Unraveling the Complexities of Aerosol-Meteorology Interactions on Snowmelt in High Mountain Asia

doi:10.21203/rs.3.rs-3645099/v1

Download PDF

Article

Unraveling the Complexities of Aerosol-Meteorology Interactions on Snowmelt in High Mountain Asia

https://doi.org/10.21203/rs.3.rs-3645099/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Snowmelt in High Mountain Asia is heavily influenced by interactions of aerosols and meteorology. However, uncertainties persist due to the complexity of these interactions, which are typically addressed using myopic approaches and 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 present a broader strategy using network theory to attribute key quantities that influence higher-order processes within snowmelt. We combine statistical and machine learning methods using observational and model data, highlighting the underappreciated relevance of coupled processes between aerosols and meteorology on snow, as well as the inconsistent representation of aerosol-meteorology interactions within major reanalyses. We find that carbonaceous aerosols and large-scale circulation emerge as the main drivers of snow interactions, emphasizing the need for their serious consideration in integrated Earth system models for the accurate assessment of water availability in developing economies.

Earth and environmental sciences/Climate sciences/Atmospheric science

Earth and environmental sciences/Climate sciences/Atmospheric science/Atmospheric chemistry

Earth and environmental sciences/Climate sciences/Cryospheric science

Earth and environmental sciences/Climate sciences/Climate change/Attribution

Earth and environmental sciences/Climate sciences/Climate change/Climate and earth system modelling

Earth and environmental sciences/Climate sciences/Climate change/Projection and prediction

The rapid acceleration of glacial snowmelt in recent decades has a critical impact on the freshwater resources that serve the livelihood of regions downstream of these glaciers ¹. The negative trends in snow cover fraction (SCF) observed in recent decades pose an imminent risk to the glacial water towers High Mountain Asia (HMA) and are particularly vulnerable to the complex interplay between the land and atmosphere and its cascading effects, impacting snowmelt and ultimately water resources ². SCF is an essential climate variable that is particularly sensitive to climate change and modulates the surface energy balance and atmospheric circulation ^3,4. High reflectivity (albedo) of snow also contributes to feedback mechanisms between the snow surface and the atmosphere by increasing surface warming and accelerating snowmelt (loss of SCF) through positive radiative forcing ^5,6.

Past studies have attributed meteorological factors such as ambient temperature, precipitation, and topography as the primary drivers of variations in SCF and snow albedo feedbacks within HMA, often overlooking the relevance of anthropogenic emissions in the vicinity of glaciers in driving these feedbacks ^7–9. Light absorbing particles (LAPs, such as black and brown carbon (BC, BrC), and dust) are a key component of these emissions that can have a considerable impact on snow albedo feedbacks through aerosol deposition on snow, with an efficacy comparable to greenhouse gases ^10–16. The interactions between meteorology and aerosols (mostly but not limited to LAPs), or aerosol-meteorology interactions (AMI) are diverse, ranging from a) altering the Earth’s radiation budget and thereby affecting atmospheric thermodynamics (b) changing cloud microphysical properties and precipitation rate, (c) modifying snow albedo feedbacks by deposition leading to snow darkening, and (d) modifying snow properties such as specific surface area that further amplify snow-albedo feedbacks ^17–23. Combined with the spatial heterogeneity in the glaciers across HMA, the non-linear interaction of these processes can either intensify or buffer the response of the climate system, confounding the net effect of AMI on the cryosphere and misattributing the relevant drivers to snowmelt ^24,2526–29. For instance, recent studies place contrasting degrees of importance on the types of LAPs and their radiative impact on snow. Although BC has been the primary candidate of study in aerosol-induced snow albedo feedbacks within the recent decades, its relevance is currently challenged by dust and BrC.

The conventional approaches of separating these non-linear interactions in complex ESMs have often been through model ensembles by 1) perturbing model parameters and assessing their sensitivities or 2) withholding selected parameters and comparing their relative impacts ^30–37. Although valuable, the heavy computational burden of these approaches compels us to investigate other possibilities. Statistical techniques ranging from regression to emulators have been suggested to provide a more simplified understanding of these interactions albeit with their limitations ^33,38–41. Recent advancements in artificial intelligence for emulating complex systems also provide a viable alternative, although they are often perceived as black boxes introducing additional complexity and biases ^42,43. The lack of consistent geophysical observations across regions like HMA, partly driven by its remoteness and complex terrain, requires the use of reanalysis products such as ERA5/CAMS-EAC4 and MERRA-2 that provide an opportunity for understanding long-term changes in HMA. In a recent study, we attempted to address AMI by quantifying the importance of second-order interactions between aerosol and meteorological variables from ERA5/CAMS-EAC4 reanalysis to satellite-based MODIS SCF using a linear regression model ⁴⁴ (hereafter as R22). Our findings show that AMI, particularly those related to carbonaceous aerosols, holds high importance for glacial regions with low snow cover (LSC) in HMA, particularly during the late snowmelt season (May-July).

Considering the coupled interconnected processes linking snow, aerosols, and meteorology, we embrace a Darwinian perspective into their interactions inspired by Harte⁴⁵, where we attempt to reduce the complexity into specific second-order non-linear interactions between aerosols and meteorology (i.e., AMI) and quantify their significance in influencing SCF. This approach resonates with a related perspective in the field of aerosol-cloud interactions (ACI), another complex phenomenon with diverse effects at different spatio-temporal scales ^46–48. Our definition of AMI has an additional layer of complexity due to its relevance to the cryosphere and the land-air interface. While the previous research explores ACI as sensitivities of different cloud properties to cloud radiative forcing, we instead account for joint dependencies between aerosol and meteorological drivers of snowmelt through a data-driven approach to quantify AMI and its relevance for cryospheric processes in regions such as HMA.

In this study, we expand on R22 to assess the significance of AMI to SCF variability over low snow cover regions in HMA during the late snowmelt season. To characterize AMI, we use a diverse set of geophysical variables spanning both aerosols and meteorology from three major state-of-the-art reanalysis datasets (from ECMWF, NASA, and NCAR) along with satellite observations. We provide a robust estimation of this significance employing two regression-based approaches, a statistical multi-linear regression, and an explainable ML model to infer the non-linear interactions between aerosols and meteorology. Our results show a significant contribution of AMI to SCF variability (at least one-fifth), with carbonaceous aerosols and large-scale circulation being the dominant processes within AMI. We compare the importance of AMI using two different model constructs and find that the higher importance of AMI from an observational construct compared to a model construct suggests its relevance and a lack of adequate representation of AMI-related processes within the three reanalyses. Furthermore, we utilize network analysis and joint histograms to visualize the quantities interacting within AMI across the reanalyses which reflect the degree of coupled parameterizations incorporated in the respective climate models. Our analysis shows that a realistic interpretation of the attribution of snowmelt drivers is heavily reliant on how AMI is represented in an ESM, and determining the relevant quantities within coupled Earth system components is key to accurate monitoring of snow hydrology over water-vulnerable regions such as HMA.

