Global climate model simulations project that the tropical Andes Mountains of South America, which are particularly vulnerable to climate change because of a reliance on snow and glacial melt for freshwater resources, will experience enhanced warming in the near future, with both higher rates of warming at higher elevations within the mountain range itself and localized enhancement of warming exceeding surrounding areas of the globe. Yet recent surface temperature changes in the tropical Andes do not show evidence for either elevation-dependent warming or regional enhancement of warming on average. However, it remains a possibility that the expected warming trends in this region have begun to manifest in other ways (e.g., in the free atmosphere or at intermediate mountain elevations). This paper proposes evidence from several reanalysis products that there has indeed been a regional enhancement of midtropospheric warming around the central Andes over the past few decades that makes this region stand out as a hot spot within the broader pantropics. This trend is generally not reproduced by historical AMIP climate model simulations, which suggests that the mechanisms through which the atmosphere is warming over the central Andes are not adequately captured by climate models. Possible explanations for localized enhancement of warming in this region are considered. On the other hand, reanalysis products do not consistently exhibit enhanced warming at intermediate mountain elevations in the central Andes as evidenced by the generally moderate rates of change in the freezing-level height, except perhaps in the highest-resolution reanalysis product.
Future global climate model simulations indicate that mountain regions will experience high rates of warming by the end of this century, with the most dramatic temperature changes at high northern latitudes and at the highest elevations within respective mountain ranges (Bradley et al. 2004, 2006; Nogués-Bravo et al. 2007). The tropical Andes Mountains of South America appear as a potentially localized hot spot of warming compared to other areas of the world in terms of both surface temperatures (Nogués-Bravo et al. 2007) and free-atmospheric temperatures (Bradley et al. 2004, 2006). Additionally, the first regional climate modeling study devoted to future changes over the tropical Andes indicated that the largest warming is expected to occur at the highest elevations (Urrutia and Vuille 2009). Note that the word tropical here is used in the broadest sense. However, there is often a distinction in the literature between the inner tropical Andes and the outer tropical (a.k.a. subtropical or extratropical) Andes. There are no consistent definitions for these terms throughout the literature—in the strictest sense, the inner tropics are defined as equatorial regions between about 5°N and 5°S (e.g., Rabatel et al. 2013), while in the broadest sense they indicate areas between about 15°N and 15°S (e.g., Vuille et al. 2008; Garreaud 2009).
The tropical Andes (Fig. 1) will be heavily impacted by these projected changes because they rely heavily on snow/glacial melt for freshwater and play host to very specialized ecosystems and biodiversity niches. In the inner tropical Andes glaciers act as major natural reservoirs of freshwater. In fact, the strong and minimally varying solar radiation in the inner tropical Andes inhibits the persistence of seasonal snow cover outside of the highest glaciated peaks, unlike in midlatitude mountain ranges like the Rockies or the Alps. Even as far south as La Paz, Bolivia (~16°S), snow cover does not endure for more than a few days (Lejeune et al. 2007; Soruco et al. 2015). Therefore, the glaciers are the only water reservoirs that change on a seasonal time scale in the inner tropical Andes (Kaser et al. 2003, 2005) and are therefore the primary natural buffers to the highly seasonal rainfall cycle and key indicators of climate change in this region (e.g., Mark et al. 2005). Meanwhile, seasonal snowpack plays an increasingly important role in controlling river runoff in the outer tropical and midlatitude Andes (Masiokas et al. 2006).
Yet while some local sites have exhibited extreme surface temperature trends, the average surface temperature rise over recent decades in the high-elevation tropical Andes of 0.1°–0.2°C decade−1 over the time frame 1981–2010 reported by Vuille et al. (2015) does not exceed the average trend over tropical land areas of 0.3°C decade−1 as calculated from the Climate Prediction Center (CPC) Global Historical Climatology Network and Climate Anomaly Monitoring System (GHCN+CAMS) global land surface air temperature analysis (Fan and van den Dool 2008). In addition, although several studies have documented recent intensification of warming with elevation in other mountainous regions of the world (Pepin et al. 2015), surface observations in the tropical Andes do not show evidence for stronger warming at higher altitudes (e.g., Vuille et al. 2003, 2015).
This is perhaps not surprising given that the manifestation of climate change in mountain regions is not necessarily a linear increase in temperatures with elevation. In fact, there have been reports of warming only at intermediate elevations in other mountain ranges, such as near the freezing line (e.g., Pepin and Lundquist 2008). Hence, it may be that the manifestations of enhanced warming in the tropical Andes are simply not captured by the current available surface observations. Therefore, two questions arise: First, have we seen any other evidence of enhanced warming in the tropical Andes in recent decades, either in the free atmosphere or at intermediate mountain elevations, and second, to what extent can we have confidence in model projections of future warming in this area?
