Browsing Tag

Climate Change

Climate Change

Acceleration of ice loss across the Himalayas over the past 40 years

Himalayan glaciers supply meltwater to densely populated catchments in South Asia, and regional observations of glacier change over multiple decades are needed to understand climate drivers and assess resulting impacts on glacier-fed rivers. Here, we quantify changes in ice thickness during the intervals 1975–2000 and 2000–2016 across the Himalayas, using a set of digital elevation models derived from cold war–era spy satellite film and modern stereo satellite imagery. We observe consistent ice loss along the entire 2000-km transect for both intervals and find a doubling of the average loss rate during 2000–2016 [−0.43 ± 0.14 m w.e. year−1 (meters of water equivalent per year)] compared to 1975–2000 (−0.22 ± 0.13 m w.e. year−1). The similar magnitude and acceleration of ice loss across the Himalayas suggests a regionally coherent climate forcing, consistent with atmospheric warming and associated energy fluxes as the dominant drivers of glacier change.


The Intergovernmental Panel on Climate Change 5th Assessment Report estimates that mass loss from glaciers contributed more to sea-level rise than the ice sheets during 1993–2010 (0.86 mm year−1 versus 0.60 mm year−1, respectively), yet uncertainties for the glacier contribution are three times greater (1). Glaciers also contribute locally to water resources in many regions and serve as hydrological buffers vital for ecology, agriculture, and hydropower, particularly in High Mountain Asia (HMA), which includes all mountain ranges surrounding the Tibetan Plateau (23). Shrinking Himalayan glaciers pose challenges to societies and policy-makers regarding issues such as changing glacier melt contributions to seasonal runoff, especially in climatically drier western regions (3), and increasing risk of outburst floods due to expansion of unstable proglacial lakes (4). Yet, substantial gaps in knowledge persist regarding rates of ice loss, hydrological responses, and associated climate drivers in HMA (2).

Mountain glaciers are known to respond dynamically to a variety of drivers on different time scales, with faster response times than the large ice sheets (56). In the Himalayas, in situ studies document significant interannual variability of mass balances (79) and relatively slower melt rates on debris-covered glacier tongues over interannual time scales (1011). Yet, the overall effects of surface debris cover are uncertain, as many satellite observations suggest similar ice losses relative to clean-ice glaciers over similar or longer periods (1213). Because of the complex monsoon climate in the Himalayas, dominant causes of recent glacier changes remain controversial, although atmospheric warming, the albedo effect due to deposition of anthropogenic black carbon (BC) on snow and ice, and precipitation changes have been suggested as important drivers (1416).

Model projections of future Himalayan ice loss and resulting impacts require robust observations of glacier response to past and ongoing climate change. Recent satellite remote sensing studies have made substantial advances with improved spatial coverage and resolution to quantify ice mass changes during 2000–2016 (121718), and former records extending back to the 1970s have been presented for several areas using declassified spy satellite imagery (131922). These long-term records are especially critical for extracting robust mass balance signals from the noise of interannual variability (6). Many studies also report the highly heterogeneous behavior of glaciers in localized regions, with some glaciers exhibiting faster rates of ice loss during the 21st century (2022). Independent analyses document simultaneously increasing atmospheric temperatures at high-elevation stations in HMA (2326). To robustly quantify the regional sensitivity of these glaciers to climate change, a reliable Himalaya-wide record of ice loss extending back several decades is needed.

Here, we provide an internally consistent dataset of glacier mass change across the Himalayan range over approximately the past 40 years. We use recent advances in digital elevation model (DEM) extraction methods from declassified KH-9 Hexagon film (27) and ASTER stereo imagery to quantify ice loss trends for 650 of the largest glaciers during 1975–2000 and 2000–2016. All aspects of the analysis presented here only use glaciers with data available during both time intervals unless specified otherwise. We investigate glaciers along a 2000-km transect from Spiti Lahaul to Bhutan (75°E to 93°E), which includes glaciers that accumulate snow primarily during winter (western Himalayas) and during the summer monsoon (eastern Himalayas), but excludes complications of surging glaciers in the Karakoram and Kunlun regions where many glaciers appear to be anomalously stable or advancing (2). Our compilation includes glaciers comprising approximately 34% of the total glacierized area in the region, which represents roughly 55% of the total ice volume based on recent ice thickness estimates (1528). This diverse dataset adequately captures the statistical distribution of large (>3 km2) glaciers, thus providing the first spatially robust analysis of glacier change spanning four decades in the Himalayas. We extract DEMs from declassified KH-9 Hexagon images for the 650 glaciers, compile a corresponding set of modern ASTER DEMs, fit a robust linear regression through every 30-m pixel of the time series of elevations, sum the resulting elevation changes for each glacier, divide by the corresponding areas, and translate the volume changes to mass using a density conversion factor of 850 ± 60 kg m−3(see Materials and Methods).


Glacier mass changes

Our results indicate that glaciers across the Himalayas experienced significant ice loss over the past 40 years, with the average rate of ice loss twice as rapid in the 21st century compared to the end of the 20th century (Fig. 1). We calculate a regional average geodetic mass balance of −0.43 ± 0.14 m w.e. year−1 (meters of water equivalent per year) during 2000–2016, compared to −0.22 ± 0.13 m w.e. year−1 during 1975–2000 (−0.31 ± 0.13 m w.e. year−1 for the full 1975–2016 interval) (see Materials and Methods). A 30-glacier moving average shows a quasi-consistent trend across the 2000-km longitudinal transect during both time intervals (Fig. 1), and subregions have similar means and distributions of glacier mass balance. Some central catchments deviate somewhat from the Himalaya-wide mean during 2000–2016 (by approximately 0.1 to 0.2 m w.e. year−1) in the Uttarakhand (~79.0° to 80.0°E), the Gandaki catchment (~83.5° to 84.5°E), and the Karnali catchment (~81° to 83°E), which has fewer larger glaciers and relatively incomplete data coverage. Similar to previous in situ and satellite-based studies (1829), we observe considerable variation among individual glacier mass balances, with area-weighted SDs of 0.1 and 0.2 m w.e. year−1during each respective interval for the 650 glaciers. This variability most likely reflects different glacier characteristics such as sizes of accumulation zones relative to ablation zones, topographic shading, and amounts of debris cover. Yet, we find that, in our survey (using a rough average of 60 glaciers per 7000-km2 subregion), local variations tend to average out and mean values are similar across most catchments.

Fig. 1Map of glacier locations and geodetic mass balances for the 650 glaciers.Circle sizes are proportional to glacier areas, and colors delineate clean-ice, debris-covered, and lake-terminating categories. Insets indicate ice loss, quantified as geodetic mass balances (m w.e. year−1) plotted for individual glaciers along a longitudinal transect during 1975–2000 and 2000–2016. Both inset plots are horizontally aligned with the map view. Gray error bars are 1σ uncertainty, and the yellow trend is the (area-weighted) moving-window mean, using a window size of 30 glaciers.

Contrasting distributions of glacier mass balances are evident when comparing between time intervals. The 1975–2000 distribution has a negative tail extending to −0.6 m w.e. year−1, while the 2000–2016 distribution is more negative, extending to −1.1 m w.e. year−1 (Fig. 2A). During the more recent interval, glaciers are losing ice twice as fast on average (Fig. 2B), though this varies somewhat between subregions. For example, we find that the average rate of ice loss has increased by a factor of 3 in the Spiti Lahaul region, and by a factor of 1.4 in West Nepal. We also compile altitudinal distributions of ice thickness change for the glaciers and create a Himalaya-wide average thickness change profile versus elevation (Fig. 2, C and D). These distributed thinning profiles are a function of both thinning by mass loss and of dynamic thinning due to ice flow. We find that the 2000–2016 thinning rate (m year−1) profile is considerably steeper, which is likely caused by a combination of faster mass loss and widespread slowing of ice velocities during the 21st century (230).

Fig. 2Comparison of ice losses between 1975–2000 and 2000–2016 for the 650 glaciers.(A) Histograms of individual glacier geodetic mass balances (m w.e. year−1) during 1975–2000 (mean = −0.21, SD = 0.15) and 2000–2016 (mean = −0.41, SD = 0.24). Shaded regions behind the histograms are fitted normal distributions. (B) Result of dividing the modern (2000–2016) mass balances by the historical (1975–2000) mass balances for each glacier, showing the resulting distribution of the mass balance change (ratio) between the two intervals (mean = 2.01, SD = 1.36). In this case, the shaded region is a fitted kernel distribution. (C) Altitudinal distributions of ice thickness change (m year−1) separated into 50-m elevation bins during the two intervals. (D) Normalized altitudinal distributions of ice thickness change. Normalized elevations are defined as (z − z2.5)/(z97.5 − z2.5), where z is elevation and subscripts indicate elevation percentiles. This scales all glaciers by their elevation range (i.e., after scaling, glacier termini = 0 and headwalls = 1), allowing for more consistent comparison of ice thickness changes across glaciers with different elevation ranges. Note the abrupt inflection point in the 2000–2016 profile at ~0.1; this is likely due to retreating glacier termini. Shaded regions in the altitudinal distributions indicate the SEM estimated as σz/nz−−√, where σz is the SD of the thinning rate for each 50-m elevation bin and nz is the number of independent measurements when accounting for spatial autocorrelation (see Materials and Methods).

