Himalayan mountain glaciers and snowpacks have a major impact on the water systems of major rivers throughout Asia and the people living in the river basins. Recent observations suggest a continuing decline in Himalayan mountain glaciers and snow cover. Bajracharya et al. (2008) observed that the Himalayan glaciers are retreating at rates ranging from 10 to 60 m per year, and many small glaciers have disappeared. Gardner et al. (2013) also showed with satellite observations the steady reduction of western China glaciers with the most rapid decline observed in the Himalayan mountain regions. Changes in the cryosphere are accompanied by documented surface warming trends over Tibet, which reveals a strong altitude dependence of surface warming with peak warming trends of 2–2.5 ∘C at 5000 m from 1961 to 2006 (Liu et al., 2009).

The last few decades also witnessed rapid growth in human population and economic activities, causing intense air pollution over the Asian region. Among the many air pollutants, black carbon (BC) aerosols have been shown to have a significant impact on global and regional climate change (Ramanathan and Carmichael, 2008). Many previous studies have linked Asian aerosols (including sulfates and BC) with monsoon systems and have demonstrated the aerosol impact on the summer rainfall (Ramanathan et al., 2005; Lau et al., 2006; Lau and Kim, 2006; Meehl et al., 2008). The BC aerosols have also been shown to have an impact on warming trends over the Himalaya–Tibet region (Ramanathan et al., 2007), on the retreat of Himalayan glaciers (Menon et al., 2010; Qian et al., 2011), and on Eurasian snow cover (Flanner et al., 2009). Observationally, using ice-core samples to reconstruct historical BC content over Tibet, Xu et al. (2009) suggested that BC is a significant contributing factor in causing the glacier change.

To date global climate models forced by historical radiative forcing scenarios (such as those in the Coupled Model Intercomparison Project Phase 5, CMIP5) have difficulty in simulating the observed record surface warming (You et al., 2015) or its anomalously strong altitude dependence in the Tibet–Himalaya region. One possible explanation as we will investigate here is that few of these earlier studies of the Himalayan and Tibetan climate change have considered the combined effects of all the following factors: BC direct heating of the atmosphere, the heating of snowpacks and glaciers by BC darkening the snow and ice, the greenhouse effect of CO2, and the surface cooling effects by aerosols other than BC.

In this study, we used a state-of-the-art global climate model to conduct a suite of model experiments to understand BC's role in the cryosphere change over the Himalaya and Tibetan region. A unique feature of the present study, compared with earlier studies, is that BC radiative forcing is constrained with multiple sources of observations (satellite observed aerosol optical depths and ground network of spectral sun photometer measurements). We also used a newly developed method to separate the BC contribution to solar absorption from other aerosols (sulfates, organics, and brown carbon) and calculate its direct radiative forcing (Bahadur et al., 2012; Xu et al., 2013). Previous studies (Ramanathan et al., 2007; Lau et al., 2010; Menon et al., 2010; Qian et al., 2011) included the effects of BC on the atmosphere and the cryosphere, but the simulated BC radiative forcing by these “standard” models used in CMIP5 is strongly biased to low values (Bond et al., 2013) due to emission inventory biases and missing physical treatments (Jacobson, 2012). As shown in Bond et al. (2013), current models are underestimating BC solar absorption over South Asia by a factor of 2 to 5. In this study, we scaled the simulated BC forcing in the climate model by factors ranging from 2 to 4 to bring the simulated values closer to the observationally constrained values (Xu, 2014). Another improvement in this study is that the simulations were conducted using a fully coupled ocean–atmosphere–land model at a high resolution of 1∘ by 1∘, in which a new land snow module is adopted (Lawrence et al., 2011) to account for BC deposition effect on snow and ice.

The observations (Robinson et al., 2012) show that the snow cover extent over the Himalayan mountain range has declined at a rate of more than 10 % per decade since the 1960s (Fig. 1). The snow cover retreat along the Himalayan mountain range is greater than in Eurasia during the same period. In situ studies on regional glaciers and snowpack also reported strong declining trends. For example, Ma and Qin (2012) used 754 stations in China to document statistically significant declining trends of spring snow for the Qinghai–Tibet Plateau for the 1951–2009 period. Consistently, permafrost degradation has been reported on the Tibetan Plateau (Cheng and Wu, 2007; Li et al., 2008).

