2.1 Data collection and preprocessing

2.1.2 Climatic data correction and evaluation

For the temperature variable, we first quantified the local vertical annual temperature lapse rates using the in situ measurements for 33 sites across the Andes (see Table S1 and Fig. S1 in the Supplement). Then, the TerraClimate temperature was corrected with these in situ records so that they could be used in the simulations (“Correction” box in Fig. 1). Lastly, the corrected TerraClimate temperature was evaluated via a comparison with the 34 in situ data (“Evaluation” box in Fig. 1). Conversely, the precipitation variable from the TerraClimate reanalysis was scaled using the mass balance measurements for 10 monitored glaciers and was evaluated for 15 glaciers (“Correction” and “Evaluation” boxes in Fig. 1). Specific data are available in Tables S3–S5.

Vertical temperature lapse rates (LRs) from the in situ records were estimated for each glaciological zone across the Andes as per Gao et al. (2012). The temperature LRs are presented in Fig. S1. These gradients were applied to correct the raw TerraClimate temperature on the glaciers (rTC_t). The corrected TerraClimate temperature at the mean elevation of glacier (cTC_t) was calculated using the following equation:

$$\begin{array}{}\text{(1)}& {\mathrm{cTC}}_{\mathrm{t}}={\mathrm{rTC}}_{\mathrm{t}}+\mathrm{\Gamma }\cdot \mathrm{\Delta }h,\end{array}$$

where Γ is the temperature LR estimated here and Δh is the elevation difference between a glacier elevation and the mean elevation of the TerraClimate grid cell where the glacier is located.

Then, we assessed the cTC_t in meteorological station locations (9–51° S) on a monthly scale, paying attention to the monthly variability in temperature as well as to the mean temperature for all the periods with data. The cTC_t monthly mean variability was evaluated using the Pearson correlation coefficient, whereas the mean temperature for the whole period considered the mean difference between cTC_t and the observed temperature (biases).

In addition, the total precipitation was scaled (cTC_p) using precipitation factors (P_f) for each glaciological zone across the Andes. (See the relationship between solid precipitation and P_f in Eq. 3.) In a second step we discriminated between snowfall and rainfall using a linear regression between the temperature thresholds to obtain the solid and liquid precipitation fraction (Maussion et al., 2019). We ran 31 simulations for 18 glaciers with mass balance measurements across the Andes using P_f values between 1 and 4 taking previous studies into account (Masiokas et al., 2016; Burger et al., 2019; Farías-Barahona et al., 2020). Ultimately, 10 glaciers were selected (see Table S3), because their simulated mass balances showed a closer standard deviation in comparison with measurements. The goal was to find the closest simulated mass balance standard deviation (simSD_mb) in comparison with the measured mass balance standard deviation (obsSD_mb) using different P_f values (Eq. 2):

$$\begin{array}{}\text{(2)}& P_{\mathrm{f}}=\mathit{\{}\mathrm{1}\le P_{\mathrm{f}}\le \mathrm{4}:{\mathrm{simSD}}_{\mathrm{mb}}\approx {\mathrm{obsSD}}_{\mathrm{mb}}\mathit{\}}.\end{array}$$

A similar methodology was proposed by Marzeion et al. (2012) and Maussion et al. (2019). The results of the closest simulated mass balance standard deviations and associated P_f are presented in Table S3. The simulated annual mass balance was evaluated on 15 monitored glaciers using a Pearson correlation coefficient and bias (as the average difference) from simulated mass balance and measured mass balance (“Evaluation” box in Fig. 1). In addition, details such as snow/rainfall partitioning are described hereafter and in the model implementation (Sect. 2.2).

Figure 2Recent glacier changes across the Andes. The glacier changes represent the mean annual differences between the periods 2000–2009 and 2010–2019 per catchment (n = 786). Panel (a) shows the distribution of the glaciological zones (11° N to 55° S), followed by the (b) volume and (c) area changes at the catchment scale. The (d) annual specific mass balances are presented in each glaciological zone (the shaded areas are the standard deviation), where the straight lines correspond to the mean geodetic mass balance (2000–2019) estimated by Hugonnet et al. (2021).

2.2 OGGM details

We ran the OGGM model (Maussion et al., 2019) for each glacier and then the results per catchment were aggregated for each of the 786 catchments across the Andes (including the three selected test catchments for a detailed analysis). OGGM is a modular and open-source numerical workflow implemented in Python that provides preprocessed datasets, such as DEMs, glacier hypsometry, and glacier flowlines, that can be used to explicitly simulate glacier mass balance and ice dynamics using calibrated parameter values for each glacier. Here, we ran OGGM from level 2, comprising the flowlines and their downstream lines. However, we used a new baseline climate time series (corrected TerraClimate) as input data. We also calibrated the mass balances and the bed inversion (ice thickness) which allowed us to obtain hydrological outputs (glacier runoff) (details at https://docs.oggm.org/en/v1.4.0/input-data.html, last access: July 2022). The spatiotemporal configuration of the model used in this study is at the glacier scale and at the monthly time step. The simulation results were analyzed at different spatial scales: by glacierized catchment, glaciological zone, and regionally.

Table 1Calibrated parameter values used in the glacier mass balance and volume simulations across the Andes (11° N to 55° S) during the period 2000–2019.

^∗ Related to zones of the tropical Andes region: inner tropics (IT) and outer tropics (OT2 and OT3).

