The team classified land cover for three time points: Landsat imagery from 2000 and 2012, and Sentinel-2 imagery from 2025. Rather than relying on manual interpretation, they applied a random forest algorithm, a machine learning method that builds hundreds of decision trees and merges their outputs to assign each pixel to a land cover class. Random forest classifiers have become a mainstay of remote sensing because they handle complex, non-linear relationships between spectral bands and land cover types while remaining robust to noise in training data. By moving the entire classification workflow into Google Earth Engine, the researchers could process large archives of satellite data in the cloud, avoiding the computational bottlenecks that traditionally limited regional land cover mapping.

The results document a landscape under growing anthropogenic pressure. Forest, grassland, and Afro-alpine ecosystems all declined over the twenty-five-year record, while farmland, built-up areas, and bare land expanded. Each of these conversions matters for different reasons. Forest loss removes the shaded, moisture-retaining habitats that support much of the region’s endemic fauna. Grassland conversion eliminates grazing grounds that sustain both wildlife and pastoral livelihoods. The contraction of Afro-alpine zones, which crown the massif and shelter specialized cold-adapted species, signals that even the most remote high-altitude habitats are not insulated from human expansion.

To assess ecosystem condition beyond simple land cover categories, the researchers turned to vegetation indices derived from the Moderate Resolution Imaging Spectroradiometer, or MODIS, sensor aboard NASA’s Terra and Aqua satellites. Two indices anchored the analysis: the Normalized Difference Vegetation Index, which exploits the contrast between red and near-infrared reflectance to estimate the density and vigor of green vegetation, and the Enhanced Vegetation Index, which corrects for atmospheric effects and soil background reflectance to provide a more stable measure of photosynthetic activity, particularly in areas with high biomass. Tracking these indices seasonally across the study period revealed a troubling trajectory: dry-season greenness declined, and the recovery of vegetation during the wet season weakened, especially after 2012. That timing suggests the ecosystem is losing its capacity to rebound between dry periods, a hallmark of accumulating stress.

The authors interpret these vegetation trends as evidence of combined pressure from land-use expansion and climatic variability. Weakened wet-season recovery is particularly significant because it indicates that rainfall is no longer translating into vegetation regrowth as effectively as it once did. Degraded soils, reduced organic matter, and the loss of deep-rooted perennial vegetation can all blunt the response of a landscape to favorable growing conditions. When that pattern coincides with the direct conversion of natural habitat to farmland and settlement, the result is a landscape that is simultaneously shrinking and functioning less efficiently.

Landscape metrics added a structural dimension to the analysis. Fragmentation increased across the study period, reflected in higher patch density, greater edge effects, and reduced connectivity among remaining habitat blocks. These are not abstract statistics. Patch density measures how many discrete fragments a habitat type has been broken into, and rising values mean continuous forest and grassland are being carved into smaller, more isolated islands. Edge effects describe the altered microclimatic and ecological conditions along fragment boundaries, where increased light, wind exposure, and human access degrade conditions for interior-dwelling species. Reduced connectivity means the stepping stones that once allowed animals and plants to move between habitat patches are disappearing, a process with well-documented consequences for gene flow, dispersal, and long-term population viability.

The biodiversity analysis drew on species occurrence records from the Global Biodiversity Information Facility and iNaturalist, two crowdsourced repositories that aggregate millions of georeferenced observations from scientists and citizen naturalists worldwide. When the researchers overlaid these records on their land cover maps, a clear pattern emerged: biodiversity observations are increasingly concentrated in remnant natural habitats. This concentration indicates habitat contraction, with species being squeezed into ever-smaller refugia as the surrounding matrix converts to agriculture and settlement. For the Bale Mountains, which support iconic endemic species such as the Ethiopian wolf and the mountain nyala, this pattern raises the vulnerability of species already restricted to narrow altitudinal and ecological ranges.

The authors are careful to note an important limitation: while the spatial associations between land cover change, vegetation stress, and biodiversity concentration are strong, causal relationships cannot be established from these data alone. Occurrence records from citizen science platforms are unevenly distributed, reflecting where people observe rather than where species actually occur, and the temporal resolution of biodiversity data does not always align with the satellite record. Nevertheless, the convergence of independent lines of evidence, from land cover classification to vegetation indices to species records, paints a coherent picture of an ecosystem under compounding stress.

What makes this study notable beyond its regional findings is the methodological integration it demonstrates. By combining cloud-based land cover mapping, time-series vegetation indices, landscape fragmentation metrics, and crowdsourced biodiversity data within a single analytical framework, the researchers show how modern geospatial tools can deliver ecosystem monitoring that would once have required years of fieldwork and expensive computation. Google Earth Engine has democratized planetary-scale analysis, allowing research teams with modest resources to process archives that span decades and continents. The Bale Mountains study illustrates how that capability can be directed at conservation questions with real urgency.

The implications for conservation are direct. The authors argue for strategies focused on habitat protection, restoring connectivity between fragmented remnants, and promoting sustainable land management practices that reduce pressure on remaining natural ecosystems. As the Bale Mountains face continued agricultural expansion and climatic variability, the satellite record now provides a baseline against which future change can be measured. Whether that record shows stabilization or continued decline will depend on decisions made in the coming years, both within Ethiopia and in the international conservation community that has long recognized the Bale Mountains as one of Africa’s most important biodiversity hotspots.

Subject of Research: Land use and land cover dynamics and their impacts on ecosystem health and biodiversity in the Bale Mountains, Ethiopia

Article Title: Geospatial and Google Earth Engine-based assessment of land use and land cover dynamics and its impacts on ecosystem health and biodiversity in the Bale Mountains, Ethiopia

Article References: Werati, G. T., Wolteji, B. N., & Sori, M. D. (2026). Geospatial and Google Earth Engine-based assessment of land use and land cover dynamics and its impacts on ecosystem health and biodiversity in the Bale Mountains, Ethiopia. Environmental Monitoring and Assessment, 198(9), Article 995. https://doi.org/10.1007/s10661-026-15847-4

Image Credits: AI Generated

DOI: 10.1007/s10661-026-15847-4

Keywords: Bale Mountains, Ethiopia, land use land cover change, Google Earth Engine, remote sensing, habitat fragmentation, biodiversity, NDVI, random forest classification, ecosystem health, Geospatial, Google

The scale of the land transformation emerges clearly from the satellite record. Between 2005 and 2024, agricultural land in and around Dehradun declined by 34 percent, forest cover fell by 14 percent, and grassland buffers diminished by 39 percent. Each of these losses matters for reasons that go beyond scenery. Farmland conversion removes the peri-urban soils that once absorbed rainfall and recharged aquifers; forest loss erodes slope stability and biodiversity habitat in a seismically and hydrologically sensitive Himalayan setting; and disappearing grasslands strip away the transitional zones that filter runoff and moderate local climate. The study attributes much of this conversion to sprawling, low-density development rather than compact growth, a pattern long associated with inefficient infrastructure provision and greater per-capita environmental burdens. In Dehradun’s case, the sprawl proceeded with little of the regulatory coordination that planners might have hoped the city’s master plan and smart city proposals would impose.

