Satellite Monitoring and Characterization of the 2010 Rockslide-Dammed Lake Gojal, North Pakistan

On January 4, 2010, a landslide blocked the Hunza River at Attabad, northern Pakistan (36.308°N, 74.820°E). The landslide destroyed the village of Attabad killing 19 people, and formed a dam approximately 1200m long, 350 meters wide, and 125 meters high. The flow of the Hunza river was blocked for 144 days, forming Lake Gojal. In addition to inundating several villages and submerging 22 km of the regionally critical Karakoram Highway, >25,000 people have been displaced or remain cut off from overland connection with the rest of the country. Lake overtopping began on May 29 via a 15m deep spillway excavated through the saddle of the dam. Remarkably, the slowly eroding natural structure remains largely intact and currently represents a new geologic feature, although a threat remains from possible catastrophic outburst flooding. We have monitored growth of the lake with multi-temporal satellite imagery collected from ASTER (Advanced Spaceborne Thermal and Reflection Radiometer) and ALI (Advanced Land Imager) sensors. We applied NASA’s ASTER Global Digital Elevation Model (GDEM) and SRTM-3 digital terrain data, along with field data obtained onsite by Schneider, and by Pakistan’s NDMA to derive volumes of the growing lake. Lake size peaked during mid-summer when it was ~22 km long, 12 km2, 119m deep, and contained 540 to 620 Mm3 water (SRTM-3 and GDEM +5m global correction estimates respectively). Our estimates indicated lake volumes three to four times higher than media quotes, and before spillover, were used to improve predictions of possible flood discharge and disaster management planning. Estimates of valley inflow based on a 31-year hydrographic history (Archer, D., 2003, Jour. Hydrology 274, 198-210) are consistent with our volume infilling estimates. As early as April 14 our volume assessments, coupled with hydrographic and seepage data were used to project a spillover date range of May 28-June 2, bracketing the actual overflow date. Additionally, we have applied vegetation indices (NDVI), landcover classifications, and image differencing change detection techniques to obtain reconnaissance level characterizations of lake-flood affected areas, including flooding of agricultural lands. Our successful prediction of lake growth and initial estimates of affected lands highlights the effectiveness of GIS methods applied to modern satellite datasets, and indicates the importance of monitoring natural hazard events with remote sensing, which can provide rapid assessments and augment onsite observations for disaster management support.