 116
- PDF: 43
- XML: 11
- Total: 170
- Supplement: 33
- BibTeX: 4
- EndNote: 4
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1