Interestingly, the long-term Met Office SCFA radiosonde station—World Meteorological Organization (WMO) station 85442 located in coastal Antofagasta, Chile, at 23.43°S, 70.44°W (yellow star in Fig. 1) with an elevation of 137.0 m above mean sea level (MSL; Met Office 2006)—exhibits a surface cooling trend and an off-surface elevated warming trend over the time period 1979–2008 (Fig. 2). The characteristic contrast between coastal cooling and warming aloft has been attributed by others to the strong modulating effect of the Pacific Ocean, which creates a sharp vertical thermal stratification in the atmosphere on the western side of the Andes (Falvey and Garreaud 2009). Nonetheless, the off-surface warming of about 0.4°C decade−1 (Fig. 2) is notable and could suggest that the free atmosphere over the adjacent mountains to the east could be experiencing similarly high rates of warming. While this observed warming is not actually at the surface, it is within the range of high-elevation surface warming reported by Vuille et al. (2015) for nearby areas in the Andes. This radiosonde record is relatively unique because it is the only observation to our knowledge that is in close proximity to the tropical Andes, which profiles vertical temperatures consistently over the past 30+ years.
In the face of such limited in situ information, one often turns to reanalysis products to provide an interpolated model-driven yet data-constrained picture of climate. For example, Schauwecker et al. (2014) showed that there are strong warming trends in the 500-hPa temperature field over the Andes around 20°S between 1979 and 2012 in two reanalysis products. Yet it is unclear whether this recent warming trend is consistent across products or reliable compared to in situ data and whether these trends are indicative of the elevated warming that is expected to occur in coming decades. Because of the scarcity of in situ observations in and over the high-elevation tropical Andes and the seeming inconsistency between recent observations and future climate projections, both a methodical examination of the available databased and model-driven products and a thorough investigation of common signals and interproduct inconsistencies are needed in this region.
Therefore, this paper considers whether reanalysis products exhibit recent manifestations of enhanced warming in the free atmosphere above or at intermediate elevations within the tropical Andes Mountains of South America and whether historical climate model simulations are consistent with reanalysis products. The following section describes the methodological approach. Section 3 describes the datasets used in this analysis. Section 4 shows the results of the analysis. Section 5 considers potential mechanisms that may be driving recent regional temperature changes. Section 6 discusses the implications of our work for the understanding of climate change in this area.
To better understand temperature changes in and around the tropical Andes Mountains of South America, this study first explores changes in key parameters in reanalysis products. Specifically, linear trends in the monthly mean 500-hPa temperature and the freezing-level height (FLH) time series are examined to determine whether there has been enhanced warming in the free atmosphere or at intermediate elevations in the tropical Andes over recent decades. The specific details of trend analysis are outlined at the end of this section. The altitude of the 0°C isotherm (i.e., FLH) was calculated by linearly interpolating the pressure-level geopotential height data for each product (either reanalysis or climate model) to the pressure level at which the corresponding pressure-level temperature data equals 0°C. We limit our analysis to the time frame 1979–2008 in order to allow direct comparison with historical climate model simulations.
Although reanalyses use a fixed model and a stable data-assimilation system (Kalnay et al. 1996), it is important to be aware that changes in the global observing system with time can create inhomogeneities, most dramatically in 1979 with the dawn of the satellite era, but even in recent decades—for example, with the addition of the SSM/I and NOAA-15 satellite (Santer et al. 1999; Trenberth et al. 2001; Bengtsson et al. 2004; Fasullo 2012). In fact homogenization efforts are ongoing in multiple reanalysis products. Owing to the fact that most reanalyses are not well suited to trend analysis because of frequent changes in data sources, it is recommended that trend analysis be performed on multiple reanalyses at once; if trends are robust across reanalyses then it is more likely, though still not completely certain, that trends are genuine (Kistler et al. 2001).
Therefore in this paper several reanalyses are analyzed in order to identify similarities between them that may be indicative of real trends. It is also for this reason that we include the National Oceanic and Atmospheric Administration (NOAA) Twentieth Century Reanalysis (20CR), version 2, which is specifically designed to be used for trend analysis because of its consistency of integrated data sources (primarily surface pressure station readings) throughout time. One caveat of 20CR, however, is that it does not assimilate radiosonde observations, which means that the free-atmospheric temperatures are largely dependent on the base model’s vertical dynamics. Another caveat is that although 20CR maintains consistent data source types, it does not maintain consistent data points—that is, data stations are added and removed at various times in the historical record. However, this is less of a problem in recent decades when the locations and number of contributing stations have not changed drastically (e.g., Oliver 2015).
From the results of the trend analysis, there appears to be an enhanced midtropospheric warming signal in the greater vicinity of the Altiplano—a region henceforth referred to as the central Andes, which is defined as the areas within the bounding box 10°–25°S, 85°–60°W (black box in Fig. 1) that are greater than 1500 m MSL—in many of the reanalysis products. Therefore, trends in 500-hPa temperatures and FLH in the central Andes are compared to the average trends over pantropical land areas—areas 30°N–30°S and >0 m MSL—as a means to see if changes in this area have been disproportionately intense. The central Andes are assessed as a tropical warming “hot spot” by testing whether this area has experienced average rates of change that are greater than one standard deviation above the global tropical overland area average. Longitude–pressure cross sections of tropospheric temperature trends across the central Andes are also examined in order to see how the vertical distribution of recent warming compares to future projections of warming in the area.