We multiply geodetic mass balances by the full glacierized area in the Himalayas between 75° and 93° longitude to estimate region-wide ice mass changes of −7.5 ± 2.3 Gt year−1 during 2000–2016, compared to −3.9 ± 2.2 Gt year−1 during 1975–2000 (−5.2 ± 2.2 Gt year−1 during the full 1975–2016 interval). Recent models using Shuttle Radar Topography Mission (SRTM) elevation data for ice thickness inversion estimate the total glacial ice mass in our region of study to be approximately 700 Gt in the year 2000 (see Materials and Methods) (1528). If this estimate is accurate, our observed annual mass losses suggest that of the total ice mass present in 1975, about 87% remained in 2000 and 72% remained in 2016.

Comparison of clean-ice, debris-covered, and lake-terminating glaciers

We study mass changes for different glacier types by separating glaciers into clean-ice (<33% area covered by debris), debris-covered (≥33% area covered by debris), and lake-terminating categories based on a Landsat band ratio threshold and manual delineation of proglacial lakes (see Materials and Methods). All three categories have undergone a similar acceleration of ice loss (Table 1), and debris-covered glaciers exhibit similar and often more negative geodetic mass balances compared to clean-ice glaciers over the past 40 years (Fig. 3). Altitudinal distributions indicate slower thinning for lower-elevation regions of debris-covered glaciers (glacier tongues where debris is most concentrated) relative to clean-ice glaciers, but comparatively faster thinning in mid- to upper elevations (Fig. 4). Lake-terminating glaciers concentrated in the eastern Himalayas exhibit the most negative mass balances due to thermal undercutting and calving (31), though they only comprise around 5 to 6% of the estimated total Himalaya-wide mass loss during both intervals.Table 1Himalaya-wide geodetic mass balances (m w.e. year−1).View this table:

Fig. 3Comparison between clean-ice (<33% debris-covered area) and debris-covered (≥33% debris-covered area) glaciers for seven subregions.Circle sizes are proportional to glacier areas, colors delineate clean-ice versus debris-covered categories, and boxplots indicate geodetic mass balance (m w.e. year−1). Box center marks (red lines) are medians; box bottom and top edges indicate the 25th and 75th percentiles, respectively; whiskers indicate q75 + 1.5 ⋅ (q75 − q25) and q25 − 1.5 ⋅ (q75 − q25), where subscripts indicate percentiles and “+” symbols are outliers.
Fig. 4Altitudinal distributions of ice thickness change (m year−1) for the 650 glaciers.Glaciers are separated by time interval (top) and category (<33% versus ≥33% debris-covered area) (bottom). (A) Altitudinal distributions of ice thickness change for clean-ice glaciers during 1975–2000 and 2000–2016. The y axes are normalized elevation as in Fig. 2. (B) Same as (A), but for debris-covered glaciers. (C) Altitudinal distributions of ice thickness change during 1975–2000 for clean-ice and debris-covered glaciers. (D) Same as (C), but for 2000–2016. (E) Altitudinal distributions of glacierized area for both glacier categories. Elevational extent of debris cover varies widely between individual glaciers, but is mostly concentrated in lower ablation zones. The clean-ice category includes 478 glaciers and the debris-covered category includes 124 glaciers.

Approximation of required temperature change

As a first approximation of the consistency between observed glacier mass balances and available temperature records, we estimate the energy required to melt the observed ice losses and conservatively estimate the atmospheric temperature change that would supply this energy via longwave radiation to the glaciers, using a simple energy balance approach (Materials and Methods). We propagate significant uncertainties associated with input from global climate reanalysis data, scaling of temperatures from coarse reanalysis grids to specific glacier elevations, and averaging of climate data over the glacierized region. Results suggest that the observed acceleration of ice loss can be explained by an average temperature ranging from 0.4° to 1.4°C warmer during 2000–2016, relative to the 1975–2000 average. This approximately agrees with the magnitude of warming observed by meteorological stations located throughout HMA, which have recorded air temperatures around 1°C warmer on average during 2000–2016, relative to 1975–2000 (Fig. 5). More comprehensive climate observations and models will be essential for further investigation, but these simple energy constraints suggest that the acceleration of mass loss in the Himalayas is consistent with warming temperatures recorded by meteorological stations in the region.

Fig. 5Compilation of previously published instrumental temperature records in HMA.(A) Regional temperature anomalies, relative to the 1980–2009 mean temperatures for each record. The yellow trend (23) from the quality-controlled and homogenized climate datasets LSAT-V1.1 and CGP1.0 recently developed by the China Meteorological Administration (CMA), using approximately 94 meteorological stations located throughout the Hindu Kush Himalayan region. The orange trend (44) is from a similar CMA dataset derived from 81 stations more concentrated on the eastern Tibetan Plateau. The blue trend (24) is from three decades of temperature data from 13 mountain stations located on the southern slopes of the central Himalayas. The black trend is the 5-year moving mean. (B) Temperature anomalies from high-elevation stations at the Chhota Shigri glacier terminus (25); Dingri station in the Everest region (26); average from the Kanzalwan, Drass, and Patseo stations (45); and average of 16 stations above 4000 m elevation on the Tibetan Plateau and eastern Himalayas (46). Here, temperature anomalies are relative to the mean of each record. The gray trend line is the 5-year moving mean. (C) Difference in mean temperature (°C) between the two intervals, i.e., the mean of the 2000–2016 interval relative to the mean of the 1975–2000 interval.


Implications for dominant drivers of glacier change in the Himalayas

The parsing of Himalayan glacier energy budgets is not a straightforward task owing to the scarcity of meteorological data, in combination with the complex climate and topography of the region (2). Furthermore, the Himalayas border hot spots of high anthropogenic BC emissions, which may affect glaciers by direct heating of the atmosphere and decreasing albedo of ice and snow after deposition (14). While improved analyses combining observations and high-resolution atmospheric and glacier energy balance models will be required to quantify these effects, the pattern of ice loss we observe has important implications regarding dominant climate influences on Himalayan glacier mass balances. Our results suggest that any drivers of glacier change must explain the region-wide consistency, the doubling of the average rate of ice loss in the 21st century compared to 1975–2000, and the observation that clean-ice, debris-covered, and lake-terminating glaciers have all experienced a similar acceleration of mass loss.

Some studies have suggested a weakening of the summer monsoon and reduced precipitation as primary reasons for negative glacier mass balances, particularly in the Everest region (16). While decreasing accumulation rates may account for a significant portion of the mass balance signal for some glaciers, an extreme Himalaya-wide decrease in precipitation would be required to explain the extensive ice losses we observe, especially given that monsoon-dominated glaciers with high accumulation rates are known to be much more sensitive to temperature than accumulation changes (532). Regional studies of precipitation trends in the Himalayas do not suggest a widespread decrease in precipitation over the past four decades (Supplementary Materials). A uniform BC albedo forcing across the Himalayas is another possible explanation, although BC concentrations measured in snow and ice in the Himalayas have been found to be spatially heterogeneous (1433), and high-resolution atmospheric models also show large spatial variability of deposited BC originating from localized emissions in regions of complex terrain (1434). Future analyses focused on quantifying the spatial patterns of BC deposition will reveal further insights, yet given the rather homogeneous pattern of mass loss we observe across the 2000-km Himalayan transect, a strong, spatially heterogeneous mechanism seems improbable as a dominant driver of glacier ice loss in the region.

Debris-covered glaciers

Similar thinning rates of debris-covered (thermally insulated) glaciers relative to clean-ice glaciers have been observed by previous studies and have been not only ascribed to surface melt ponds and associated ice cliffs acting as localized hot spots to concentrate melting but also attributed to declining ice flux causing dynamic thinning and stagnation of debris-covered glacier tongues (2). While we cannot yet directly deconvolve relative contributions from these processes, we find that average thinning rates for debris-covered glaciers are slower than clean-ice glaciers at low elevations (glacier tongues where debris is most concentrated), which agrees with reduced melt rates from field studies. In turn, debris-covered glaciers tend to have comparatively faster thinning at mid-range elevations, where debris cover is sparser and also where the majority of total glacierized area resides (Fig. 4). Models of debris-covered glacier processes suggest that this pattern of thinning may cause a reduction in down-glacier surface gradient, which, in turn, reduces driving stress and ice flux and explains why debris-covered ablation zones become stagnant (35). We also find that clean-ice glaciers exhibit a much more pronounced steepening of the thinning profile over time, compared to debris-covered glaciers. It may be that both glacier types experience a uniform thinning phase followed by a changing terminus flux and retreat phase, but the clean-ice glaciers are in a later phase of response to recent climate change (36).