Spatial Distribution and Seasonality of SCF

The motivation behind this analysis lies in the difference in SCF across satellites and reanalyses. In Fig. 1a, we show the average spatial distribution of SCF in the late snowmelt season across four data sources (MODIS from satellite, ERA5-Land, MERRA-2, and MATCHA as reanalyses, see Methods). SCF is very high in ERA-Land which can be attributed to excessive snowfall in the ECMWF snow model, while extremely low SCF in MERRA-2 can be attributed to the high snow depth specified in its land model to consider 100% SCF, leading to lower SCF ⁴⁹. This disparity in SCF across different datasets alludes to diverse model representations of processes driving SCF which we try to leverage in this work. SCF in MATCHA closely resembles that of MODIS, which is a possible result of CLM-SNICAR coupling within MATCHA effectively constraining SCF.

Figure 1b shows the monthly time series distribution of spatially averaged SCF across the datasets for high and low snow cover (HSC and LSC) regions (see Methods). A seasonal cycle can be observed across the four datasets, with high variability in both regions. As in Fig. 1a, ERA5-Land shows a consistently high median and interquartile range (IQR) for both regions across all months, except in the late snowmelt period (especially July), when the median is similar across the datasets. The largest differences with respect to MODIS occur in the snow accumulation months (DJF), especially in the HSC regions. The median SCF from MERRA-2 is closest to MODIS SCF across all months, especially in LSC regions but the former has a low IQR compared to the other datasets.

A cursory glance at the half-decadal trends of maximum SCF over the six glacier regions (not shown here) shows that the maximum decrease in SCF occurs in the glaciers of the Himalayas (HIM) and Kunlun Shan (KLN) which lie in the vicinity of pollution (anthropogenic LAPs emitted from the Indo-Gangetic Plain and dust emitted from the Taklamakan Desert) ^50,51.

Spatial Distribution and Seasonality of MET and AER Predictors

As in Fig. 1, we now turn our attention to the spatio-temporal distribution of selected predictors across aerosols and meteorology in Fig. 2. The selection The meteorological variables are based on prognostic variables that have been used in snow models with black carbon and dust mixing ratios at the surface being some of the most relevant aerosol quantities for snow-darkening ^52–54.

In Fig. 2a, we explore the average spatial distribution of the chosen aerosol and meteorology variables during the snowmelt period across the three reanalyses. Surface BC mixing ratios are higher in CAMS-EAC4 than in MERRA-2 and MATCHA despite emission input in MATCHA utilizing Copernicus Atmosphere Monitoring Service (CAMS) emissions (CAMS-GLOB-ANT v4.2) ⁵⁵. High values of surface BC are primarily concentrated in the vicinity of the glacier regions pointing to pollution sources from nearby Asian countries. For surface dust mixing ratios, MERRA-2 shows higher values compared to CAMS-EAC4 and MATCHA. High values are mostly concentrated in northern HMA due to its proximity to desert regions. Skin temperature is consistent across datasets, with low values in ERA5 and MATCHA over the Tibetan Plateau. The wind speed at 10 m is higher in MERRA-2 and MATCHA compared to ERA5, especially near southern HMA. Specific humidity at 2 m shows a similar distribution across datasets with higher values in MERRA-2 and MATCHA concentrated at the geographical extremities of HMA.

In Fig. 2b, we show the monthly time series distribution of these predictors across datasets for HSC and LSC regions separately. Both MET and AER variables show a pronounced seasonal cycle in both regions. Surface BC is lowest in the late snowmelt season (May-July) with a high median from CAMS-EAC4 compared to the other datasets. Surface dust is high during May-July, with a higher median and IQR from MERRA-2. SKT is consistent across all three datasets for both regions, with a strong seasonal cycle that anti-correlates to that of SCF in Fig. 1. Wind speed at 10 m however shows a weak correlation with SCF, the effect of which can be buffered by dominant variables like temperature and moisture that drive SCF ²⁵. Restating our Darwinian approach to the interaction problem, we are considering the importance of the associations between our predictors and SCF rather than their individual relationships, which can often be misleading due to buffering by the dominant drivers such as temperature or moisture.

Quantifying Importance of AMI to SCF

A brief overview of the importance approach is provided below and is extensively covered in the Methods section. We consider 22 different predictors spanning aerosols (AER), meteorology (MET), and elevation (ELEV) from the three reanalyses and regress them on SCF using 1) a statistical multi-linear regression and 2) a machine learning (ML) algorithm called eXtreme Gradient Boosting (XGBoost) (see Supplementary Methods). Interactions between AER, MET, and ELEV are defined as second-order terms in each algorithm. We focus specifically on the interacting terms between aerosols and meteorology, defined as aerosol-meteorology interactions (AMI) for our work (see Eq. 1 in Methods). Our analysis is based on the importance estimates from the regression algorithms, which denote the sensitivity of the 22 predictors and their higher-order terms to the target variable (SCF). This sensitivity is quantified by two metrics, relative importance (RI) from the multi-linear regression and Shapely contribution (SHAPc) from the ML model. We also use two model constructs on this regression framework to distinguish between the importance of AMI from an observational (Obs-Model) and reanalysis (Mod-Mod) point-of-view. The key point to note is that the target variable (SCF) in the regression is used from two sources, 1) satellite data from MODIS, and 2) each reanalysis model.

In Fig. 3, we show the RI and SHAPc distributions of AMI and MMI only in the Obs-Model (3b) and Model-Model construct (3c) for LSC regions in the late snowmelt season. RI and SHAPc importances for MMI are higher across all datasets with an average contribution of 50–70% (both RI and SHAPc) to SCF variability. AMI shows a consistent magnitude across all datasets in the Obs-Model construct with an average RI of 10–20% and SHAPc of 20–40%, indicating a significant contribution to SCF variability. In the Model-Model construct, RI and SHAPc for AMI are lower by an average of 10% than the Obs-Model construct. The variability of AMI across both constructs and datasets is higher than the difference in the average AMI importance (for both RI and SHAPc). A non-parametric Mann-Whitney test of the AMI distributions shows a significant difference (95% level) for both constructs across the three datasets. AMI is thus significant for both constructs, and the decrease in AMI in the Model-Model construct suggests second or higher-order interaction terms reflecting missing or unresolved processes that drive SCF. The large variability in the spatiotemporal distribution of SCF (Fig. 1) in the late snowmelt season, combined with the difference in AMI importance across both constructs suggests the disparity of AMI-related processes that drive SCF in each reanalysis dataset. SHAPc values for AMI are higher in both constructs compared to RI, which can be due to the ability of XGBoost to capture the non-linear interactions to a fuller extent, compared to the MLR model, where the interactions are restrictive in its definition (only non-square product terms).

Process Drivers of AMI

We further decompose the AMI importance in both the constructs (Fig. 3a and 3d) by meteorology (with five sub-groups of variables) and chemistry (with four sub-groups of variables). Among AER variables, carbonaceous aerosols contribute significantly to AMI (average 18%), surpassing dust (average 11%) in both the constructs and metrics. Within MET variables, circulation-related variables contribute the highest (average 13%) followed by cloud cover (average 10%).

