Change of topographic control on the extent of cirque glaciers since the Little Ice Age

Transcription

1 GEOPHYSICAL RESEARCH LETTERS, VOL. 33, L24505, doi: /2006gl028204, 2006 Change of topographic control on the extent of cirque glaciers since the Little Ice Age J. I. López-Moreno, 1 David Nogués-Bravo, 2 Javier Chueca-Cía, 3 and Asunción Julián-Andrés 4 Received 18 September 2006; revised 8 November 2006; accepted 22 November 2006; published 28 December [1] This paper analyses the influence of topography on the extent of cirque glaciers for four different stages since the Little Ice Age (LIA). The study focuses on a Pyrenean massif affected by a marked glacial shrinkage. Modeled distributed probabilities of glacier development during each stage from terrain characteristics enabled us to: (1) quantify the percentage of variance of the glacier distribution explained by topography; (2) assess the influence of individual variables on ice extent at each stage; and (3) identify those areas most vulnerable to ice degradation during the subsequent stage. Results show that topography has a greater influence on ice extent during times of warmer climate. In addition, the relative role of the different topographic variables changed over time. The probabilities of ice existence predicted by the models for subsequent stages are in good agreement with the observed paths of glacier retreat. Thus, topography should be considered a key factor in further research into the impact of climate change on glacier evolution. Citation: López-Moreno, L. I., D. Nogués-Bravo, J. Chueca-Cía, and A. Julián-Andrés (2006), Change of topographic control on the extent of cirque glaciers since the Little Ice Age, Geophys. Res. Lett., 33, L24505, doi: /2006gl Introduction [2] The near-ubiquitous worldwide glacier retreat that is currently underway [IUGG (CSS)-UNEP-UNESCO, 2005] has been highlighted as one of the most tangible and irrefutable arguments in support the fact that global climate has warmed since the beginning of the industrial era. In fact, it is widely recognized that glaciers provide valuable information in terms of monitoring climatic variability in mountainous areas where instrumental data are scarce. The usefulness of glacial bodies as indicators of climate change is based on their high sensitivity, especially that of smaller glaciers, to anomalies in precipitation and temperature [Carrivick and Brewer, 2004]. However, it must be remembered that topography exerts a noticeable influence on the response of glaciers to the observed regional-scale climate 1 Department of Geosciences, University of Fribourg, Fribourg, Switzerland. 2 Macroecology and Conservation Unit, University of Évora, Évora, Portugal. 3 Departamento de Geografía y Ordenación del Territorio, Facultad de Ciencias Humanas y de la Educación, Huesca, Spain. 4 Departamento de Geografía y Ordenación del Territorio, Facultad de Filosofía y Letras, Zaragoza, Spain. Copyright 2006 by the American Geophysical Union /06/2006GL [Bishop et al., 2001; Evans, 2006a]. Thus, the evolution of different glaciers within a single massif may well show marked differences according to the characteristics of the local terrain [Allen, 1998; Benn and Lehmkul, 2000]. In this context, research on topographic controls of glacial dynamics is necessary to reduce the existing noise when climate evolution and glacial changes are related. [3] López-Moreno et al. [2006] demonstrated non-linear relationships between terrain characteristics and the distribution of ice bodies within the Spanish Pyrenees. Thus, Generalized Additive Models (GAMs), which represent an improvement upon the linear regression approach [Hastie and Tibshirani, 1987], are proved to be highly accurate in establishing non-linear relationships between the extent of glaciers within the Maladeta Massif and the topographic variables of altitude, slope, potential incoming solar radiation, and terrain mean curvature. [4] In the present study, we used GAMs to clarify the effect of topography on glacier development during four stages since the Little Ice Age (LIA): (maximum LIA), 1957, 1981, and The used models are binomial (existence or absence of ice as response variable), and they provided probabilities of ice existence, using topographic variables as predictors. This approach enabled us to address the following key issues related to the impact of climate change on glaciers in mountainous regions: (1) the temporal stability or instability of topographic controls on glacier extent; (2) the identification of variables that show temporally varying influences on glacier development; and (3) assessment of the proposal that topography can be used as a reliable indicator of those areas that are most vulnerable to ice degradation under the assumption of a future warming climate. [5] The present study area comprises cirque glaciers of the Maladeta Massif within the Pyrenees (41 N, northeast Iberian Peninsula). Information on the exact location and physiographic characteristics of the massif is given by Chueca et al. [2005]. The Maladeta Massif is an excellent setting within which to apply the proposed methodology, as it contains the southernmost examples of glaciers within Europe. As in the rest of the Pyrenees, the surface extent and volume of these glaciers have decreased dramatically since the LIA; the survival of the remaining ice bodies is in a critical situation [López-Moreno et al., 2006]. 2. Data and Methods [6] The basis of the proposed methodology is the establishment of robust models that relate different topographic characteristics (predictors) to the existence of ice cover during each stage (response). Obtained models for each L of5