Comparison with previous studies in the Himalayas

To compare our results with previous remote sensing studies that derive mass changes from various sensors (including Hexagon, SRTM, SPOT5, ICESat, and ASTER), we restrict our results to the approximate geographical regions covered by each corresponding study (12131722) and then compute area-weighted average geodetic mass balances. In addition, we compare individual glacier mass balances for the Everest and Langtang Himal regions, where mass changes were previously estimated using declassified Corona and Hexagon imagery (131920). Despite factors such as variable spatial resolutions, distinct void-filling methods, heterogeneous spatial and temporal coverages, and different definitions of glacier boundaries, we find that our average mass balances generally agree with previous analyses and overlap within uncertainties (table S1). However, because of the significant variability of individual glacier mass changes within subregions, our results also highlight the importance of sampling a large number of glaciers to obtain a robust average trend for any given area.

Comparison with benchmark mid-latitude glaciers and global average

To gain perspective on mass losses from these low-latitude glaciers in the monsoonal Himalayas, we compare our results with benchmark mid-latitude glaciers in the European Alps, as well as with a global average mass balance trend (fig. S1) (37). The Alps contain the most detailed long-term glaciological and high-elevation meteorological records on Earth, and the climatic sensitivity and behavior of these European glaciers are well understood compared to glaciers in HMA. Air temperatures in the Alps show an abrupt warming trend beginning in the mid-1980s, and Alpine mass balance records display a concurrent acceleration of ice loss, with a continual strongly negative mass balance after that time. Himalayan weather station data indicate a more gradual warming trend, with the strongest warming beginning in the mid-1990s (fig. S1, A and B). We find that mass balance in the Himalayas is less negative compared to the Alps and the global average, despite close proximity to a known hot spot of increasing BC emissions with rapid growth and accompanying combustion of fossil fuels and biomass in South Asia (38). The concurrent acceleration of ice loss observed in both the Himalayas and Europe over the past 40 years coincides with a distinct warming trend beginning in the latter part of the 20th century, followed by the consistently warmest temperatures through the 21st century in both regions.


Our analysis robustly quantifies four decades of ice loss for 650 of the largest glaciers across a 2000-km transect in the Himalayas. We find similar mass loss rates across subregions and a doubling of the average rate of loss during 2000–2016 relative to the 1975–2000 interval. This is consistent with the available multidecade weather station records scattered throughout HMA, which indicate quasi-steady mean annual air temperatures through the 1960s to the 1980s with a prominent warming trend beginning in the mid-1990s and continuing into the 21st century (2326). We suggest that degree-day and energy balance models focused on accurately quantifying glacier responses to air temperature changes (including energy fluxes and associated feedbacks) will provide the most robust estimates of glacier response to future climate scenarios in the Himalayas.



U.S. intelligence agencies used KH-9 Hexagon military satellites for reconnaissance from 1973 to 1980. A telescopic camera system acquired thousands of photographs worldwide, after which film recovery capsules were ejected from the satellites and parachuted back to Earth over the Pacific Ocean. With a ground resolution ranging from 6 to 9 m, single scenes from the mapping camera cover an area of approximately 30,000 km2 with overlap of 55 to 70%, allowing for stereo photogrammetric processing of large regions. Images were scanned by the U.S. Geological Survey (USGS) at a resolution of 7 μm and downloaded via the Earth Explorer user interface ( Digital elevation models were extracted using the Hexagon Imagery Automated Pipeline methodology, which is coded in MATLAB and uses the OpenCV library for Oriented FAST and Rotated BRIEF (ORB) feature matching, uncalibrated stereo rectification, and semiglobal block matching algorithms (27). The majority of the KH-9 images here were acquired within a 3-year interval (1973–1976), and we processed a total of 42 images to provide sufficient spatial coverage (fig. S2).


The ASTER instrument was launched as part of a cooperative effort between NASA and Japan’s Ministry of Economy, Trade and Industry in 1999. Its nadir and backward-viewing telescopes provide stereoscopic capability at 15-m ground resolution, and a single DEM covers approximately 3600 km2. Approximately 26,000 ASTER DEMs were downloaded via the METI AIST Data Archive System (MADAS) satellite data retrieval system (, a portal maintained by the Japanese National Institute of Advanced Industrial Science and Technology and the Geological Survey of Japan. To use all cloud-free pixels (including images with a high percentage of cloud cover), no cloud selection criteria were applied when downloading the images. We used the Data1.l3a.demzs geotiff product, which has a spatial resolution of 30 m. After downloading, the DEMs were subjected to a cleanup process: For a given scene, any saturated pixels (i.e., equal to 0 or 255) in the nadir band 3 (0.76 to 0.86 μm) image were masked in the DEM. Next, the SRTM dataset was used to remove any DEM values with an absolute elevation difference larger than 150 m (relative to SRTM), which effectively eliminated the majority of errors caused by clouds. While more sophisticated cloud masking procedures are possible, the ASTER shortwave infrared detectors failed in April 2008, making cloud detection after this time impossible using standard methods. We examined existing cloud masks derived using Moderate Resolution Imaging Spectroradiometer images as another option ( However, these are not optimized for snow-covered regions and often misclassify glacier pixels as clouds. Instead, our large collection of multitemporal ASTER scenes, the SRTM difference threshold, and our robust linear trend fitting algorithm [see description of Random Sample Consensus (RANSAC) in the “Trend fitting of multitemporal DEM stacks” section] effectively excluded any remaining erroneous cloud elevations after the initial threshold. As a final step, all ASTER DEMs were coregistered to the SRTM using a standard elevation–aspect optimization procedure (39). We did not apply fifth-order polynomial correction procedures to the ASTER DEMs for satellite “jitter” effects and curvature bias as done in some previous studies (18). We found that while these types of corrections may reduce the overall average elevation error in a scene, the polynomial fitting can be unwieldy and may introduce unwanted localized biases. By stacking many ASTER DEMs (with 20.5 being the average number of observations per pixel stack during the ASTER trend fitting, see fig. S3E) and using a robust fitting procedure, we found that biases do not correlate across overlapping scenes, and thus tend to cancel out one another. Furthermore, the elevation change results from this portion of our study overlap within uncertainties with Brun et al. (18) (Supplementary Materials) who did perform polynomial corrections. This suggests that for a large-scale regional study using a high number of overlapping ASTER scenes, the satellite jitter and curvature bias corrections have a relatively minimal impact on the final results.

Glacier polygons

To delineate glaciers during all portions of the analysis, we used manually refined versions of the Randolph Glacier Inventory (RGI 5.0) (40). Starting with the original RGI dataset, we edited the glacier polygons to reflect glacier areas during 1975, 2000, and 2016. For the 1975 edit, we used a combination of Hexagon imagery, the Global Land Survey (GLS) Landsat Multispectral Scanner mosaic (GLS1975), and glacier thickness change maps derived from differencing the Hexagon and modern ASTER DEMs, which are particularly useful for debris-covered glacier termini that often have spectral characteristics indistinguishable from surrounding terrain. Debris-covered areas for each glacier were delineated using a Landsat DN TM4/TM5 band ratio with a threshold of 2.0, and glaciers with ≥33% debris cover were assigned to the debris-covered category. For the 2000 edit, we used the GLS2000 Landsat Enhanced Thematic Mapper Plus mosaic, along with glacier thickness change maps derived from differencing ASTER DEMs. For the 2016 edit, we used a custom mosaic of Landsat 8 imagery with acquisition dates spanning 2014–2016. The individually edited glacier polygons were used for all ice volume change and geodetic mass balance computations. The polygons were also used during alignment of the DEMs, where the shapefiles were converted to raster masks with a dilation (morphological operation) of 250 m on the binary rasters. This effectively excluded unstable terrain surrounding the glaciers during the DEM alignment process, such as steep avalanching slopes and unstable moraines.