The prevalence of carbonaceous aerosols can be attributed to increased surface BC and total aerosol optical depth (AOD) in the vicinity of the LSC regions during the pre-monsoon season (April-May). This includes wheat crop residue burning in the northern part of the Indian subcontinent leading to potential interactions with large-scale atmospheric circulation in the subsequent months (late snowmelt season) ^56–59. Multi-model intercomparison of global aerosols has also reported that carbonaceous aerosols contribute an average of 70% to aerosol-induced absorption, a key process in AMI ⁶⁰. A decrease in importance can be seen across the subgroups and datasets in AMI from the Obs-Model to the Model-Model construct for both metrics, which can be attributed to the absence of sufficient processes driving SCF in the model representation of the three datasets.

The use of the two metrics RI and SHAPc highlights two aspects of these interactions. SHAPc captures the bulk non-linear effect of AMI and is more consistent across datasets while RI seems to capture more of the specific or local second-order representations of AMI especially those seen in the Model-Model construct.

Emergent Connections within AMI

In Fig. 4, we show the importance of individual interactions within AMI using network diagrams in both model constructs ⁶¹. The five larger nodes (circles) represent the AER predictors while the small nodes represent the MET predictors. The connections (edges) between them are weighted (edge widths) by the pairwise importance metric (RI and SHAPc) and represent the interactions between AER and MET. In the Obs-Model construct (Fig. 4a), we observe a higher number of strong interactions (moderate - very high) whereas the Model-Model construct exhibits fewer but more specific strong interactions. A notable strong interaction consistently observed across the reanalyses in the Obs-Model construct involves total AOD and surface dust with meteorology variables related to circulation (viz. geopotential height at 300 and 500 hPa, mean sea level pressure), temperature (skin temperature), and radiation (surface sensible heat flux) corresponding to ACrI, ATI, and ARI (see Methods). Interactions are sparser in the Model-Model construct (Fig. 4b), indicating specific AMI processes captured by RI and SHAPc within each reanalysis model. Interactions with temperature (ATI) are predominant in SHAPc across the three datasets, while RI reveals interactions spread across cloud cover (ACCI), radiation (ARI), and precipitation variables (AMoI). SHAPc captures the overall (bulk) sensitivity of AMI to SCF, whereas RI focuses on the sensitivity of specific (local) second-order interactions within AMI.

For ERA5/CAMS-EAC4, RI shows strong interactions of total AOD and carbonaceous aerosols with low cloud cover (ACCI) and surface sensible heat flux (ARI). In contrast, for MERRA-2 and MATCHA, we observe interactions of carbonaceous and dust aerosols with precipitation (AMoI). SHAPc on the other hand, highlights a strong interaction of near-surface temperature with aerosol variables (ATI) in the Model-Model construct consistently across the three reanalyses.

The network diagrams reflect the complexity of each reanalysis model. The progression of importance, particularly SHAPc from ERA5/CAMS-EAC4 to MATCHA in both constructs signifies the degree of coupling incorporated in the three datasets, considering the absence of coupling between aerosols and meteorology in ERA5, and its presence in MERRA-2 and MATCHA. The ability to visualize the strength of each interaction within AMI through the network diagrams points toward relevant processes that drive SCF in the study period, which is otherwise difficult to interpret from the seasonality of the selected predictors in Fig. 2. Variables such as surface dust and wind speed at 10 m (U10 and V10) show significant differences in the late snowmelt season (from Fig. 2) indicating an apparent weak correlation with SCF which we discussed can be obscured by the buffering of dominant patterns by other predictors. An example of this can be seen in Fig. 4, where the low to moderate interactions related to wind variables at 10 m (U10 and V10) in the Obs-Model construct are not apparent, suggesting the misattribution of AMI processes related to near-surface wind in the Model-Model construct.

Observable AMI

Considering the significance of AMI in regulating SCF in the late snowmelt season for LSC regions in HMA, it would be useful to interpret what these interactions represent through relationships between the predictors and SCF. Although having multiple predictors from satellite observations would be ideal for exploring AMI to its fullest extent, we consider four such variables from satellite observations, MAIAC AOD, MODIS LST, and IMERG PRECIP, and visualize their relationship with MODIS SCF. Exploring the relationship between predictors in the Obs-Obs construct will provide a basis of ground truth for relative comparison with findings using the same predictors in the other two constructs.

In Fig. 5, we show the relationship between MODIS SCF and the predictors MODIS LST and IMERG PRECIP weighted by MAIAC AOD for LSC regions during May-July. We observe an exponential decay of MODIS SCF with MODIS LST above 0^oC, dominated by high values of MAIAC AOD, especially at higher LST and lower SCF. Such behavior points to the radiative effects of absorbing aerosols causing warmer temperatures (high LST) and accelerated snowmelt (low SCF) ⁶². To assess the sensitivity of SCF with LST at different AOD values, we use a mutual information-based metric to quantify non-linear sensitivity (shown by the bars in Fig. 5) ⁶³. The strongest sensitivity between MODIS LST and SCF occurs for low to moderate values of AOD (values within 0.07–0.17). Compared to satellite observations, MATCHA shows a similar relationship between SCF and LST compared to the other two datasets (for both Obs-Model and Model-Model constructs). Given that MATCHA is the only framework among the three datasets with coupling between snow, radiation, and LAPs, this similarity confirms the ability of parameterizations in MATCHA to represent AMI better than datasets from MERRA-2 or ERA5/CAMS-EAC4. In the Model-Model construct, MERRA-2 shows the strongest relationship between SCF and LST at moderate to very high values of AOD (0.17–5.7) suggesting an overestimated aerosol loading in LSC regions during the late snowmelt season that might contribute to the lower SCF values seen in Fig. 1 (for MERRA-2). ERA5/CAMS-EAC4 on the other hand, has stronger SCF-LST sensitivities for values below high AOD (< 0.3), which can allude to a lack of coupling within ERA5 based on aerosol radiative feedbacks in the model that translates to an absence of strong SCF-LST dependencies to the AOD distribution in CAMS-EAC4.

We also see an exponential decay between MODIS SCF and IMERG PRECIP in the Obs-Obs construct, with strong SCF-PRECIP dependency at low to moderate values of AOD (0.7–0.17). This is a potential representation of wet scavenging of absorbing aerosols by precipitation that minimizes the radiative effect of absorbing aerosols in accelerating snowmelt (thus higher SCF in the low-moderate AOD range, compared to when AOD is very high, where SCF is minimal) ⁶⁴. As seen for the SCF-LST relationship, MATCHA exhibits the most similarity in the SCF-PRECIP relationship compared to the other two datasets, with strong sensitivity of SCF to PRECIP in the low-moderate AOD range, indicating the degree of coupling in MATCHA compared to MERRA-2 and ERA5/CAMS-EAC4. MERRA-2 reflects the underestimation of SCF as seen in Fig. 1 and is dominated mostly by moderate to very high values of AOD (0.08–6.75) compared to the other datasets, suggesting overestimated AOD in LSC regions during May-July. High SCF in ERA5/CAMS is dominated by low to moderate AOD (0.04–0.08) which can indicate lower-than-usual aerosol loading in the study region during the late snowmelt period.

