Gahar Lake’s future mapped using remote sensing and climate models

High in the Zagros Mountains of western Iran, at an elevation of 2,350 meters inside the Oshtorankuh Protected Area, one of the country’s most celebrated alpine lakes is quietly disappearing. A new study published in the open-access journal Heliyon reports that Gahar Lake, famous for its extraordinary transparency and striking turquoise water, has shrunk from approximately 190 hectares in 2001 to just 64 hectares in 2023, a loss of roughly 66 percent of its surface area in a little over two decades. The finding places the small mountain lake ahead of even Iran’s most notorious cases of lake collapse, including the better-known shrinkage of Lake Urmia, and suggests that high-altitude freshwater ecosystems may be more acutely vulnerable to hydrological stress than their larger lowland counterparts.

The research, led by Sanaz Vahidimanesh and colleagues including Ali Haghizadeh, Mahdi Soleimani-Motlagh, and Payam Amouzegari, is notable not only for the severity of the decline it documents but also for its methodological scope. The team fused more than two decades of imagery from three satellite systems, Landsat 7, Landsat 8, and the European Space Agency’s Sentinel-2, within Google Earth Engine, a cloud-based platform that provides access to over forty years of global satellite archives and parallel computing power. The authors report that the platform reduced processing times that would previously have taken days on local hardware to minutes, an increasingly important consideration for environmental monitoring in data-scarce regions where ground-based measurements are sparse or absent altogether.

To separate genuine water bodies from cloud, shadow, snow, and surrounding vegetation, the researchers computed three spectral indices and compared their performance. The Normalized Difference Water Index, or NDWI, exploits the contrast between green and near-infrared reflectance; the Modified NDWI, or MNDWI, substitutes a shortwave infrared band, which suppresses interference from built surfaces and vegetation; and the Normalized Difference Moisture Index, or NDMI, tracks moisture content in soil and plants rather than open water. The comparison proved decisive. NDWI, which is highly sensitive to soil moisture and riparian vegetation, produced a mean error of 22.4 hectares and even misclassified snowpack as water during a wet year, inflating the apparent lake area to an implausible 385 hectares in 2012. NDMI performed worst of all in dry periods, registering values close to zero when the lake was at its lowest. The MNDWI, applied to 10-meter Sentinel-2 imagery, emerged as the most reliable estimator, with a standard error of just 7.71 hectares during the Sentinel-2 era.

The trend statistics reinforce the picture of an ecosystem under sustained pressure. Applying the Mann-Kendall test, a nonparametric method robust to outliers and distribution-free time series, the team found a statistically significant declining trend at the 90 percent confidence level for the MNDWI-derived series from the Landsat 7 period, with a Z-statistic of −1.95. Sen’s slope estimator, which computes the median rate of change across all data pairs, put the average loss at −1.57 hectares per year. The spatial maps are equally stark: the lake’s autumn extent contracted from roughly 190 hectares in 2001 to 78 hectares in 2016, rebounded briefly to 123 hectares in 2017 after favorable snowfall, and then fell again to 64 hectares by 2023. Limited streamflow records from 2012 to 2019 show a corresponding downward trend in inflow.

What makes the study more than a retrospective is its forward-looking component. The researchers coupled the satellite record with projections of future climate using the LARS-WG stochastic weather generator, a semi-empirical model that produces synthetic daily precipitation, temperature, and solar radiation series by downscaling General Circulation Model output to the local scale. They used the EC-EARTH global climate model under RCP8.5, the high-emissions representative concentration pathway in which radiative forcing reaches 8.5 watts per square meter by 2100 and atmospheric carbon dioxide climbs toward 1,000 parts per million. The choice of a worst-case scenario was deliberate: the team wanted an upper bound of risk to inform proactive conservation, and they note that historical cumulative carbon emissions have tracked RCP8.5 closely through the 2020s, making it a realistic near-term reference.