Trend fitting of multitemporal DEM stacks

Glacier polygons were processed individually—all DEMs from a given time interval (1975–2000 or 2000–2016) that overlap a polygon were selected and resampled to the same 30-m resolution using linear interpolation. The overlapping DEMs were sampled with a 25% extension around each glacier to include nearby stable terrain for alignment and uncertainty analysis (fig. S4). After ensuring that there is no vertical bias, the aligned DEMs were sorted in temporal order as a three-dimensional matrix, and linear trends were fit to every pixel “stack” (i.e., along the third dimension of the matrix) using the RANSAC method. During each RANSAC iteration, a random set of two elevation pixels per stack were selected. A linear trend was fit to these two values, and then all remaining elevation pixels were compared to the trend. Any elevation pixels within 15 m of the trend line were marked as inliers. This process was repeated for 100 iterations, and the iteration with the greatest number of inliers was selected. A final linear fit was performed using all inliers from the best iteration, and this trend was used for each pixel stack’s thickness change estimate. The thickness change maps were subjected to outlier removal using thresholds for maximum slope, maximum thickness change, minimum count per pixel stack, minimum timespan per pixel stack, maximum SD of inlier elevations per pixel stack, and maximum gradient of the thickness change map (fig. S3). In addition, the thickness change pixels were separated into 50-m elevation bins, and pixels falling outside the 2 to 98% quantile range were excluded. Any bins with less than 100 pixels were removed and then interpolated using the two adjacent bins. Before computing ice volume change for the glaciers, the final thickness change maps were visually inspected, any remaining erroneous pixels (which occurred almost exclusively in low-contrast, snow-covered accumulation zones) were manually masked, and a 5 × 5 pixel median filter was applied. We did not attempt to perform seasonality corrections, as no seasonal snowfall records are available and because nearly all the Hexagon DEMs were acquired during winter, thus minimizing any seasonality offsets between regions. For the 1975–2000 interval, we used the Hexagon DEMs and sampled ASTER thickness change trends at the start of the year 2000. For regions with multiple overlapping Hexagon DEMs, we used the same RANSAC method. During the 1975–2000 interval, only two DEMs were available for most glaciers. In these cases, the RANSAC iterations were unnecessary, and we simply differenced the two available DEMs. We did not use SRTM for any thickness change estimates; thus, no correction for radar penetration was necessary.

Mass changes

To compute (mean annual) ice volume changes for individual glaciers, all thickness change pixels falling within a glacier polygon were transformed to an appropriate projected WGS84 UTM coordinate system (zones 43 to 46, depending on longitude of the glacier). Pixel values (m year−1) were then multiplied by their corresponding areas (pixel width × pixel height) and summed together. The resulting ice volume change was then divided by the average glacier area to obtain a glacier thickness change. We used the average of the initial and final glacier areas for a given time interval and excluded slopes greater than 45° to remove any cliffs and nunataks. Last, the glacier thickness change was multiplied by an average ice-firn density (41) of 850 kg m−3 and then divided by the density of water (1000 kg m−3) to compute glacier geodetic mass balance in m w.e. year−1. Because of cloud cover, shadows, and low radiometric contrast, some glacier accumulation zones had gaps in the DEMs and resulting thickness change maps. This is particularly evident in the Hexagon DEMs for the Spiti Lahaul region owing to extensive cloud cover. To fill these gaps, we tested two different void-filling methods for comparison. In the first method, we defined a circular search area with a radius of 50 km around the center of a given glacier. All thickness change pixels from glaciers in this surrounding area were binned (into 50-m elevation bins, and following the same outlier-removal procedure given in the preceding section), and any missing data in the glacier were set to this “regional bin” mean value at the corresponding elevation. In the second method, we filled data gaps using an interpolation procedure, where voids in an individual glacier were linearly interpolated using bin values at upper and lower elevations relative to the missing data (those belonging to the same glacier), and assumed zero change at the highest elevation bin (headwall). Both methods yielded similar results (table S1). In addition, no obvious trends were apparent when geodetic mass balance was plotted versus percent data coverage or glacier size (fig. S5). While smaller glaciers exhibited more scatter, the average mass balance was similar for all glacier sizes. These observations indicate that our representative sample of glaciers, while biased toward larger glaciers, adequately captures the statistical distribution of glacier mass balances in the Himalayas.

To calculate regional geodetic mass balances, we separated glaciers into four subregions (Spiti Lahaul, West Nepal, East Nepal, and Bhutan) as defined by Brun et al. (18). We then calculated the average mass balance for each of these four subregions, weighted by individual glacier areas. Last, we calculated a final average mass balance for the Himalayas, weighted by the total glacierized area (from the RGI 5.0 database) in each of the four subregions, between 75° to 93° longitude. Because of the relatively homogeneous mass balance distribution, we found that this approach resulted in similar values (well within the uncertainties) compared to simply calculating the area-weighted average mass balance of the 650 measured glaciers. To obtain the total mass changes in Gt year−1, we multiplied each subregion mass balance by its total glacierized area and then summed the results from all subregions to get Himalaya-wide totals of −3.9 Gt year−1 for 1975–2000 and −7.5 Gt year−1 for 2000–2016. To calculate contributions to sea-level rise, we used a global ocean surface area of 361.9 × 106 km2 (fig. S4G).

To estimate the total ice mass present in our region of study, we used ice thickness estimates from Kraaijenbrink et al. (15), who used the Glacier bed Topography version 2 model to invert for ice thickness (28) with input from the SRTM DEM (acquired in February of 2000). The ice thickness estimates from (15) did not include glaciers smaller than 0.4 km2, and to estimate the additional mass contribution from these smallest glaciers (along with any other glaciers that are missing thickness estimates), we fit a second-order polynomial to the logarithm of glacier volumes versus the logarithm of glacier areas and evaluated this fit equation for any glaciers without volume data (fig. S6). We then converted glacier volume to mass using a density value of 850 kg m−3. Over our region of study, the ice volumes from the thickness data amounted to 649 Gt, with an additional contribution of 35 Gt from the fitting procedure, for a total of 684 Gt.

Uncertainty assessment

We quantified statistical uncertainty for individual glaciers using an iterative random sampling approach. For a given glacier, the SD of elevation changes from the surrounding stable terrain (σz) was first calculated. For any missing thickness change pixels within the glacier polygon, we also included an extrapolation uncertainty σe. This accounts for additional error in regions with incomplete data, i.e., those glacier regions filled by extrapolating thickness changes from surrounding glaciers, or linear interpolation assuming zero change at the headwall, as described in the previous section. We found that in the Himalaya-wide altitudinal distributions, the maximum SD of thickness change in any 50-m elevation bin above 5000 m is 0.56 m year−1. Nearly all regions with incomplete data coverage are above this elevation, resulting from poor radiometric contrast for snow-covered glacier accumulation zones. We thus conservatively set σe equal to 0.6 m year−1. We then combined both sources of error to get σp for every individual thickness change pixelσp=σ2z+σ2e−−−−−−√(1)

To account for spatial autocorrelation, we started with a normally distributed random error field (with a mean of 0 and an SD of 1) the same size as the thickness change map and then filtered it using an n-by-n moving window average to add spatial correlation, where n is defined as the spatial correlation range divided by the spatial resolution of the thickness change map. We used 500 m for the spatial correlation range, which is a conservative value based on semivariogram analysis in mountainous regions from previous studies (182142). The resulting artificial error field En (now with spatial correlation) is scaled by the σp values and added to the thickness change map ΔH as follows, where (xy) are pixel coordinatesΔHE(x,y)=ΔH(x,y)+En(x,y)⋅σp(x,y)σn(2)

If thickness change data exist at a given pixel location (xy) on the glacier, σn is the SD of the set of all En values where data exist (i.e., where σe is equal to zero). Conversely, if thickness change data do not exist at a given pixel location (xy) on the glacier, σn is the SD of the set of all En values where data do not exist (i.e., where σe is equal to 0.6 m year−1). In this way, the second term of Eq. 2 assigns larger uncertainties to regions with incomplete data. Last, all glacier thickness change pixels in ΔHE were summed together to compute a volume change with the introduced error. This procedure was repeated for 100 iterations, and the volume change uncertainty σΔV was computed as the SD of the resulting distribution (fig. S4). For region-wide volume change estimates, we conservatively assumed total correlation between glaciers and computed region-wide uncertainty as follows, where g is the total number of glaciers (~17,400)σΔV region=∑1gσΔV(3)

For glaciers where thickness change data are not available, a measure of uncertainty is still required to factor into the final regional uncertainty estimate. For these glaciers, we estimated σΔVas (42)σΔV=σ2z region⋅Acor5⋅A−−−−−−−−−−−√(4)Acor=π⋅L2(5)

In this case, σz region is the region-wide SD of elevation change over stable terrain (0.42 m year−1) (fig. S7), Acor is the correlation area, L is the correlation range (500 m), and A is the glacier area. Last, all σΔV and σΔV regjon estimates were combined with an area uncertainty (43) of 10% and a density uncertainty (41) of 60 kg m−3 using standard uncorrelated error propagation.

