Temporal glacier area changes correlated with the El Niño/La Niña Southern Oscillation using satellite imagery

We used Landsat satellite images to determine the areal change between 1988 and 2010 of the Condoriri glacier in Bolivia and found that the area decreased by 41% during that period. Moreover, the interannual pattern of recession and expansion of the glacier coincided with warm and cold phases of El Niño/La Niña-Southern Oscillation (ENSO), respectively. Because the glacier recedes more during El Niño events than it expands during La Niña events, the net result is a retreating trend, which, if it continues, means that the glacier will disappear completely by 2035. ENSO frequency increased during the latter part of the 20th century, and ENSO events may become more frequent with continued climate change. Therefore, it is urgent to take measures to adapt Bolivian water management to the loss of glacier meltwater.


INTRODUCTION
Mountain glaciers are a key element of the water cycle in Andean countries such as Peru and Bolivia.In the semiarid Altiplano region where the annual rainfall is approximately 500 mm, meltwater from mountain glaciers is the main water resource for densely populated cities of Bolivia such as La Paz, its capital, and El Alto.During the dry season (May-October), when the monthly rainfall is less than 10 mm, the rivers are largely fed by meltwater, which is almost entirely from glaciers.The loss of glaciers, therefore, would threaten both the health of the region's ecosystems and water resources in the lowlands.
Mountain glaciers around the world have been retreating since the early 20th century with few exceptions (e.g., Paul et al., 2004;Bolch et al., 2012).Low-latitude tropical Andean glaciers at high elevations are particularly sensitive and respond very rapidly to climate variations.For example, the Chacaltaya glacier (16°21'S, 68°07'W) in Bolivia lost 93% of its ice volume between 1940 and 1998 (Francou et al., 2003).Although the cause of such tremendous changes are not yet well understood, recent research suggests that they are linked to both long-term climate change (Vincent et al., 2005) and sub-decadal climate variability known as El Niño/La Niña-Southern Oscillation (ENSO) (Arnaud et al., 2001).
Climate influences the rate of glacial melting in the tropics in complex ways, through a combination of factors such as precipitation, temperature, humidity, and sunshine duration.Meteorological observations suggest that among these factors, changes in precipitation and cloudiness in the latter part of the 20th century were less important than warming temperatures (Coudrain et al., 2005).The temperature in the tropical Andes increased by 0.10-0.11°Cdecade −1 between 1939 and 1998, and by 0.32-0.34°Cdecade −1 between 1974 and 1998 (Vuille and Bradley, 2000).Thus, the rate of warming in the last quarter of the 20th century was approximately triple the rate earlier in the century.Simulations by the HadAM3 regional climate model with IPCC-SRES scenarios A2 and B2 have projected significant warming in the 21st century (2-7°C warmer during 2071-2100 compared with observations during , and this warming is likely to accelerate the rate of glacial retreat in the tropical Andes (Urrutia and Vuille, 2009).In addition, the frequency of ENSO events has increased since the late 1970s (Lee and McPhaden, 2010), and, furthermore, they are likely to become even more frequent with continued global warming (Yeh et al., 2009).More frequent ENSO events would also accelerate the rate of glacier retreat.
Understanding how glaciers are likely to respond to longterm climate change as well as to sub-decadal ENSO variability is of broad scientific and socioeconomic interest.For example, the Chacaltaya glacier used to host the highest ski resort in the world, thus its near-disappearance affected not only ecosystems associated with the glacier but also the livelihoods of many people and the tourism industry in Bolivia.In this study, our aim was to evaluate long-term and interannual melting trends of the tropical Andean Condoriri glacier by using satellite observations of the glacier's area obtained between 1988 and 2010 and, further, to examine the relationship between the interannual variability of the glacial area and ENSO.

STUDY AREA
Bolivia is a landlocked country in the center of South America.Because one-third of its land area is above 3500 m, its climate is influenced not only by latitude but also by altitude.Meltwater from the Condoriri glacier, which is approximately 40 km northwest of La Paz and extends in elevation from 4800 to 5600 m a.s.l.(Figure 1), is one of the main water sources for the cities of El Alto and La Paz (Quiroga et al., 2013).The glacier is shaped like a condor with open wings, and, according to 2010 estimates, it spreads over an area of approximately 5 km 2 .This glacier is classified as a summer-accumulation-type glacier, and accumulation and ablation occur simultaneously in summer.Between 2001 and 2003, a glacial lake formed at the glacier front (16°10'52''S, 68°15'43''W) (Figure 2).More than 1000 glacial meltwater lakes have formed in Bolivia in recent years as a result of the dramatic retreat of the Condoriri glacier and other mountain glaciers (Coudrain et al., 2005).
During the 20th century, Andean climate variations were not uniform; instead, on a decadal time scale they were apparently strongly governed by sea surface temperature (SST) variations in the equatorial Pacific driven by ENSO (Vuille, 1999).During the warm phase of ENSO (El Niño), a precipitation deficit is observed in the Andes, and during the cooler and more humid phase (La Niña), precipitation increases.