Download Print Version | Download XLSX

The required input data for running the model are as follows: air temperature and precipitation time series, as well as glacier outlines and surface topography for specific years. From these input data we computed annual simulated processes such as the surface mass balance, glacier volume and area, monthly glacier melt (snow and ice), and rainfall on glaciers (Fig. 1). Modeled processes, such as the surface mass balance and glacier volume, were calibrated (Fig. 2; Table 1). The calibration procedure of the parameters was applied per glacier to match the simulated mass balance of 2000–2019 to the geodetic mass balance product from Hugonnet et al. (2021). The simulated glacier volume was calibrated using the Farinotti et al. (2019) product at a glaciological zone scale to fit the Glen A parameter. In other words, the same Glen A parameter was used for each glaciological zone.

First, using a glacier outline and topography, OGGM estimates the flow lines and catchments per glacier, and then the flow lines are calculated using a geometrical algorithm (adapted from Kienholz et al., 2014). Assuming a bed shape, it estimates the ice thickness based on mass conservation and shallow-ice approximation (Farinotti et al., 2009; Maussion et al., 2019). After these numerical steps, it is possible to determine the area and volume per glacier. The mass balance is implemented using a precipitation phase partitioning and a temperature-index approach (Braun and Renner, 1992; Hock, 2003; Marzeion et al., 2012). The monthly mass balance mb_i at an elevation z is calculated as follows:

where ${\mathrm{TC}}_{\mathrm{p}\mathrm{i}}^{\mathrm{snow}}$ is the TerraClimate solid precipitation before being scaled by the precipitation correction factor (P_f), M_f is the glacier's temperature melt factor, and cTC_t  i is the monthly corrected TerraClimate temperature. P_f and M_f parameters are related to the snow/ice onset (T_melt) and precipitation fraction ($T_{\mathrm{i}}^{\text{snow}}$ and $T_{\mathrm{i}}^{\text{rain}}$). Their values are different across the Andes. T_melt is the monthly air temperature above which snow/ice melt is assumed to occur (0 °C for the Dry and Wet Andes and 2.1 °C for the tropical Andes). $T_{\mathrm{i}}^{\text{snow}}$ is calculated as a fraction of the total precipitation (cTC_p) where 100 % is obtained if cTC_t  i ≤ $T_{\mathrm{i}}^{\text{snow}}$ (0 °C for the Dry and Wet Andes and 2.1 °C for the tropical Andes) and 0 % if cTC_t  i ≥ $T_{\mathrm{i}}^{\text{rain}}$ (2 °C for the Dry and Wet Andes and 4.1 °C for the tropical Andes), using a linear regression between these temperature thresholds to obtain the solid and liquid precipitation fraction. Here, M_f was calibrated for each glacier individually using the previously described glacier volume change datasets (Hugonnet et al., 2021). The calibrated parameter values are summarized by glaciological zone in Table 1.

2.2.1 Model setup, calibration, and validation

The input data are as follows: the corrected monthly TerraClimate precipitation (cTC_p) and temperature (cTC_t), glacier outlines were obtained from RGI v6.0 (RGI Consortium, 2017), assuming the glacier outlines of all glaciers were made for the year 2000. The surface topography data were sourced from NASADEM (Crippen et al., 2016). NASADEM has a spatial resolution of 1 arcsec (∼ 30 m), and the data were acquired in February 2000 (NASA JPL, 2020). The simulated glacier volume was calibrated using the Farinotti et al. (2019) product at a glaciological zone scale to fit the Glen A parameter (Fig. 2; Table 1).

Lastly, the simulated mass balance was evaluated in comparison with in situ mass balance observations (Marangunic et al., 2021; WGMS, 2021). Although the OGGM outputs are in calendar years and the observations are in hydrological years, we consider it essential to evaluate the interannual performance (Pearson correlation, p value, variance, RMSE, and bias from average difference) and the cumulative mass balance since the year 2000.

Table 2Mean annual changes in glacier area and volume, glacier runoff, and climate between the periods 2000–2009 and 2010–2019 at the glaciological zone scale across the Andes (11° N to 55° S).

Download Print Version | Download XLSX

3.1 Climatic variations on glaciers across the Andes during the period 2000–2019

The climate associated with 786 Andean glacierized catchments (11° N to 55° S) presents a mean corrected TerraClimate temperature (cTC_t) of −0.2 ± 2.2 °C and a mean annual corrected TerraClimate precipitation (cTC_p) of 2699 ± 2006 mm yr⁻¹ between 2000 and 2019. The various glaciological regions show significant climatic differences, with contrasting extreme values between the tropical Andes and Wet Andes in terms of mean annual precipitation (939 ± 261 and 3751 ± 1860 mm yr⁻¹, respectively) and also mean annual temperature between the Dry Andes and tropical Andes (−3.7 ± 1.4 °C and 1.3 ± 0.8 °C, respectively). Certain glaciological zones highlight very negative and positive mean annual temperature values such as Dry Andes 2 (−4.8 °C) and Wet Andes 2 (1.9 °C), as well as lower and higher cumulative precipitation values such as Dry Andes 1 (447 mm yr⁻¹) and Wet Andes 5 (6075 mm yr⁻¹). Meanwhile, variations in climate between the periods 2000–2009 and 2010–2019 across the Andes show a cumulative precipitation decrease of −9 % (−234 mm yr⁻¹) and a mean annual temperature increase of 0.4 ± 0.1 °C. Between these two periods, precipitation decreases primarily in the Dry Andes (−256 mm yr⁻¹; −23 %) and Wet Andes (−337 mm yr⁻¹; −9 %), and increases in the tropical Andes (44 mm yr⁻¹; 5 %), whereas the temperature increases between 0.3 and 0.4 °C in all regions. At the glaciological zone scale, only the tropical Andes and Dry Andes 1 (12 %) show a cumulative increase in precipitation, whereas a larger decrease in precipitation is found in Dry Andes 2 (−32 %) and Dry Andes 3 (−27 %). The mean annual temperature increases in all zones, especially the inner tropics (+0.6 °C) followed by Wet Andes 3 (+0.5 °C). A summary of the variations in climate by glaciological zone is presented in Table 2.

