Refine
Language
- English (16)
Is part of the Bibliography
- yes (16)
Keywords
- Accuracy Asseessment (2)
- Digital Elevation Model (2)
- Error (2)
- Expression (2)
- Extraction (2)
- GEDI (2)
- ICESat-2 (2)
- Landscape Response (2)
- Passive Microwave (2)
- Patterns (2)
Institute
The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) with its land and vegetation height data product (ATL08), and Global Ecosystem Dynamics Investigation (GEDI) with its terrain elevation and height metrics data product (GEDI Level 2A) missions have great potential to globally map ground and canopy heights. Canopy height is a key factor in estimating above-ground biomass and its seasonal changes; these satellite missions can also improve estimated above-ground carbon stocks. This study presents a novel Sparse Vegetation Detection Algorithm (SVDA) which uses ICESat-2 (ATL03, geolocated photons) data to map tree and vegetation heights in a sparsely vegetated savanna ecosystem. The SVDA consists of three main steps: First, noise photons are filtered using the signal confidence flag from ATL03 data and local point statistics. Second, we classify ground photons based on photon height percentiles. Third, tree and grass photons are classified based on the number of neighbors. We validated tree heights with field measurements (n = 55), finding a root-mean-square error (RMSE) of 1.82 m using SVDA, GEDI Level 2A (Geolocated Elevation and Height Metrics product): 1.33 m, and ATL08: 5.59 m. Our results indicate that the SVDA is effective in identifying canopy photons in savanna ecosystems, where ATL08 performs poorly. We further identify seasonal vegetation height changes with an emphasis on vegetation below 3 m; widespread height changes in this class from two wet-dry cycles show maximum seasonal changes of 1 m, possibly related to seasonal grass-height differences. Our study shows the difficulties of vegetation measurements in savanna ecosystems but provides the first estimates of seasonal biomass changes.
The Upper Indus Basin (UIB), which covers a wide range of climatic and topographic settings, provides an ideal venue to explore the relationship between climate and topography. While the distribution of snow and glaciers is spatially and temporally heterogeneous, there exist regions with similar elevation-snow relationships. In this work, we construct elevation-binned snow-cover statistics to analyze 3415 watersheds and 7357 glaciers in the UIB region. We group both glaciers and watersheds using a hierarchical clustering approach and find that (1) watershed clusters mirror large-scale moisture transport patterns and (2) are highly dependent on median watershed elevation. (3) Glacier clusters are spatially heterogeneous and are less strongly controlled by elevation, but rather by local topographic parameters that modify solar insolation. Our clustering approach allows us to clearly define self-similar snow-topographic regions. Eastern watersheds in the UIB show a steep snow cover-elevation relationship whereas watersheds in the central and western UIB have moderately sloped relationships, but cluster in distinct groups. We highlight this snow-cover-topographic transition zone and argue that these watersheds have different hydrologic responses than other regions. Our hierarchical clustering approach provides a potential new framework to use in defining climatic zones in the cyrosphere based on empirical data.
Insolation differences play a primary role in controlling microclimate and vegetation cover, which together influence the development of topography. Topographic asymmetry (TA), or slope differences between terrain aspects, has been well documented in small-scale, field-based, and modeling studies. Here we combine a suite of environmental (e.g., vegetation, temperature, solar insolation) and topographic (e.g., elevation, drainage network) data to explore the driving mechanisms and markers of TA on a global scale. Using a novel empirical TA analysis method, we find that (1) steeper terrain has higher TA magnitudes, (2) globally, pole-facing terrain is on average steeper than equator-facing terrain, especially in mid-latitude, tectonically quiescent, and vegetated landscapes, and (3) high-elevation and low-temperature regions tend to have terrain steepened toward the equator. We further show that there are distinct differences in climate and vegetation cover across terrain aspects, and that TA is reflected in the size and form of fluvial drainage networks. Our work supports the argument that insolation asymmetries engender differences in local microclimates and vegetation on opposing terrain aspects, which broadly encourage the development of asymmetric topography across a range of lithologic, tectonic, geomorphic, and climatic settings.