Data
We used 47 satellite images (Path/Row: 001/071) with a 30-m spatial resolution from Landsat-5 TM and Landsat-7 ETM+ sensors acquired from 1988 to 2010.We used radiometrically and geometrically corrected Landsat images from Instituto Nacional de Pesquisas Espaciais of Brazil (http://www.dgi.inpe.br/CDSR/index.php).Monthly precipitation data averaged over 2000-2009 indicate that in Bolivia, 90% of the annual precipitation occurs from October to March.Therefore, we selected images from the dry season (May-October) for analysis, to avoid having to extract pixels showing snow on the ground surface and so there would not be snow on the glacier.In addition, our selected images had cloud cover no greater than 15%.
For comparisons with glacial area changes and precipitation, we used El Niño and La Niña events identified from SST anomalies based on the running 30-year mean in the Niño 3 region (4°N-4°S, 90°-150°W).The Japan Meteorological Agency (JMA) defines an El Niño event as when the 5-month running mean SST anomaly equals or exceeds +0.5°C for more than six months.Correspondingly, a La Niña event is defined by this value being less than or equal to −0.5°C for more than six months.We used data on SST anomalies in the Niño 3 region from the JMA website (http://www.data.jma.go.jp/gmd/cpd/data/elnino/ index/nino3irm.html),and air temperature and precipitation data obtained at La Paz by the Servicio Nacional de Meteorología e Hidrología -Bolivia.

Methods
Snow and ice have high spectral reflectance values in the visible band and low values in the mid-infrared band, and their spectral characteristics are widely used for glacier classification.In this study, we used the Normalized Difference Snow Index (NDSI), calculated from the midinfrared (1.55-1.75μm) and green (0.52-0.60 μm) bands.This index is a simple but accurate way to distinguish glacial areas (e.g., Gunawardhana and Kazama, 2012;Paul et al., 2004).
Many previous studies have used an NDSI threshold of 0.40 for mapping snow cover and glacier area (e.g., Dozier, 1989).However, in our study area, if the 0.40 NDSI threshold is adopted, part of the glacial lake is incorrectly identified as glacier.In their study in the Qilian mountain range of China, Xiao et al. (2001) adopted an NDSI threshold of 0.40, but most very bright white areas likely to be snow and ice in their study area had NDSI values of 0.70 or more.The spectral reflectance values of glaciers and snow vary according to season and region.Therefore, we used a NDSI threshold of 0.70 to identify the Condoriri glacier.Moreover, by visually interpreting true color images and comparing them with NDSI images on a GIS display, we found that estimates of snow and glacier area based on true-color images showed good agreement with NDSI estimates when we used an NDSI threshold of 0.70.

Long-term change
Under the present climate, the Condoriri glacier is shrinking rapidly (Figure 3).Our results showed that the area of the glacier was 9.4 km 2 in September 1988 and 5.5 km 2 in September 2010.Thus, the glacier lost 41% of its area from 1988 to 2010.Based on the linear trend fitted to the data (red line in Figure 4a), the loss occurred at an estimated average rate of 0.23 km 2 year −1 , which is significant at the 95% confidence level (Figure 4a).If the linear trend of the glacial retreat is extrapolated into the future, the glacier will virtually disappear by 2035.However, large glaciers that through melting have become several smaller, discontinuous glaciers may begin to melt even faster and lose area at a greater rate (Paul et al., 2004).Therefore, the Condoriri glacier may disappear earlier than is suggested by this projection.frequent in the equatorial Pacific, but since 1976, the frequency of warm, El Niño events has increased, resulting in abnormally warm conditions.

