Remote Sensing and Hydrology 2000 (Proceedings of a symposium held at Santa Fe, New Mexico, USA, April 2000). IAHS Publ. no. 267, 2001. 99 Modelling runoff from large glacierized basins in the Karakoram Himalaya using remote sensing of the transient snowline ANDREW T. LOWE Cold Regions Research Centre, Wilfrid Laurier University, Waterloo, Ontario NIL 3C5, Canada DAVID N. COLLINS School of Environment and Life Sciences, University ofsalford, Salford Crescent, Manchester M5 4WT, UK e-mail: d.n.collins@salford.ac.uk Abstract A glacier runoff model has been developed with a view to forecasting discharge from large remote glacierized basins in the Karakoram Himalaya. The model requires modest data input. Energy inputs are represented by positive degree-days, determined for various elevations from lapse rates, and depletion of snow cover and consequent rise of the transient snowline (TSL) during summer are determined from close-interval satellite monitoring by the Advanced Very High Resolution Radiometer (AVHRR) sensor. Coupled with a digital elevation model, position of the TSL is used to partition glacierized areas into snow-covered and exposed ice portions, to which differential melt rates are then applied. The degree-day factor applied over bare ice is substantially higher than that over snow. Model performance evaluations indicated a good fit between measured and simulated runoff for the Hunza River at Danyore Bridge for both spring and summer of 1989. Both the model and the spatial and temporal resolution of the satellite platforms are well suited to and appropriate for the scale of the large glacierized high mountain basins of the Himalaya. Key words AVHRR; glacier hydrology; glacierized basin; Karakoram mountains; runoff modelling INTRODUCTION Much of the total flow of the upper Indus River is supplied from snow and ice melt in headwater basins in the Karakoram mountains. Runoff, supplied to the Indus River from snowmelt in spring and ice melt later in the season, is important for water resources in the semiarid valleys of the Karakoram, and downstream across the arid lowlands of the Indus plain, where meltwaters provide power for industry and irrigation for agriculture. The aim was to develop a glacier runoff model suitable and appropriate for simulating (and ultimately forecasting) runoff from large glacierized basins in the Himalaya. Movement of the transient snowline (TSL) is critical in modelling glacier runoff, as the relationship between elevation of TSL and basin hypsometry determines the area of ice exposed to melting and, together with energy input, influences the seasonal pattern of runoff formation (Collins & Lowe, 1997). The model is simple in nature but conceptually sound incorporating TSL migration derived from the National Oceanic and Atmospheric Administration (NOAA) and Defence Meteorological Satellite Program
100 Andrew T. Lowe & David N. Collins (DMSP) satellites. The approach avoids the need to treat in detail all the physical processes and parameters that affect melting over snow and ice surfaces in glacier basins. The model was constructed to rely on only modest data input, because field measurements are difficult to achieve in remote Himalayan basins. The model design uses remotely sensed snow cover information, terrain data expressed as a digital elevation model (DEM), and lapse-rate dependent degree-day data to simulate runoff from large glacierized basins. In the model, snow-covered area and seasonal rise of the TSL are derived from close-interval satellite monitoring. Coupled with the DEM, movement of the TSL is used to partition glacier basins into snow-covered and bare-ice portions. STUDY AREA The Hunza basin, at about 36 N 75 E, in the Karakoram Himalaya has an approximate area of 13 200 km 2, of which perennial snow and ice cover is about 28.5%. Runoff from the Hunza basin has been gauged at Danyore Bridge located on the Hunza River immediately above the confluence with the Gilgit River, which in turn is tributary to the upper Indus. MODEL DESCRIPTION The simple conceptual model developed for the computation of runoff from large glacier basins in the Karakoram is illustrated in Fig. 1. In this model, data inputs consist of satellite imagery, digital elevation data and climate data. Satellite data are processed, corrected and then classified, providing images of snow cover and TSL position. Figure 1 shows the importance of the DEM in (a) registration of the satellite images, (b) calculation of TSL elevation, (c) construction of elevation zones, (d) calculation of basin/ice hypsometry, and (e) correction of temperature and precipitation data for elevation. The