Sensitivity of region-wide glacier mass change estimates

We further tested the sensitivity of our region-wide estimates to potential biases, including (i) the exclusion of small glaciers, (ii) incomplete data coverage for many glacier accumulation zones during 1975–2000, and (iii) void-filling technique. First, we note that our geodetic mass balance analysis only includes glaciers larger than 3 km2. This is because mass balance uncertainties increase with decreasing glacier size, and we find that uncertainties often exceed the magnitude of mass changes for glaciers smaller than ~3 km2. To test whether the neglected small glaciers appreciably affect the result, we also computed mass balances using all available glaciers (i.e., all glaciers with ≥33% data coverage, including those smaller than 3 km2). We find that including the full set of smaller glaciers changes the region-wide geodetic mass balance estimates by a maximum of 0.04 m w.e. year−1 (fig. S4G). Next, we note that the Hexagon DEMs in particular have poor data coverage over glacier accumulation zones (figs. S8 and S9). However, the vast majority of thinning occurs in glacier ablation zones, and the amount of thinning decreases with elevation in a quasi-linear fashion, especially in mid- to upper regions of the glaciers where data gaps are most common. Thus, we hypothesize that we can extrapolate and interpolate with reasonable confidence over accumulation areas. To test the robustness of this assumption, we used the 2000–2016 glacier change data. The ASTER data over this interval have superior radiometric contrast and adequately capture elevation change trends for most accumulation zones. We first set all 2000–2016 thickness change pixels to be empty where the 1975–2000 data are missing to simulate the same data gaps over accumulation zones as in the 1975–2000 data. We then performed the same geodetic mass balance calculations and found that the region-wide geodetic mass balance only changes by 0.01 m w.e. year−1 (fig. S4G, comparing test 3 to test 1). Last, we performed two separate void-filling methods for all tests (see the “Mass changes” section for descriptions of void-filling methods) and observed a maximum change in geodetic mass balance of 0.04 m w.e. year−1. Overall, the relatively small impact of each test suggests that our results are robust to the exclusion of small glaciers, incomplete data coverage over glacier accumulation zones, and void-filling technique.


Supplementary material for this article is available at

Fig. S1. Comparison of Himalayan temperature trends and regional mass balance with benchmark mid-latitude glaciers and a global average trend.

Fig. S2. Coverage of glacierized area in the Himalayas.

Fig. S3. Trend fit examples for two large glaciers using ASTER DEMs during 2000–2016, histograms of ASTER pixel counts and timespans per stack (glacier averages), and outlier thresholds.

Fig. S4. Illustration of uncertainty estimation procedure for a single iteration/glacier and Himalaya-wide sensitivity tests.

Fig. S5. Geodetic mass balances during 1975–2000 and 2000–2016 plotted against various parameters.

Fig. S6. Log-log plot of glacier volumes versus areas used to estimate the total ice mass present in our region of study.

Fig. S7. Analysis of elevation change for nonglacier pixels (stable terrain) during both intervals.

Fig. S8. Thickness change maps used in the analysis.

Fig. S9. Thickness change maps for the three remaining Himalayan regions.

Table S1. Geodetic mass balance comparisons with prior studies.

References (4770)

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.

  1. . M. Maurer1,2,*
  2. J. M. Schaefer1,2
  3. S. Rupper3 and 
  4. A. Corley1

Students urged for active role in nature conservation


“The idea behind celebration of World Earth Day was to create awareness about the need for preserving and renewing the threatened ecological balances upon which all life on Earth depends. However, that idea remained confined to a handful of population around the globe. As a result of which an average of 60% of the population of fish, birds, mammals, amphibians and reptiles have dwindled in between the celebration of the first World Earth Day in 1970 to 2014,” said Mubina Akhtar, noted journalist and wildlife activist, while speaking on this year’s World Earth Day theme “Protecting the Species” at the NKC Auditorium, University of Science and Technology, Meghalaya, organized by Earth Science Department.  “An incalculable number of species have disappeared forever, and a large number are threatened with extinction. In Assam, 14 species–comprising seven plant and seven animal species–have now become threatened that includes the siya nahar, lady’s slipper orchid, Cycas and animal species like Assamese day geck, tokay gecko, black soft-shelled turtle etc.  In almost all cases, the threats to wildlife can be traced to human activities. Habitat destruction is the main cause for wildlife extinction in India” she said and added “When the natural habitat of animals is destroyed, it leads to a decline in their primary food supply. With loss of breeding and nesting grounds, their numbers get drastically reduced. In the case of plants, if their natural habitat is destroyed, then their survival is at risk.” 

“We depend on various plant and animal species for livelihood support but unsustainable use has led to rapid decline in those species. We tend to forget that ‘each species is unique and has been created as a consequence of evolutionary process. Therefore every species has a natural right to exist.’

Akhtar appealed to the students for a more active and defined role in creating awareness for preservation of natural ecosystems as she released the 6th edition of the Department’s Wall Magazine dedicated to this year’s World Earth Day theme “Protecting the Species” on the occasion. 

The programme started with welcome speech by Dr. E. Al Huda Head of the Department of Earth Science. 

The Vice Chancellor, USTM, DR. P.K. Goswami in his speech also referred to human interventions that led to imbalances in nature and the urgency need to preserving the same. Dr. R.K. Sharma, Adviser to USTM detailed on the history of events that led to the celebration of World Earth Day.

Earlier, Nitu T. Upadhya, Assistant Professor, felicitated the guests. Dr. Lalit Saikia, Assistant Professor, Department of Earth Science offered the vote of thanks.


Climate change threatens one in three Bangladeshi children

Rohingya refugee man with child returns to his shelter in Kutupalong camp during heavy rain in Cox’s Bazar, 

REUTERS/Mohammad Ponir Hossain

Worsening storms, sea level rise and other threats could drive worsening poverty, hunger, early marriage and child labour

Nearly one in three children in Bangladesh are at risk from cyclones, flooding and other climate change-linked disasters, the United Nations warned on Friday, urging more help for one of the world’s worst hit countries.

More than 19 million children live in the most disaster-prone districts of low-lying Bangladesh, according to a new report from the U.N. children’s agency UNICEF.

In addition, longer-term changes such as rising sea levels are pushing families deeper into poverty and forcing some from their homes, disrupting children’s education and access to health services, UNICEF said.

“Children who miss out on good nutrition or on education, who are uprooted from their homes, or who are forced into exploitative labour, will fail to fulfil their potential as citizens,” said the author of the report, Simon Ingram.

The call comes weeks after schoolchildren around the world walked out of classes to protest against global government inaction on climate change.

Global temperatures are on course to rise by 3 degrees to 5 degrees Celsius (5.4 degrees to 9 degrees Fahrenheit) this century, far overshooting a global target of limiting the increase to 2C or less, the U.N. World Meteorological Organization says.

That is bringing growing risks from extreme weather – including worsening droughts, floods, fires and storms – as well as threats of worsening hunger, poverty and water shortages, scientists say.

Bangladesh ranked ninth in the Global Climate Risk Index 2019, which said it was the seventh worst hit by climate change between 1998 and 2017, with 37 million people affected.

UNICEF said Bangladesh had already done much to reduce the exposure of poorer communities to cyclones and other threats, notably through the construction of shelters.

But it called for more focus on the specific needs of children threatened by the effects of climate change, including food shortages and increased migration to cities as flooding and drought make some rural areas uninhabitable.

That should include making schools and health facilities in flood-prone areas more resilient and introducing stronger measures to protect children affected by climate-induced disasters against exploitation and abuse, said Ingram.

Nurul Qadir, a senior official at Bangladesh’s Ministry of Environment, Forest and Climate Change, told the Thomson Reuters Foundation the government was already addressing the issues raised in the report.

“Right now, we are going to schools across the country to make children aware about climate change and how it can be tackled,” he said.

The UNICEF study found 12 million children in Bangladesh live near rivers that regularly burst their banks. Another 4.5 million live in coastal areas vulnerable to cyclones and 3 million are at risk from drought, it said.

These risk factors are forcing people from rural areas into cities, where children are at greater risk of being pushed into forced labour or early marriage.

“They face danger and deprivation in the cities, as well as pressure to go out to work despite the risk of exploitation and abuse,” said UNICEF Bangladesh representative Edouard Beigbeder.

Climate change threatens one in three Bangladeshi children

(Source: Thomson Reuters Foundation)


Widespread losses of pollinating insects revealed across Britain

A widespread loss of pollinating insects in recent decades has been revealed by the first national survey in Britain, which scientists say “highlights a fundamental deterioration” in nature.

The analysis of 353 wild bee and hoverfly species found the insects have been lost from a quarter of the places they were found in 1980. A third of the species now occupy smaller ranges, with just one in 10 expanding their extent, and the average number of species found in a square kilometre fell by 11.