Our cTC_t evaluation is statistically significant (p < 0.01) at 32 meteorological stations with a mean temperature bias of 0.4 °C and a mean correlation of 0.96. These results are available in Fig. S2 and Table S2. The regional results show a larger bias in the tropical Andes (mean = 2.1 °C for four stations) with a meteorological station mean elevation of 4985 $\mathrm{m}\mathrm{a}.\mathrm{s}.\mathrm{l}.$, where cTC_t cannot represent the mean monthly temperature. However, cTC_t represents well the maximum temperatures in spring/summer and the minimum temperatures in winter. The lowest bias is observed in the Wet Andes and Dry Andes. The Wet Andes, with a meteorological station mean elevation of 813 $\mathrm{m}\mathrm{a}.\mathrm{s}.\mathrm{l}.$, shows good results in terms of reproducing the mean monthly temperature in most stations, with a minimum correlation higher than 0.86. In the Dry Andes, with a meteorological station mean elevation of 3753 $\mathrm{m}\mathrm{a}.\mathrm{s}.\mathrm{l}.$ (18 stations) and bias of 0.2 °C, the cTC_t reproduces the mean monthly temperature very well. However, in some stations, such as La Frontera and Estrecho Glacier (29° S), the mean cTC_t is warmer than 6 °C, whereas in other stations, such as El Yeso Embalse (33.7° S) and Cipreses glacier (34.5° S), the mean cTC_t is colder than 6 °C. (The detailed cTC_t evaluation based on bias and Pearson's correlation can be found in Fig. S2 as well as Tables S1 and S3.) The cTC_t presented a mean bias of 2.1 °C in the tropical Andes and a mean bias of 0.2 °C in the Dry Andes and Wet Andes in comparison with in situ measurements.

3.2 Glaciological changes across the Andes during the period 2000–2019

The 36 % of the total glacierized surface area across the Andes (11° N to 55° S) is simulated to obtain the annual glacier area and glacier volume, as well as the monthly glacier runoff (glacier melting and rainfall on glaciers).

Considering mean values for the periods 2000–2009 and 2010–2019, the glacier volume and surface area in the Andean catchments show a decrease of −8.3 % (−59.1 km³) and −2.2 % (−245 km²), respectively. This corresponds to a mean annual mass balance difference between the two periods of −0.5 ± 0.3 $\mathrm{m}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$ (Fig. 2d). A decrease in glacier volume and surface is seen in 93 % of the catchments (n = 724), whereas 7 % of the catchments (n = 65) present an increase in glacier volume and surface. The loss in glacier volume (Fig. 2b) is largest (−47.8 km³, −9 %) in the Wet Andes, followed by the tropical Andes (−5.9 km³, −7 %) and Dry Andes (−5.4 km³, −6 %). Similarly, a larger decrease in the glacier surface area (Fig. 2c) is observed in the Wet Andes (−144.4 km², −2 %), followed by the tropical Andes (−55.5 km², −4 %) and lastly the Dry Andes (−45.2 km², −3 %). As expected, the correlation between both glacier change variables is consistent at the zone scale, showing a positive correlation between the changes in area and volume (r = 0.9).

When estimating the mass balance, it is interesting to check the calibrated melt factors (M_f) of the temperature-index model in order to evaluate its possible regionalization, i.e., to evaluate the spatial coherence (see Fig. 3 and Table 1). The mean calibrated melt factor values decrease from the tropical Andes toward the Wet Andes (tropical Andes: 0.5 ± 0.3 $\mathrm{mm}{\mathrm{h}}^{-\mathrm{1}}\mathrm{°}{\mathrm{C}}^{-\mathrm{1}}$; Dry Andes: 0.6 ± 0.2 $\mathrm{mm}{\mathrm{h}}^{-\mathrm{1}}\mathrm{°}{\mathrm{C}}^{-\mathrm{1}}$; Wet Andes: 0.2 ± 0.1 $\mathrm{mm}{\mathrm{h}}^{-\mathrm{1}}\mathrm{°}{\mathrm{C}}^{-\mathrm{1}}$). The lowest mean temperatures estimated in the Dry Andes imply higher factor values to reach the calibrated mass loss in the few months in which the temperatures exceed 0 °C. The opposite can be observed in the Wet Andes, where low factor values are associated with a greater number of months with temperatures exceeding 0 °C. We obtain very similar values in contiguous zones, with the lowest values found in the Wet Andes (monthly mean below 179 mm °C⁻¹), followed by the tropical Andes (monthly mean below 434 mm °C⁻¹) and the Dry Andes (monthly mean below 479 mm °C⁻¹). The largest melt factor values are found in the Dry Andes where Dry Andes 2 (monthly mean = 479 mm °C⁻¹) presents the lowest mean temperatures across the Andes (−4.8 °C between 2000 and 2019). The lowest melting factor values are calibrated in the Wet Andes where Wet Andes 1 (monthly mean = 103 mm °C⁻¹) shows high mean temperatures (1.8 °C between 2000 and 2019). Despite this, a lower correlation between the melt factors and mean temperature for the 2000–2019 period is estimated (r = −0.5; p = 0.08). Conversely, the correlation between the melt factors and mean precipitation for the 2000–2019 period is high (r = −0.8; p = 0.002).

Figure 3Statistics for the calibrated melt factors per glacier at the glaciological zone scale across the Andes. Shown are the mean (×), median (circles), mode (squares), and 25th as well as 75th percentile (lines) values for 13 179 glaciers.