2 Figure 1. Changing extent of glaciers upon the Maladeta Massif during the four considered stages. stage enable us to determine the degree of topographic influence on glacier distribution, isolate the role of each of the considered predictors, and identify those areas that are most vulnerable to glacial retreat. [7] Accordingly, the first step was to reconstruct the extent of glaciers for a number of different periods for which we have reliable information. This was possible for the following stages: (maximum LIA), 1957, 1981, and The extent of glaciers during was determined from dated geomorphological features (prominent and well preserved moraine arcs), while the extent of glaciers during the other stages was mapped from aerial photographs. Once georectified, aerial photographs are highly reliable sources of information of this kind. The most recent set of aerial photographs were taken during the period ; these were updated to 2004 from observations carried out during fieldwork surveys. [8] The reconstructed glacial extents for each stage were matched with a Digital Elevation Model with 10 m cell-size resolution, providing information on the detailed topographic characteristics over both glaciated and non-glaciated terrain. For all glacial stages, we used the same set of variables that were considered to be significant predictors of glacier development in the area [López-Moreno et al., 2006]: i) altitude, ii) slope magnitude, iii) terrain mean curvature (half the sum of the principal curvatures at a point on a surface), and iv) potential mean annual solar radiation. These variables were selected as they are widely considered as being the most significant in accounting for glacier development and/or the location of the Equilibrium Line Altitude (ELA) [Allen, 1998; Brocklehurst and Whipple, 2004; Carrivick and Brewer, 2004]. In calculating potential solar radiation, we used a method that takes into account the effects of terrain complexity (shadowing and reflection), trajectory of the Sun throughout the day (hourly), the Earth Sun distance (monthly), atmospheric extinction, and daily solar position. GAMs were used to assess the probability of ice cover according to terrain characteristics. GAMs employ a regression technique that supports non- Gaussian error distributions and non-linear relationships between predictor and response variables [Guisan and Zimmermann, 2000]. In this case, the response variable is a binomial that indicates the presence or absence of ice cover. To evaluate the relative roles of each predictor in the final model, we calculated the change in total variance explained when each term was dropped from the analysis [Wood and Augustin, 2002]. Ten percent of the cells within the study area were randomly selected for the analyses. Then, following the method of Guisan and Zimmermann [2000], the sample was split into two subsets: a calibration subset (70% of the sample) and a validation subset (30% of the sample). Continuous probability maps were compiled after applying the statistical models (obtained in the calibration process) to the entire area. The predictive capacity of the models was assessed from the Kappa values, K, drawn from a confusion matrix (summary of the accuracy and faults when the existence or absence of ice is predicted for different probability thresholds) obtained from the validation dataset [Fielding and Bell, 1997]. Kappa values vary between 0 and 1, with K < 0.4 considered to indicate poor predictive capacity, 0.4 < K < 0.75 representing good predictive capacity, and K > 0.75 indicating excellent predictive capacity [Landis and Koch, 1997]. Additional information on the use of GAMs to model glacial extent and the validation procedure is given by López-Moreno et al. [2006]. [9] The location of the ELA for each stage was determined using the classic Kurowski method [Carrivick and Brewer, 2004], and climatic characteristics for the area were obtained from instrumental series and dendroclimatic reconstructions. Additional detailed information in this regard is given by Chueca et al. [2005]. 