The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) with its land and vegetation height data product (ATL08), and Global Ecosystem Dynamics Investigation (GEDI) with its terrain elevation and height metrics data product (GEDI Level 2A) missions have great potential to globally map ground and canopy heights. Canopy height is a key factor in estimating above-ground biomass and its seasonal changes; these satellite missions can also improve estimated above-ground carbon stocks. This study presents a novel Sparse Vegetation Detection Algorithm (SVDA) which uses ICESat-2 (ATL03, geolocated photons) data to map tree and vegetation heights in a sparsely vegetated savanna ecosystem. The SVDA consists of three main steps: First, noise photons are filtered using the signal confidence flag from ATL03 data and local point statistics. Second, we classify ground photons based on photon height percentiles. Third, tree and grass photons are classified based on the number of neighbors. We validated tree heights with field measurements (n = 55), finding a root-mean-square error (RMSE) of 1.82 m using SVDA, GEDI Level 2A (Geolocated Elevation and Height Metrics product): 1.33 m, and ATL08: 5.59 m. Our results indicate that the SVDA is effective in identifying canopy photons in savanna ecosystems, where ATL08 performs poorly. We further identify seasonal vegetation height changes with an emphasis on vegetation below 3 m; widespread height changes in this class from two wet-dry cycles show maximum seasonal changes of 1 m, possibly related to seasonal grass-height differences. Our study shows the difficulties of vegetation measurements in savanna ecosystems but provides the first estimates of seasonal biomass changes.
High Mountain Asia (HMA) is dependent upon both the amount and timing of snow and glacier meltwater. Previous model studies and coarse resolution (0.25° × 0.25°, ∼25 km × 25 km) passive microwave assessments of trends in the volume and timing of snowfall, snowmelt, and glacier melt in HMA have identified key spatial and seasonal heterogeneities in the response of snow to changes in regional climate. Here we use recently developed, continuous, internally consistent, and high-resolution passive microwave data (3.125 km × 3.125 km, 1987–2016) from the special sensor microwave imager instrument family to refine and extend previous estimates of changes in the snow regime of HMA. We find an overall decline in snow volume across HMA; however, there exist spatially contiguous regions of increasing snow volume—particularly during the winter season in the Pamir, Karakoram, Hindu Kush, and Kunlun Shan. Detailed analysis of changes in snow-volume trends through time reveal a large step change from negative trends during the period 1987–1997, to much more positive trends across large regions of HMA during the periods 1997–2007 and 2007–2016. We also find that changes in high percentile monthly snow-water volume exhibit steeper trends than changes in low percentile snow-water volume, which suggests a reduction in the frequency of high snow-water volumes in much of HMA. Regions with positive snow-water storage trends generally correspond to regions of positive glacier mass balances.
High Mountain Asia (HMA) is dependent upon both the amount and timing of snow and glacier meltwater. Previous model studies and coarse resolution (0.25° × 0.25°, ∼25 km × 25 km) passive microwave assessments of trends in the volume and timing of snowfall, snowmelt, and glacier melt in HMA have identified key spatial and seasonal heterogeneities in the response of snow to changes in regional climate. Here we use recently developed, continuous, internally consistent, and high-resolution passive microwave data (3.125 km × 3.125 km, 1987–2016) from the special sensor microwave imager instrument family to refine and extend previous estimates of changes in the snow regime of HMA. We find an overall decline in snow volume across HMA; however, there exist spatially contiguous regions of increasing snow volume—particularly during the winter season in the Pamir, Karakoram, Hindu Kush, and Kunlun Shan. Detailed analysis of changes in snow-volume trends through time reveal a large step change from negative trends during the period 1987–1997, to much more positive trends across large regions of HMA during the periods 1997–2007 and 2007–2016. We also find that changes in high percentile monthly snow-water volume exhibit steeper trends than changes in low percentile snow-water volume, which suggests a reduction in the frequency of high snow-water volumes in much of HMA. Regions with positive snow-water storage trends generally correspond to regions of positive glacier mass balances.