Download

To test our results we evaluated the simulated mass balance evaluation for the 15 glaciers that can be found in Figs. S3 and S4, as well as Tables S4 and S5. The in situ data show a mean negative mass balance (−832 ± 795 $\mathrm{mm}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$) between 2000 and 2019 greater than our mean simulated mass balance (−647 ± 713 $\mathrm{mm}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$) in the same glaciers. The evaluation results give a mean Pearson correlation of 0.67 (except for the Agua Negra, Ortigas 1, Guanaco, and Amarillo glaciers, for which it shows either no correlation or a negative correlation) with an underestimation of the mean simulated mass balance of 185 $\mathrm{mm}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$ (bias); 40 % of the glaciers present a correlation equal to or greater than 0.7. In terms of the best results by glaciological region, in the tropical Andes, the Conejeras glacier has a high correlation (r = 0.9) and bias (1104 $\mathrm{mm}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$), whereas in the Dry Andes, the Piloto Este, Paula, Paloma Este, and Del Rincón glaciers display a high correlation (r ≥ 0.8) and a mean bias of 351 $\mathrm{mm}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$. In the Wet Andes, the Mocho–Choshuenco and Martial Este glaciers show a moderate correlation (r = 0.5) and a lower overestimation of the simulated mass balance (−118 $\mathrm{mm}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$). Model limitations are observed on the Zongo glacier (r = 0.3 and bias = −224 $\mathrm{mm}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$) in the tropical Andes. In Dry Andes 1, no correlation is observed in the three monitored glaciers (Guanaco, Amarillo, and Ortigas 1), mainly because sublimation is very high on these glaciers, reaching 81 % of the annual ablation (MacDonell et al., 2013). On the other hand, sublimation is lower southward in Dry Andes 2 with 7 % of the annual ablation (Ayala et al., 2017). For the tropical zone, sublimation is close to 13 % in the outer tropics (Sicart et al., 2005) and 5 % in the inner tropics (Favier et al., 2004). However, sublimation is implicitly included in the model through the calibrated melt factor values, which are derived from measured mass balance data by Hugonnet et al. (2021). As a result, our estimates of snow/ice melt in the Dry Andes 1 zone tend to be overestimated.

The details of the glacier changes in the 786 Andean catchments, which are larger in the Wet Andes followed by the tropical Andes and then the Dry Andes, are available in the Supplement.

Figure 4Recent glacier runoff components across the Andes. The total glacier melt and rainfall on glaciers represent the mean differences between the periods 2010–2019 and 2000–2009 per catchment (n = 786). Panel (a) shows the distribution of the glaciological zones (11° N to 55° S), followed by (b) glacier melt and (c) rainfall on glaciers at the catchment scale. The (d) total annual glacier melt is presented in each glaciological zone. “G. melt” and “Rainfall” refer to changes in (b) glacier melt and (c) rainfall on glaciers, respectively. “G. melt” on the y axis in (d) refers to cumulative annual glacier melt by glaciological zone.

3.3 Changes in glacier runoff across the Andes during the period 2000–2019

Due to glacier changes across the Andes, high glacier runoff variations are observed from glacier melt and rainfall on glaciers (Fig. 4). The mean annual glacier melt in all catchments for the period 2000–2019 was 696 m³ s⁻¹. At the regional scale, the Wet Andes shows the largest mean annual glacier melt in the Andes (583.5 m³ s⁻¹), followed by the Dry Andes (59.9 m³ s⁻¹) and then the tropical Andes (52.7 m³ s⁻¹). However, if we look at the mean annual glacier melt changes between the periods 2000–2009 and 2010–2019, we see an increase of 12 % (86.5 m³ s⁻¹) across the Andes, where 84 % (n = 661) of catchments show an increase and 12 % (n = 95) of them present a decrease. As Table S6 shows, an increase in glacier melt is observed in catchments with a higher glacier elevation, larger glacier size, lower mean temperature, and higher mean precipitation compared with catchments that show either a decrease in glacier melt or no changes at all. These latter catchments also show the largest decrease in precipitation (−10 % to −14 %).

The mean annual glacier melt changes show the largest percentage increase in the tropical Andes (40 %, 21 m³ s⁻¹), followed by the Dry Andes (36 %, 21.7 m³ s⁻¹) and the Wet Andes (8 %, 4.8 m³ s⁻¹). In addition, significant differences are observed for the different zones: for instance, the inner tropics in the tropical Andes presents the largest increase (73 % with only 4.1 m³ s⁻¹), followed by Dry Andes 1 (62 % with only 1.8 m³ s⁻¹) in the Dry Andes. In the Wet Andes, the larger percentage of increase in the mean annual glacier melt changes is observed in Wet Andes 5 (14 % with 4.1 m³ s⁻¹), showing a lower percentage in comparison with the inner tropics and Dry Andes 1 zones; however, its absolute increase in glacier melt is equal to or greater than 4.1 m³ s⁻¹. These results per glaciological zone are summarized in Table 2. Related to the previously described glacier changes (see Sect. 3.2) between the periods 2000–2009 and 2010–2019, at the glaciological zone scale, we logically find a high negative correlation between the glacier melt and glacier volume changes in the tropical Andes and Dry Andes (r = −0.9) as well as the Wet Andes (r = −1).

In addition, the mean annual rainfall on glaciers across the Andes is 387 m³ s⁻¹ for the period 2000–2019. The Wet Andes has the largest amount of annual rainfall (372.7 m³ s⁻¹), followed by the tropical Andes (10.5 m³ s⁻¹) and Dry Andes (4.2 m³ s⁻¹) with the lowest contribution of rainfall.