The most direct human consequence documented by the team is the deterioration of groundwater, the resource on which much of the city depends. Total dissolved solids in the city’s groundwater increased by more than 70 percent over the study period. Water hardness rose from 142.5 milligrams per liter to 197.4 milligrams per liter, climbing toward and past thresholds that consumers and regulators find concerning. Nitrate concentrations were found to frequently exceed permissible limits, a classic signature of sewage infiltration and fertilizer leaching, and the researchers recorded widespread microbial contamination across monitored wells. For a city perched above porous Himalayan foothill aquifers, these numbers describe more than a technical problem. They indicate that the same ground on which houses and roads now stand is transmitting waste-derived contaminants directly into the drinking water supply, with potential consequences for public health that the indicators only begin to capture.

Air quality tells a parallel story of growth outpacing environmental capacity. Across Dehradun’s major urban corridors, PM10 concentrations increased by 45 to 60 percent and PM2.5 concentrations by 40 to 80 percent over the study period. Particulate matter at these sizes penetrates deep into the lungs and enters the bloodstream, and epidemiological research has linked long-term exposure to cardiovascular and respiratory disease. In a valley city where topography can trap pollutants during temperature inversions, rising particulate loads compound an already vulnerable atmospheric setting. The study’s statistical modeling makes the connection explicit: built-up expansion correlated strongly with PM2.5 levels, with a correlation coefficient of 0.84, among the strongest associations reported in the analysis. In other words, the spatial pattern of new construction does not merely coincide with dirtier air; the two move together closely enough that land-use policy is, in effect, air-quality policy.

Water and air are not the only media under stress. Noise monitoring revealed that levels exceeded regulatory standards at more than 65 percent of the locations surveyed, a burden concentrated along the transport corridors where traffic, construction, and commercial activity converge. Chronic noise exposure is associated with sleep disturbance, elevated stress hormones, and cardiovascular risk, yet it is among the least regulated dimensions of urban environmental quality in fast-growing Indian cities. Meanwhile, the city’s material flows have swelled beyond its capacity to process them. Municipal solid waste generation reached 280 tonnes per day, but processing efficiency stands at only 70 percent, meaning roughly a third of the daily waste stream is unaccounted for in terms of treatment, with predictable consequences for land, water, and air. Sewage treatment capacity, likewise, remains below the volume of wastewater the city actually generates, leaving untreated effluent to find its way into groundwater and surface channels.

The statistical backbone of the study gives these observations their force. By correlating built-up expansion against a suite of environmental indicators, the researchers demonstrated that degradation is not randomly distributed but systematically tied to the geography and pace of development. Groundwater TDS correlated with built-up growth at 0.82, and nitrate concentration at 0.77, alongside the 0.84 coefficient for PM2.5. In environmental research, correlation coefficients of this magnitude across independent indicators are notable, because they suggest a common driver rather than a collection of unrelated local problems. The authors frame this as evidence that unregulated spatial growth functions as an integrated stressor: the same expansion that paves over recharge zones also concentrates traffic, generates waste, and outpaces the sewerage network. Addressing any single indicator in isolation, the analysis implies, will fail so long as the underlying land-use trajectory remains unchecked.

Set against the Sustainable Development Goals, the picture darkens further. SDG 11 calls for cities that are inclusive, safe, resilient, and sustainable, with specific targets covering air quality, waste management, water and sanitation, and access to green public space. The study finds substantial deviations from these targets in Dehradun on nearly every dimension it measured: particulate pollution rising rather than falling, waste processing short of full coverage, sewage treatment lagging behind generation, water quality deteriorating, and the green and agricultural land base contracting. The researchers’ assessment of urban infrastructure performance, drawing on census data, municipal datasets, and environmental monitoring records, indicates that infrastructure augmentation has simply not kept pace with the physical growth of the city. This is a pattern familiar across Himalayan hill towns, where steep terrain, limited flat land, and ecological fragility magnify the costs of every unplanned extension of the urban edge.

The methodological approach itself carries lessons for other cities in the region. By combining freely available Landsat and Sentinel imagery with supervised classification and post-classification change detection, the team produced a two-decade change record without requiring expensive bespoke data collection. Anchoring the satellite analysis in census data, municipal records, and environmental monitoring allowed the researchers to move from mapping to causally suggestive statistics, connecting the where of urban growth to the what of environmental harm. For city governments with limited analytical budgets, this kind of indicator-based geospatial framework offers a replicable template: it can flag which corridors are degrading fastest, which aquifers are failing, and which SDG targets are most off-track, providing an evidence base for prioritizing interventions before conditions become irreversible.

The authors’ prescription is correspondingly integrated rather than piecemeal. They call for strengthened land-use regulation, ecological conservation of the remaining forest and grassland buffers, augmentation of water, sewage, and waste infrastructure, and climate-resilient urban planning to steer Dehradun back toward sustainable development. None of these measures is novel in itself; what the study adds is quantitative proof that in Dehradun they are interdependent. Every year of continued sprawl simultaneously worsens the air residents breathe, the water they drink, and the noise they endure, while eroding the natural systems that could buffer the city against floods, landslides, and a warming climate. As Himalayan foothill cities across India and neighboring countries absorb some of the world’s fastest urban growth, Dehradun’s two-decade satellite record stands as a clear-eyed account of what unmanaged expansion costs, and a data-driven argument for why the next two decades must be different.