In the interpretation of results from reanalysis products, it is often difficult to separate the effects of actual data versus data-assimilation methodologies and model dynamics. Therefore, we perform a synonymous trend analysis and hot spot assessment of tropical tropospheric temperature and FLH changes in several simulations from phase 5 of the Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012) from the Atmospheric Model Intercomparison Project (AMIP), which are constrained only by observational records of sea surface temperatures (SSTs) and sea ice. A comparison to AMIP simulations allows us to distinguish the effects of SST forcing on atmosphere-only models and gives us a sense for how well the AMIP simulations capture recent climate change in the area of interest. Results may differ across reanalyses and AMIP simulations for a number of reasons—namely, 1) the assimilation of data, particularly satellite-based temperatures, 2) the potentially different SST datasets utilized, and 3) differences in the physical parameterization in the underlying models, which will be especially important in areas that are poorly constrained by observations. Note that there is a well-documented tendency for GCMs to overestimate historical upper-tropospheric (~200 mb; 1 mb = 1 hPa) tropical temperature trends compared to observations (e.g., Fu et al. 2011; Po-Chedley and Fu 2012). While there is some evidence that constraining atmospheric GCMs (AGCMs) with prescribed SSTs reduces this discrepancy (Mitchell et al. 2013), it is still an important issue to be aware of when analyzing tropospheric temperature trends.
To calculate the trend with time, a linear fit is calculated for the raw monthly time series data at each product/model grid cell or station. All time series data are truncated at whole-year intervals, so that a reported trend from 1979 to 2008 is computed including all months from January 1979 through December 2008. Note that because we are working only with whole years, the slope of the linear fit line for the deseasonalized monthly time series is negligibly different from the slope for the raw monthly time series. The slope of the linear fit line yields the annual rate of change in units per year. The total change over the time period of interest is simply the unit trend multiplied by the number of years (e.g., 1979–2008 → 30 yr). Throughout the paper we report the total trend in 500-mb temperature and OLR because the annual rates of change would be quite small, whereas we report the annual rate of change in FLH because the total trend would be quite large. The symbol Δ is used to indicate either the total change or the rate of change depending on the variable—note the reported units to differentiate the two. Trends are computed only if at least 75% of the time series is complete (nonmissing).
To determine whether trends are significant, we apply the Mann–Kendall test (Mann 1945; Sen 1968), a nonparametric, distribution-free method, which tests the null hypothesis of trend absence against the alternative of trend. This test has been shown to work well even with nonnormal and incomplete time series (Yue and Pilon 2004). Throughout this paper we use a significance level of 0.05, meaning that the time series has a linear slope that has a 95% chance of being different from zero.
3. Data and products
a. Elevation data
We use ETOPO1 elevation data at 1-arc-min (1/60°) resolution (Amante and Eakins 2009), averaged in a simple manner to determine the approximate model or product-based gridcell elevations. For spatial longitude–latitude plots, the average elevation of the product-based grid cells is determined by averaging all ETOPO1 points within each grid cell (not including points along the edges of the cells). A thin yellow contour is shown on all spatial plots to denote grid cells with average elevations above 1500 m, which indicates mountainous regions. For meridionally averaged longitude–pressure plots, a red line shows the maximum ETOPO1-derived average product/model gridcell elevation within that meridional range and is also not a real feature but rather a guide for the maximum product/model topography within that meridional range.
b. Reanalysis products and AMIP climate models
Monthly mean output is examined in a suite of six reanalysis products, the details of which, including abbreviations that are used throughout this paper, are outlined in Table 1. This is an adapted version of the table provided on the University Corporation for Atmospheric Research (UCAR) climate data guide website (Dee et al. 2015). Most of the reanalyses examined here are based on AGCMs, except for CFSR, which is based on a coupled atmosphere–ocean–land surface–sea ice model. Although all of the models underlying the reanalyses use terrain-following coordinates, there are some differences in the interpolation rules to convert to pressure surfaces in the final product. Therefore, some products or models (e.g., MERRA) have missing information at low-altitude pressure levels that may happen to intersect with high-elevation land, while other products (e.g., CFSR) generate output at all pressure levels. Monthly mean output is also examined for a suite of six AGCMs, the details of which are outlined in Table 2, which were run for the 30-yr historical time period 1979–2008 in accordance with the AMIP time-varying SST protocol. The results for the AMIP models are ensemble averages over all available realizations for each model.