In terms of the variations in the mean annual rainfall on glaciers between the periods 2000–2009 and 2010–2019, we observe a reduction of −2 % (−7.6 m³ s⁻¹) across the Andes, showing a reduction in 41 % of the catchments (n = 322), whereas the largest proportion of the catchments (51 %; n = 403) show an increase. Table S6 shows that the catchments with the larger increase in rainfall on glaciers are concentrated in the same latitude range as the catchments with an increase in glacier melt. These catchments have similar glacier elevations and glacier sizes. The catchments that do not show variations in rainfall on glaciers are concentrated in the Dry Andes region, where the rainfall contributes less to the glacier runoff volume.

Figure 5Recent annual specific mass balance, surface area, and volume variations in the La Paz, Maipo, and Baker catchments from 2000 to 2019. Panels (a)–(c) show the mass balance for each simulated glacier (blue line), as well as the weighted mean mass balance per catchment (red line). The mean geodetic mass balance and its error for the period 2000–2019 are also presented (green bar). (d–f) presents the total glacierized area per catchment (blue line). The total area from RGI v6.0 and the simulated area percentage are also presented. Panels (g)–(i) exhibit the total volume per catchment (green line). The surface area and volume have both been normalized to make it easier to compare the evolution between the catchments.

Download

At the glaciological region scale, the mean annual rainfall on glaciers decreases in the Wet Andes (−3 %, 10.1 m³ s⁻¹) but increases in the tropical Andes (23 %, 2.4 m³ s⁻¹) and Dry Andes (3 %, 0.1 m³ s⁻¹). In addition, large differences are observed in the glaciological zones (Table 2): e.g., Dry Andes 1 in the Dry Andes has the largest percentage increase (106 % with only 0.2 m³ s⁻¹), followed by the inner tropics (74 % with only 0.4 m³ s⁻¹) in the tropical Andes. In the Wet Andes, the larger increase (by percent) in the mean annual rain on the glaciers is observed in Wet Andes 5 (6.6 % with 2.1 m³ s⁻¹). Other zones, such as Wet Andes 2 and Wet Andes 6, show large absolute reductions (−11.7 m³ s⁻¹ and −4.5 m³ s⁻¹, respectively).

The changes in glacier melt and rainfall on glaciers observed in the tropical Andes, Dry Andes, and Wet Andes are summarized in Table 2 and are available, in the Supplement, for the 786 Andean catchments.

3.4 Hydro-glaciological behavior at the catchment scale during the period 2000–2019

In this section, we focus on three Andean catchments: La Paz (16° S, tropical Andes), Maipo (33° S, Dry Andes), and Baker (47° S, Wet Andes) (see locations in Figs. 2 and 4), where previous glaciological observations and simulations of glacier evolution and water production have been carried out, and in situ records are also available. Detailed results for each of the 786 catchments and glaciers included are available in the dataset provided in the Supplement.

Table 3Hydro-glaciological changes and variations in climate between the periods 2000–2009 and 2010–2019 for the three selected catchments.

Download Print Version | Download XLSX

3.4.1 Glaciological variations in the selected catchments: La Paz (16° S), Maipo (33° S), and Baker (47° S)

Figure 5 shows the annual specific mass balance in the three catchments (2000–2019). The mean over the study period is negative, and there is a negative trend for the annual values toward 2019. For instance, for the Maipo catchment, we estimate a mean annual mass balance of −0.29 ± 0.14 $\mathrm{m}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$, a slightly more negative balance in the Baker catchment (−0.47 ± 0.19 $\mathrm{m}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$), and in the glaciers in the La Paz catchment, there is a greater loss of −0.56 ± 0.19 $\mathrm{m}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$. In addition, when considering the annual mass balance values, it is possible to note differences between the catchments. The La Paz catchment shows mostly negative annual mass balance values over the whole period, while in the Baker and Maipo catchments the mass balances are predominantly negative after 2004 and 2009, respectively. Considering the total area and volume changes per catchment in the periods 2000–2009 and 2010–2019, an overall reduction is observed in each of the three catchments. For the La Paz catchment, considering 86 % (14 km²) of glacierized area in 2000 (mean glacierized elevation of 5019 $\mathrm{m}\mathrm{a}.\mathrm{s}.\mathrm{l}.$) and 20 glaciers, the glacierized surface area and volume decrease by −7 % (−1 km²) and −11 % (−0.1 km³), respectively. For the Maipo catchment, with a larger percentage of simulated glacierized surface area in 2000 (99 %, with mean elevation of 4259 $\mathrm{m}\mathrm{a}.\mathrm{s}.\mathrm{l}.$) and a greater number of glaciers (n = 225), the area and volume decrease by −1 % (−4.2 km²) and −5 % (−1 km³), respectively. For the Baker catchment, which contains the largest glacierized surface area of the three catchments in 2000, we simulated 66 % of this glacierized area (1514 km², with mean elevation of 1612 $\mathrm{m}\mathrm{a}.\mathrm{s}.\mathrm{l}.$) and 1805 glaciers: this area shrank by approximately −2 % (−36.7 km²), losing close to −11 % (−9.3 km³) of its volume. These results are summarized in Table 3.

Figure 6Hydro-glaciological responses and climate variations in the La Paz, Maipo, and Baker catchments from 2000 to 2019. The first row (a–c) presents the mean annual glacier runoff (purple line = ice melt + snowmelt + rainfall on glaciers), the mean annual glacier melt (blue line = ice melt + snowmelt), and the annual specific mass balance (red line). The other rows (d–i) show the mean total annual precipitation and mean annual temperature with the mean annual amount for the periods 2000–2019 (black line), 2000–2009 (blue line), and 2010–2019 (red line).