Several satellite observations since the year of 2001 provided additional record in snow cover extent. The observed trend over this shorter period (2001–2012) is less significant (Fig. S1a in the Supplement) and negative trends are only found along some portion of the Himalayan range. Consistently, the 5 km MODIS data set (Fig. S1b) also shows that the snow cover extent averaged over the entire Tibet region only has a slight decrease. But as other studies have pointed out (Pu et al., 2007; Pu and Xu, 2009), the highest altitudes of 5750–6000 m exhibit larger negative trends (-6 % decade-1). We note that at a shorter timescale (10 years), the snow cover trend is heavily influenced by natural variability and less significant. For example, during 1980–1991 (Fig. S1c) or 1990–2001 as shown in Fig. 5 of Menon et al. (2010), the declining trends are much larger. Nevertheless, the declining trend in the 40–50 year timescale (Fig. 1) is more robust and warrants further investigation of its causes, which is the main objective of this study.

To understand the causes of the observed trends of snow reduction over the multi-decadal timescale, we conducted global climate model simulations, in which BC emissions, CO2 concentration or SO4 emissions are increased instantaneously from preindustrial to present-day levels. Figure 2 (left column) shows the simulated change in snow fraction due to the increase in BC, CO2, and SO4 aerosols. The pattern of snow cover decline in the BC model simulation captures the broad features of the observed decline (Fig. 1), with the largest snow reduction along the mountain range. The Tibetan Plateau on average showed a reduction in snow fraction of 1.9 % due to BC. The snow fraction shrinks by 2.9 % due to present-day CO2. Along the Himalayan mountain range, where the near-permanent snow cover exists, the reduction of snow fraction exceeds 10 % in both BC and CO2 cases.

Menon et al. (2010) attempted to simulate the snow reduction trends during the 1990s but the spatial distribution of the observed trend was not well captured mainly due to the coarse resolution of the model. Qian et al. (2011) also acknowledged their model's limitation in representing the snow cover climatology and therefore may have biases in estimating BC impact on snow. It is well known that global models tend to overestimate the snow cover of the Tibetan Plateau, and one potential reason is that the blocking effect for the moisture transport crossing the Himalayas is too small due to the coarse resolution of the global models and too much snowfall is simulated (Ménégoz et al., 2013). This limitation can partly be overcome with models using higher spatial resolutions. The modeling work presented here is a major step forward in terms of spatial resolution (about 1∘ by 1∘), as opposed to earlier studies (2.8∘ by 2.8∘ in Flanner et al., 2009, and Qian et al., 2011; 4∘ by 5∘ in Menon et al., 2010), which helps to better resolve the complex topography in this region. As a result of increased spatial resolution and also the improved land scheme, the biases in snow cover simulation is significantly reduced from its earlier model versions (Lawrence et al., 2011; also contrast Fig. S2c with Fig. 2 of Qian et al., 2011). However, we note that the precipitation over the Tibetan Plateau is still overestimated (Fig. S2b), and future studies, especially using regional climate models with even higher resolutions, are needed to improve the fidelity of model simulations of snowpack and glaciers over this topography-complicated region.

One consequence of the snow cover reduction is the decrease in surface albedo, which provides a positive feedback mechanism to localized warming. Such a surface albedo change in response to sea-ice loss has been observationally detected (Kay and L'Ecuyer, 2013; Pistone et al., 2014) and is important in explaining amplified Arctic warming. Flanner et al. (2011) also used observations during recent decades to calculate the surface albedo feedback in Northern Hemisphere large-scale snow-covered regions. Our simulations show that surface albedo over the Tibet region decreased by over 2 % (Fig. 2, right column) in response to BC. The maximum reduction occurs right along the Himalayan mountain range and part of the Tibet–Sichuan mountain regions.

The surface albedo decrease due to CO2 shares a similar spatial pattern with BC (Fig. 2, right column) but with a smaller magnitude (Table 2b). Moreover, the snow depth reduction in response to CO2 is only 30 % of that due to BC (Table 2c and Fig. 2), and this highlights the larger effect of BC in causing the regional cryospheric change over the Himalayas and Tibet. Not surprisingly, in the simulations the snow cover and surface albedo are increasing in response to cooling aerosols like SO4 (Fig. 2), but in a smaller magnitude than that of the decreases due to BC and CO2.