Subject of Research: Urban sprawl and its environmental impacts in Dehradun, India, assessed through geospatial and environmental indicator analysis against SDG 11

Article Title: Assessing urban sprawl, environmental vulnerability, and SDG 11 alignment using geospatial and environmental indicator-based analysis in Dehradun, India

Article References: Bharmoria, R., Nainwal, R., Chaturvedi, P., & Belwal, H. (2026). Assessing urban sprawl, environmental vulnerability, and SDG 11 alignment using geospatial and environmental indicator-based analysis in Dehradun, India. SN Social Sciences, 6(10), Article 541. https://doi.org/10.1007/s43545-026-01827-5

Image Credits: AI Generated

DOI: 10.1007/s43545-026-01827-5

Keywords: urban sprawl, Dehradun, land use land cover change, remote sensing, groundwater quality, air pollution, noise pollution, SDG 11, sustainable cities, Himalayan foothills, environmental indicators, urban planning

The research team, led by scientists at the University of Barishal with a collaborator at the University of Dhaka, used the Google Earth Engine platform to process multi-temporal Landsat imagery spanning 1990 to 2024, drawing on Landsat 5 Thematic Mapper data for the earlier epochs and Landsat 8 OLI/TIRS data for recent years. All images were restricted to the peak summer period from March to June with cloud cover below ten percent, ensuring that comparisons across decades captured the same seasonal window. Land cover was mapped into four classes using a Random Forest classifier, with spectral indices such as the Normalized Difference Vegetation Index, the Normalized Difference Built-up Index, the Modified Normalized Difference Water Index and a bare-soil index added as input bands to sharpen class separation. The number of decision trees in the classifier was tuned independently for each year, and validation was rigorous: overall classification accuracy exceeded 92 percent in every epoch, Kappa coefficients stayed above 89 percent, and ten-fold cross-validation scores topped 95 percent.

The land-cover results are stark. Built-up area expanded by 92.37 square kilometers, an increase of 194.54 percent, spreading from the dense southern wards of the city toward the north. This growth came almost entirely at the expense of vegetation, which fell by 45.99 percent, and water bodies, which collapsed by 75.57 percent. The losses were not gradual or uniform. Vegetation declined most sharply between 1994 and 1999, a period of major infrastructure development, while water bodies suffered their two worst losses in 1990 to 1994 and again between 2019 and 2024, when half of the remaining surface water disappeared in just five years. A large water body in the city’s central-eastern zone, visible in the 1990 imagery, was progressively filled with bare land as development pushed into previously aquatic terrain. Bare land and other surfaces, often a signature of active land-clearing and sand-filling ahead of construction, increased by 228 percent over the study period.

Land surface temperature, retrieved from the thermal bands using a radiative transfer-based method with emissivity corrections derived from vegetation cover, tells the corresponding story. In 1990, nearly 78 percent of the study area registered temperatures below 25 degrees Celsius and no location exceeded 35 degrees. By 2024, the sub-25-degree zones had vanished entirely, areas above 40 degrees covered 70.06 percent of the city, and 4.44 percent of the land surface exceeded 45 degrees. The turning point came between 2003 and 2007, when the share of the city hotter than 30 degrees jumped from 6.41 percent to 76 percent in a single epoch. The team validated its temperature retrievals against NASA’s MODIS Terra product, finding correlation coefficients between 0.71 and 0.79, and confirmed that no abrupt discontinuity appeared at the 2007-to-2014 transition between the two Landsat sensors, indicating that the observed warming signal, which exceeds 15 degrees Celsius in urban areas, dwarfs any cross-sensor calibration differences.

Each land-cover class warmed, but not equally. Water bodies, once the city’s thermal anchor with a mean surface temperature of 22.69 degrees in 1990, climbed to 36.44 degrees by 2024, a sign that their cooling efficiency is eroding under cumulative thermal stress. Vegetation, still the coolest class, rose from a mean of 23.96 degrees to 39.28 degrees. Urban surfaces escalated from 25.99 degrees to a mean of 42.48 degrees, with a maximum of 52 degrees, while bare land proved the hottest category of all, averaging 42.23 degrees and peaking at a scorching 53.12 degrees. The widening thermal gap between natural and built surfaces is the classic fingerprint of an intensifying surface urban heat island, in which concrete, asphalt and exposed soil absorb and re-emit solar radiation that vegetation and water once dissipated through evapotranspiration and high thermal inertia.

The heat island analysis itself, calculated as the difference between mean urban surface temperature and that of an equally sized rural buffer ring around the city, revealed a troubling diurnal shift. Using separate daytime and nighttime MODIS observations from 2003 to 2024, the researchers found that both day and night heat island intensity rose steadily, but in recent years the nighttime values surpassed the daytime ones. This means Dhaka is no longer merely heating up under the sun; it is failing to cool down after dark. The likely culprits are the high thermal mass of building materials that store heat through the day and release it slowly at night, combined with anthropogenic heat from traffic, industry and air conditioning. For residents, persistent nighttime heat is particularly dangerous because it denies the body the overnight recovery period that normally protects against heat stress.

To move beyond description and identify causes, the team sampled 1,000 random points across the 2024 scene and ran both Pearson correlation analysis and an Ordinary Least Squares multiple regression, with Variance Inflation Factor diagnostics to guard against multicollinearity. The built-up index emerged as the strongest positive correlate of surface temperature, with a correlation coefficient of 0.869, while vegetation showed a strong negative correlation of minus 0.749. Building height, nighttime light density and population density all correlated positively with heat, and proximity to major roads and industrial zones added further warming. The full regression model explained 80.1 percent of the variation in land surface temperature, with the built-up index, population density and nighttime lights as the dominant positive drivers, and distance to water and vegetation both raising temperatures as those cooling features receded. One intriguing anomaly, a positive regression coefficient for the vegetation index despite its negative bivariate correlation, was traced to mixed-pixel effects in Dhaka’s hyper-dense fabric, where small pocket gardens and roadside trees are spectrally entangled with surrounding hot concrete, masking their independent cooling signal at medium resolution.

The authors frame their findings as a diagnostic blueprint rather than a lament. They recommend adapting the 3-30-300 rule as a spatial benchmark for Dhaka’s neighborhoods, so that every resident can see at least three trees from home, every ward achieves at least 30 percent tree canopy cover, and everyone lives within 300 meters of quality green space. They call for strict enforcement against the illegal filling of wetlands and canals, restoration of the Buriganga, Turag, Shitalakshya and Balu rivers and internal drainage channels, and deployment of Sponge City infrastructure such as retention ponds and permeable pavements to boost evaporative cooling. Cool-roof ordinances with reflective coatings for industrial zones and high-density hotspots, plus green walls and passive ventilation retrofits in the historic neighborhoods of Old Dhaka, round out the mitigation package.

Because the entire workflow runs on free, open-access satellite archives and publicly available demographic datasets within Google Earth Engine, the framework is deliberately replicable for other rapidly urbanizing cities of the Global South facing similar pressures. With the United Nations projecting that 68 percent of humanity will live in urban areas by 2050, and with Dhaka’s own population approaching twenty million, the study offers both a warning and a method: the drivers of deadly urban heat can be measured, ranked and targeted, but only if cities act before their last cool zones disappear the way Dhaka’s did.