UK pollinating insects survey: losers and winners – in pictures

A small group of 22 bee species known to be important in pollinating crops such as oilseed rape saw a rise in range, potentially due to farmers increasingly planting wild flowers around fields. However, the scientists found “severe” declines in other bee species from 2007, coinciding with the introduction of a widely used neonicotinoid insecticide, which has since been banned.

Researchers have become increasingly concerned about dramatic drops in populations of insects, which underpin much of nature. Some warned in February that these falls threaten a “catastrophic collapse of nature’s ecosystems”, while studies from Germany and Puerto Rico have shown plunging numbers in the last 25 to 35 years.Advertisement

The study, published in the journal Nature Communications, is based on more than 700,000 sightings made by volunteers across Britain from 1980 to 2013. These are used to map the range of each species of bee and hoverfly over time. The data did not allow the assessment of numbers of insects, but some researchers think populations have fallen faster than range.

Pollinating insects are vital to human food security, as three-quarters of crops depend on them. They are also crucial to other wildlife, both as food and as pollinators of wild plants. “The declines in Britain can be viewed as a warning about the health of our countryside,” said Gary Powney at the Centre for Ecology and Hydrology in Wallingford, who led the research.

He called for more volunteers to take part in the UK Pollinator Monitoring Scheme: “Their contribution is vital for us to understand what is happening in our landscape.” Another recent study found that allotments, weedy corners and fancy gardens can all be urban havens for bees.

The biggest factor in the decline in pollinators is likely to be the destruction of wild habitats and use of pesticides as farming has intensified. But the analysis also revealed a particularly big drop of 55% in the range of upland bee and hoverfly species, and significant falls in northern Britain, which may result from climate change making conditions too warm.

Among the bees whose range has shrunk are the formerly widespread red-shanked carder bee, whose extent fell by 42%, and the large shaggy bee, whose range fell 53%. But the lobe-spurred furrow bee, which was once rare, has expanded its range fivefold and is now considered an important crop pollinator in England.

Powney said the increased range of the bees most commonly pollinating crops is good news and might be a result of more oilseed rape being grown, as well as wildflower margins being planted. But he also warned: “They are a relatively small group of species. Therefore, with species having declined overall, it would be risky to rely on this group to support the long-term food security for our country. If anything happens to them in the future there will be fewer other species to ‘step up’.”

Prof Dave Goulson, at the University of Sussex and not part of the latest research, said: Previous studies have described declines in UK butterflies, moths, carabid beetles, bees and hoverflies – this new study confirms that declines in insects are ongoing.”

If the losses of upland and northern species are due to climate change, “then we can expect far more rapid declines of these species in the future, as climate change has barely got started”, he said. Goulson also said the start of more rapid declines in southern bees after 2007 coincided with the first use of now-banned neonicotinoid pesticides.

Roy van Grunsven, at the Dutch Butterfly Conservation project, said the decline in numbers of insects was very likely to be a lot higher than the shrinking of their range: “Going from flowery meadows full of bees to intensive agriculture with a few individuals in a road verge does not result in a change in distribution, but of course is a huge change in [numbers].”

Matt Shardlow, of the conservation charity Buglife, said unless the pesticide approval process was improved to help bee safety and green subsidies were targeted to create corridors that connect wild spaces, we can expect the declines to continue or worsen.

by Damian Carrington, Environment editor, The Guardian


Rainy week for Assam, Arunachal, Meghalaya, Nagaland, Manipur, Mizoram and Tripura

The northeastern states of India that is Assam, Arunachal Pradesh, Meghalaya, Nagaland, Manipur, Mizoram and Tripura are currently rain deficient. Although on and off rains were going on over parts of Northeast India, particularly over Assam and Arunachal Pradesh, but the intensity and spread remained on the lower side over most of the states. Thus, it failed to bring down the rain deficiency.

According to Skymet Weather, since the last few days, an Anti-Cyclone has been persisting over North Bay of Bengal, due to which moisture feed remained restricted over Northeast India.

Now a Cyclonic Circulation has formed over Assam and adjoining area. Along with this, the Anti-Cyclone is also moving away. Therefore, the moisture feed will increase over Northeast India in the form of southwesterly humid winds.

As a result, rain and thundershowers will now increase over all the states of Northeast India and these on and off weather activities will continue for at least one week. Isolated heavy spells also cannot be ruled out during this time.

We expect lightning strikes accompanied with hailstorm activity in few districts of Meghalaya, Nagaland, Manipur, Mizoram and Tripura. Thus, we can say that a rainy week is ahead for Northeast India. In the wake of these rains, by the end of March, we expect rain deficiency of the northeastern states to reduce to some extent.

Indigenous no-state people

Climate change is not killing Muga, but tea pesticides

No action has been taken against tea growers who have been responsible for death of lakhs of Muga worms last month at Dhakuakhana in Lakhimpur district.

After massive loss in Muga rearing, the farmers of Dhenukhona-Hiloidhari village in Dhakuakhana sub-division have filed case against a number of tea gardens.

Rampant uses of pesticides in nearby tea gardens killed Muga worms while they were feeding on Som trees last month. Muga is a unique and exclusive golden silk textiles produced in Assam.

Muga larva and worm rearing is outdoor activity and polluted air affects it easily.

Muga is very sensitive to the odour of toxic chemicals, temperature and humidity.

The phenomenon has been taking place in the past years also. But Muga farmers protested rampant uses of pesticides last month.

Affected Muga farmers have filed case against Duleshwar Gogoi, Pranab Gogoi, Min Gogoi, Hemanta Gogoi, Ruhiteshwar Gogoi and Nitya Gogoi, who are tea garden owners.

Ghilamora Police has already processed the inquiry, but no effective action has been taken so far.

After fall and finally death of lakhs of worms, the farmers lodged a joint FIR in Ghilamara Police Station. Police had registered the case.

Dead Muga worms. Photo credit – Amit Paban Bora

But police claim that they have no suitable law or act to take action against tea planters. The farmers of Dhenukhona have a cultivation of Som trees covering an area of thousand hectares and Muga has been a traditional livelihood and occupation for many local people.

It has been mentioned in the FIR that around 5 million worm feed on Som trees have died. Muga farmers brought seed worth Rs 1 million and it could have produced silk of hundred crores.

Amit Paban Bora, a local social activist and writer, alleged that as soon as the case was registered it came to light that there is no government protection to save the golden silk heritage of Assam.

Tea planters have violated all norms of uses of pesticides and Muga cultivation is now in peril.

It may be mentioned that it was earlier a section of people had blamed climate change to be the reason behind the death of Muga worms.

Indigenous no-state people

The Brahmaputra

A Picture is Worth a Thousand Words

Brahmaputra River


LAST UPDATED: Feb 19, 2019 See Article HistoryAlternative Titles: Jamuna, Tsangpo, Ya-lu-tsang-pu Chiang, Yarlung Zangbo Jiang

Brahmaputra River, Bengali Jamuna, Tibetan Tsangpo, Chinese (Pinyin) Yarlung Zangbo Jiang or (Wade-Giles romanization) Ya-lu-tsang-pu Chiang, major river of Central and South Asia. It flows some 1,800 miles (2,900 km) from its source in the Himalayas to its confluence with the Ganges (Ganga) River, after which the mingled waters of the two rivers empty into the Bay of Bengal.

Brahmaputra River
Brahmaputra RiverBrahmaputra River.Encyclopædia Britannica, Inc.

Along its course the Brahmaputra passes through the TibetAutonomous Region of China, the Indian states of Arunachal Pradesh and Assam, and Bangladesh. For most of its length, the river serves as an important inland waterway. It is not, however, navigable between the mountains of Tibet and the plains of India. In its lower course the river is both a creator and a destroyer—depositing huge quantities of fertile alluvial soil but also causing disastrous and frequent floods.

Tsangpo (Brahmaputra) River
Tsangpo (Brahmaputra) RiverTsangpo (Brahmaputra) River flowing through the Himalayas in the Tibet Autonomous Region of China.© Dmitriy Sarbash/Fotolia

Physical Features


The Brahmaputra’s source is the Chemayungdung Glacier, which covers the slopes of the Himalayas about 60 miles (100 km) southeast of Lake Mapam in southwestern Tibet. The three headstreams that arise there are the Kubi, the Angsi, and the Chemayungdung. From its source the river runs for nearly 700 miles (1,100 km) in a generally easterly direction between the Great Himalayas range to the south and the Kailas Range to the north. Throughout its upper course the river is generally known as the Tsangpo (“Purifier”); it is also known by its Chinese name (Yarlung Zangbo) and by other local Tibetan names.

The Brahmaputra and Ganges river basins and their drainage network.
The Brahmaputra and Ganges river basins and their drainage network.Encyclopædia Britannica, Inc.