a. Reanalysis products
Since in situ measurements of surface temperatures do not yet support the idea that the tropical Andes are experiencing regionally enhanced warming (Vuille et al. 2015), we turn to reanalysis products to determine how free-atmosphere temperatures and the FLH have been changing in the tropical Andes. In an extension of the analysis done by Schauwecker et al. (2014), total trends in the 500-mb temperature field (ΔT500mb) are examined over the global tropics in six reanalysis products over the time period 1979–2008. Looking at the spatial distribution of midtropospheric temperature changes in the different reanalyses, we see that the central Andes—specifically around the Bolivian Altiplano—stand out when compared to the rest of the pantropics (Fig. 3), though the spatial extent and magnitude of warming trends vary between the different reanalysis products. In fact, the average ΔT500mb over the central Andes is greater than one standard deviation above the pantropical land-area average trends in most of the reanalyses examined here, except for CFSR and 20CR, the latter of which still comes very close to that threshold (Table 3). Note that because of the wide range of values between the various products (e.g., Table 3), the panels of figures displaying the spatial distribution of trends throughout this paper (e.g., Fig. 3) may have different contour scales, which is necessary to make regional differences more visually apparent. It is also evident from looking at the area-averaged time series of 500-mb temperatures that the trend over the central Andes is much steeper—with an interproduct average ΔT500mb of 0.84°C compared to 0.50°C over pantropical land areas—and is much more uniform across the different reanalysis products (Fig. 4). Therefore, the central Andes have likely been a hot spot in recent decades in terms of midtropospheric warming.
On the other hand, the majority of reanalyses do not show the central Andes as being a hot spot in terms of FLH. The reanalyses examined here indicate an average rate of change in the FLH (ΔFLH) over the central Andes ranging anywhere from about 2 to 7 m yr−1 over the 30-yr time frame of 1979–2008 (Table 4). R1 is the exception to this because it shows negative (though insignificant) changes in FLH. Overall, the Andes do not stand out as an area of enhanced FLH change with respect to the rest of the tropics (Fig. 5), except in CFSR where there are extremely local rises in FLH in the central Andes. Upon close inspection of Fig. 5f, the CFSR product shows both the high central Andes and the high Himalayas as being areas of extremely local FLH rise. In fact, if we adjust the minimum mountain elevation from 1500 to 3000 m MSL, both the central Andes and the tropical Himalayas (30°–25°N) qualify as hot spots of FLH change in CFSR. It is, however, worthwhile to note that there is not a strong correlation between elevation and FLH rise in those mountain ranges—rather, local dynamics appear to determine the locations of the strongest changes in FLH.
Yet in the majority of reanalyses, the central Andes display ΔFLH that are within one standard deviation of the pantropical land-area average (Table 4) and therefore do not meet this study’s qualification for a hot spot. Note that while MERRA does indicate that the central Andes qualify as a hot spot of FLH change, the center of FLH rise activity is actually located over the Chaco lowlands and not in the Andes themselves (Fig. 5). Nonetheless, although the central Andes have not likely been a hot spot of FLH activity, moderate rises are still very important in this region because the FLH determines the extent of snowpacks and/or glaciers (e.g., Bradley et al. 2009), as well as potentially how much precipitation is received as snow versus rain (e.g., Shook and Pomeroy 2012).
When we look at the longitude–pressure cross sections of the temperature trends ΔT over the central Andes (meridionally averaged from 10° to 25°S), it is clear that the vertical distribution of warming also varies greatly between the different reanalyses (Fig. 6). Whereas R1, R2, and MERRA show an elevated hot spot of warming above or around the mountains, CFSR and 20CR show more of an altitudinal downward extension of upper-level warming over the mountain peaks. The latter behavior more closely resembles the vertical profile of warming projected by climate model simulations of the future (Bradley et al. 2004, 2006). It is worthwhile to note that the results from 20CR must be interpreted with caution considering that the free-atmospheric temperatures are not heavily influenced by data because 20CR does not assimilate radiosondes. Although there are very few radiosonde observations to assimilate in this region, the other reanalysis products have the advantage of assimilating satellite-retrieved temperature profiles. Still, it is interesting that 20CR produces similar spatial patterns in temperature trends to the other reanalysis products that do assimilate radiosondes. This could indicate that 20CR actually does a good job of simulating free-atmosphere temperature trends in the absence of the assimilation of satellite and radiosonde data.
Regarding radiosondes, the SCFA sounding record does not show significant warming around 500 mb (Fig. 2) as one might expect from the reanalyses (Fig. 3). To the best of our knowledge, this particular radiosonde record is assimilated into all of the reanalysis products, except for 20CR of course. Therefore, the discrepancy between this station and the reanalysis products that assimilate it could be due to a few different possibilities. First, it could be that the base models underlying the reanalysis products are generating spurious midtropospheric warming that is not in alignment with reality. Second, the off-surface warming seen at this coastal station could extend up and along the mountains, such that there is midtropospheric warming localized only to the mountains, while the reanalysis products produce a pattern of warming that is too broad. Finally, it could be that other data (particularly satellite-derived temperature profiles) exhibit warming trends that override this particular station in the data assimilation schemes of the reanalyses. Although the SCFA station is certainly a contributor to the trends, there are a multitude of other databased contributors (short-term radiosonde records in other areas, satellite records, surface observations, etc.) and model-based factors (e.g., the ability of the base models to advect temperature anomalies) that could be driving the trends that manifest in the reanalyses. Therefore, the SCFA radiosonde record alone cannot support the idea of a regional warming hot spot, and we caution against in-depth interpretation of this record when comparing to reanalysis products. Furthermore, since even 20CR shows regional midtropospheric warming in the vicinity of the central Andes, it is unlikely that the SCFA station is causing major differences between 20CR and the other products.