glacier runoff model uses basin/ice hypsometry and TSL elevation to partition glacier basins into snow-covered and bare ice portions to which degree-day energy inputs are applied. Field temperature data, corrected for elevation, are used in the model to generate melt. A full description of the glacier runoff model, with details of melt and drainage routines, is given in Lowe (1999). The model has been developed to simulate daily runoff from large glacierized basins in the Karakoram and, in this paper, is applied to the Hunza basin. The areal extent of the Hunza basin necessitates the use of remote sensing for determining the position of the TSL. Low cost, wide swath, visible spectrum sensors (e.g. AVHRR and DMSP Operational Linescan System (OLS)) on satellites with daily repeat cycles provide update frequency of clear-sky scenes and resolution appropriate to the scale of large basins such as those of the Karakoram in which field measurement programmes are modest. CONCEPTUAL FRAMEWORK The conceptual framework for the interaction of TSL with basin hypsometry is described fully in Collins (1998). The position of the TSL partitions the surface of a
Modelling runoff from large glacierized basins in die Karakoram Himalaya 101 glacierized basin into snow-covered and snow-free areas, and hence also glaciers into ice- and snow-covered portions. As the winter snow cover on a glacier is removed by melting in spring and summer, the TSL moves up-glacier exposing the underlying bare ice. Thermal production of melt depends mainly on the albedo of the snow/ice surface. Therefore, as the snow is removed and the low albedo glacier ice area expands as the TSL ascends, melt production is increased dramatically. Increasing runoff between June and August effectively results from the rising TSL exposing an increasing area of bare ice to melt. This offsets the seasonal decline in radiation input after the June solstice reducing the melt rate per unit area. The elevation of the TSL only falls, or moves down-glacier should summer snowfall occur, i.e. when snow falls below the TSL position to date. Precipitation is effectively partitioned spatially as snow or rain by the elevation of the 0 C isotherm in the atmosphere. Satellite data Climate data Snow cover image Lapse rates DEM Transient snowline elevation Basin hypsometry Corrected climate data Glacier runoff model Simulated runoff Performance model Fig. 1 Diagram showing the modelling scheme used for computing runoff from large glacierized basins, and the primary datasets used in the glacier runoff model. SPATIAL EXTENT OF SNOW COVER AND ELEVATION OF THE TSL AVHRR scenes in Fig. 2 show snow cover in the Hunza basin on 8 May, 1 June and 26 July 1989, on which dates 88%, 66% and 47% of the basin area respectively remained above the TSL (Table 1). In this period, the regional TSL rose from an elevation within the range 3200-3300 m a.s.l. to 4550 m a.s.l. in late July, reaching a maximum at 5000 m a.s.l. by mid-august. Ascent of the TSL led to the snow-covered portion of the Hunza basin declining from 88% on 8 May to 28% of the overall basin area on 15 August. Exposure of glacier ice by the seasonal rise of the TSL is clearly shown in the snow cover maps in Fig. 2. On 8 May, with the TSL at 3280 m a.s.l., only 168.7 km 2 of ice was exposed to melt. By 1 June the TSL had risen a further 770 m to 4050 m a.s.l, increasing the area of exposed glacier ice to 805.4 1cm 2. A further rise of
102 Andrew T. Lowe & David N. Collins Fig. 2 AVHRR scenes acquired on (a) 8 May, (b) 1 June, and (c) 26 July 1989, showing snow cover in the Hunza basin, together with respective snow cover maps derived from the imagery (d), (e), (f). Grey areas show the expansion of bare ice area. Table 1 Regional TSL elevations in the Hunza basin calculated from AVHRR imagery acquired between April and July 1989, with percentage of basin area which was snow-covered, percentage of glacier ice which was snow-free, and total area of exposed bare ice at the time of each image acquisition. Date of image acquisition Regional TSL elevation (m a.s.l.) Percentage of basin area snow-covered (%) Percentage of glacier area snow-free (%) Area of bare ice exposed (km 2 ) 19/04/1989 3350 86.26 5.98 209.28 28/04/1989 3285 87.71 4.86 170.25 08/05/1989 3280 87.80 4.81 168.70 01/06/1989 4050 66.49 22.98 804.52 26/07/1989 4550 47.12 43.60 1526.32 15/08/1989 5000 28.42 68.74 2406.54 the TSL to 4550 m a.s.l. by 26 July expanded the exposed ice area to 1526.3 km 2. Maximum TSL elevation of 5000 m a.s.l. was reached by 15 August, with 69% of the ice-covered portion of the basin exposed, a total of 2406.5 km 2 of glacier ice.