High Mountain Asia (HMA) – encompassing the Tibetan Plateau and surrounding mountain ranges – is the primary water source for much of Asia, serving more than a billion downstream users. Many catchments receive the majority of their yearly water budget in the form of snow, which is poorly monitored by sparse in situ weather networks. Both the timing and volume of snowmelt play critical roles in downstream water provision, as many applications – such as agriculture, drinking-water generation, and hydropower – rely on consistent and predictable snowmelt runoff. Here, we examine passive microwave data across HMA with five sensors (SSMI, SSMIS, AMSR-E, AMSR2, and GPM) from 1987 to 2016 to track the timing of the snowmelt season – defined here as the time between maximum passive microwave signal separation and snow clearance. We validated our method against climate model surface temperatures, optical remote-sensing snow-cover data, and a manual control dataset (n = 2100, 3 variables at 25 locations over 28 years); our algorithm is generally accurate within 3–5 days. Using the algorithm-generated snowmelt dates, we examine the spatiotemporal patterns of the snowmelt season across HMA. The climatically short (29-year) time series, along with complex interannual snowfall variations, makes determining trends in snowmelt dates at a single point difficult. We instead identify trends in snowmelt timing by using hierarchical clustering of the passive microwave data to determine trends in self-similar regions. We make the following four key observations. (1) The end of the snowmelt season is trending almost universally earlier in HMA (negative trends). Changes in the end of the snowmelt season are generally between 2 and 8 days decade−1 over the 29-year study period (5–25 days total). The length of the snowmelt season is thus shrinking in many, though not all, regions of HMA. Some areas exhibit later peak signal separation (positive trends), but with generally smaller magnitudes than trends in snowmelt end. (2) Areas with long snowmelt periods, such as the Tibetan Plateau, show the strongest compression of the snowmelt season (negative trends). These trends are apparent regardless of the time period over which the regression is performed. (3) While trends averaged over 3 decades indicate generally earlier snowmelt seasons, data from the last 14 years (2002–2016) exhibit positive trends in many regions, such as parts of the Pamir and Kunlun Shan. Due to the short nature of the time series, it is not clear whether this change is a reversal of a long-term trend or simply interannual variability. (4) Some regions with stable or growing glaciers – such as the Karakoram and Kunlun Shan – see slightly later snowmelt seasons and longer snowmelt periods. It is likely that changes in the snowmelt regime of HMA account for some of the observed heterogeneity in glacier response to climate change. While the decadal increases in regional temperature have in general led to earlier and shortened melt seasons, changes in HMA's cryosphere have been spatially and temporally heterogeneous.
Studies of glaciers generally require precise glacier outlines. Where these are not available, extensive manual digitization in a geographic information system (GIS) must be performed, as current algorithms struggle to delineate glacier areas with debris cover or other irregular spectral profiles. Although several approaches have improved upon spectral band ratio delineation of glacier areas, none have entered wide use due to complexity or computational intensity.
In this study, we present and apply a glacier mapping algorithm in Central Asia which delineates both clean glacier ice and debris-covered glacier tongues. The algorithm is built around the unique velocity and topographic characteristics of glaciers and further leverages spectral and spatial relationship data. We found that the algorithm misclassifies between 2 and 10% of glacier areas, as compared to a similar to 750 glacier control data set, and can reliably classify a given Landsat scene in 3-5 min.
The algorithm does not completely solve the difficulties inherent in classifying glacier areas from remotely sensed imagery but does represent a significant improvement over purely spectral-based classification schemes, such as the band ratio of Landsat 7 bands three and five or the normalized difference snow index. The main caveats of the algorithm are (1) classification errors at an individual glacier level, (2) reliance on manual intervention to separate connected glacier areas, and (3) dependence on fidelity of the input Landsat data.
Snowfall comprises a significant percentage of the annual water budget in High Mountain Asia (HMA), but snow water equivalent (SWE) is poorly constrained due to lack of in-situ measurements and complex terrain that limits the efficacy of modeling and observations. Over the past few decades, SWE has been estimated with passive microwave (PM) sensors with generally good results in wide, flat, terrain, and lower reliability in densely forested, complex, or high-elevation areas. In this study, we use raw swath data from five satellite - sensors the Special Sensor Microwave/Imager (SSMI) and Special Sensor Microwave Imager/Sounder (SSMIS) (1987-2015, F08, F11, F13, F17), Advanced Microwave Scanning Radiometer - Earth Observing System (AMSR-E, 2002-2011), AMSR2 (2012-2015), and the Global Precipitation Measurement (GPM, 2014-2015) - in order to understand the spatial and temporal structure of native sensor, topographic, and land cover biases in SWE estimates in HMA. We develop a thorough understanding of the uncertainties in our SWE estimates by examining the impacts of topographic parameters (aspect, relief, hillslope angle, and elevation), land cover, native sensor biases, and climate parameters (precipitation, temperature, and wind speed). HMA, with its high seasonality, large topographic gradients and low relief at high elevations provides an excellent context to examine a wide range of climatic, land-cover, and topographic settings to better constrain SWE uncertainties and potential sensor bias. Using a multi-parameter regression, we compare long-term SWE variability to forest fraction, maximal multiyear snow depth, topographic parameters, and long-term average wind speed across both individual sensor time series and a merged multi-sensor dataset. In regions where forest cover is extensive, it is the strongest control on SWE variability. In those regions where forest density is low (<5%), maximal snow depth dominates the uncertainty signal. In our regression across HMA, we find that forest fraction is the strongest control on SWE variability (75.8%), followed by maximal multi-year snow depth (7.82%), 90th percentile 10-m wind speed of a 10-year December-January-February (DJF) time series (5.64%), 25th percentile DJF 10-m wind speed (5.44%), and hillslope angle (5.24%). Elevation, relief, and terrain aspect show very low influence on SWE variability (<1%). We find that the GPM sensor provides the most robust regression results, and can be reliably used to estimate SWE in our study region. While forest cover and elevation have been integrated into many SWE algorithms, wind speed and long-term maximal snow depth have not. Our results show that wind redistribution of snow can have impacts on SWE, especially over large, flat, areas. Using our regression results, we have developed an understanding of sensor specific SWE uncertainties and their spatial patterns. The uncertainty maps developed in this study provide a first-order approximation of SWE-estimate reliability for much of HMA, and imply that high-fidelity SWE estimates can be produced for many high-elevation areas. (C) 2016 Elsevier Inc. All rights reserved.