Subject of Research: Spatiotemporal dynamics and drivers of surface urban heat island intensity in Dhaka, Bangladesh, assessed with geospatial remote sensing

Article Title: Spatiotemporal dynamics and drivers of surface urban heat island intensity in Dhaka, Bangladesh using geospatial remote sensing

Article References: Saha, S. C., Satil, U. A., Paul, N., Nath, D. C., Nepa, F. R., Alam, M. M. T., & Goswami, S. (2026). Spatiotemporal dynamics and drivers of surface urban heat island intensity in Dhaka, Bangladesh using geospatial remote sensing. Discover Geoscience, 4(1), Article 396. https://doi.org/10.1007/s44288-026-00767-0

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00767-0

Keywords: urban heat island, land surface temperature, Dhaka, remote sensing, Google Earth Engine, land use land cover change, Landsat, urbanization, climate resilience, vegetation loss, water bodies, Bangladesh

The research, led by Rajib Pegu of Sankardeva Mahavidyalaya with colleagues from institutions across Assam, Arunachal Pradesh, and New Delhi, harnessed the archival power of the Landsat program, which has been imaging the Earth’s surface since the 1970s. By processing multi-temporal Landsat imagery on the Google Earth Engine platform, a cloud-computing service that has revolutionized large-scale environmental analysis, the team was able to classify land cover and retrieve surface temperatures for multiple epochs spanning 1995 to 2024 without downloading or manually processing petabytes of raw data.

The technical pipeline behind the study reflects the current state of the art in remote sensing. Land use and land cover maps were generated using a Random Forest classifier, a machine learning algorithm that builds hundreds of decision trees from training samples and votes on the most probable class for each pixel. Random forest classification has become a workhorse of land cover mapping because it handles noisy, high-dimensional satellite data robustly and resists overfitting. For temperature retrieval, the researchers turned to the radiative transfer approach, converting the thermal infrared radiation measured by Landsat’s sensors into true land surface temperature, with corrections for atmospheric effects and for the emissivity of different surfaces, which they estimated using the Normalized Difference Vegetation Index, a standard proxy for vegetation density.

Land surface temperature is not the same as the air temperature reported by weather stations. Satellites measure the radiative skin of the ground itself: rooftops, asphalt, bare soil, crop fields, forest canopies, and water. Because these surfaces absorb and release heat very differently, land surface temperature is an extraordinarily sensitive indicator of what a landscape is made of. Dark, impervious materials like asphalt and concrete absorb solar radiation during the day and re-emit it slowly, while vegetated surfaces cool themselves through evapotranspiration, the process by which plants release water vapor and shed heat. Water bodies, with their enormous heat capacity, remain the coolest surfaces of all.

The Guwahati data bore this physics out with striking clarity. Built-up areas recorded the highest mean surface temperature of any land cover class, at 36.11 degrees Celsius, while water bodies registered the lowest at 27.29 degrees Celsius, a gap of nearly nine degrees across the same city on the same days. This thermal stratification is the surface signature of the urban heat island effect, the phenomenon first described by Luke Howard in London in the early nineteenth century and quantified for modern cities by climatologist T. R. Oke in the 1970s. What makes the Guwahati study notable is that it documents this effect unfolding in a mid-sized city of the Global South, where urban growth is fastest and adaptive capacity is often thinnest.

To establish that the warming was real and not an artifact of a few unusually hot years, the team applied the Mann-Kendall trend test, a non-parametric statistical method that detects monotonic trends in time series without assuming any particular distribution, together with Sen’s slope estimator, which provides a robust measure of the magnitude of change. The result was a statistically significant warming trend at the conventional p-value threshold of 0.05, confirming that Guwahati’s surface has been heating steadily across the three-decade record rather than merely fluctuating.

The statistical core of the study came from a pixel-level analysis of the 2024 dataset. The researchers drew 5000 randomly sampled pixels and examined how vegetation greenness, expressed through the NDVI, related to surface temperature. Pearson’s correlation analysis revealed a significant inverse relationship, with a correlation coefficient of negative 0.492, meaning that greener pixels were systematically cooler. Linear regression showed that vegetation alone explained approximately 24.2 percent of the spatial variability in surface temperature, a substantial fraction given how many other factors, from elevation to soil moisture to building geometry, also shape the thermal landscape. A one-way analysis of variance confirmed that temperature differences among land cover classes were highly significant, with a p-value below 0.001.

These numbers carry a message that extends well beyond one Indian city. Guwahati sits at the gateway to Northeast India, a region of extraordinary biodiversity, and its growth mirrors that of hundreds of secondary cities across Asia and Africa where urban expansion is consuming wetlands, farmland, and forest fringes. Previous work on Guwahati had already flagged declining groundwater levels and shifting heat island patterns, and studies from Pune, Jaipur, Delhi, Mumbai, and the East Kolkata Wetlands have documented similar dynamics. The new study adds a rigorous, three-decade quantitative baseline that links landscape change to thermal change in a single coherent framework.

The implications for planning are direct. Because vegetation explains roughly a quarter of the spatial variation in surface temperature, protecting and expanding urban green cover is not merely an aesthetic preference but a measurable cooling strategy. The authors point to the conservation of urban vegetation, wetlands, and other natural landscapes as priorities, and frame their findings as evidence for climate-responsive urban planning aligned with the United Nations Sustainable Development Goals on sustainable cities, climate action, and life on land. In a warming century, the coolest surfaces a city owns are the ones it has not yet paved.

The study also demonstrates how the tools of planetary observation have been democratized. A team of researchers at regional Indian institutions, using free Landsat archives and the Google Earth Engine platform, produced an analysis that would once have required a supercomputing facility and a large budget. As urbanization accelerates across the Global South, this kind of accessible, reproducible satellite monitoring offers city governments a way to see, in near real time, the thermal consequences of every hectare of green space lost, and every hectare of concrete gained.