Modelling runoff from large glacierized basins in die Karakoram Himalaya 103 OUTPUT The model was used to compute runoff in the Hunza River during the spring and summer of 1989. Modelled flow was well simulated by comparison with measured runoff. An example of one model run is given in Fig. 3. Nash-Sutcliffe efficiency 2 2 coefficients E and E i and percentage volume deviation values (D v ) indicate a close similarity between computed and measured runoff. Higher values of E 2 and E 2 ]n indicate a closer fit between modelled and actual runoff, a value of 1 indicating perfect fit. For the simulations conducted, E 2 values range between 0.71 and 0.82, and E 2 \ n between 0.80 and 0.94 (Lowe, 1999). These model efficiencies compare favourably with those from previous studies modelling snow-covered and glacierized basins, which generally lie between 0.54 and 0.90 (Fountain & Tangborn, 1985). Values of A, in the model described here are sufficiently low to indicate that the model has simulated runoff in the Hunza basin well. Volumetric deviations between simulated and measured runoff for all model runs were in the range -0.54 to +3.13 (Lowe, 1999). High coefficients of determination, R 2, for model runs, were in the range 0.74 to 0.85, indicating a good ability by the model to simulate the measured data (Lowe, 1999). - MEASURED -SIMULATED R = 0.85 E!, = 0.80 D V = + 2. 7 0 May June July 1989 Fig. 3 Measured and simulated runoff at Danyore Bridge in the Hunza River, between 21 May and 4 August 1989. DISCUSSION Remote sensing provides a means of observing the evolution of the snow-covered area and TSL migration during summer in mountain basins. Remote sensing of the changing snow cover at repeat cycles of several days is required if the "volatile" position of the TSL is to be monitored during a melt season. Low-cost, wide swath, visible spectrum sensors on-board the NOAA and DMSP satellites with daily repeat cycles can provide clear-sky scenes, with frequency and resolution appropriate for monitoring the regional TSL at the scale of the large glacierized basins in the Himalaya. Both the NOAA and DMSP satellites have a daylight repeat cycle of 24 h,
104 Andrew T. Lowe & David N. Collins so that, even in cloud-prone mountainous areas, seasonal changes in snow-covered area and position of the TSL can be observed at close intervals in time. Moreover, cost is reduced considerably as resolution is traded off against enhanced repeat cycle frequency and wider swath for the AVHRR and DMSP OLS sensors. Overall the study indicates that close-interval monitoring of the depletion of the snowpack, and the subsequent rise of the TSL, is a pre-requisite for modelling and ultimately forecasting of runoff from large glacierized basins in the Karakoram. Incorporation of the seasonal rise of the TSL and subsequent expansion of low albedo glacier ice within the model is conceptually important. By modelling the rise of the TSL and exposure of glacier ice, the contribution of melt from glacier ice to the total basin discharge is better represented than by the application of other runoff models to glacierized basins (Lowe, 1999). Good representation of the contribution made to runoff by glaciers is especially important in the Karakoram because runoff arising from the glacierized portion of such basins dominates total basin discharge more than in the case for the European Alps (Collins & Lowe, 1997). Models that inadequately represent or neglect melting from glaciers, and/or fail to consider the influence of the seasonal rise of the TSL, may result in poor forecasts of flow from glacierized basins in the Karakoram mountains. The model is a conceptually sound alternative to other runoff models for simulating and forecasting runoff from large glacier basins in the Himalaya. CONCLUSION A glacier runoff model that is suitable and appropriate for simulating (and forecasting) runoff from large glacierized basins in the Karakoram Himalaya has been developed. The model is simple in nature and conceptually sound, incorporating TSL migration obtained from the NOAA and DMSP satellites. The modelling scheme used here, and the spatial and temporal resolution of the satellite platforms advocated for use to monitor the TSL, are appropriate in both spatial and temporal scale, for large glacierized basins. Further data are required to calibrate the model more fully and thereby enable its subsequent use as a predictive tool in water resource management. Acknowledgements The authors gratefully acknowledge receipt of financial support from the Natural Environment Research Council (grant GR9/222 to DNC). The Leverhulme Trust funded the continuation of this research through a post-doctoral Study Abroad Fellowship to ATL. REFERENCES Collins, D. N. (1998) Outburst and rainfall-induced peak runoff events in highly-glacierised Alpine basins. Processes 12, 2369-2381. Collins, D. N. & Lowe, A. T. (1997) Remote sensing of the seasonal ascent of the transient snowline for analysis of variability of runoff from iarge glacierized basins in the Karakoram Mountains. In: Proc. Brit. Hydrol. Soc. Sixth National Hydrology Symp. (Salford, UK), 5.19-5.26. Institute of Hydrology, Wallingford, UK. Fountain, A. G. & Tangborn, W. V. (1985) Overview of contemporary techniques for the prediction of runoff from glacierized areas. In: Techniques for Prediction of Runoff from Glacierized Areas (ed. by G. J. Young), 27-41. IAHS Publ. no. 149. Hydrol. Lowe, A. T. (1999) Runoff modelling from large glacierized basins in the Karakoram Himalayas using remote sensing of the transient snowline. PhD Thesis, University of Manchester, Manchester, UK.