In Tibet the Tsangpo receives a number of tributaries. The most important left-bank tributaries are the Raka Zangbo (Raka Tsangpo), which joins the river west of Xigazê (Shigatse), and the Lhasa (Kyi), which flows past the Tibetan capital of Lhasa and joins the Tsangpo at Qüxü. The Nyang Qu (Gyamda) River joins the river from the north at Zela (Tsela Dzong). On the right bank a second river called the Nyang Qu (Nyang Chu) meets the Tsangpo at Xigazê.

Tsangpo (Brahmaputra) River: shoals
Tsangpo (Brahmaputra) River: shoalsShoals in the Tsangpo (Brahmaputra) River, Tibet Autonomous Region, China.© Lukas Hlavac/

After passing Pi (Pe) in Tibet, the river turns suddenly to the north and northeast and cuts a course through a succession of great narrow gorges between the mountainous massifs of Gyala Peri and Namjagbarwa (Namcha Barwa) in a series of rapids and cascades. Thereafter, the river turns south and southwest and flows through a deep gorge (the “Grand Canyon” of the Tsangpo) across the eastern extremity of the Himalayas with canyon walls that extend upward for 16,500 feet (5,000 metres) and more on each side. During that stretch the river enters northern Arunachal Pradesh state in northeastern India, where it is known as the Dihang (or Siang) River, and turns more southerly.

The Dihang, winding out of the mountains, turns toward the southeast and descends into a low-lying basin as it enters northeastern Assam state. Just west of the town of Sadiya, the river again turns to the southwest and is joined by two mountain streams, the Lohit and the Dibang. Below that confluence, about 900 miles (1,450 km) from the Bay of Bengal, the river becomes known conventionally as the Brahmaputra (“Son of Brahma”). In Assam the river is mighty, even in the dry season, and during the rains its banks are more than 5 miles (8 km) apart. As the river follows its braided 450-mile (700-km) course through the valley, it receives several rapidly rushing Himalayan streams, including the Subansiri, Kameng, Bhareli, Dhansiri, Manas, Champamati, Saralbhanga, and Sankosh rivers. The main tributaries from the hills and from the plateau to the south are the Burhi Dihing, the Disang, the Dikhu, and the Kopili.

Sibsagar, India: temple
Sibsagar, India: templeShaiva temple at Sibsagar near the Brahmaputra River, Assam, India.Foto Features

The Brahmaputra enters the plains of Bangladesh after turning south around the Garo Hills below Dhuburi, India. After flowing past Chilmari, Bangladesh, it is joined on its right bank by the Tista Riverand then follows a 150-mile (240-km) course due south as the Jamuna River. (South of Gaibanda, the Old Brahmaputra leaves the left bank of the main stream and flows past Jamalpur and Mymensingh to join the Meghna River at Bhairab Bazar.) Before its confluence with the Ganges, the Jamuna receives the combined waters of the Baral, Atrai, and Hurasagar rivers on its right bank and becomes the point of departure of the large Dhaleswari River on its left bank. A tributary of the Dhaleswari, the Buriganga (“Old Ganges”), flows past Dhaka, the capital of Bangladesh, and joins the Meghna River above Munshiganj.

Tista River
Tista RiverTista River, a major tributary of the Brahmaputra River, flowing through the Siwalik Hills, northeastern India.Anupam Manur

The Jamuna joins with the Ganges north of Goalundo Ghat, below which, as the Padma, their combined waters flow to the southeast for a distance of about 75 miles (120 km). After several smaller channels branch off to feed the Ganges-Brahmaputra delta to the south, the main body of the Padma reaches its confluence with the Meghna River near Chandpur and then enters the Bay of Bengal through the Meghna estuary and lesser channels flowing through the delta. The Meghna forms the eastern limit of the Sundarbans, a vast tract of forest and saltwater swamp that constitutes much of the Ganges-Brahmaputra delta. The growth of the delta is dominated by tidal processes.

The Ganges-Brahmaputra system has the third greatest average discharge of the world’s rivers—roughly 1,086,500 cubic feet (30,770 cubic metres) per second; approximately 700,000 cubic feet (19,800 cubic metres) per second of the total is supplied by the Brahmaputra alone. The rivers’ combined suspended sediment load of about 1.84 billion tons per year is the world’s highest.


The climate of the Brahmaputra valley varies from the harsh, cold, and dry conditions found in Tibet to the generally hot and humid conditions prevailing in Assam state and in Bangladesh. Tibetan winters are severely cold, with average temperatures below 32 °F (0 °C), while summers are mild and sunny. The upper river valley lies in the rain shadow of the Himalayas, and precipitation there is relatively light: Lhasa receives about 16 inches (400 mm) annually.

The Indian and Bangladeshi parts of the valley are governed by the monsoon (wet, dry) climate, though it is somewhat modified there compared with other parts of the subcontinent; the hot season is shorter than usual, and the average annual temperature ranges from 79 °F (26 °C) in Dhuburi, Assam, to 85 °F (29 °C) in Dhaka. Precipitation is relatively heavy, and humidity is high throughout the year. The annual rainfall—between 70 and 150 inches (1,780 and 3,810 mm)—falls mostly between June and early October; however, light rains also fall from March to May.


The course of the Brahmaputra has changed continually over time. The most spectacular of these changes was the eastward diversion of the Tista River and the ensuing development of the new channel of the Jamuna, which occurred in 1787 with an exceptionally high flood in the Tista. The waters of the Tista suddenly were diverted eastward into an old abandoned course, causing the river to join the Brahmaputra opposite Bahadurabad Ghat in Mymensingh district. Until the late 18th century the Brahmaputra flowed past the town of Mymensingh and joined the Meghna River near Bhairab Bazar (the path of the present-day Old Brahmaputra channel). At that time a minor stream called the Konai-Jenai—probably a spill channel of the Old Brahmaputra—followed the course of today’s Jamuna River (now the main Brahmaputra channel). After the Tista flood of 1787 reinforced it, the Brahmaputra began to cut a new channel along the Konai-Jenai and gradually converted it after 1810 into the main stream, now known as the Jamuna.

Ganges-Brahmaputra delta cyclone
Ganges-Brahmaputra delta cycloneSatellite image of the Ganges-Brahmaputra delta cyclone, November 12, 1970.NOAA

Along the lower courses of the Ganges and Brahmaputra and along the Meghna, the land undergoes constant erosion and deposition of silt because of the shifts and changes in these active rivers. Vast areas are subject to inundation during the wet monsoon months. The shifts in the course of the Jamuna since 1787 have been considerable, and the river is never in exactly the same place for two successive years. Islands and sizable newly deposited lands (chars) in the river appear and disappear seasonally. The chars are valuable to the economy of Bangladesh as additional cultivable areas.

In Tibet the waters of the Brahmaputra are clear because little silt is carried downstream. As soon as the river enters Assam, however, the silt load becomes heavy. Because of the speed and volume of water in the northern tributaries that flow down from the rain-soaked Himalayan slopes, their silt load is much heavier than that carried by the tributaries crossing the hard rocks of the old plateau to the south. In Assam the deep channel of the Brahmaputra follows the southern bank closer than the northern. This tendency is reinforced by the silt-laden northern tributaries pushing the channel south.

Another important feature of the river is its tendency to flood. The quantity of water carried by the Brahmaputra in India and Bangladesh is enormous. The river valley in Assam is enclosed by hill ranges on the north, east, and south and receives more than 100 inches (2,540 mm) of rainfall annually, while in the Bengal Plain heavy rainfall—averaging 70 to 100 inches—is reinforced by the huge discharge of the Tista, Torsa, and Jaldhaka rivers. Extensive flooding is virtually an annual occurrence in the Brahmaputra valley during the summer monsoon. In addition, tidal surges accompanying tropical cyclones sweeping inland from the Bay of Bengal periodically bring great destruction to the delta region. One such storm—the Ganges-Brahmaputra delta cyclone (also called the Bhola cyclone) of November 1970—caused an estimated 300,000 to 500,000 deaths and inundated a vast area. In the 21st century the delta has also been affected by rising sea levels as a result of global warming.

Plant and animal life

Along the upper reaches of the Brahmaputra (Tsangpo) on the high Plateau of Tibet, the vegetation is mainly xerophytic (drought-resistant) shrubs and grasses. As the river descends from Tibet, increased precipitation supports the growth of forests. Forests of sal (genus Shorea)—a valuable timber tree that is also utilized to cultivate the lac insect, which produces the resin used to make shellac—are found in Assam. At even lower elevations, tall reed jungles grow in the swamps and depressed water-filled areas (jheels) of the immense floodplains. Around towns and villages in the Assam Valley, the many fruit trees yield plantains, papayas, mangoes, and jackfruit. Bamboo thickets abound throughout Assam and Bangladesh. Nipa palms (Nypa fruticans) and other halophytic (salt-tolerant) flora predominate in the delta region’s mangrove swamps.