3. Results [10] Figure 1 shows the changing extent of glaciers in the Maladeta Massif during the four considered stages. Table 1 provides information on the elevation of the ELA, the progressive decrease in the extent of the ice surface, and the annual surface-loss rate for the intervals between the reconstructed ice extents. During the maximum of the LIA ( ), 21 glaciers and glacierettes made up a combined 730 ha of ice. The extents of the different ice bodies were highly variable: glaciers upon north-facing slopes extended down to 2400 m a.s.l., whereas those on southfacing slopes were smaller, although well developed ice bodies extended below 2800 m a.s.l. The altitude of the regional ELA at this time is estimated to have been around 2840 m a.s.l. [11] By 1957, the surface extent of ice had decreased a 61% of that measured during the LIA (286 ha), representing an annual rate of ice degradation of 3.5 ha yr 1. Chueca et al. [2005] related this observed decrease in ice extent to a progressive increase of temperature (ELA increased in elevation to 3030 m a.s.l.) and a dominance of negative Table 1. Changes in the ELA and Extent of Ice Cover During the Intervals Between the Considered Stage 1820/ ELA (+190 m) (+5 m) (+60 m) Surface covered by ice (Decrease in surface) (Decrease in percentage) (444 ha) (61%) (41 ha) (14%) (70 ha) (28%) Surface-loss rate 3.5 ha yr 1 1.7hayr ha yr 1 2of5

3 Table 2. Kappa Values Obtained From the Validation Procedure Kappa Values Threshold Mean value Figure 2. Predicted probabilities of the extent of glacier development for the four considered stages. precipitation anomalies. The glacier retreat rate decreased during the period when ice covered 245 ha ( 0.13 ha yr 1 ), a 14% less than This period was relatively cold and wet, and the ELA increased a small amount to 3035 m. In 2004, the ELA was at 3095 m a.s.l., 60 m higher than the elevation during The climate during the period was characterized by positive anomalies of summer temperature and negative precipitation anomalies. The surface area covered by ice in 2004 was 175 ha. It means that glaciated area decreased a 28% in only 23 years, representing a loss rate of 3.1 ha yr 1 over the period Bearing in mind these general trends, Figure 1 reveals significant differences in the trends for individual glaciers. For example, during the period some glaciers remained extremely stable whereas others retreated markedly or disappeared completely. [12] Figure 2 shows the calculated probabilities of glacier development for the four stages, based on the terrain characteristics as predictors. The predicted probabilities decreased significantly over the period from the LIA until the present. In terms of the four stages, the upper and middle sectors of glaciers have higher predicted probabilities, while glacial termini show values near 0. [13] Kappa values (K) (Table 2) highlight marked differences in the abilities of the topographic variables to predict the extent of the glaciers over time. Thus, for predictions for the LIA maximum, K values are around , indicating large deviations in the confusion matrix. For 1957 and 1981, K values increase to within the range , and values for 2004 are close to 0.7. These trends represent a progressive increase (from poor to good or very good) in the predictive capacity of the topographic variables in explaining glacier development; in other words, the data demonstrate a clear temporal increase in the influence of topography on glacial extent. [14] A number of findings on the relationships between topography and glacier dynamics can be derived from Table 3, which shows the influence of each variable on the extent of ice cover. The influence of each variable was quantified for each stage by comparing the change in total variance when each term was dropped from the models. Explained variance using all predictors varies from 44% during the LIA to 77% in The progressive increase of the total explained variance confirms the above conclusion based on an interpretation of Table 2. For the entire set of models, the ranking of variables by explained variance in terms of explaining glacier development (decrease of variance when a variable is dropped from the model divided by the total explained variance, see relative contributions n Table 2) is as follows: mean curvature (explained variance decreases by 3 8% when the variable is dropped from the Table 3. Reduction in the Explained Variance Obtained When Each Predictor Variable is Removed From the Model a All variables Curvature Radiation Slope Altitude Explained variance Decrease Relative contribution* Explained variance Decrease Relative contribution Explained variance Decrease Relative contribution Explained variance Decrease Relative contribution a Relative contribution is obtained dividing the decrease of the explained variance when a term is dropped by the total explained variance. 3of5