We also visualized the relationship between SCF and total cloud cover fraction from MODIS through both constructs, as shown in Fig. 5. We did not find a clear separation across the different segments of the AOD distribution, as in LST and PRECIP. This indicates the complexity of ACCI (and ACI) which cannot be decomposed through a direct relationship between SCF and cloud cover ²⁵. The definition of cloud cover (fraction) across the three reanalyses and MODIS is also different (see Supplementary Table 2) which makes it difficult to perform an equivalent comparison across the datasets.

Through our analysis, we substantiated the importance of AMI in driving SCF variability over low snow-covered glaciers in HMA during the late snowmelt season. A top-down Darwinian and data-driven approach is used to quantify cryospheric interactions through AMI and to unravel the non-linear relationships between aerosols and meteorology. While interactions within meteorology (MMI) are the dominant driver of SCF (~ 60% contribution), drivers related to AMI impact an average of 20% variability in SCF. We establish the robustness of this importance using two regression-based algorithms to capture the non-linear interactions within AMI. The comparison between two model constructs, one based on observations and the other on reanalysis, reveals that the importance of AMI to SCF is underestimated within the reanalyses when compared to satellites. This suggests that although AMI significantly drives snowmelt, the reanalyses fail to adequately capture these interactions. Decomposing AMI into its constituent variables further highlights the heterogeneous representations of AMI in each reanalysis model, providing a better understanding of snow-related quantities that each model emphasizes. Two aspects of the sensitivity of AMI to SCF were found 1) a bulk component due to carbonaceous aerosols and circulation, and 2) an individual (local) component due to total AOD and surface dust, and their interactions with meteorological variables related to circulation, temperature, and radiation.

Reanalysis from NCAR (MATCHA) displays the strongest resemblance to the relationships between aerosols, temperature, precipitation, and SCF from satellites, considering that the degree of coupling parameterizations interfacing the atmosphere and land (cryosphere) is highest in MATCHA due to the inclusion of feedbacks in the model between aerosol, radiation, and snow. The variation in AMI across the reanalyses is also lower than the difference from both constructs (Obs-Model to Model-Model, Fig. 3), indicating that the coupling within MATCHA is far from ideal. Thus, the need for parameterizations that represent the feedbacks between snow, aerosol emissions, and the ambient meteorological conditions is necessary to consider in the development of future ESMs.

The synergistic approach of statistical regression and machine learning proves to be a viable and more economical alternative to expensive feedback separation methods conventional in ESMs. Although the proposed methods have their own inherent biases, particularly the multi-collinearity within predictors and interpretability of machine learning algorithms, they prove to be capable of highlighting the relevance of AMI in driving SCF over HMA. The consistent importance of AMI across the algorithms and model constructs underscores the potential in our methodology to capture non-linear relationships within the cryosphere and other Earth system processes, ensuring that they are not merely products of statistical inference, but rooted in physics-informed insights.

Our results emphasize the necessity to 1) incorporate relevant non-linear interactions between aerosol and meteorological variables within ESMs in the prediction of snow hydrology, 2) inform specific variables that need to be assimilated in the design of observing systems, and 3) include more observable variables in future phases of the Coupled Model Intercomparison Project (CMIP) across both aerosols and meteorology to assess their co-variability. From our observations in Fig. 3 and the individual interactions within the networks in Fig. 4, we emphasize the incorporation of variables related to large-scale atmospheric circulation, near-surface temperatures, as well as improved proxies of absorbing aerosols within the CMIP and ESM outputs. Considering that the future of ESMs is directed toward an Integrated Earth System Model and Analysis (IESM, IESA) with an emphasis on observationally constrained coupled chemistry meteorological models (CCMMs), the joint assimilation of AMI-relevant variables is a necessity for this development ^65,66. Diagnosing and quantifying interactions is critical in reducing uncertainties in Earth system prediction and in minimizing the misattribution of observed environmental changes ⁶⁷. Accounting for AMI in ESMs has the potential to improve medium-range and sub-seasonal to seasonal forecasts of high-impact weather, particularly water cycle extremes that can help increase the resilience of vulnerable populations in regions such as HMA ⁶⁸.

While our study primarily focuses on low snow-covered regions in HMA and the late snowmelt season, further assessments of AMI in other glacier regions and during the snow accumulation period in HMA are needed. It is also important to note that interpretability frameworks within explainable machine learning are dependent on the prediction of the ML models. Thus, a combination of other interpretability techniques and more complex ML algorithms (i.e., deep neural networks) can also add to a thorough understanding of the Earth system interactions. Additional observational datasets also need to be explored to improve the robustness of our findings.

Reanalyses

ERA5 and its land counterpart ERA5-Land are the most recent fifth-generation reanalysis datasets from ECMWF (European Centre for Medium-Range Weather Forecasts) available from 1979 to the present ^69,70. MERRA-2 is the most recent reanalysis from NASA GMAO (Global Modeling and Assimilation Office) with data available from 1980 to the present ⁷¹. Both of these datasets are observationally constrained by assimilating multiple satellites and in-situ data. A relevant difference between the two reanalysis frameworks is 1) the inclusion of online coupling between aerosol and radiation in MERRA-2, whereas in ECWMF, the radiation scheme uses an aerosol climatology; and 2) a separate reanalysis (CAMS-EAC4) for atmospheric composition from ECMWF, while aerosol products are already simulated in MERRA-2 ^72,73. In addition to these publicly available reanalyses, we also use a recent, regional reanalysis product from NCAR (National Center for Atmospheric Research) called MATCHA (Model for Atmospheric Transport and Chemistry in Asia) consisting of ~ 16 years of hydrometeorological and aerosol fields over HMA, generated using the WRF-Chem v3.9.1 (Weather Research and Forecasting Model with Chemistry) coupled with the CLM-SNICAR (Community Land Model – Snow Ice Coupled with Aerosol and Radiation) model ^74–77. The model framework in MATCHA couples variables between aerosols and meteorology in two ways 1) the Rapid Radiative Transfer Model for General Circulation Models (RRTMG) allows for online interaction between simulated aerosols and radiation and 2) the use of SNICAR with CLM is an additional component in MATCHA that modifies snow albedo due to deposition of LAPs ^31,78,79. Similar to the reanalyses from ECWMF and NASA, MATCHA is observationally constrained by decadal assimilation of daily MODIS AOD and MOPITT CO products to constrain the concentration and deposition of LAPs in Asia. These reanalyses encompass different meteorological models and representations of aerosol processes that make them suitable candidates for understanding the representation of interactions in each model framework. The general characteristics of these datasets are available in Supplementary Table 1.

A total of 22 variables (6 aerosol, and 15 meteorology-related) from the three reanalysis datasets in addition to elevation were selected as predictors that can potentially drive SCF. The meteorological variables, defined hereafter as MET include a) temperature (2-m temperature, skin temperature), b) cloud cover (total, low, mid, and high-level cloud cover fraction), c) dynamic circulation (mean sea level pressure, geopotential height at 500 hPa and 300 hPa, 10-m zonal and meridional winds), d) radiative surface fluxes (surface sensible and latent heat), and e) moisture (2-m specific humidity, daily accumulated total precipitation). We note that temperature, precipitation, surface radiative fluxes and cloud cover are important factors across snow variability studies ^80–84. The dynamic circulation variables are chosen considering the association of wind-driven processes and atmospheric teleconnections on SCF ^85–87.