Subject of Research: The relationship between land use land cover change and land surface temperature dynamics in Guwahati, India, analyzed with Landsat imagery and Google Earth Engine

Article Title: Urbanization-driven thermal changes: a remote sensing approach to land surface temperature (LST) and land use land cover (LULC) dynamics in Guwahati, Assam, India

Article References: Pegu, R., Mazumder, T., Borah, H., Bora, K., & Hazarika, P. (2026). Urbanization-driven thermal changes: a remote sensing approach to land surface temperature (LST) and land use land cover (LULC) dynamics in Guwahati, Assam, India. Environmental Science and Pollution Research, 33(28), 14266-14281. https://doi.org/10.1007/s11356-026-38169-x

Image Credits: AI Generated

DOI: 10.1007/s11356-026-38169-x

Keywords: land surface temperature, land use land cover change, urban heat island, Guwahati, remote sensing, Landsat, Google Earth Engine, random forest classification, NDVI, urbanization, Mann-Kendall test, Assam

The research team, led by Samkelisiwe Khanyile, combined post-classification change detection in a geographic information system with chi-square statistical testing and a spatial clustering index. Their raw material was the South African National Land Cover dataset, built from Landsat 5 Thematic Mapper imagery for 1990 and Landsat 8 Operational Land Imager imagery for 2020, both at a 30-meter resolution. The original 73 and 72 land cover classes were collapsed into 12 comparable categories, allowing every one of nearly 2.9 million pixels to be traced from its 1990 state to its 2020 state. The resulting transition matrix, a 12-by-12 table of conversions, became the evidentiary backbone for asking whether landscape change in a post-mining region follows logic, chance, or policy failure.

The statistical verdict was unambiguous. The chi-square test returned a value of 11,319,251.69 with 121 degrees of freedom and a p-value below 0.001, and 141 of the 144 possible transitions proved significant. Cramér’s V, a measure of effect size, reached 0.60, indicating that the transitions substantially reshaped the composition of the landscape rather than nudging it at the margins. In plain terms, the land did not drift randomly into new uses. Institutional, economic, and planning forces drove the conversions, which means the inequities embedded in those conversions are also the product of decisions, not accidents.

The headline numbers tell a story of paradox. On one hand, 54.53 square kilometers of former mining land, 37.8 percent of the mining footprint recorded in 1990, successfully transitioned to natural vegetation, a conversion that occurred far more often than chance would predict and that points to genuine revegetation efforts. On the other hand, the wider landscape degraded. Natural vegetation shrank by 150.33 square kilometers, planted vegetation plummeted by 47.6 percent, waterbodies and wetlands declined by more than 12 percent, and barren land surged by 376.4 percent. Site-level rehabilitation, in other words, could not counteract landscape-scale ecological loss, a finding that echoes international evidence that point-source restoration rarely offsets cumulative regional damage.

Urbanization compounded the pressure. Built-up formal areas grew by 60.74 square kilometers, commercial areas nearly doubled with a 37.41 square kilometer expansion, and informal settlements exploded by 42.91 square kilometers, an astonishing 847.89 percent increase. Informal settlements also showed a persistence rate of only 45.42 percent, meaning they are constantly forming, dissolving, and reforming. Crucially, the spatial clustering index derived from the chi-square residuals reached 99,240.66, far above the 1,000-to-5,000 range expected from random distributions. The changes concentrated in Soweto, Tsakane, Simunye, Daveyton, and Mohlakeng, townships historically designated for Black residents under apartheid and deliberately sited near mining operations.

That geographic concentration is where the study’s environmental justice argument becomes sharpest. Environmental justice scholarship has long held that hazards are disproportionately placed near marginalized populations, but demonstrating this with quantitative landscape evidence has been rare in mineralized urban contexts of the Global South. Here, three lines of evidence converge: the transitions are systematic, they cluster in previously disadvantaged areas, and they include residential development on contaminated land. The belt hosts more than 270 tailings storage facilities covering roughly 400 square kilometers, some of them radioactive because of uranium co-occurring with gold. Provincial authorities identified 157 of 374 mine residue sites as radioactive, spanning 220 square kilometers.

Perhaps the most damning finding concerns the buffer zones that were supposed to keep people away from that contamination. In 2003, the Gauteng Department of Agriculture and Rural Development established pollution buffers around mine residue sites, later codified at distances of 500 to 1,500 meters under the 2012 Gauteng Pollution Buffers Policy, amended in 2017. Yet the analysis found that 0.80 square kilometers of mining land was converted to formal and informal built-up areas, placing residents directly adjacent to potentially contaminated ground despite explicit regulatory protection. The study also documents the reverse flow: 41.55 square kilometers of natural vegetation was converted to mining-related land use, showing that extraction continued consuming green space even as the sector contracted overall, with mining land declining only marginally from 144.25 to 143.28 square kilometers.

The authors frame this as a policy-implementation gap with two possible readings, both unjust. Either mining companies successfully defended their land holdings through regulatory power, monopolizing urban land while housing-poor populations are pushed onto the highest-risk zones, or mining actors strategically permitted settlement in buffer zones to monetize their land assets. Either way, the injustice is both spatial and temporal: apartheid-era mining footprints continue to structure environmental exposure and development opportunity three decades after the regulatory regime that produced them was formally dismantled. South Africa’s post-apartheid legal architecture, including the Constitution’s environmental rights clause, the National Environmental Management Act, the Mineral and Petroleum Resources Development Act with its mandatory rehabilitation provisions, and the Spatial Planning and Land Use Management Act, has not translated into protection on the ground.

The sustainability implications reach beyond South Africa. The findings show direct misalignment with Sustainable Development Goal 11 on sustainable cities, since urban growth in the mining belt has been predominantly informal and hazard-adjacent rather than planned, and with SDG 15 on terrestrial ecosystems, given the vegetation losses and barren land expansion. Gauteng’s position on a watershed divide between the Crocodile, Olifants, and Vaal river systems makes the wetland losses particularly troubling, as the landscape loses its capacity to retain and filter water precisely when demand is escalating. Acid mine drainage, soil contamination, and radioactive dust remain persistent threats that satellite imagery cannot capture, which the authors acknowledge as a limitation alongside their reliance on two time points and pre-classified 30-meter data with 88 to 96 percent accuracy.

What distinguishes this study is its replicable methodological template: transition matrices validated by chi-square testing, effect sizes, and clustering indices, all benchmarked against regulatory buffer zones. Because the changes are systematic, the authors argue, they can be deliberately reversed through targeted intervention. Their prescriptions include stronger regulatory enforcement with meaningful penalties, affordable housing provision to relieve pressure on contaminated land, participatory rehabilitation planning that centers community voices, and accountability mechanisms grounded in independently verifiable landscape-scale evidence rather than self-reported site compliance. A periodic land cover monitoring framework benchmarked against the provincial buffer policy could, they suggest, turn satellite pixels into enforceable metrics. The Witwatersrand’s golden scar will not heal on its own, but for the first time its growth, its healing patches, and its human costs have been measured with a precision that regulators can no longer ignore.