The Tibet region has witnessed increasing surface temperature by 0.3 ∘C per decade – more than twice the global average (Wang et al., 2008). One feature of the surface-warming trend over Tibet is that the warming magnitude increases significantly with altitude (Liu et al., 2009). To understand this anomalous feature, we show in Fig. 3 the tropospheric temperature responses (as a function of altitude and latitude) to BC as well as CO2 and SO4. BC-induced heating rate (Fig. 3a) is more concentrated over the Northern Hemisphere (NH) due to larger emissions there from industrial activities, consistent with radiative forcing distribution (Table 1). The notable feature of BC response is the elevated warming at altitudes of 4000 to 8000 m and 30 to 60∘ N (Fig. 3b), in particular over the Tibetan Plateau. The CO2 warming pattern (Fig. 3b) features an amplified warming at the surface of the Arctic and in the upper tropical troposphere. CO2-induced warming in the upper atmosphere (500 hPa) over Tibet is 1 ∘C, larger than BC-induced warming of 0.5 ∘C, but the vertical gradient is much smaller (Fig. S3). SO4 cooling features an even stronger north–south asymmetry (north cooling; south slightly warming) but is more confined to the surface (Fig. S3). The temperature response in the troposphere is associated with strong meridional circulation change. The mechanisms behind the free-atmosphere circulation change, especially for the SO4 case that does not have strong atmospheric forcing, are discussed in detail in Xu and Xie (2015).

Ground-based observations have shown that the last three decades were subject to a factor of 2 greater warming in the high-altitude interior of the Tibetan Plateau than at the edge of the plateau and at lower altitudes. The observations in Liu et al. (2009) were made between 1965 and 2006 from ground meteorological stations on the Tibetan Plateau region, and they revealed clear altitude dependence in the daily-minimum surface temperature (purple line in Fig. 4). The vertical profile of temperature change based on daily-average measurement from another reanalysis data set (Fig. 5) also reveals similar altitude dependence. What is driving the larger warming at high-altitude regions?

Figure 4 shows the model-simulated change of the daily-minimum surface temperature as a function of elevation due to three different forcing agents (CO2, SO4 and BC). The surface temperature responses are calculated from all of the model grid cells over the Tibetan Plateau and the surrounding region (20–50∘ N, 70–110∘ E) to capture the altitude variation in this region. As shown in Fig. 4, the altitude dependence of the surface warming is mostly determined by the response to BC forcing (red dots). At altitudes below 1000 m the warming is minimal, but with increasing altitudes the magnitude of the warming increases up to 2 ∘C at 5000 m. The dependence of the surface warming on altitude is much smaller in the CO2 case, which only increased from 1.2 ∘C warming at low altitudes to 1.6 ∘C warming at higher altitudes (yellow dots).

The combined temperature response (black triangles in Fig. 4) by adding the individual trends due to BC, CO2, and SO4 is largely consistent with the observed trend. To test the additivity of the temperature response, we conducted another simulation in which all of the three forcings were imposed simultaneously. The warming profile simulated by the combined anthropogenic forcing experiment largely agrees with the sum of the individual responses within 30 % (Fig. 5). Some non-linearity is expected as discussed in other modeling studies (Ming and Ramaswamy, 2009). The agreement of the simulated and the observed warming profiles provides a qualitative estimate of the relative contributions of BC, CO2, and SO4. Over the entire Tibetan Plateau, CO2-induced surface warming is 1.3 ∘C, compared to the BC-induced warming of 0.84 ∘C (Table 1b). Almost half of the surface warming at the highest altitudes (around 5000 m) is due to BC.

A potential complexity arises due to the internal variability of the climate systems, which has been shown to be important in determining decadal trends over individual regions (Deser et al., 2012). To examine the role of natural variability, we calculated the temperature trend from all 350 consecutive 45-year periods out of 394 years of simulation (i.e. year 1–45, year 2–46,..., year 350–394). The 80 % (10th to 90th percentile) probability range of temperature change is shown in Fig. 5 (light-blue shading). The magnitude of the warming rarely exceeds 0.5 ∘C in any 45-year period in the long-term preindustrial control simulations without any external forcing. Therefore we infer that the vertical gradient of the temperature trend found in the simulations is very unlikely due to natural variability. One further concern is that the internal variability deduced from the long-term preindustrial control simulation can be model dependent. However, comparing a 30-member ensemble of CESM1 simulation (the same spatial resolution and configuration as in our control simulation) with the 38-member CMIP5, Kay et al. (2015) found that the ensemble spread in the 30-year trend of surface temperature in CESM1 ensemble is statistically the same with the spread in trends within CMIP5.