It is likely that the elevated warming portrayed in Fig. 6 contributes to the rising FLH in the central Andes. With the exception of ERA-Interim, rises in FLH are generally larger in the higher resolution, with CFSR exhibiting extreme local rises in FLH of close to 30 m yr−1 (Table 4). However, the maximum FLH change should be interpreted with caution because extreme values could be the result of grid-scale numerical artifacts. Furthermore, such extreme rises in FLH are likely unrealistic and would outpace glacier recession trends in the region (e.g., Vuille et al. 2008; Rabatel et al. 2013). Nonetheless, the general theme of higher, more local rises in FLH in the higher-resolution products suggests that higher-resolution base models may be required to resolve the feedback between the high topography of the mountains and the regional atmospheric circulation patterns that are important for driving climate change in this region. It also suggests that we should probably see higher rates of warming at higher elevations within the Andes. However, the surface station data available from the local environmental agencies of Andean countries does not provide evidence for elevation-dependent warming, except perhaps on the lower-elevation eastern slopes (Vuille and Bradley 2000; Vuille et al. 2003, 2015).
This discrepancy between the observational data and reanalysis products could be due to several different factors. For example, it could be that the assimilation of new data sources over time into reanalyses produces an artificial temperature rise at higher elevations. However, it is difficult to determine whether the reanalysis products exhibit strong biases related to changes in the observing system. While the time series of reanalysis midtropospheric temperatures (Fig. 4) do not show any clear jumps that would indicate assimilation of drastically new data sources, there is a gradual change in both the trend and the range of variability during and after the 1990s that is present on both a regional (Fig. 4a) and a pantropical (Fig. 4b) scale. Nonetheless, the 20CR product, which maintains consistent data sources over time, also shows a localization of warming trends, which gives more confidence that the regional enhancement of warming is not simply an artifact of data assimilation.
The lack of elevation-dependent signals in the station data presented in the aforementioned studies may also be simply due to averaging over large areas. For instance, this study finds evidence for possible elevated warming in the region centralized around the Altiplano, while the results of Vuille et al. (2015) were an average of stations over a large meridional extent. It may be that some subregions do in fact exhibit elevation-dependent temperature trends but that they are outweighed by the surrounding areas that go into the overall average. This is not surprising given than prior studies have shown that temperature trends in the Andes are highly dependent on the both the geographic location (latitude, elevation, and aspect) and the time period analyzed (e.g., Vuille et al. 2015). There is also the possibility that the surface stations have simply not felt the effects of elevated warming yet because of natural interannual variability and local insulating mechanisms.
One caveat of this study is that it is challenging to validate the elevated warming trends exhibited by the reanalysis products with data because there are few long-term sounding readings in the tropical Andes and most individual satellite observations are not long enough to span a 30-yr climatological time period. For instance, the Atmospheric Infrared Sounder (AIRS) satellite instrument data (which plays an important role in the MERRA reanalysis) is only available from 2003. However, the consistency of a qualitative trend in the reanalysis products examined here, the presence of a dipping down of warming in the most internally consistent product with time (20CR), and the evidence of extreme elevated warming from the Antofagasta sounding station give us confidence that that there is a real elevated warming trend above the mountains in the central tropical Andes.
b. Historical AMIP climate simulations
Since it is difficult to tease apart the model-driven aspects of reanalysis products, we investigate temperature trends in several AMIP climate model simulations. Figure 7 shows the spatial pattern of ΔT500mb in all the AMIP models. In contrast to the reanalyses there is no localized hot spot of warming around the central Andes in any of these simulations (Table 3). The rates of midtropospheric warming in the central Andes are highly comparable to the rates of change over other pantropical land areas and are often even lower than the surrounding areas. In terms of FLH, the AMIP models examined here also do not show any hot spot–type behavior around the central Andes (Table 4), and they tend to underestimate trends in FLH on the whole as shown by the smaller color-bar range in Fig. 8. And while the AMIP simulations do show elevation-dependent warming, with stronger warming at higher elevations and cooling at lower elevations over the western Andean slopes, consistent with observations (Falvey and Garreaud 2009), they do not show a prominent localized warming over the central Andes in the midtroposphere (Fig. 9). In GEOS-CCM and CMCC-CM there are slight indications of a dipping down of the upper-level warming over the mountain peaks (Fig. 9), but not to the extent that we might expect given the future climate model projections.