Subject of Research: Mining-driven land use and land cover change and environmental justice along the Witwatersrand mining belt, Gauteng, South Africa, 1990–2020

Article Title: Mining Legacies and Environmental Injustice: Systematic Mining-Led LULC Change in the Witwatersrand Mining Belt, Gauteng, South Africa (1990–2020)

Article References: Khanyile, S., Tsoriyo, W., & Mahamuza, P. (2026). Mining Legacies and Environmental Injustice: Systematic Mining-Led LULC Change in the Witwatersrand Mining Belt, Gauteng, South Africa (1990–2020). Environmental Management, 76(10), Article 336. https://doi.org/10.1007/s00267-026-02543-9

Image Credits: AI Generated

DOI: 10.1007/s00267-026-02543-9

Keywords: land use land cover change, mining, environmental justice, Witwatersrand, Gauteng, remote sensing, GIS, chi-square analysis, informal settlements, mine rehabilitation, tailings storage facilities, sustainable development goals

The research team, led by Gaositwe Lillian Modutlwe of the University of Botswana together with colleagues at Jagannath University in Bangladesh and the University of Nebraska-Lincoln, chose Gaborone for good reason. The city sits between the Kgale and Oodi Hills in southeastern Botswana, west of the Notwane River, and covers approximately 190 square kilometers at an altitude of about 1,000 meters. With a mean annual rainfall of just 500 millimeters, it is a quintessentially semi-arid urban environment, which makes the loss of vegetated land particularly consequential. According to Botswana’s 2021 Central Statistical Report, the city is home to 246,325 people, more than 10 percent of the country’s entire population, packed into 58,476 households. Much of that population arrived through migration from the rest of the country, drawn by Gaborone’s role as the national economic engine, with industries ranging from diamond cutting and polishing to textiles, brewing, and printing.

To capture how this growth reshaped the land, the team turned to the Landsat archive, the longest continuous record of Earth’s surface collected by the US Geological Survey since 1972. They downloaded imagery for 1991, 2001, 2011, and 2021, deliberately aligning each snapshot with a national census year so that land change could later be compared against demographic data. Cloud and shadow masking was applied across the different Landsat sensors, from the Thematic Mapper of the early 1990s to the Operational Land Imager of today, and the study area was clipped to Gaborone’s administrative boundary. All of the data acquisition, preprocessing, and initial classification ran inside Google Earth Engine, a cloud-computing platform that has become a workhorse for large-scale environmental monitoring because it avoids the overfitting pitfalls that often plague local machine learning setups.

The classification itself relied on random forest, an ensemble algorithm that builds hundreds of decision trees and lets them vote on what each pixel represents. The researchers trained the model on six land cover classes: water, built-up, agriculture, urban forest, vegetation, and bare ground. Crucially, they fed the classifier not just raw spectral bands but six derived spectral indices, each designed to highlight a specific surface property. The Enhanced Vegetation Index captured vegetation vigor while accounting for topographic effects, the Normalized Burn Ratio flagged burnt areas, the Normalized Difference Moisture Index identified agricultural zones under moisture stress, the Normalized Difference Water Index isolated open water, the Normalized Difference Built-up Index delineated urban surfaces, and the Normalized Difference Bareness Index improved detection of exposed soil. A 30-meter digital elevation model from the Shuttle Radar Topographic Mission added terrain information to the mix.

The random forest configuration was tuned through iterative testing, ultimately running 300 decision trees with the square root of the available predictor variables considered at each split. Validation used 600 randomly selected reference points per map, with ground truth drawn from high-resolution imagery in Google Earth Pro going back to 1991. The results were strikingly good: overall accuracy exceeded the widely accepted 80 percent threshold in every year, peaking at 97 percent with a kappa coefficient of 0.96 in 1991, and dipping only to 87 percent with a kappa of 0.84 in 2021. Water was classified perfectly in every epoch, while bare ground proved the hardest class, frequently confused with agriculture and built-up surfaces, a common problem in semi-arid landscapes where bare soil and fallow fields share nearly identical spectral signatures.

With the historical maps in hand, the team then did something most previous Gaborone studies had not: they predicted the future. For this they used a one-dimensional convolutional neural network implemented in TensorFlow Keras on Google Colab. The classified maps from the four census years were exported from Earth Engine and fed into a sequential model consisting of six convolutional layers, each paired with ReLU activation, followed by max pooling to compress dimensionality and extract dominant temporal features. A dropout rate of 0.2 after the second convolutional layer guarded against overfitting, and the flattened outputs passed through two fully connected dense layers before a softmax output layer assigned class probabilities. The choice of a 1D-CNN was deliberate: unlike recurrent architectures, it extracts local temporal dependencies from short time series with far lower computational cost, and it performs well even when training data is limited, which matters when your entire history consists of four observation dates.

The network’s performance held up under scrutiny. Overall accuracy across the modeled years ranged from 0.82 to 0.88, comfortably above the 80 percent benchmark. Water again achieved precision of 1, built-up areas scored consistently above 0.89 on precision, recall, and F1 measures, and vegetation was reliably predicted with scores between 0.82 and 1. The weak spots were bare ground, which managed a precision of only 0.33 in 1991, and agriculture, which dipped to 0.43 in 2011, both largely attributable to limited training samples for those classes. When the researchers compared predicted against observed land areas for 2001, 2011, and 2021, the agreement was close, with the largest discrepancy for built-up land being just 4.35 square kilometers in 2021, and bigger deviations for agriculture and bare ground in 2011, likely stemming from spectral similarity between those classes and uneven sample sizes.

The projections for 2031 tell a clear story. The model expects built-up area to grow from roughly 76.8 square kilometers, or 31 percent of the city, to 99.6 square kilometers, or 41 percent, with expansion continuing the historical pattern of spreading northward from the area above the Gaborone dam. Vegetation is projected to decline further, and bare ground is expected to shrink from about 16.1 to 8.8 square kilometers, presumably as even the empty fringes get absorbed into the urban fabric. The gain and loss analysis shows built-up land increasing by 7.5 percent from 1991 to 2001, 6.3 percent from 2001 to 2011, and 2.4 percent from 2011 to 2021, a decelerating trend that the model expects to reverse with a 9.3 percent gain by 2031. The steepest losses fell on urban forest, down 8.9 percent between 2001 and 2011, and vegetation, down 8.8 percent between 1991 and 2001.