4 Figure 3. Percentage of cells from which ice had disappeared at the subsequent stage. Calculations were made according to varying ranges of predicted probability of ice cover. model), slope (11 15%), potential incoming solar radiation (27 45%), and altitude (57 80%). [15] A number of interesting points arise from the temporal evolution of these data. First, the influence of altitude tends to decrease with the time, from 80% of the explained variance during the LIA, 70 % in 1957, 67 % in 1981, and just 57 % in Second, the opposite trend is recorded for mean curvature and solar radiation. The influence of the former increases from just 2.9% during the LIA to approximately 8% in 2004, while the contribution of solar radiation increases from 27% during the LIA to 45% in [16] Finally, Figure 3 shows the percentage of cells that changed from glaciated to non-glaciated within different ranges of probabilities predicted for each stage. We draw two conclusions from this figure: (1) the number of cells where ice cover remains increases as predicted probabilities are higher; and (2) marked differences occur among different stages in terms of the percentage of ice-degradation cells according to predicted probabilities. This latter trend probably reflects the differing lengths and climatic characteristics of each period. For example, the long interval from 1820 to 1957 featured markedly warm and dry anomalies, and ice disappeared almost completely in cells with probabilities less than 0.2, most of ice disappeared for values of , and sharp decreases in the rates of ice retreat were recorded for values above 0.5. The interval was wet and cold, leading to a degree of stabilization of glacial retreat. The highest rate of ice disappearance occurred for the lowest predicted probabilities and vice versa; however, the development of climatic conditions suited to glacier development meant that half of the most vulnerable ice cells (probability < 0.1) remained glaciated. The interval was a brief but critical period for glacier development because of conditions of high temperatures and low precipitation. Consequently, the ice disappeared from more than 70% of cells with probabilities of less than 0.3. For cells with probabilities above 0.5, the degree of ice persistence increased rapidly. 4. Discussion and Conclusions [17] One of the main results obtained in this work is the detection of clear temporal changes in the degree to which topography controls glacier development in the study area. This pattern of change led to a progressive increase in the influence of terrain characteristics on the extent of ice bodies. This conclusion is also supported by an observed temporal increase in model accuracy when using topographic predictors from the maximum LIA to the present. These observations introduce a source of non-linearity in the response of glaciers to climate, a relationship that it is already complex in itself [Klinge et al., 2003; Evans, 2006a, 2006b]. The relative importance of the considered predictors has also changed with time: the relative influence of terrain mean curvature and exposure to solar radiation has increased over time, whereas the influence of altitude has decreased since the LIA maximum. The limited ability of topography to predict ice distribution in the period suggests that during periods of glacial advance (as in the LIA), the ice was able to occupy surfaces that are usually not favourable for glacier development, e.g., slopes strongly exposed to the Sun and convex surfaces. The thickness of ice in such sites was probably significantly less than that in the most favourable locations. Moreover, the existence of wide areas above the ELA enabled several glaciers to develop incipient glacial tongues that covered some of the cirque terrain; this occurred via ice flux dynamics rather than the control of topographic or climatic factors. These factors are the reasons behind the large amount of unexplained variance and the poor reliability of the predictions. [18] As the climate warmed, the glaciers retreated into classic cirque basins. Logically, the retreat was especially rapid in areas with adverse topography where melting rates were higher and the volume of accumulated ice was lower. Cirque basins tend to have high terrain curvature (hence, explaining a rise in that parameter), but most importantly, some of these basins may be strongly illuminated by the sun or reflected radiation, whereas in others nearby peaks and ridges can produce strong shadows (hence, explaining a rise over time in the radiation parameter s influence). Once the cirque glaciers in sun-exposed basins are burnt out, we predict that what will remain will be small, highly resistant, deeply shadowed cirque glaciers. These conditions explain a reduction of the response of remnant glaciers to climatic signals which are closely related to the elevation parameter: sensible and latent heat fluxes, amount of precipitation, and ratio of precipitation as snow or rain [López-Moreno et al., 2006]. Thus, the decreasing contribution of altitude over time accounts for the locations of glaciers, which benefit from the increasing influence of topographic conditions