Hence, in contrast to the fact that GCMs reasonably represent the global average tropical midtropospheric warming (Fu et al. 2011; Po-Chedley and Fu 2012; Mitchell et al. 2013), the AGCMs examined here fail to capture the regional enhancement of free-atmospheric temperature changes over the central Andes in comparison to reanalyses. This could indicate that the regional manifestations of elevated warming are not well captured by the GCMs, meaning that future projections may be underestimating the amount of change in this area, which could be very problematic for water resources. Or perhaps the shortcomings of the AMIP simulations are due to the very tendency of GCMs to exaggerate upper-tropospheric temperature change on average. A full exploration of these possibilities lies outside the scope of this paper because the AMIP models examined here are not the same as the ones that were shown to predict enhanced tropospheric warming over the tropical Andes (Bradley et al. 2004) or the ones that were examined for assessing historical tropical tropospheric temperature trends (Fu et al. 2011; Po-Chedley and Fu 2012; Mitchell et al. 2013).
Those theories aside, the discrepancy between the AMIP simulations and reanalyses in midtropospheric hot spot–type warming could be due to uncertainties or deficiencies in the observational SST datasets used to force the AGCMs (e.g., Flannaghan et al. 2014). It could also be due to differences of model resolution (e.g., Mitchell et al. 2013). However, in contrast to the general pattern in the reanalysis products, the higher-resolution AMIP climate models do not necessarily show more localized warming or even higher magnitudes of midtropospheric warming over the central Andes (Table 3). Alternatively, it could indicate that the AGCMs examined here either respond too slowly to observed SST forcings or diffuse anomalous sea surface heating too quickly throughout the tropical troposphere.
5. Potential mechanisms for enhanced warming
Assuming that the elevated warming exhibited in the reanalysis products is not a spurious trend (which still remains a possibility, though unlikely), the following question is raised: What is driving localized enhancement of warming in the Andes, and why do the AMIP climate models not capture it? Below, we consider several possible mechanisms and suggest areas for future research and more rigorous testing across reanalyses, models, and observations based on the plausibility of each potential driver.
a. Sea surface temperatures
One might hypothesize that the broad pattern of midtropospheric warming seen in several of the reanalysis products (Fig. 3) is due to changes in the mean state and/or internal variability of large-scale SSTs. On interannual time scales, El Niño–Southern Oscillation (ENSO) is known to play a major role in the interannual variability of temperature and precipitation in the tropical Andes (Vuille et al. 2000a,b). However, Vuille et al. (2015) show a low correspondence between tropical Andean station temperature trends and ENSO over recent decades. Furthermore, a quick look into 20CR shows that the spatial patterns of seasonal ΔT500mb do not resemble the ENSO composite maps (not shown). So it is not simply the case that a shift in ENSO behavior (say a shift toward more El Niño–like conditions) has resulted in the recent pattern of midtropospheric warming over the central Andes.
Alternatively, the Pacific Ocean has recently entered a negative phase in both its decadal variability and its multidecadal variability (Trenberth and Fasullo 2013; England et al. 2014; Meehl et al. 2014; Steinman et al. 2015), which is likely to play some role in changing temperatures over the tropical Andes. Vuille et al. (2015) argued that the shift of the Pacific decadal oscillation (PDO) into a negative phase drives much of the temperature variability in coastal and low-elevation (<2000 m) sites along the western slopes of the tropical Andes, which have exhibited insignificant temperature trends over recent decades. They also point out that the higher-elevation sites seem to be insulated from the effects of Pacific Ocean variability because of the high thermal vertical stratification of the atmosphere along the western Andean slopes (e.g., Falvey and Garreaud 2009) because they continue to warm at a rate of anywhere from 0.1° to 0.2°C decade−1. This may partly explain why the Andes stand out as a hot spot of warming over recent decades because they are relatively insulated from the effects of the long-term variability in the Pacific Ocean, which modulates temperatures in much of the surrounding regions. It is also possible that the remote effect of changes in the Pacific Ocean’s mean state or internal variability could be felt by the tropical Andes via other teleconnections. For example, a decrease in the cross-Pacific temperature gradient could result in a weakening of Walker circulation (e.g., Vecchi and Soden 2007), which would likely increase tropospheric temperatures over the Andes.
b. Regional atmospheric circulation patterns
It is also possible that changes in regional circulation patterns and/or water vapor flux are driving temperature changes in the Andes. According to the NOAA interpolated outgoing longwave radiation product (Liebmann and Smith 1996) the subtropical Andes and the Chaco lowlands have been experiencing increases in outgoing longwave radiation (OLR) over recent decades (Fig. 10f, indicated with a black box) which could potentially indicate decreases in water vapor and/or cloudiness. One possible explanation for this trend could be the recent intensification and southward shift of the South Pacific anticyclone (SPAC) as documented in Falvey and Garreaud (2009). This change, which is a consistent feature of future climate model simulations (Stocker et al. 2013), acts to cool the western coast of South America and may induce moisture divergence from the Andes. Another explanation could be a weakening of the Chaco low pressure system to the east of the Bolivian Altiplano resulting in less uplift over the Chaco lowlands and less moisture transport to the subtropical Andes.