Indian rhinoceros
Indian rhinocerosIndian rhinoceros (Rhinoceros unicornis) in Kaziranga National Park, Assam, India.© Jeremy Richards/

The most-notable animal of the swamps in Assam is the one-horned rhinoceros, which has become extinct in other parts of the world; Kaziranga National Park (designated a UNESCO World Heritage sitein 1985) provides a refuge for the rhinoceros and for other wildlife in the valley, including elephants, Bengal tigers, leopards, wild buffalo, and deer. Numerous varieties of fish include the pabda (Omdok pabda), chital (Notopterus chitala), and mrigal (Cirrhinus cirrhosus).


The people living in the different sections of the Brahmaputra valley are of diverse origin and culture. North of the Great Himalayas, the Tibetans practice Buddhism and speak the Tibetan language. They engage in animal husbandry and cultivate the valley with irrigation water taken from the river.

Stupa on the bank of the Tsangpo (Brahmaputra) River, Tibet Autonomous Region, China.
Stupa on the bank of the Tsangpo (Brahmaputra) River, Tibet Autonomous Region, China.© Naomi Duguid/Asia Access

The ancestry of the Assamese includes peoples speaking Tibeto-Burman languages from the surrounding highlands and peoples from the lowlands of India to the south and west. The Assamese language is akin to Bengali, which is spoken in West Bengal state in India and in Bangladesh. Since the late 19th century a vast number of immigrants from the Bengal Plain of Bangladesh have entered Assam, where they have settled to cultivate vacant lands, particularly the low floodplains. In the Bengal Plain itself the river flows through an area that is densely populated by the Bengali people, who cultivate the fertile valley. The hilly margins of the plain are inhabited by the tribal Garo, Khasi, and Hajong of Meghalayastate in India.


Irrigation and flood control

Flood-control schemes and the building of embankments were initiated after 1954. In Bangladesh the Brahmaputra embankment running west of the Jamuna River from north to south helps to control floods. The Tista Barrage Project is both an irrigation and a flood-protection scheme.

Buriganga River, Dhaka, Bangladesh
Buriganga River, Dhaka, BangladeshBoat traffic on the Buriganga (“Old Ganges”) River, Dhaka, Bangladesh.© Dmitry Chulov/

Until the 21st century, little power had been harnessed along the Brahmaputra, although the estimated potential was great—some 12,000 megawatts in India alone. An increasing number of hydroelectric stations have been completed in Assam, most notably the Kopili Hydel Project in the south of the state. Another major project, the Ranganadi plant, has been built in Arunachal Pradesh, which has considerably more generating capacity than the Kopili station. In addition, a giant hydropower installation in Tibet on the Tsangpo River became fully operational in late 2015.

Navigation and transport

Near Lhazê (Lhatse Dzong) in Tibet, the river becomes navigable for about 400 miles (640 km). Coracles (boats made of hides and bamboo) and large ferries ply its waters at 13,000 feet (4,000 metres) above sea level. The Tsangpo is spanned in several places by suspension bridges.

Because it flows through a region of heavy rainfall in Assam and Bangladesh, the Brahmaputra is more important for inland navigation than for irrigation. The river has long formed a waterway between the Indian states of West Bengal and Assam, although, on occasion, political conflicts have disrupted the movement of traffic through Bangladesh. The Brahmaputra is navigable throughout the Bengal Plain and Assam upstream to Dibrugarh, 700 miles (1,100 km) from the sea. In addition to all types of local craft, powered launches and steamers easily travel up and down the river, carrying bulky raw materials, timber, and crude oil.

The Brahmaputra remained unbridged throughout its course in the plains until the Saraighat Bridge—carrying both road and rail traffic—was opened in 1962 near Guwahati, Assam. A second crossing in Assam, the Kalia Bhomora road bridge near Tezpur, was opened in 1987. Ferries, however, have continued as the most important—and in Bangladesh the only—means of crossing the Brahmaputra. Sadiya, Dibrugarh, Jorhat, Tezpur, Guwahati, Goalpara, and Dhuburi are important towns and crossing points in Assam, while Kurigram, Rahumari, Chilmari, Bahadurabad Ghat, Phulchari, Sarishabari, Jagannathganj Ghat, Nagarbari, Sirajganj, and Goalundo Ghat are major crossing points in Bangladesh. The railheads are located at Bahadurabad Ghat, Phulchari, Jagannathganj Ghat, Sirajganj, and Goalundo Ghat.

Study And Exploration

The upper course of the Brahmaputra was explored as early as the 18th century, although it remained virtually unknown until the 19th century. The explorations of the Indian surveyor Kinthup (reported in 1884) and of J.F. Needham in Assam in 1886 established the Tsangpo River as the upper course of the Brahmaputra. Various British expeditions in the first quarter of the 20th century explored the Tsangpo upstream in Tibet to Xigazê, as well as the river’s mountain gorges. More-recent scientific work has concentrated on understanding the hydrology of the Brahmaputra for watershed management and flood-hazard mitigation.

Indigenous no-state people

Kerala Floods Match Climate Change Forecasts



  1. A scientist said Kerala floods can’t be attributed to climate change only. India’s average annual temperatures are set to rise 1.5 C to 3 C

Once-a-century rains that have pounded Kerala and displaced 1.3 million people are in line with the predictions of climate scientists, who warn that worse is to come if global warming continues unabated.

The monsoon rains upon which farmers in the southern state depend for their food and livelihoods dumped two-and-a-half times the normal amount of water across the state last week, according to meteorologists in the country.

It is difficult to attribute any single extreme weather event — such as the Kerala flooding — to climate change, said Roxy Mathew Koll, a climate scientist at the Indian Institute of Tropical Meteorology in Pashan, near Mumbai.  

At the same time, “our recent research shows a three-fold increase in widespread extreme rains during 1950-2017, leading to large-scale flooding,” he told AFP.

Across India, flooding caused by heavy monsoons rainfall claimed 69,000 lives and left 17 million people without homes over the same period, according to a study he co-authored, published last year in Nature Communications. 

In Kerala, all 35 of the state’s major reservoirs were brimming with rain water by August 10, forcing local authorities to open the sluice gates on the Idukki Dam for the first time in 26 years.

“These floods that we are seeing in Kerala right now are basically in line with climate projections,” said Kira Vinke, a scientist at the Potsdam Institute for Climate Impact Research in Germany.

“If we continue with current levels of emissions — which is not unlikely — we will have unmanageable risks,” she told AFP. 

The weather patterns behind these destructive downpours are well understood, even if the fingerprint of global warming is still hard to distinguish from what scientists call “natural variability”.

Rapid warming in the Arabian Sea and nearby landmass causes monsoon winds to fluctuate and intensify for short spans of three-to-four days, Koll explained.

During those periods, moisture from the Arabian Sea is dumped inland.

South Asia’s ‘hotspots’

“Over the last decade, due to climate change, the overheating of landmass leads to the intensification of monsoon rainfalls in central and southern India,” said monsoon expert Elena Surovyatkina, a professor at the Russian Academy of Sciences, and a senior scientist at PIK. 

The changes observed so far have occurred after an increase in Earth’s average surface temperature of only one degree Celsius (1.8 degrees Fahrenheit) above pre-industrial levels.

On current trends, India’s average annual temperatures are set to rise 1.5 C to 3 C compared to that benchmark by mid-century, according to a World Bank report entitled “South Asia’s Hotspots”.

“If no corrective measures are taken, changing rainfall patterns and rising temperatures will cost India 2.8 percent of its GDP and will drag down living standards of half its population by 2050,” the World Bank said in a statement. 

The 196-nation Paris climate treaty calls for capping global warming at “well below” 2 C (3.6 F), and 1.5 C if possible. 

But voluntary national pledges to reduce greenhouse gas emissions, even if respected, would still see temperatures rise at least 3 C. 

Flooding is not the only problem India’s burgeoning — and highly vulnerable — population will face as a consequence of global warming.

“What we will see with climate change in India is that the wet season is going to be wetter and the dry season drier,” said Vinki. 

“Already we are observing that the monsoon is becoming harder to predict with traditional methods.”

If manmade carbon emissions continue unabated, some regions in northeast India could literally become unlivable by the end of the century due to a deadly combination of heat and humidity during heatwaves, recent research has projected.

Indeed, larges swathes of south Asia, including the Ganges-Brahmaputra Basin, could approach the threshold for survivability outdoors.1 COMMENT

Coastal cities, meanwhile, are especially vulnerable to sea level rise, driven by melting ice sheets and expanding ocean water, on the one hand, and subsidence due to over-development and the depletion of water tables, on the other.

(Kerala has to rebuild itself after the worst floods in over a century. Hundreds have died and lakhs are homeless. You can help.)