High Mountain Asia provides water for more than a billion downstream users. Many catchments receive the majority of their yearly water budget in the form of snow - the vast majority of which is not monitored by sparse weather networks. We leverage passive microwave data from the SSMI series of satellites (SSMI, SSMI/S, 1987-2016), reprocessed to 3.125 km resolution, to examine trends in the volume and spatial distribution of snow-water equivalent (SWE) in the Indus Basin. We find that the majority of the Indus has seen an increase in snow-water storage. There exists a strong elevation-trend relationship, where high-elevation zones have more positive SWE trends. Negative trends are confined to the Himalayan foreland and deeply-incised valleys which run into the Upper Indus. This implies a temperature-dependent cutoff below which precipitation increases are not translated into increased SWE. Earlier snowmelt or a higher percentage of liquid precipitation could both explain this cutoff.(1) Earlier work 2 found a negative snow-water storage trend for the entire Indus catchment over the time period 1987-2009 (-4 x 10(-3) mm/yr). In this study based on an additional seven years of data, the average trend reverses to 1.4 x 10(-3). This implies that the decade since the mid-2000s was likely wetter, and positively impacted long-term SWE trends. This conclusion is supported by an analysis of snowmelt onset and end dates which found that while long-term trends are negative, more recent (since 2005) trends are positive (moving later in the year).(3)
Snow meltwaters account for most of the yearly water budgets of many catchments in High Mountain Asia (HMA). We examine trends in snow water equivalent (SWE) using passive microwave data (1987 to 2009). We find an overall decrease in SWE in HMA, despite regions of increased SWE in the Pamir, Kunlun Shan, Eastern Himalaya, and Eastern Tien Shan. Although the average decline in annual SWE across HMA (contributing area, 2641 x 10(3) km(2)) is low (average, -0.3%), annual SWE losses conceal distinct seasonal and spatial heterogeneities across the study region. For example, the Tien Shan has seen both strong increases in winter SWE and sharp declines in spring and summer SWE. In the majority of catchments, the most negative SWE trends are found in mid-elevation zones, which often correspond to the regions of highest snow-water storage and are somewhat distinct from glaciated areas. Negative changes in SWE storage in these mid-elevation zones have strong implications for downstream water availability.
Determining the optimal grid resolution for topographic analysis on an airborne lidar dataset
(2019)
Digital elevation models (DEMs) are a gridded representation of the surface of the Earth and typically contain uncertainties due to data collection and processing. Slope and aspect estimates on a DEM contain errors and uncertainties inherited from the representation of a continuous surface as a grid (referred to as truncation error; TE) and from any DEM uncertainty. We analyze in detail the impacts of TE and propagated elevation uncertainty (PEU) on slope and aspect.
Using synthetic data as a control, we define functions to quantify both TE and PEU for arbitrary grids. We then develop a quality metric which captures the combined impact of both TE and PEU on the calculation of topographic metrics. Our quality metric allows us to examine the spatial patterns of error and uncertainty in topographic metrics and to compare calculations on DEMs of different sizes and accuracies.