Interannual changes
In addition to the consistent long-term retreating trend of the Condoriri glacier, we found clear interannual differences in glacial area changes (Figure 4).These glacier area changes apparently correspond to changes in the Southern Oscillation regime.This phenomenon relates to the intensities of El Niño and La Niña events, associated with the recession and expansion of glacial area, respectively.Following extraordinary warm events in 1991-1992, 1997-1998, 2002-2003, and 2009-2010, the Condoriri glacier experienced large reductions in area (Figure 4a).In contrast, following La Niña events (blue shading in Figure 4a), the glacier halted its shrinkage or even showed an increase.The expansion during La Niña events, however, was insufficient to reverse the long-term retreating trend.Note that temporary snow cover may have been misidentified as glacier, possibly accounting for the abrupt changes seen in glacial area in some cases.
On average, near-surface summer temperatures over the Altiplano are 0.7-1.3°Chigher during El Niño events than during La Niña events (Vuille et al., 2000).In contrast, interannual precipitation variability showed no clear association with either glacier area or ENSO events (Figure 4b).Ronchail (1998) showed that many dry periods in the Bolivian Altiplano are unrelated to El Niño events.Vuille (1999) used the Southern Oscillation Index (SOI), which is based on sea-level pressure differences between Tahiti and Darwin, to analyze precipitation patterns during ENSO events and found no significant relationship between SOI anomalies and precipitation amounts in the Altiplano.In our results, however, La Niña events appear to correlate with minor increases in rainfall amounts (Figure 4b).The SST anomaly during December to February is correlated more strongly with air temperature (R = 0.59) than with precipitation (R = 0.33).
The intensity of the SST anomaly did not appear to correlate with the rate of change in the glacier area.For example, the 1997-1998 El Niño event, one of the strongest recorded in the 20th century, caused severe drought in southern Bolivia, and the SST anomaly during that event was approximately three times the magnitude of the anomaly during other El Niño events (Figure 4c).However, in our satellite imagery results, we did not see a dramatic reduction in glacier area during or following the 1997-1998 event, perhaps because of the rather weak relationship between precipitation and ENSO.Vuille (1999) found that extreme precipitation can occur during extreme ENSO phases as well as during normal years.Unfortunately, the temporal resolution of the available satellite images is insufficient to detect such detailed variations in glacier area.However, where ground measurements and continuous observations are lacking, estimation using satellite imagery is the only practical way to monitor the area of whole glaciers.Therefore, the apparently strong relationship that we found between ENSO events and glacier area change indicates that a more detailed investigation is needed.

CONCLUDING REMARKS
Using satellite images of 1988-2010, we showed that the Condoriri glacier has dramatically retreated during the past few decades.If the retreat continues at the same rate, the glacier may completely disappear by 2035.Aside from the long-term retreating trend, the glacial area also showed interannual patterns of recession and expansion that correspond to warm and cold phases of ENSO, respectively.Although the estimated rates of glacier expansion during La Niña events were greater than the long-term retreat rate, the impacts of the overall long-term retreating trend and shortterm El Niño events outweighed the effect of La Niña on the glacial area.Thus, the net effect was a loss of area.
Veettil (2012) analyzed fluctuations in surface areas of ice caps in Ecuador on satellite images and found that El Niño events are followed by glacier ablation with a lag time of 1-3 months, whereas La Niña events are followed by positive mass balance changes.He also suggested that a small fragmented glacier responds quickly to La Niña events.This study by Veettil (2012) underlines the usefulness of satellite monitoring for studying changes in the surface area of glaciers.
Although our results suggest a strong temporal relationship between the area changes of Condoriri glacier and ENSO events, it is not possible to attribute the longterm retreating trend to natural or greenhouse gas forcing on the basis of our results.Lee and McPhaden (2010) showed that the warming of the central equatorial Pacific region during the past three decades is primarily due to the more intense El Niño events in the region.Therefore, the effect of natural or greenhouse gas forcing, if it exists, is too small to be compared with the effects of El Niño events.However, Yeh et al. (2009) have suggested that the increasing frequency of El Niño events in the central equatorial Pacific region is associated with a change in the thermocline structure under greenhouse gas forcing; for this reason, El Niño events are likely to become more frequent with continued global warming.Given the social, ecological, and economic importance of glaciers, further studies examining the relationship between glacier mass balance and meteorological parameters are needed.

Figure 1 .
Figure 1.The location of the Condoriri glacier in Bolivia.

Figure 2 .
Figure 2. A glacial lake formed at the foot of the glacier between 2001 and 2003.
Figure 4. Time series of (a) glacier area, (b) monthly mean temperature (line) and precipitation (blue bars) during November-April (rainy season) at La Paz, and (c) the SST anomaly in the Niño 3 region.El Niño and La Niña phases are also shown.The equation in (a) is for the red regression line.