Aerosol variables, defined as AER hereafter, consist of aerosol optical depth (AOD) and surface mass mixing ratios grouped by the following species: a) carbonaceous (hydrophilic and hydrophobic BC and organic matter (OM), b) dust (DU), c) sulphate (SU), and d) others (sea-salt surface mixing ratio and total AOD at 550 nm). Note that MATCHA does not separate carbonaceous aerosols into hydrophobic and hydrophilic components and uses an internal mixing assumption of different aerosol species from the time of emissions. Supplementary Table 2 provides an overview of the variables used in the reanalyses.

In addition to aerosol and meteorological variables, we used elevation (defined as ELEV) from the Global Multi-resolution Terrain Elevation Data (GMTED 2010) as a static predictor to represent topography and its related interactions ^9,88. Although snow hydrology is found to be sensitive to not only elevation but also other topographical factors like aspect, slope, and shadowing effects, for this study we use only elevation as a common static predictor to represent topographical interactions across three datasets ⁸⁹.

Satellite Data

We use MODIS-based level 3 daily satellite products with a horizontal resolution of 0.05^o, namely snow SCF from MOD10C1/MYD10C1 Collection 6.1, AOD at 550 nm from MODIS processed using the MAIAC algorithm (MCD19A2CMG version 6.1), and land-surface temperature (LST) (MOD11C1/MYD11C1 Collection 6.1) ^90–94. MODIS LST contains daily data for both day and night, averaged to a daily estimate of LST. MODIS LST is used as a surrogate variable for skin temperature from each reanalysis ⁹⁵. SCF and LST products from MODIS contain products from both satellites Terra and Aqua, which were averaged to a single quantity for this study. We also use daily accumulated precipitation from IMERG (post-processed final runs) with a spatial resolution of 0.1^o to represent precipitation over HMA ⁹⁶.

Regridding the Datasets

The finer pixels of the predictors from both reanalysis and satellites were spatially averaged to 0.75^o, considering that AER variables from CAMS-EAC4 are available only at 0.75^o. Hourly to 3-hourly products from each reanalysis (ERA5-CAMS-EAC4, MERRA-2, and MATCHA) were averaged to daily data between the years 2003 and 2018. An exception is the daily accumulated precipitation from the three datasets, which was calculated by aggregating the hourly products.

Glacier Regions

A total of 6 glacier regions (GRs) are defined for HMA following the classification in Randolph Glacier Inventory version 6.0 ⁹⁷. A total of 15 second-order glacier regions were aggregated into 6 major GRs for this study ⁴⁴. The geographical extent of the glacier basins over HMA is shown in Fig. 1a. GRs marked in red (blue) denote regions of high snow cover or HSC (low snow cover or LSC) and have been identified using the methodology described in R22. We specifically focus on the late snowmelt season, i.e., the months of May-July across the years 2003–2018 when AMI is found to be significant in LSC (blue) regions (R22). The spatio-temporal mean (standard deviation) across LSC regions during 2003–2018 is 2.4% − 5.5% (5.6% − 9.1%)

Regression Framework to Estimate the Importance of AMI

We regress the target variable (daily SCF) on 22 predictors spanning aerosols (AER), meteorology (MET), and elevation (ELEV) following the equation,

$${Y}_{SCF}^{{ }^{s,t}}=\underset{1}{\underset{⏟}{\sum _{p} {\alpha }_{p}{X}_{p}^{{ }^{s,t}}}}+\underset{2}{\underset{⏟}{\sum _{p,q} {\alpha }_{pq}{X}_{p}^{{ }^{s,t}}{X}_{q}^{{ }^{s,t}}}}+\underset{3}{\underset{⏟}{\alpha \mathbb{O}\left({X}^{{ }^{s,t}}\right)}}$$

where α is the importance of the predictors (X) in regulating SCF(Y), p and q denote sets of different types of predictors (AER, MET, and ELEV) with the subscripts s and t denoting the spatio-temporal dependency of the quantities. Term 1 is the linear sensitivity of SCF as a linear combination of AER, MET, and ELEV variables. The second and third terms introduce non-linear effects that account for interactions between different predictors. Term 2 focuses on product interactions grouped by, 1) aerosol-meteorology interactions (AMI); 2) aerosol-elevation interactions (AEI); 3) meteorology interactions (MMI); and 4) elevation-meteorology interactions (MEI) with AMI as the primary focus in this study. In contrast, term 3 points toward higher-order unresolved processes extending beyond second-order product interactions in the second term. We select daily products of the target and predictor variables over a 0.75^o by 0.75^o grid, grouped by 6 glacier regions (GR) and 3 months within the late snowmelt season (May-July).

The importance α is estimated using two distinct metrics, relative importance (RI) from multi-linear regression and Shapely contribution (SHAPc) from ML (Supplementary Methods). The calculation of these metrics using the regression algorithms is explained in the Supplementary Information. RI quantifies the importance of terms 1 and 2 in Eq. 1, where we fit the linear (term 1) and product terms (term 2) in a multi-linear regression model, while SHAPc is a bulk measure of the importance of the three terms. While the regression model is an additive combination of individual and product terms, the ML model is trained only on the individual inputs. Both metrics of importance are bivariate, reflecting the joint sensitivity of SCF to a predictor in the presence of another predictor. Importance of AMI can be thus be interpreted as the impact of MET predictors in the presence of AER variables, which we characterize further based on the MET variables, 1) aerosol-radiation interactions (ARI), 2) aerosol-temperature interactions (ATI), 3) aerosol-cloud cover interactions (ACCI), 4) aerosol-circulation interactions (ACrI), and 5) aerosol-moisture interactions (AMoI).

Our notion of importance draws parallels to the chain rule representation of the extensively studied ACI in the context of cloud radiative forcing, with cloud fraction (cover) as one of the dependencies. While the product of sensitivities in the chain rule formulation may not fully capture the non-linear feedback (interaction) between its dependencies, we can draw a direct link between the importance of ACCI and ACI-cloud cover sensitivity from past studies.

Leveraging Model Constructs

We perform our analysis based on three model constructs: 1) the Observation-to-Model (Obs-Model) Construct, where MODIS SCF is the target variable, 2) the Model-to-Model (Model-Model) Construct, where SCF from each reanalysis dataset is the target against corresponding predictors, and 3) the Observation-to-Observations (Obs-Obs) Construct, where we chose a set of variables directly observable through satellites (MODIS SCF, MODIS LST, MAIAC AOD, and IMERG PRECIP) to explore the non-linear relationships between SCF and its predictors depicted in Eq. 1.

The regression approach from Eq. 1 is reserved only for the Obs-Model and Model-Model constructs, while the Obs-Obs construct is solely used to elucidate AMI that can be observed through satellites. This is due to the lack of a diverse range of predictors available from observations with consistent spatiotemporal coverage, a gap that can be bridged by reanalyses.