within the cirques where remaining glaciers are located. Moreover, the elevation parameter s decreased role could be also related with the steep slopes of small remnant glaciers, whereby a given incremental rise in ELA reduces the area by less than the same incremental change imposed on a larger valley glacier. [19] We obtained a strong relationship between predicted probabilities and the spatial patterns of subsequently observed glacier retreat. It is logical that during long periods of time that are unfavourable for glacier development, an increased percentage of cells are affected by ice degradation, especially within the lower probability ranges. Unfortunately, we considered just a limited number of reconstructed glacial stages, and the establishment of robust links between climatic conditions and the meaning of the predicted probabilities in terms of expected glacial shrinkage is therefore open to improvement. This would be an 4of5

5 interesting issue to pursue in other study areas where a greater number of glacial stages are reconstructed. The possibility of acquiring data on ice surfaces using remote sensing at high spatial and temporal resolutions [i.e., Kieffer and Kargel, 2000; Kargel et al., 2005] provides a potential solution to this challenge in the near future. In any case, with the information provided herein and taking into account the probabilities predicted for 2004 (i.e., the number of cells in Figure 2D with probability < 0.3), it is reasonable to deduce a high risk of total disappearance of some of the studied ice bodies and noticeable reductions in the size of other glaciers if the climate conditions experienced over the past two decades continue into the future. [20] Acknowledgments. This study was supported by the following research project: Study of the dynamics of glaciers in the Aragonais Pyrenees (H9007-CMA), funded by the Gobierno de Aragón. References Allen, T. R. (1998), Topographic context of glaciers and perennial snowfields, Glacier National Park, Montana, Geomorphology, 21, Benn, D. I., and F. Lehmkuhl (2000), Mass balance and equilibrium-line altitudes of glaciers in high mountain environments, Quat. Int., 65/66, Bishop, M., R. Bonk, U. Kamp, and F. Shroeder (2001), Terrain analysis and data modeling for alpine glacier mapping, Polar Geogr., 25(3), Brocklehurst, S. H., and K. X. Whipple (2004), Hypsometry of glaciated landscapes, Earth Surf. Processes Landforms, 29, Carrivick, J. L., and T. R. Brewer (2004), Improving local estimations and regional trends of glacier equilibrium line altitudes, Geogr. Ann., 86a, Chueca, J., A. Julián, M. A. Saz-Sánchez, J. Creus-Novau, and J. I. López- Moreno (2005), Responses to climatic changes since the Little Ice Age on Maladeta Glacier (Central Pyrenees), Geomorphology, 68, Evans, I. S. (2006a), Local aspect asymmetry of mountain glaciation: A global survey of consistency of favoured directions for glacier numbers and altitudes, Geomorphology, 73, Evans, I. S. (2006b), Glacier distribution in the alps: Statistical modelling of altitude and aspect, Geogr. Ann., 88a, Fielding, A. H., and J. F. Bell (1997), A review of methods for the assessment of prediction errors in conservation presence/absence models, Environ. Conserv., 24, Guisan, A., and N. E. Zimmermann (2000), Predictive habitat distribution models in ecology, Ecol. Modell., 135, Hastie, T., and R. Tibshirani (1987), Generalised additive model: Some applications, J. Am. Stat. Assoc., 82, IUGG (CCS)-UNEP-UNESCO (2005), Fluctuations of Glaciers , vol. VIII, 307 pp., World Glacier Monit. Serv., Zurich. (Available at Kargel, J. S., et al. (2005), Multispectral imaging contributions to global land ice measurements from space, Remote Sens. Environ., 99, Kieffer, H. H., and S. S. Kargel (2000), New eyes in the sky measure glaciers and ice sheets, Eos Trans. AGU, 81(265), Klinge, M., J. Börner, and F. Lehmkuhl (2003), Climate pattern, snow and timberlines in the Altai Mountains, central Asia, Erdkunde, 57, Landis, J. R., and G. C. Koch (1997), The measurement of observed agreement for categorical data, Biometrics, 33, López-Moreno, J. I., D. Nogués-Bravo, J. Chueca-Cía, and A. Julián-Andrés (2006), Glacier development and topographic context, Earth Surf. Processes Landforms, 31(12), Wood, S. N., and N. H. Augustin (2002), GAMs with integrated model selection using penalized regression splines applications to environmental modelling, Ecol. Modell., 157, J. Chueca-Cía, Departamento de Geografía y Ordenación del Territorio, Facultad de Ciencias Humanas y de la Educación, Huesca, Spain. (jchueca@unizar.es) A. Julián-Andrés, Departamento de Geografía y Ordenación del Territorio, Facultad de Filosofía y Letras, Zaragoza, Spain. (ajulian@ unizar.es) J. I. López-Moreno, Department of Geosciences, University of Fribourg, Fribourg CH-1700, Switzerland. (ignacio.lopez@unifr.ch) D. Nogués-Bravo, Macroecology and Conservation Unit, University of Évora, Évora, Portugal. (dnogues@bi.ku.dk) 5of5