Each of these theories is difficult to validate in the reanalysis products examined here because most show OLR trends that are at odds with or spatially inconsistent with the NOAA OLR product (Fig. 10). For example, MERRA shows the exact opposite trend in the area of interest (enhanced Chaco low pressure system due to a westward shift of the South American low-level jet). In the case of MERRA, this suggests that the internal model dynamics generate an incorrect mechanism in order to match the observed warming trends. This may be because MERRA exhibits negative biases in the overall climatological representation of precipitation during this time period (i.e., MERRA is much too dry) as compared to observations (Quadro et al. 2013), which might cause it to develop a runaway positive feedback between precipitation, atmospheric stability, and temperature changes.
c. EDW mechanisms
It is worth coming back to the general theme that higher-resolution reanalyses produce more localized and stronger trends in FLH over the tropical Andes in addition to the extremely local FLH rises exhibited by CFSR (Table 4). This, combined with the fact that future climate model simulations project more warming at higher elevations in the Andes, leads us to believe that local elevation-dependent feedbacks cannot be entirely discounted. A recent review by Pepin et al. (2015) found evidence for elevation-dependent warming (EDW) in many different mountain systems around the world and outlined several different mechanisms that explain why higher-elevation areas might experience more warming than lower-elevation areas. We examine a few of these mechanisms here and discuss whether and/or to what extent they may contribute to warming in the Andes.
1) Albedo–radiative feedback
One such mechanism is the albedo–radiative feedback mechanism whereby melting of snow and ice reduces the albedo of the surface causing more radiation to be absorbed at the surface, which increases melt rates and results in a positive feedback loop. This is unlikely to be a major actor in the inner tropical Andes because the strong and minimally varying solar radiation inhibits the persistence of snow cover on seasonal time scales (Kaser et al. 2003, 2005; Lejeune et al. 2007). On the other hand, long-term changes in snow cover may be important in the subtropical Andes where snowpacks cover more surface area and last longer. There have been a few studies of changes in snow cover in the extratropical Andes (south of about 30°S), though most report insignificant trends over the past century (Prieto et al. 2001; Masiokas et al. 2006, 2012). More recently, Vuille et al. (2015) report that surface temperatures in the extratropical Andes (defined in their analysis as 18°–42°S) have been decreasing at an average rate of −0.05°C decade−1 over the time period 1981–2010, which could suggest that snow-cover changes have been minimal over recent decades in this region. Nonetheless, a comprehensive analysis of snow-cover trends over recent decades in the vicinity of the warming exhibited by the reanalysis products seems warranted in order to better understand the potential for albedo feedbacks in this region.
2) DLR–water vapor feedback
Another possible candidate that could be responsible for the elevated warming we see in the tropical Andes is the feedback between downward longwave radiation (DLR) and water vapor. Both observations and models have shown that changes in absolute humidity at high elevations can result in a greatly enhanced greenhouse effect owing to the nonlinear relationship between humidity and DLR (Philipona et al. 2005; Rangwala et al. 2009, 2010; Rangwala 2013). In the 20CR product, there is some evidence for this mechanism playing a role because the eastern slopes of the central Andes show an extremely high correlation of >0.9 between 500-mb pressure vertical velocity and DLR. Increasing vertical motion could lead to more convergence and uplift of water vapor thereby trapping longwave radiation, which is consistent with warming temperatures. Although the NOAA satellite OLR product (Fig. 10f) shows this to be an area of increasing OLR, this does not necessarily contradict the mechanism in question because increases in OLR do not necessarily indicate decreases in clouds/water vapor (e.g., Vuille et al. 2003). Hence, this feedback mechanism is still a viable candidate for contributing to rising temperatures in the region.
3) Cloud–radiative feedbacks
Changes in clouds associated with changes in regional circulation are a strong candidate for inducing elevated warming over the mountains. Yet the complexity of cloud feedbacks and the lack of long-term observations in the Andes make this a difficult mechanism to test. According to the NOAA interpolated satellite OLR product, most of the central and southern part of the continent, including the central and subtropical Andes, has been experiencing increases in OLR (Fig. 10f), which could indicate decreases in cloudiness and therefore less precipitation. However, changes in OLR could also be influenced by changes in surface temperature and water vapor content, perhaps even to a greater extent than changes in clouds (especially in nonconvective regions and seasons), which means that this mechanism is heavily linked to the previous mechanism.
The broad region of increased OLR is captured to some extent in the reanalysis products except for an incorrect band of increased cloudiness or no significant change on the eastern slopes of the Andes and into the Chaco lowlands (Fig. 10). The AMIP climate models on the other hand generally show insignificant OLR trends or trends that are too weak in this region. This inconsistency may indicate that the climate models do not correctly simulate decreases in cloudiness and/or water vapor that are associated with less blocking of intense solar radiation, which heats up the midtroposphere. At the same time, the tendency of reanalyses to produce an OLR trend that is at odds with observations on the eastern slopes of the Andes and into the Chaco lowlands requires an explanation.