Importance estimates in the Obs-Model construct encompass all three terms in Eq. 1 (especially the unresolved stochastic processes driving SCF in the third term) and are the closest approximation to an observable estimate of AMI. The Model-Model construct sheds light on the representation of cryospheric processes driven by AMI in each reanalysis. Comparing the Obs-Model and Model-Model constructs can thus provide insights into the processes in the third term of Eq. 1 and highlight any potential misattribution of importance estimated in any of the terms within the Model-Model construct.

Data, Materials, and Software Availability

Satellite and reanalysis data except MATCHA are available in the public domain. ERA5/ERA5-Land data were downloaded from the Copernicus Climate Data Store available at https://cds.climate.copernicus.eu/cdsapp#!/search?type=dataset&text=ERA5. CAMS-EAC4 data were downloaded from the Atmospheric Data Store available at https://ads.atmosphere.copernicus.eu/cdsapp#!/dataset/cams-global-reanalysis-eac4?tab=overview. MERRA-2 and IMERG Final Run data were downloaded from NASA GES DISC available at https://disc.gsfc.nasa.gov/. MODIS SCF and LST and MAIAC AOD were downloaded from NASA Earthdata https://search.earthdata.nasa.gov/search. The MATCHA dataset is planned to be archived for public access in the High Mountain Asia (HMA) project within the NSIDC (National Snow and Ice Data Center).

Acknowledgments

This work is supported by a NASA HiMAT2 grant (#NNH19ZDA001N). HiMAT2 is an interdisciplinary effort to understand the cryospheric and hydrological state of HMA. We also acknowledge the National Center for Atmospheric Research (NCAR) (sponsored by the National Science Foundation (NSF)) for assisting this ongoing study. This work is in tandem with the goals of the Aerosol subgroup under HiMAT2 which seeks to quantify the deposition of aerosols over snow in HMA. We would also like to thank Miguel Hilario and Debottama Das for discussions about this work.

Author Contributions

CR contributed to the conceptualization, methodology, formal analysis, investigation, visualization, writing-original draft, reviewing, and editing. CH contributed to the funding acquisition, and writing – review and editing. RK contributed to the funding acquisition, and writing-review and editing. AFAJ contributed to the conceptualization, methodology, supervision, and writing-reviewing and editing.

Competing Interests

The authors declare no competing interests.