Perhaps the most compelling finding is the statistical link between people and pavement. Linear regression and Pearson correlation analysis revealed an almost perfect positive relationship between population growth and built-up expansion, with a correlation coefficient of 0.98, while vegetation showed a strong negative correlation of -0.95 and agriculture -0.73. The built-up class gained about 0.0003 units of area for every unit of population increase, and vegetation lost area at essentially the same rate. The analysis also uncovered relationships among the land classes themselves: built-up expansion correlated negatively with vegetation at -0.97, confirming that the city is growing directly at the expense of green cover, while bare ground and water showed a negative relationship of -0.83, suggesting that some areas classified as bare ground were formerly water bodies that dried out.

The authors are candid about the limitations. Four observation dates is a thin temporal foundation for any deep learning model, and the population variable served as a proxy for a bundle of interacting socioeconomic processes rather than being explicitly integrated into the prediction itself. Future work, they suggest, should incorporate longer time series, higher-resolution imagery, and ancillary data such as infrastructure networks and socioeconomic indicators, and should embed demographic drivers directly into the modeling pipeline. Still, the implications of the 2031 forecast are hard to ignore. Expanding impervious surfaces in an already hot, dry city threaten to intensify urban heat islands, alter runoff patterns, accelerate land degradation, and fragment the remaining pockets of indigenous vegetation, with knock-on effects for biodiversity, water security, and public health. The study’s value lies in giving Botswana’s land managers a quantified, spatially explicit preview of where those pressures will land, and a demonstrated, transferable machine learning framework for doing the same in other rapidly growing semi-arid cities.

Subject of Research: Machine learning-based modelling of historical and future land use and land cover change in Gaborone, Botswana

Article Title: Modelling historical and future land use and land cover change using random forest and TensorFlow 1D-convolution neural networks in Gaborone, Botswana

Article References: Modutlwe, G. L., Kombani, L., Rahman, M. M., & Rana, M. M. S. P. (2026). Modelling historical and future land use and land cover change using random forest and TensorFlow 1D-convolution neural networks in Gaborone, Botswana. Discover Geoscience, 4(1), Article 296. https://doi.org/10.1007/s44288-026-00662-8

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00662-8

Keywords: land use land cover change, random forest, 1D-CNN, TensorFlow, remote sensing, Gaborone, Botswana, urbanization, Landsat, Google Earth Engine, population growth, semi-arid cities

The study, published in Discover Geoscience, assessed ecological environmental quality in Rupandehi District from 1993 to 2023 using the Remote Sensing Ecological Index, a composite measure built entirely from openly available Landsat satellite data. The researchers, led by Gaurav Parajuli and including collaborators from the University of Kansas, Tribhuvan University, and other institutions, processed their imagery in Google Earth Engine, a cloud-based platform that makes it feasible to analyze decades of satellite data without downloading enormous archives. Their analysis code has been shared in a public repository, making the entire workflow reproducible by anyone with an internet connection.

The Remote Sensing Ecological Index, first proposed to overcome the limitations of single-indicator assessments, combines four ecological dimensions into a single score ranging from zero to one. Greenness is captured by the Normalized Difference Vegetation Index, which exploits the fact that healthy vegetation reflects near-infrared light strongly while absorbing red light. Wetness is derived from the tasseled cap transformation of satellite reflectance, a mathematical rotation of the spectral bands that isolates the moisture content of soil and vegetation. Dryness is measured by averaging a soil index and an index-based built-up index, effectively flagging impervious and barren surfaces. Heat is represented by land surface temperature, retrieved from the thermal bands of Landsat 5 and Landsat 8 using a single-channel algorithm that corrects for atmospheric radiance and surface emissivity.

Because these four indicators are measured on different scales, the team normalized each one before feeding them into a Principal Component Analysis, a statistical technique that derives variable weights from the data itself rather than relying on subjective assignments. The first principal component explained between 86 and 92 percent of the total variance in every year analyzed, an unusually stable result that confirmed the four indicators were coherently capturing a single underlying gradient of ecological quality. Crucially, the loading structure behaved exactly as theory predicts: greenness and wetness contributed positively, while heat and dryness acted as negative influences, in all four study years.

The resulting index revealed a non-linear trajectory. The district-wide mean RSEI rose from 0.59 in 1993 to 0.635 in 2004, then fell to 0.55 in 2013, before recovering to 0.67 in 2023, an overall improvement of roughly 13 percent. The mid-period decline coincided with a wave of migration from mountainous districts such as Palpa, Gulmi, and Syangja, along with road construction on the Siddhartha Highway, expanding impervious surfaces, and deforestation. Between 2004 and 2013, ecological deterioration swept across 700.82 square kilometers, about 53 percent of the study area, while only 24 percent improved. By the final decade, the balance had shifted: improvement covered 35 percent of the district and deterioration 33 percent, and the lowest quality class became virtually nonexistent by 2023.

Underneath the district average, the land itself was being rearranged. Built-up area expanded from just 13.32 square kilometers, less than one percent of the district, in 1993 to 156.16 square kilometers, over ten percent, in 2023. Agricultural land declined from 838.91 to 787.00 square kilometers over the same period, and water bodies shrank from 18.02 to 11.80 square kilometers. Barren land dropped dramatically, from 122.45 square kilometers to just 26.47. Forest cover fluctuated, dipping around 2013 before recovering, and ended the period with a slight net increase, a pattern the authors attribute in part to Nepal’s community forestry program, which has driven substantial forest regeneration across the Terai region.

To map these land cover classes, the researchers turned to a Support Vector Machine, a supervised machine learning algorithm that finds the optimal hyperplane separating classes in multidimensional spectral space while maximizing the margin between them. Using 1,099 training points and 585 validation points for each of the four classification years, the classifier achieved overall accuracies ranging from 0.795 in 1993 to 0.839 in 2023, with corresponding Kappa values of 0.743 and 0.799. Water was consistently the most reliably classified category, while dry bare land beside rivers proved the most troublesome, its reflectance sometimes resembling agricultural fields or herbaceous cover, a limitation the authors acknowledge as inherent to medium-resolution imagery.

The spatial structure of ecological quality was as telling as its average value. Higher index values clustered in the forested Chure hills of the north, while the lowest values concentrated in the urban centers of Butwal and Bhairahawa, where impervious surfaces dominate and urban heat island effects are most likely; the Tinau river corridor at Butwal also has naturally sparse vegetation that lowers the index independently of urbanization. Spatial autocorrelation analysis, performed on 1,302 sampling points resampled to a one-kilometer grid to avoid inflating the statistics, yielded Global Moran’s I values between 0.391 and 0.558 across the four years. Local cluster maps showed persistent high-high clusters in the northern forests and steadily growing low-low clusters along the Siddhartha Highway corridor, confirming that ecological decline is not random but tightly bound to the geography of development.