Using lidar data with point density of ∼10 pts m−2 covering Santa Cruz Island in southern California, we are able to generate DEMs and uncertainty estimates at several grid resolutions. Slope (aspect) errors on the 1 m dataset are on average 0.3∘ (0.9∘) from TE and 5.5∘ (14.5∘) from PEU. We calculate an optimal DEM resolution for our SCI lidar dataset of 4 m that minimizes the error bounds on topographic metric calculations due to the combined influence of TE and PEU for both slope and aspect calculations over the entire SCI. Average slope (aspect) errors from the 4 m DEM are 0.25∘ (0.75∘) from TE and 5∘ (12.5∘) from PEU. While the smallest grid resolution possible from the high-density SCI lidar is not necessarily optimal for calculating topographic metrics, high point-density data are essential for measuring DEM uncertainty across a range of resolutions.
Improving semi-automated glacier mapping with a multi-method approach: applications in central Asia
(2015)
Studies of glaciers generally require precise glacier outlines. Where these are not available, extensive manual digitization in a geographic information system (GIS) must be performed, as current algorithms struggle to delineate glacier areas with debris cover or other irregular spectral profiles. Although several approaches have improved upon spectral band ratio delineation of glacier areas, none have entered wide use due to complexity or computational intensity.
In this study, we present and apply a glacier mapping algorithm in Central Asia which delineates both clean glacier ice and debris-covered glacier tongues. The algorithm is built around the unique velocity and topographic characteristics of glaciers and further leverages spectral and spatial relationship data. We found that the algorithm misclassifies between 2 and 10% of glacier areas, as compared to a similar to 750 glacier control data set, and can reliably classify a given Landsat scene in 3-5 min.
The algorithm does not completely solve the difficulties inherent in classifying glacier areas from remotely sensed imagery but does represent a significant improvement over purely spectral-based classification schemes, such as the band ratio of Landsat 7 bands three and five or the normalized difference snow index. The main caveats of the algorithm are (1) classification errors at an individual glacier level, (2) reliance on manual intervention to separate connected glacier areas, and (3) dependence on fidelity of the input Landsat data.
Determining the optimal grid resolution for topographic analysis on an airborne lidar dataset
(2019)
Digital elevation models (DEMs) are a gridded representation of the surface of the Earth and typically contain uncertainties due to data collection and processing. Slope and aspect estimates on a DEM contain errors and uncertainties inherited from the representation of a continuous surface as a grid (referred to as truncation error; TE) and from any DEM uncertainty. We analyze in detail the impacts of TE and propagated elevation uncertainty (PEU) on slope and aspect.
Using synthetic data as a control, we define functions to quantify both TE and PEU for arbitrary grids. We then develop a quality metric which captures the combined impact of both TE and PEU on the calculation of topographic metrics. Our quality metric allows us to examine the spatial patterns of error and uncertainty in topographic metrics and to compare calculations on DEMs of different sizes and accuracies.
Using lidar data with point density of ∼10 pts m−2 covering Santa Cruz Island in southern California, we are able to generate DEMs and uncertainty estimates at several grid resolutions. Slope (aspect) errors on the 1 m dataset are on average 0.3∘ (0.9∘) from TE and 5.5∘ (14.5∘) from PEU. We calculate an optimal DEM resolution for our SCI lidar dataset of 4 m that minimizes the error bounds on topographic metric calculations due to the combined influence of TE and PEU for both slope and aspect calculations over the entire SCI. Average slope (aspect) errors from the 4 m DEM are 0.25∘ (0.75∘) from TE and 5∘ (12.5∘) from PEU. While the smallest grid resolution possible from the high-density SCI lidar is not necessarily optimal for calculating topographic metrics, high point-density data are essential for measuring DEM uncertainty across a range of resolutions.