Kraaijenbrink, P. D. A., Stigter, E. E., Yao, T. & Immerzeel, W. W. Climate change decisive for Asia’s snow meltwater supply. Nat. Clim. Chang. 11, 591–597 (2021).
Mudryk, L. et al. Historical Northern Hemisphere snow cover trends and projected changes in the CMIP6 multi-model ensemble. The Cryosphere 14, 2495–2514 (2020).
Cohen, J. & Entekhabi, D. The influence of snow cover on northern hemisphere climate variability. Atmosphere-Ocean 39, 35–53 (2001).
Cerveny, R. S. & Balling Jr., R. C. The impact of snow cover on diurnal temperature range. Geophysical Research Letters 19, 797–800 (1992).
Flanner, M. G., Shell, K. M., Barlage, M., Perovich, D. K. & Tschudi, M. A. Radiative forcing and albedo feedback from the Northern Hemisphere cryosphere between 1979 and 2008. Nature Geosci 4, 151–155 (2011).
Robock, A. Ice and Snow Feedbacks and the Latitudinal and Seasonal Distribution of Climate Sensitivity. Journal of the Atmospheric Sciences 40, 986–997 (1983).
Bonekamp, P. N. J., Wanders, N., van der Wiel, K., Lutz, A. F. & Immerzeel, W. W. Using large ensemble modelling to derive future changes in mountain specific climate indicators in a 2 and 3°C warmer world in High Mountain Asia. International Journal of Climatology 41, E964–E979 (2021).
Sorg, A. B., Tobias; Stoffel, Markus; Solomina, Olga; Beniston, Martin. Climate change impacts on glaciers and runoff in Tien Shan (Central Asia). Nature Climate Change 2, 725–731 (2012).
Pepin, N. et al. Elevation-dependent warming in mountain regions of the world. Nature Clim Change 5, 424–430 (2015).
Warren, S. G. & Wiscombe, W. J. A Model for the Spectral Albedo of Snow. II: Snow Containing Atmospheric Aerosols. Journal of the Atmospheric Sciences 37, 2734–2745 (1980).
Qian, Y. et al. Light-absorbing particles in snow and ice: Measurement and modeling of climatic and hydrological impact. Adv. Atmos. Sci. 32, 64–91 (2015).
He, C. et al. Black carbon radiative forcing over the Tibetan Plateau. Geophysical Research Letters 41, 7806–7813 (2014).
Kang, S., Zhang, Y., Qian, Y. & Wang, H. A review of black carbon in snow and ice and its impact on the cryosphere. Earth-Science Reviews 210, 103346 (2020).
Shindell, D. & Faluvegi, G. Climate response to regional radiative forcing during the twentieth century. Nature Geosci 2, 294–300 (2009).
Sarangi, C. et al. Dust dominates high-altitude snow darkening and melt over high-mountain Asia. Nat. Clim. Chang. 10, 1045–1051 (2020).
Brown, H. et al. Brown Carbon Fuel and Emission Source Attributions to Global Snow Darkening Effect. Journal of Advances in Modeling Earth Systems 14, e2021MS002768 (2022).
Ramanathan, V., Crutzen, P. J., Kiehl, J. T. & Rosenfeld, D. Aerosols, Climate, and the Hydrological Cycle. Science 294, 2119–2124 (2001).
He, C. Modelling light-absorbing particle–snow–radiation interactions and impacts on snow albedo: fundamentals, recent advances and future directions. Environ. Chem. 19, 296–311 (2022).
Andreae, M. O. & Rosenfeld, D. Aerosol–cloud–precipitation interactions. Part 1. The nature and sources of cloud-active aerosols. Earth-Science Reviews 89, 13–41 (2008).
Painter, T. H. et al. Impact of disturbed desert soils on duration of mountain snow cover. Geophysical Research Letters 34, (2007).
Lohmann, U. & Feichter, J. Global indirect aerosol effects: a review. Atmospheric Chemistry and Physics 5, 715–737 (2005).
Flanner, M. G., Zender, C. S., Randerson, J. T. & Rasch, P. J. Present-day climate forcing and response from black carbon in snow. Journal of Geophysical Research: Atmospheres 112, (2007).
Huang, H., Qian, Y., He, C., Bair, E. H. & Rittger, K. Snow Albedo Feedbacks Enhance Snow Impurity-Induced Radiative Forcing in the Sierra Nevada. Geophysical Research Letters 49, e2022GL098102 (2022).
Sand, M., Samset, B. H., Tsigaridis, K., Bauer, S. E. & Myhre, G. Black Carbon and Precipitation: An Energetics Perspective. Journal of Geophysical Research: Atmospheres 125, e2019JD032239 (2020).
Stevens, B. & Feingold, G. Untangling aerosol effects on clouds and precipitation in a buffered system. Nature 461, 607–613 (2009).
Sakai, A. & Fujita, K. Contrasting glacier responses to recent climate change in high-mountain Asia. Sci Rep 7, 13717 (2017).
Ragettli, S., Immerzeel, W. W. & Pellicciotti, F. Contrasting climate change impact on river flows from high-altitude catchments in the Himalayan and Andes Mountains. Proceedings of the National Academy of Sciences 113, 9222–9227 (2016).
Kapnick, S. B., Delworth, T. L., Ashfaq, M., Malyshev, S. & Milly, P. C. D. Snowfall less sensitive to warming in Karakoram than in Himalayas due to a unique seasonal cycle. Nature Geosci 7, 834–840 (2014).
Bonekamp, P. N. J., de Kok, R. J., Collier, E. & Immerzeel, W. W. Contrasting Meteorological Drivers of the Glacier Mass Balance Between the Karakoram and Central Himalaya. Frontiers in Earth Science 7, (2019).
Usha, K. H., Nair, V. S. & Babu, S. S. Modeling of aerosol induced snow albedo feedbacks over the Himalayas and its implications on regional climate. Clim Dyn 54, 4191–4210 (2020).
Archer-Nicholls, S., Lowe, D., Schultz, D. M. & McFiggans, G. Aerosol–radiation–cloud interactions in a regional coupled model: the effects of convective parameterisation and resolution. Atmospheric Chemistry and Physics 16, 5573–5594 (2016).
Barthlott, C., Zarboo, A., Matsunobu, T. & Keil, C. Impacts of combined microphysical and land-surface uncertainties on convective clouds and precipitation in different weather regimes. Atmospheric Chemistry and Physics 22, 10841–10860 (2022).
Carslaw, K. S. et al. Large contribution of natural aerosols to uncertainty in indirect forcing. Nature 503, 67–71 (2013).
Gettelman, A. Putting the clouds back in aerosol–cloud interactions. Atmospheric Chemistry and Physics 15, 12397–12411 (2015).
Moch, J. M. et al. Aerosol-Radiation Interactions in China in Winter: Competing Effects of Reduced Shortwave Radiation and Cloud-Snowfall-Albedo Feedbacks Under Rapidly Changing Emissions. Journal of Geophysical Research: Atmospheres 127, e2021JD035442 (2022).
Stein, U. & Alpert, P. Factor Separation in Numerical Simulations. Journal of Atmospheric Sciences 50, 2107–2115 (1993).
Zhou, M. et al. The impact of aerosol–radiation interactions on the effectiveness of emission control measures. Environ. Res. Lett. 14, 024002 (2019).
Johnson, J. S. et al. Evaluating uncertainty in convective cloud microphysics using statistical emulation. Journal of Advances in Modeling Earth Systems 7, 162–187 (2015).
Lee, L. A. et al. The magnitude and causes of uncertainty in global model simulations of cloud condensation nuclei. Atmospheric Chemistry and Physics 13, 8879–8914 (2013).
Xiao, Y., Ke, C.-Q., Shen, X., Cai, Y. & Li, H. What drives the decrease of glacier surface albedo in High Mountain Asia in the past two decades? Science of The Total Environment 863, 160945 (2023).
Wall, C. J. et al. Assessing effective radiative forcing from aerosol–cloud interactions over the global ocean. Proceedings of the National Academy of Sciences 119, e2210481119 (2022).
Castelvecchi, D. Can we open the black box of AI? Nature News 538, 20 (2016).
Lipton, Z. C. The Mythos of Model Interpretability: In machine learning, the concept of interpretability is both important and slippery. Queue 16, 31–57 (2018).
Roychoudhury, C., He, C., Kumar, R., McKinnon, J. M. & Arellano Jr., A. F. On the Relevance of Aerosols to Snow Cover Variability Over High Mountain Asia. Geophysical Research Letters 49, e2022GL099317 (2022).
Harte, J. Toward a Synthesis of the Newtonian and Darwinian Worldviews. Physics Today 55, 29–34 (2002).
Feingold, G. et al. New approaches to quantifying aerosol influence on the cloud radiative effect. Proceedings of the National Academy of Sciences 113, 5812–5819 (2016).
Mülmenstädt, J. & Feingold, G. The Radiative Forcing of Aerosol–Cloud Interactions in Liquid Clouds: Wrestling and Embracing Uncertainty. Curr Clim Change Rep 4, 23–40 (2018).
Seinfeld, J. H. et al. Improving our fundamental understanding of the role of aerosol – cloud interactions in the climate system. Proceedings of the National Academy of Sciences 113, 5781–5790 (2016).
Orsolini, Y. et al. Evaluation of snow depth and snow cover over the Tibetan Plateau in global reanalyses using in situ and satellite remote sensing observations. The Cryosphere 13, 2221–2239 (2019).
Chen, S. et al. Comparison of dust emissions, transport, and deposition between the Taklimakan Desert and Gobi Desert from 2007 to 2011. Sci. China Earth Sci. 60, 1338–1355 (2017).