Download

3.4.2 Hydrological contribution of glaciers in the selected catchments: La Paz (16° S), Maipo (33° S), and Baker (47° S)

The La Paz, Maipo, and Baker catchments display large climatic and glaciological differences over the period 2000–2019. For instance, contrasting cumulative precipitation amounts can be found between the Baker and La Paz catchments (2224 ± 443 and 791 ± 100 mm yr⁻¹, respectively), while the La Paz and Maipo catchments present the maximum difference in mean annual temperature (1.4 ± 0.5 °C and −4.1 ± 0.5 °C, respectively) (Fig. 6). At a seasonal scale, precipitation in the Maipo and Baker catchments is concentrated in fall and winter (April to September), even if the latter catchment also has a significant amount of precipitation in summer. Conversely, precipitation in the La Paz catchment mainly occurs in spring and summer (October to March). In addition, the La Paz and Baker catchments are characterized by the warmest temperatures (> 0 °C) in spring and summer; the warmest temperatures for the Maipo catchment occur in summer. Variations in the climatic conditions are observed between 2000–2009 and 2010–2019. For instance, a decrease in cumulative precipitation is observed in the Maipo (−30 %; −454 mm yr⁻¹) and Baker catchments (−2 %; −52 mm yr⁻¹), but an increase can be seen in the La Paz catchment (4 %, 30 mm yr⁻¹). The mean annual temperature increases in the three catchments (+0.5 °C in La Paz and Baker, +0.4 °C in Maipo).

The glacier runoff simulation, which considers the glacier melt (ice melt and snowmelt) and rainfall on glaciers (liquid precipitation), shows strong differences between the catchments (Fig. 6). Over the period 2000–2019, the glaciers in the Baker catchment have the highest mean annual glacier melt (94 ± 19.6 m³ s⁻¹), followed by those in the Maipo (15.1 ± 4.2 m³ s⁻¹) and La Paz catchments (0.5 ± 0.2 m³ s⁻¹). The rainfall on glaciers contributes 30 % to glacier runoff in the Baker catchment (41 ± 10.1 m³ s⁻¹); a lower value is found in the La Paz catchment with 17 % (0.1 m³ s⁻¹) followed by the Maipo catchment with 5 % (0.8 ± 0.3 m³ s⁻¹), which is the lowest contribution of rainfall on glaciers in these catchments. The simulations of glacier runoff changes between the periods 2000–2009 and 2010–2019 for the three catchments show an increase in glacier melt and rainfall on glaciers. The largest relative increase in mean annual glacier melt is observed in the Maipo catchment with 37 % (4.7 m³ s⁻¹), followed by the La Paz catchment with 21 % (0.09 m³ s⁻¹) and the Baker catchment with 10 % (9 m³ s⁻¹). Meanwhile, the largest relative increase in the mean annual rainfall on glaciers is observed in the La Paz catchment (15 %; 0.01 m³ s⁻¹), followed by the Baker catchment (11 %; 4.3 m³ s⁻¹) and lastly the Maipo catchment (2 %; 0.02 m³ s⁻¹). The results for the glacier melt and rainfall on glaciers are summarized in Table 3.

Figure 7Monthly hydro-glaciological responses and climate variations in the La Paz, Maipo, and Baker catchments from 2000 to 2019. The first (a–c) and second (d–f) rows present the mean monthly glacier melt and rainfall on glaciers (black line) and the mean amounts per year during the periods 2000–2009 (red lines) and 2010–2019 (blue lines). In the last row (g–i), the climographs show the mean monthly precipitation (blue bars) and temperature (red line) for the period 2000–2019.

Download

In Fig. 7, at a mean monthly temporal scale for the period 2000–2019, the glacier melt simulation presents a short maximum during summer (January to February) in the Maipo and Baker catchments. In contrast, peaks in the La Paz catchment are extended during spring and summer (November to March) highlighting the so-called transition season (between September and November) where there is a low amount of rainfall on glaciers and glacier melt progressively increases. In the Baker catchment, melting begins earlier in September, while in Maipo it begins later (November). The interannual variability in glacier melt over the periods 2000–2009 and 2010–2019 shows a larger contribution from the glacier in the period 2010–2019 for the Maipo catchment. Furthermore, the simulated rainfall on glaciers is larger mainly during the summer season in all catchments, with more rainfall in the La Paz catchment (December to February) after the transition season.

Figure 8Monthly simulated glacier runoff (glacier melt + rainfall on glaciers) and discharge measurements in the La Paz, Maipo, and Baker catchments from 2000 to 2019. The results for the glacier melt (blue lines) and rainfall on glacier calculations (gray lines) are presented, as well as the discharge measurement (black line) and its standard deviation (gray area). The mean annual glacier runoff contribution (as a percentage) and the mean glacier runoff contribution (as a percentage) from January to March are shown in parentheses. The values are normalized by the mean river discharge.

Download

For the mean annual discharge measurements in each catchment and the mean annual simulated glacier runoff (glacier melt and rainfall on glaciers) between 2000 and 2019 (Fig. 8), we estimate that the largest glacier runoff contribution is in the Baker catchment (24 %), followed by the La Paz (22 %) and Maipo catchments (14 %), where all catchments present a similar proportion of glacierized surface area (7.5 %–8.2 %). If we consider the summer season only (January to March), the glacier runoff contribution is highest in the Baker catchment (43 %), followed by the Maipo (36 %) and La Paz catchments (18 %), where the larger percentage of glacier melt is found in the Maipo catchment (34 %) and the larger percentage of rainfall on glaciers is displayed in the Baker catchment (12 %). Unlike the Maipo and Baker catchments, which present a maximum glacier runoff contribution in the summer season, the La Paz catchment shows the largest glacier runoff contribution (45 %) in the transition season (September to November).