The projections for 2021 to 2040 are troubling for a basin already in deficit. Spring precipitation, critical for feeding the mountain snowmelt that sustains the lake, is projected to decline by up to 15 percent in May relative to the previous climate decade. Minimum temperatures rise in every month of the year, and maximum temperatures climb sharply in late winter and early spring, with increases of 52 percent projected for March, 33 percent for November, 13 percent for April, and 10 percent for October. Warmer winters mean less snow accumulating on the high slopes of Oshtorankuh and more precipitation falling as rain, which runs off quickly rather than being released slowly as meltwater through spring. In the Zagros region, previous research has already documented annual snow cover declines of 0.25 to 0.68 percent, a trend that directly erodes the spring meltwater contribution to downstream lakes. The study’s authors argue that this mechanism, reduced snow accumulation combined with rising evaporative demand from warmer air, forms a self-reinforcing feedback loop that accelerates water loss from high-altitude lacustrine systems.

Climate, however, is only part of the story, and the researchers are candid about the limits of attribution. The Borujerd-Dorud plain downstream supports roughly 12,500 hectares of intensive irrigated agriculture, and during the dry season from May to September, surface water from the river and its tributaries feeding the lake basin is withdrawn for farming. Seasonal tourism within the protected area adds further pressure, along with waste disposal and disturbance of endemic species. The lake’s own hydrology is precarious: it receives an inflow of 280 liters per second against an outflow of 570 liters per second, a balance sustained only by underground springs and seasonal streams. Because there are no concurrent groundwater observations and the streamflow record is short, the team could not quantitatively separate the share of the lake’s decline caused by climate variability from that caused by human water use, a distinction that integrated socio-hydrological modeling efforts in South Asia have shown is essential. They recommend that future work apply process-based models such as SWAT or WEAP to disentangle the two drivers.

The lake’s origins are as dramatic as its decline. Geologists believe Gahar Lake was formed by a major seismic event, sitting as it does directly on the main active Zagros fault system, likely created through fault displacement and subsequent landslide activity. It comprises two basins, the Great Gahar, up to 28 meters deep and 1,800 meters long, and the smaller, shallower Small Gahar, two kilometers away. The lake’s remarkable clarity, allowing observers to see fish at the bottom, and its vivid turquoise color, produced by light refraction, have made it an ecological and aesthetic treasure, and in heavy snowfall years it freezes over entirely. Its two-part structure, fed by springs and snowmelt rather than large rivers, is precisely the kind of fragile, closed hydrological regime that the study suggests makes mountain lakes faster responders to climatic shifts than the large lowland lakes where most attention has been focused.

The researchers also acknowledge the technical uncertainties inherent in stitching together three different sensor families. Differences in spatial resolution between 30-meter Landsat and 10-meter Sentinel-2 imagery, variations in sensor calibration and spectral response functions, and persistent cloud and topographic shadow problems in mountainous terrain all complicate trend detection. Intriguingly, the combined NDWI series across all three satellites showed no significant trend even though the Landsat-only series showed a strong one, a discrepancy the authors attribute to these multi-sensor artifacts. They mitigated the problem with consistent cloud masking using the QA_PIXEL band for Landsat and the Scene Classification Layer for Sentinel-2, and with Level-2 surface reflectance products that have been atmospherically corrected. In the absence of local ground-truth measurements, the convergence of three independent indices and two sensor families on the same declining trend serves as internal validation.

The implications extend well beyond a single lake. A collapse of Gahar Lake would threaten endemic fish populations and habitat for migratory birds, degrade ecosystem services within a nationally protected area, and undercut the fishing and tourism economies of Lorestan Province, whose communities depend on the lake’s appeal. More broadly, the study offers a replicable template: an open, cloud-based, multi-sensor remote sensing workflow combined with statistically downscaled climate projections, accessible to water managers in regions that lack extensive monitoring infrastructure. The authors compare the 66 percent loss at Gahar with the 64 percent reduction documented at Iran’s Maharloo Lake, the 56 percent decline of Mighan Wetland, and far milder contractions at Turkish lakes such as Burdur, which lost about 42 percent of its area. Gahar’s proportional loss exceeds them all, a warning that some of the world’s most beautiful and least-studied freshwater systems are drying fastest, and that the coming two decades, under the emission trajectory the study models, will test whether proactive conservation can outpace the climate.