High Mountain Asia (HMA) - encompassing the Tibetan Plateau and surrounding mountain ranges - is the primary water source for much of Asia, serving more than a billion downstream users. Many catchments receive the majority of their yearly water budget in the form of snow, which is poorly monitored by sparse in situ weather networks. Both the timing and volume of snowmelt play critical roles in downstream water provision, as many applications - such as agriculture, drinking-water generation, and hydropower - rely on consistent and predictable snowmelt runoff. Here, we examine passive microwave data across HMA with five sensors (SSMI, SSMIS, AMSR-E, AMSR2, and GPM) from 1987 to 2016 to track the timing of the snowmelt season - defined here as the time between maximum passive microwave signal separation and snow clearance. We validated our method against climate model surface temperatures, optical remote-sensing snow-cover data, and a manual control dataset (n = 2100, 3 variables at 25 locations over 28 years); our algorithm is generally accurate within 3-5 days. Using the algorithm-generated snowmelt dates, we examine the spatiotemporal patterns of the snowmelt season across HMA. The climatically short (29-year) time series, along with complex interannual snowfall variations, makes determining trends in snowmelt dates at a single point difficult. We instead identify trends in snowmelt timing by using hierarchical clustering of the passive microwave data to determine trends in self-similar regions. We make the following four key observations. (1) The end of the snowmelt season is trending almost universally earlier in HMA (negative trends). Changes in the end of the snowmelt season are generally between 2 and 8 days decade 1 over the 29-year study period (5-25 days total). The length of the snowmelt season is thus shrinking in many, though not all, regions of HMA. Some areas exhibit later peak signal separation (positive trends), but with generally smaller magnitudes than trends in snowmelt end. (2) Areas with long snowmelt periods, such as the Tibetan Plateau, show the strongest compression of the snowmelt season (negative trends). These trends are apparent regardless of the time period over which the regression is performed. (3) While trends averaged over 3 decades indicate generally earlier snowmelt seasons, data from the last 14 years (2002-2016) exhibit positive trends in many regions, such as parts of the Pamir and Kunlun Shan. Due to the short nature of the time series, it is not clear whether this change is a reversal of a long-term trend or simply interannual variability. (4) Some regions with stable or growing glaciers - such as the Karakoram and Kunlun Shan - see slightly later snowmelt seasons and longer snowmelt periods. It is likely that changes in the snowmelt regime of HMA account for some of the observed heterogeneity in glacier response to climate change. While the decadal increases in regional temperature have in general led to earlier and shortened melt seasons, changes in HMA's cryosphere have been spatially and temporally heterogeneous.
High Mountain Asia (HMA) - encompassing the Tibetan Plateau and surrounding mountain ranges - is the primary water source for much of Asia, serving more than a billion downstream users. Many catchments receive the majority of their yearly water budget in the form of snow, which is poorly monitored by sparse in situ weather networks. Both the timing and volume of snowmelt play critical roles in downstream water provision, as many applications - such as agriculture, drinking-water generation, and hydropower - rely on consistent and predictable snowmelt runoff. Here, we examine passive microwave data across HMA with five sensors (SSMI, SSMIS, AMSR-E, AMSR2, and GPM) from 1987 to 2016 to track the timing of the snowmelt season - defined here as the time between maximum passive microwave signal separation and snow clearance. We validated our method against climate model surface temperatures, optical remote-sensing snow-cover data, and a manual control dataset (n = 2100, 3 variables at 25 locations over 28 years); our algorithm is generally accurate within 3-5 days. Using the algorithm-generated snowmelt dates, we examine the spatiotemporal patterns of the snowmelt season across HMA. The climatically short (29-year) time series, along with complex interannual snowfall variations, makes determining trends in snowmelt dates at a single point difficult. We instead identify trends in snowmelt timing by using hierarchical clustering of the passive microwave data to determine trends in self-similar regions. We make the following four key observations. (1) The end of the snowmelt season is trending almost universally earlier in HMA (negative trends). Changes in the end of the snowmelt season are generally between 2 and 8 days decade 1 over the 29-year study period (5-25 days total). The length of the snowmelt season is thus shrinking in many, though not all, regions of HMA. Some areas exhibit later peak signal separation (positive trends), but with generally smaller magnitudes than trends in snowmelt end. (2) Areas with long snowmelt periods, such as the Tibetan Plateau, show the strongest compression of the snowmelt season (negative trends). These trends are apparent regardless of the time period over which the regression is performed. (3) While trends averaged over 3 decades indicate generally earlier snowmelt seasons, data from the last 14 years (2002-2016) exhibit positive trends in many regions, such as parts of the Pamir and Kunlun Shan. Due to the short nature of the time series, it is not clear whether this change is a reversal of a long-term trend or simply interannual variability. (4) Some regions with stable or growing glaciers - such as the Karakoram and Kunlun Shan - see slightly later snowmelt seasons and longer snowmelt periods. It is likely that changes in the snowmelt regime of HMA account for some of the observed heterogeneity in glacier response to climate change. While the decadal increases in regional temperature have in general led to earlier and shortened melt seasons, changes in HMA's cryosphere have been spatially and temporally heterogeneous.