4.1 Comparison with previous studies across the Andes

Huss and Hock (2018) studied 12 Andean catchments across the Andes (1980–2000 and 2010–2030) and estimated an increase in glacier runoff in the tropical Andes (Santa and Titicaca catchments) and the Dry Andes (Rapel and Colorado catchments). Our results are consistent with these estimates. We show an increase in glacier melt of 40 % and 36 % in both regions, respectively, between the periods 2000–2009 and 2010–2019. However, in the Wet Andes, Huss and Hock (2018) did not estimate any changes in glacier runoff on the western side of the Andes (Biobio catchment), and instead found a decrease (Río Negro catchment) and an increase (Río Santa Cruz catchment) in glacier runoff on the eastern side of the Andes. Our results for this region show an increase in glacier melt of 8 % and a decrease in rainfall on glaciers of −3 %.

Based on local reports in the tropical Andes, the catchment associated with the Los Crespos glacier (catchment id: 6090223080) on the Antisana volcano shows a small decrease in the glacier area of −1 % between the periods 2000–2009 and 2010–2019, which is in agreement with Basantes-Serrano et al. (2022). Their study estimated that almost half of the glacier area (G1b, G2-3, G8, G9, and G17) had a positive mass balance during the period 1998–2009 with the largest glacier presenting a mass balance of 0.36 ± 0.57 $\mathrm{m}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$, in agreement with our mass balance estimation at the catchment scale of 0.2 ± 0.5 $\mathrm{m}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$ (2000–2009). However, in this region, the corrected TerraClimate temperature cannot reproduce the magnitude of the monthly temperature variation (see Fig. S2). This limits the effectiveness of the parameter values used in the model to accurately simulate the melting onset and the amount of solid and liquid precipitation. Furthermore, the mass balance simulation is performed through the temperature-index model which does not take the sublimation process into account; and in addition, it runs at a monthly time step thereby limiting the relevant processes that occur hourly. On the other hand, the catchments that contain the Zongo glacier (catchment id: 6090629570) and the Charquini glacier (catchment id: 6090641570) display results that are consistent with the observations of others (Rabatel et al., 2012; Seehaus et al., 2020; Autin et al. 2022). In addition, our simulated mass balance evaluation on the Zongo glacier shows a low bias (−0.2 $\mathrm{m}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$) with regard to the observations. In the Dry Andes, the catchments associated with the Pascua Lama area (catchment id: 6090836550 and catchment id: 6090840860), the Tapado glacier (catchment id: 6090853340), and the glaciers of the Olivares catchment (catchment id: 6090889690) show consistent results in terms glaciological variations in comparison with the observations (Rabatel et al., 2011; Malmros et al., 2016; Farías-Barahona et at., 2020; Robson et al., 2022). In the Wet Andes, the catchments associated with the Chilean side of the Monte Tronador (catchment id: 6090945100) and the Martial Este and Alvear glaciers in Tierra del Fuego (catchment id: 6090037770) show results that are consistent with previous reports (Rabassa 2010; Ruiz et al., 2017). Despite this, it is possible that our methodology could overestimate precipitation in some catchments; for example, the cumulative precipitation associated with the Nevados de Chillán catchment (catchment id: 6090916140) was estimated at 4023 mm yr⁻¹.

At the glaciological region scale, previous studies have reported a large decrease in the percentage of glacier area in the tropical Andes of −29 % (2000–2016) (Seehaus et al., 2019, 2020), followed by the Dry Andes between −29 % and −30 % (Rabatel et al., 2011; Malmros et al., 2016) although for a longer time period. In the Wet Andes, Meier et al. (2018) reported a −9 % decrease in the glacier area (1986–2016). Our simulations are consistent with these observed glacier area reductions. In addition, Caro et al. (2021) estimated a similar trend across the Andes between 1980 and 2019 (tropical Andes: −41 %; Dry Andes: −39 %; Wet Andes: −24 %). On the other hand, we found high correlations between the mean annual climatic variables and annual mass balance. In the Dry Andes, this correlation was high with precipitation (r = 0.8 ± 0.1; p < 0.05), and in the Wet Andes, temperature was correlated with mass balance (r = −0.7 ± 0.1; p < 0.05) as previously observed by Caro et al. (2021). These correlations between precipitation or temperature with the annual mass balances for each catchment across the Andes can be reviewed in Fig. S5 and Table S7.