While each of these aforementioned EDW mechanisms are likely drivers of the enhanced midtropospheric warming over the central Andes that is exhibited by reanalyses, it is important to be aware that changes in radiation emissions or absorptions, water vapor, clouds, and winds are all intertwined. Therefore, it may be that some combination of these mechanisms is at play. In addition, although future climate model simulations predict both free-atmosphere warming and EDW in the tropical Andes, it is not clear which is the primary driver and how closely the two are linked. Furthermore, the lack of recent EDW signals in surface station data (Vuille et al. 2015) does not necessarily preclude the possibility that free-atmospheric warming will drive EDW in the near future. These points need to be more thoroughly examined and will be the subject of future work.
6. Conclusions and discussion
This paper presents evidence from multiple reanalysis products that the central Andes Mountains of South America have experienced rates of midtropospheric warming that are much higher than the rates exhibited by other pantropical land areas over the past few decades. This behavior is consistent with and perhaps a prelude to the projected enhanced warming in this region (Bradley et al. 2004, 2006; Nogués-Bravo et al. 2007). Whether the vertical structure of warming is also localized or is more of a regional drawdown of upper-level warming as projected by future climate model simulations (Bradley et al. 2004, 2006) is still unclear. In addition, this paper reveals that there is no consistent evidence that warming is enhanced at intermediate mountain elevations around the FLH in the central Andes compared to pantropical land areas. Nonetheless, it seems that the central Andes have been and will likely continue to be a hot spot for global climate change.
Historical AMIP climate model simulations generally do not exhibit the same regional enhancement of midtropospheric warming over the central Andes, which ties into the greater issue of GCMs’ ability to simulate free-atmosphere temperature changes and leads us to suspect that the models do not adequately capture the mechanisms that have enhanced warming in this area over recent decades. When we examine possible mechanisms that might be responsible for elevated warming over the central Andes, we find that a variety of both large-scale and local mechanisms are likely drivers. In particular, large-scale SST patterns and changes in regional atmospheric circulation patterns could result in the broad pattern of warming exhibited in many of the reanalyses, while EDW-type feedback mechanisms may drive more localized enhancement of warming along the mountain peaks. Moving forward, it is important that we work toward a better understanding of the mechanisms that enhance warming in this region in order to improve their representation in climate models.
Although there has been a recent surge in the amount of regional climate modeling efforts over South America (see Solman 2013 for a review), to our knowledge there has been only one regional climate model simulation over the historic time frame of interest here—the Swedish Meteorological and Hydrological Institute (SMHI) Rossby Centre ran a regional atmospheric model (RCA4) that is based on the numerical weather prediction model known as HIRLAM (Unden et al. 2002), with approximately 50-km resolution for the Coordinated Regional Climate Downscaling Experiment (CORDEX) South America domain over the time period 1979–2005. A preliminary examination of the trends in the 500-mb temperatures (not shown) indicates that this model does not show the central Andes as a hot spot for warming over that time period. This reinforces the conclusion that increased resolution alone is not enough to rectify the inability of models to capture recent Andean warming. Nonetheless, there is certainly added value to increasing resolution in this area because of the dramatic topography, which is oversmoothed in global climate models. Improving the resolution will likely improve the representation of the low-level jet and the upslope winds. Since this is but one regional climate model, it is impossible to know whether other regional climate models might be better able to represent the physical dynamics that contribute to the recent temperature trends. It also highlights the need for additional historical regional climate modeling efforts over South America.
In summation, this study reveals how much remains to be done to properly understand recent Andean temperature changes. These efforts are important because the impacts of climate change in mountain regions are far reaching, including everything from shifts in vegetation and biodiversity belts to the loss of cryospheric environments (i.e., glaciers and/or seasonal snowpack). The disappearance of many low-altitude Andean glaciers (e.g., Rabatel et al. 2013) is a sobering testament to this. The decline of these natural runoff buffers is expected to have dramatic economic consequences for Andean countries where rapid population growth and resource exploitation are placing extra demands on freshwater resources (e.g., Vergara et al. 2007). The loss of wetland areas (e.g., Bury et al. 2013) poses an additional challenge to water resource management in the tropical Andes. At the same time, there have been clear shifts in ecosystem and species distributions in the tropical Andes that cannot always keep pace with the dramatic changes in climate (Feeley et al. 2011; Rapp et al. 2012; Forero-Medina et al. 2011; Bury et al. 2013). Therefore, a better understanding of recent climate change and better representation in models is essential because it will ultimately help regional agriculture and water resources managers plan for the future in this highly vulnerable area.
Met Office SCFA radiosonde station—World Meteorological Organization (WMO) station 85442—data are collected by the Met Office. We use a version of the data record provided by the Department of Atmospheric Science at the University of Wyoming (http://weather.uwyo.edu/upperair/sounding.html). The authors acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 2 of this paper) for producing and making available their model output. The authors thank Dr. Valentina Aquila, NASA GSFC, for providing the GEOS-CCM AMIP output. A. Russell acknowledges support from NSF through the Water, Climate and Health IGERT at Johns Hopkins University (Grant 1069213).