Gul, C. et al. Black carbon concentration in the central Himalayas: Impact on glacier melt and potential source contribution. Environmental Pollution 275, 116544 (2021).
Marshall, S., Roads, J. O. & Glatzmaier, G. Snow Hydrology in a General Circulation Model. Journal of Climate 7, 1251–1269 (1994).
Maskey, S., Uhlenbrook, S. & Ojha, S. An analysis of snow cover changes in the Himalayan region using MODIS snow products and in-situ temperature data. Climatic Change 108, 391–400 (2011).
Yamazaki, T., Kondo, J., Sakuraoka, T. & Nakamura, T. A one-dimensional model of the evolution of snow-cover characteristics. Annals of Glaciology 18, 22–26 (1993).
Soulie, A. et al. Global Anthropogenic Emissions (CAMS-GLOB-ANT) for the Copernicus Atmosphere Monitoring Service Simulations of Air Quality Forecasts and Reanalyses. https://essd.copernicus.org/preprints/essd-2023-306/ (2023) doi:10.5194/essd-2023-306.
Das, S. et al. Linkage between the absorbing aerosol-induced snow darkening effects over the Himalayas-Tibetan Plateau and the pre-monsoon climate over northern India. Theor Appl Climatol 147, 1033–1048 (2022).
Kumar, R. et al. Influences of the springtime northern Indian biomass burning over the central Himalayas. Journal of Geophysical Research: Atmospheres 116, (2011).
Lau, K. M., Kim, M. K. & Kim, K. M. Asian summer monsoon anomalies induced by aerosol direct forcing: the role of the Tibetan Plateau. Clim Dyn 26, 855–864 (2006).
Ramanathan, V. et al. Atmospheric brown clouds: Hemispherical and regional variations in long-range transport, absorption, and radiative forcing. Journal of Geophysical Research: Atmospheres 112, (2007).
Sand, M. et al. Aerosol absorption in global models from AeroCom Phase III. https://acp.copernicus.org/preprints/acp-2021-51/acp-2021-51.pdf (2021) doi:10.5194/acp-2021-51.
Inglis, A., Parnell, A. & Hurley, C. B. Visualizing Variable Importance and Variable Interaction Effects in Machine Learning Models. Journal of Computational and Graphical Statistics 31, 766–778 (2022).
Charlson, R. J. et al. Climate Forcing by Anthropogenic Aerosols. Science 255, 423–430 (1992).
Kraskov, A., Stögbauer, H. & Grassberger, P. Estimating mutual information. Phys. Rev. E 69, 066138 (2004).
Gryspeerdt, E., Stier, P., White, B. A. & Kipling, Z. Wet scavenging limits the detection of aerosol effects on precipitation. Atmospheric Chemistry and Physics 15, 7557–7570 (2015).
Bocquet, M. et al. Data assimilation in atmospheric chemistry models: current status and future prospects for coupled chemistry meteorology models. Atmospheric Chemistry and Physics 15, 5325–5358 (2015).
Thriving on Our Changing Planet: A Decadal Strategy for Earth Observation from Space. (National Academies Press, 2018). doi:10.17226/24938.
National Academies of Sciences, E., and Medicine. Next Generation Earth Systems Science at the National Science Foundation. (The National Academies Press, 2022). doi:10.17226/26042.
NOAA Science Advisory Board. A Report on Priorities for Weather Research. 119 https://sab.noaa.gov/wp-content/uploads/2021/12/PWR-Report_Final_12-9-21.pdf (2021).
Hersbach, H. et al. The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society 146, 1999–2049 (2020).
Muñoz-Sabater, J. et al. ERA5-Land: a state-of-the-art global reanalysis dataset for land applications. Earth System Science Data 13, 4349–4383 (2021).
Gelaro, R. et al. The Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2). Journal of Climate 30, 5419–5454 (2017).
Inness, A. et al. The CAMS reanalysis of atmospheric composition. Atmospheric Chemistry and Physics 19, 3515–3556 (2019).
Randles, C. A. et al. The MERRA-2 Aerosol Reanalysis, 1980 Onward. Part I: System Description and Data Assimilation Evaluation. Journal of Climate 30, 6823–6850 (2017).
Flanner, M. G. et al. SNICAR-ADv3: a community tool for modeling spectral snow albedo. Geosci. Model Dev. 14, 7673–7704 (2021).
Grell, G. A. et al. Fully coupled “online” chemistry within the WRF model. Atmospheric Environment 39, 6957–6975 (2005).
Oleson, K. W. et al. Technical description of version 4.0 of the community land model (CLM). NCAR Tech. Note NCAR/TN-478 + STR 257, 1–257 (2010).
Skamarock, C. et al. A Description of the Advanced Research WRF Version 3. (2008) doi:10.5065/D68S4MVH.
Mlawer, E. J., Taubman, S. J., Brown, P. D., Iacono, M. J. & Clough, S. A. Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. Journal of Geophysical Research: Atmospheres 102, 16663–16682 (1997).
Kumar, R., Barth, M. C., Pfister, G. G., Naja, M. & Brasseur, G. P. WRF-Chem simulations of a typical pre-monsoon dust storm in northern India: influences on aerosol optical properties and radiation budget. Atmospheric Chemistry and Physics 14, 2431–2446 (2014).
Duan, A. & Wu, G. Change of cloud amount and the climate warming on the Tibetan Plateau. Geophysical Research Letters 33, (2006).
Ohmura, A., Kasser, P. & Funk, M. Climate at the Equilibrium Line of Glaciers. Journal of Glaciology 38, 397–411 (1992).
Shi, X., Déry, S. J., Groisman, P. Y. & Lettenmaier, D. P. Relationships between Recent Pan-Arctic Snow Cover and Hydroclimate Trends. Journal of Climate 26, 2048–2064 (2013).
Södergren, A. H., McDonald, A. J. & Bodeker, G. E. An energy balance model exploration of the impacts of interactions between surface albedo, cloud cover and water vapor on polar amplification. Clim Dyn 51, 1639–1658 (2018).
Wang, W., Huang, X., Deng, J., Xie, H. & Liang, T. Spatio-Temporal Change of Snow Cover and Its Response to Climate over the Tibetan Plateau Based on an Improved Daily Cloud-Free Snow Cover Product. Remote Sensing 7, 169–194 (2015).
Jiang, X. et al. Impacts of ENSO and IOD on Snow Depth Over the Tibetan Plateau: Roles of Convections Over the Western North Pacific and Indian Ocean. Journal of Geophysical Research: Atmospheres 124, 11961–11975 (2019).
Mott, R., Vionnet, V. & Grünewald, T. The Seasonal Snow Cover Dynamics: Review on Wind-Driven Coupling Processes. Frontiers in Earth Science 6, (2018).
You, Q. et al. Review of snow cover variation over the Tibetan Plateau and its influence on the broad climate system. Earth-Science Reviews 201, 103043 (2020).
Danielson, J. J. & Gesch, D. B. Global multi-resolution terrain elevation data 2010 (GMTED2010). (US Department of the Interior, US Geological Survey Washington, DC, USA, 2011).
Hao, D. et al. A parameterization of sub-grid topographical effects on solar radiation in the E3SM Land Model (version 1.0): implementation and evaluation over the Tibetan Plateau. Geoscientific Model Development 14, 6273–6289 (2021).
Hall, D. K. & Riggs, G. A. MODIS/Terra Snow Cover Daily L3 Global 0.05Deg CMG, Version 61. (2021) doi:10.5067/MODIS/MOD10C1.061.
Hall, D. K. & Riggs, G. A. MODIS/Aqua Snow Cover Daily L3 Global 0.05Deg CMG, Version 61. (2021) doi:10.5067/MODIS/MYD10C1.061.
Lyapustin, A. MODIS/Terra + Aqua AOD and Water Vapor from MAIAC, Daily L3 Global 0.05Deg CMG V061. (2023) doi:10.5067/MODIS/MCD19A2CMG.061.
Wan, Z., Hook, S. & Hulley, G. MYD11C1 MODIS/Aqua Land Surface Temperature/Emissivity Daily L3 Global 0.05Deg CMG V006. (2015) doi:10.5067/MODIS/MYD11C1.006.
Wan, Z., Hook, S. & Hulley, G. MOD11C1 MODIS/Terra Land Surface Temperature/Emissivity Daily L3 Global 0.05Deg CMG V006. (2015) doi:10.5067/MODIS/MOD11C1.006.
Jin, M. & Dickinson, R. E. Land surface skin temperature climatology: benefitting from the strengths of satellite observations. Environ. Res. Lett. 5, 044004 (2010).
Huffman, G. J. et al. GPM IMERG Final Precipitation L3 1 day 0.1 degree x 0.1 degree V06. (2014) doi:10.5067/GPM/IMERGDF/DAY/06.
Pfeffer, W. T. et al. The Randolph Glacier Inventory: a globally complete inventory of glaciers. Journal of Glaciology 60, 537–552 (2014).

No competing interests reported.

SRSupplementaryCR.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Unraveling the Complexities of Aerosol-Meteorology Interactions on Snowmelt in High Mountain Asia

Status:

Version 1

Abstract

Figures

Introduction

Results

Spatial Distribution and Seasonality of SCF

Spatial Distribution and Seasonality of MET and AER Predictors

Quantifying Importance of AMI to SCF

Process Drivers of AMI

Emergent Connections within AMI

Observable AMI

Discussion

Methods

Reanalyses

Satellite Data

Regridding the Datasets

Glacier Regions

Regression Framework to Estimate the Importance of AMI

Leveraging Model Constructs

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1