Note that the climate simulations shown in Figs. 4 and 5 are driven by the instantaneous increase in present-day forcing. Since the real forcing trends were time dependent, we further analyzed a set of 20th century transient simulation output from the same model (Fig. 6), as part of CMIP5. The relative contributions of CO2 and SO4 to the simulated warming are consistent between the two sets of simulations (the instantaneous forcing and transient forcing). But the trends estimated from the 20th century transient forcing simulations are smaller than the quasi-equilibrium response to instantaneous forcing (parentheses in Table 1b). The reason is that only 70 % of SO4 forcing and about 60 % of CO2 forcing in the transient simulation were applied after 1960. The standard BC forcing (red solid line in Fig. 6) only lead to a weak warming, not exceeding the range of natural variability. As a result, the combined all-forcing responses (black triangles) did not capture the altitude dependence in observations very well.

Only when we adjusted historical BC forcing using the same scaling factors constrained by present-day observations, transient BC forcing induced a robust warming and amplification over high altitudes (red dots in Fig. 6), similar to what is shown in the instantaneous forcing experiment. However, note that the historical time dependence of the BC forcing is more uncertain, and we were also only able to produce one ensemble of adjusted BC simulation that was more subject to the influence of decadal variability. Therefore, the response to the instantaneous present-day BC forcing seems a more reliable indicator of the BC effects. While the absolute values of the warming profile need more model tests, our inference regarding the relative role of BC and CO2 in the observed decrease in snow cover as well as the major role of BC on the altitude dependence of the warming trends is robust.

Both CO2 and BC contribute to the elevated warming at 5 km as shown in Fig. 4. However, BC is mostly responsible for the vertical gradient of the simulated warming trend. Most of the BC aerosols in the region are emitted over India and China and subsequently transported to the Tibetan Plateau and the Himalayan mountain range. The physical mechanisms for the amplified warming at higher altitude due to BC are at least threefold.

The observed surface warming over the Tibetan and Himalayan region of about 0.5 ∘C at sea level to about 2–2.5 ∘C at 5000 m (from 1961 to 2006) has been an outstanding feature of climate trends. The more than 2 ∘C warming is close to the peak warming trend observed anywhere on the planet. For comparison, the Arctic warming associated with large sea-ice retreat during this period is 1.2 ∘C.

The high-resolution coupled ocean–atmosphere model in this study was able to attribute the observed warming trends and their high-altitude enhancement to imposed increases in CO2, BC, and SO4 aerosols. The simulated changes with all forcing imposed were consistent with the observations. The key to the success is that we obtained the BC forcing from the reconstruction of ground-based and satellite-based observations. The imposed BC forcing was about 2 to 4 times (depending on the regions) larger than that simulated by the models using bottom-up emission inventories. The analysis of model simulations highlights that the high-altitude warming due to BC is as large as CO2 warming over the Tibetan Plateau and the elevated warming profile is unique in BC responses.

The observed record warming is accompanied by retreat of glaciers and snow cover as well as thinning of the snowpacks. In response to the preindustrial to the present-day increase in BC emissions, the annual averaged snow fraction over the Tibetan Plateau is reduced by more than 6 % (relatively), and the snow depth by approximately 19 %. The surface albedo decreases by more than 5 % along the Himalayan mountain range and 1.4 % over the entire Tibet region, providing a positive local feedback to the enhanced local warming. In stark contrast, despite having 5 times larger effect in global mean temperature than BC, over Tibet CO2 impact is only 1.5 times stronger in snow cover decrease, and only one-third in snow depth decrease. We conclude that BC is instrumental in causing snow retreat and its effects are manifested simultaneously through a threefold process: (i) direct atmospheric heating, (ii) darkening of the snow surface, and (iii) the snow albedo feedback. It is important to note that, without the scaling factor we applied to bring the model BC forcing to the observationally constrained values, the impact of BC on the observed temperature trends would have been marginal. This perhaps explains why the models used in IPCC assessments have not simulated the role of BC in the large warming trend over the Himalayas.