4.2 Comparison of our results with previous studies in the three selected catchments

In the La Paz catchment, Soruco et al. (2015) evaluated the mass balance of 70 glaciers (1997–2006) and their contribution to the hydrological regime. In the present study, we simulated a less negative mass balance (−0.56 ± 0.19 $\mathrm{m}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$ vs. −1 $\mathrm{m}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$) considering a larger glacierized area due to the use of RGI v6.0 (with 14.1 km² in comparison with 8.3 km²). Our estimation of the mean annual glacier runoff (22 %) is larger than the previous estimation close to 15 % (Soruco et al., 2015). This may be due to the fact that we have considered a warmer 2010–2019 period than the one observed in Soruco et al. (2015). Unlike the previous report, we estimated a larger glacier runoff contribution during the wet season (26 % from October to March) and increasing in the transition season (45 % from September to November). This increase in glacier runoff contribution given by the model agrees with the larger glacier mass loss observed by Sicart et al. (2007) and Autin et al. (2022) during this season. In the Maipo catchment, we identified a slightly smaller glacierized area (325 km² for the year 2000; −14 %) compared with Ayala et al. (2020) because they considered rock glaciers from the Chilean glacier inventory. In addition, we observed a more negative mass balance after 2008, coinciding with the mega-drought period characterized by a decrease in precipitation and an increase in temperature (Garreaud et al., 2017). The hydrological response to this negative mass balance trend is an increase in glacier runoff since 2000 that is concentrated between December and March. The modeled mean annual glacier runoff contribution estimation is close to 15 %, reaching 36 % in summer (January to March), and is close to Ayala et al.'s (2020) estimation (16 % at the annual scale for the period 1955–2016). However, this comparison between our results and previous studies in the Maipo and La Paz catchments is limited due to the utilization of different inputs, spatial resolutions, time steps, and workflows in the simulated processes where some processes, such as mass balance of all glaciers, were not done. Lastly, in the Baker catchment, Dussaillant et al. (2012) stated that catchments associated with the Northern Patagonian Icefield (NPI) are strongly conditioned by glacier melting. In this respect, Huss and Hock (2018) did not identify glacier runoff changes between the periods 1980–2000 and 2010–2030 – they only considered 183 km² of the glacierized area (−12 % until 2020) – whereas we estimated a 10 % and 11 % increase in glacier melt and rainfall on glaciers, respectively, taking a larger glacierized area (1514 km²; −2 % until 2020) into account. The relevance of the rainfall on glaciers with regard to the glacier runoff estimated here is close to 30 % (including glaciers east of the NPI) which is confirmed by Krogh et al. (2014), who estimated that over 68 % of the total precipitation at the catchment scale in the east NPI (León and Delta catchments) corresponds to rainfall.

4.3 Simulation limitations

Limitations in the simulations result from different sources: (1) the quality/accuracy of the input data; (2) the calibration of the precipitation and melt factors; and (3) the model itself, including its structure and the processes that are not represented. Regarding the evaluation of the corrected TerraClimate temperature using meteorological observations in the tropical Andes, the corrected TerraClimate data do not reproduce either the low monthly temperature or the higher temperature in specific months which have a mean bias of 2.1 °C (e.g., Llan_Up-2 9° S, Zongo at glacier station 16° S). These differences found in the corrected TerraClimate data limit the capacity of the ice and snow melting module to accurately simulate the months in which melting can occur. On the other hand, in the Dry Andes and Wet Andes, the corrected TerraClimate temperatures are closer to the in situ observations (mean bias = 0.2°C) and present a reliable monthly distribution. This results in model parameter values that are in better agreement with the values used in other studies. Other limitations come from RGI v6.0 because some glaciers are considered as only one larger glacier. For example, in the Dry Andes (catchment id: 6090889690) two large glaciers, the Olivares Gamma and the Juncal Sur, form one (even larger) glacier. These glaciers could underestimate the simulated change in glacier area, limiting the performance of the volume module which depends on the glacier geometry and bedrock shape.

Furthermore, we applied different precipitation factor values in the tropical Andes (1), Dry Andes (1.9–4), and Wet Andes (2.3–4) in order to increase the annual mass balance. These values are in agreement with former studies, for example, similar values were used in the Dry Andes (Masiokas et al., 2016; Burger et al., 2019; Farías-Barahona et al., 2020). Values that are too high could lead to an overestimation of precipitation on some glaciers. However, to confirm that the precipitation factor produces realistic precipitation values, we adjusted the standard deviation of the simulated mass balance to the observed mass balance, a method similar to that proposed in Marzeion et al. (2012) and Maussion et al. (2019). Additional uncertainties in the calibrated melt factors come from the climate and geodetic mass balance datasets used to run and calibrate the model. Indeed, the melting temperature threshold establishes the onset of melting and influences the number of months in which it occurs. On the other hand, the geodetic mass balance defines the accumulated gain or loss per glacier over the calibration period, which in this case spans 20 years. Based on our evaluation of the corrected TerraClimate temperature and simulated mass balance, we correctly reproduce the seasonal melt distribution, associated with a mean underestimated overall annual mass balance of 185 $\mathrm{mm}\mathrm{w}.\mathrm{e}.{\mathrm{yr}}^{-\mathrm{1}}$ which, however, is correlated with the in situ data (r = 0.7). According to Rounce et al. (2020), similar results of glacier surface mass balance could be due to different combinations of model parameters. For instance, a wetter (or dryer) and warmer (or colder) parameter set – where high (or low) precipitation factors are compensated by high (or low) temperature biases – can lead to similar recent glacier mass changes and projections. Conversely, the implications for glacier runoff are likely to be significant for both recent and future simulations. In a wetter (or dryer) and warmer (or colder) scenario, there would be increased (or decreased) precipitation and melt, resulting in larger (or smaller) glacier runoff contribution. To address this, we obtained realistic values for precipitation and temperature based on in situ spatially distributed measurements and on our field experiences on monitored Andean glaciers. Furthermore, our evaluation of simulations in the three selected catchments enabled us to estimate glacier runoff amounts in the same order of magnitude as that of previous reports. However, caution must be exercised when using the calibrated melt factors estimated in the tropical Andes. This is because the temperature in this region was overestimated by an average of 2.4 °C, impacting the calibration of the melt factor values. These values should be lower than those estimated here (see Fig. 3).

With regard to the structural limitations of the model, it would be relevant to distinguish between ice and snowmelt when simulating the glacier melt with two melt factors. In addition, the sublimation on the glacier surface is very relevant in some glaciers located in the tropical Andes and Dry Andes 1 (Rabatel et al., 2011; MacDonell et al., 2013). However, the OGGM model does not incorporate these processes in glacier runoff and mass balance simulations.