Perhaps the study’s most important insight is what the district average conceals. When the researchers calculated the mean index value for each land cover class, forest scored highest, rising from 0.76 in 1993 to 0.87 in 2023, agricultural land held moderate to high values, and built-up areas remained consistently low throughout all thirty years. The district-wide mean stayed relatively high in 2023 partly because built-up pixels still occupy only a small share of the total area, even though their number grew nearly tenfold. In other words, ecological quality reflects the combined influence of every land cover class rather than any single one, and a district-scale average can mask severe localized degradation within rapidly urbanizing pockets.

The authors are careful about the limits of their work. The 30-meter resolution of Landsat imagery cannot resolve fine-scale ecological variation, only four indicators were used, and factors such as population density, precipitation, and economic activity were not included in the model. A single post-monsoon scene per year, chosen because monsoon clouds block optical imaging, introduces seasonal bias that multi-date composites could reduce. Still, the study offers a transferable, low-cost template for ecological monitoring in data-scarce regions, and its implications reach beyond Nepal. As the United Nations projects the global population to reach about 9.7 billion by 2050, the Rupandehi experience suggests that unplanned urban growth erodes local ecological quality even as forest regeneration and the conversion of barren land can partly offset the damage at the landscape scale, a balance that land-use planners everywhere will need to strike with their eyes on the satellites.

Subject of Research: Assessing the impact of urbanization-driven land use and land cover change on ecological environmental quality in Rupandehi District, Nepal, using the Remote Sensing Ecological Index

Article Title: Impact of urbanization driven land use and land cover change on ecological environmental quality in Rupandehi Nepal assessed using the Remote Sensing Ecological Index

Article References: Parajuli, G., Regmi, Y., Silwal, A., Parajuli, B., Pokhrel, B., Acharya, T. D., Karki, A. M., & Shrestha, S. (2026). Impact of urbanization driven land use and land cover change on ecological environmental quality in Rupandehi Nepal assessed using the Remote Sensing Ecological Index. Discover Geoscience, 4(1), Article 298. https://doi.org/10.1007/s44288-026-00650-y

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00650-y

Keywords: Remote Sensing Ecological Index, urbanization, land use land cover change, ecological environmental quality, Nepal, Google Earth Engine, Landsat, Principal Component Analysis, Support Vector Machine, spatial autocorrelation, land surface temperature, community forestry

The research team, led by scientists at the University of Dhaka with collaborators in the United States, analyzed Landsat satellite imagery from 1994, 2004, 2014 and 2024, covering the four districts of Sylhet, Sunamganj, Maulvibazar and Habiganj. Using an unsupervised classification algorithm known as ISODATA, they grouped the satellite pixels into five major land use and land cover categories, then refined their analysis with two spectral indices: the normalized difference vegetation index, or NDVI, which tracks the health and density of green vegetation, and the normalized difference water index, or NDWI, which distinguishes surface water from dry land. The results were stark. Settlement areas expanded by 191.75 percent over the three decades, while forest cover declined by 25.65 percent, a shift the authors attribute to intensified deforestation and the conversion of wooded land into agriculture, infrastructure and housing.

What makes the study unusual among urban heat island investigations in Bangladesh is its insistence on ground truth. Most satellite-based studies of land surface temperature rely on remote data alone, but the researchers took a Eurolab ST9269 thermometer into the field, inserting its probe 1.2 to 2.5 centimeters into the soil at 100 observation points spread across different land cover types. Each location was logged with a handheld GPS. When the field measurements were compared with the satellite-derived temperatures, the agreement was strong: the classification achieved an overall accuracy of 89 percent with a Kappa coefficient of 0.86, and the correlation between field and satellite temperatures yielded a coefficient of determination of 0.80. The root mean square error of about 3.5 degrees and a mean absolute error of 3.12 degrees suggested consistent, modest deviations rather than wild outliers, likely stemming from atmospheric variability and the inherent differences between near-surface probe readings and radiometric measurements from orbit.

The temperature story that emerges from the satellite record is remarkable in its trajectory. In 1994, the coolest thermal class, below 20 degrees Celsius, dominated the division, covering roughly 1.1 million hectares. By 2024, that class had disappeared entirely. Meanwhile, the 25 to 27.5 degree band, which occupied a mere 638 hectares in 1994, ballooned to nearly 489,000 hectares by 2024. The 27.5 to 30 degree class grew from just 129 hectares to more than 218,000 hectares over the same period. The hottest categories, above 30 degrees, first appeared in 2014 and continued to expand through 2024, covering more than 16,500 hectares in the 30 to 32.5 degree range and nearly 2,600 hectares above 32.5 degrees. In effect, the entire thermal distribution of the region has shifted upward, with cooler surfaces progressively converted into mid-range and high-temperature zones.

To examine the urban heat island effect in detail, the team focused on four administrative centers, or Sadar Upazilas: Sylhet, Habiganj, Maulvi Bazar and Sunamganj. In 1994, none of them showed any sign of heat island formation, with dominant temperatures well below 20 degrees. By the mid-2000s, however, the 25 to 27.5 degree class began expanding around urban cores, marking the first emergence of localized heat islands near places like Dhopagul, Sylhet Airport and Shahjalal Upashahar. By 2024, the warming had intensified and spread outward into neighborhoods such as Haldarpara, Jalalabad and the Stadium Area in Sylhet Sadar, where surface temperatures reached 32.5 degrees. Similar patterns unfolded in the other three districts: Maulvi Bazar Sadar saw zones above 30 degrees grow to nearly 900 hectares, Sunamganj Sadar recorded a small but critical patch exceeding 32.5 degrees, and Habiganj Sadar developed its own hotspots around Shaistaganj and Keshabpur. In every case, the coolest temperature classes contracted dramatically, in some instances from tens of thousands of hectares to only a few thousand.

The statistical heart of the study lies in a Pearson correlation analysis of 18 variables spanning four decades of data. The results quantify with unusual precision what urban climatologists have long suspected. Built-up and agricultural surfaces showed strong positive correlations with the highest temperature classes, with coefficients ranging from 0.92 to 0.98 for settlement areas. Forests and water bodies, by contrast, displayed strong negative correlations with the same hot zones, ranging from minus 0.89 to minus 1.00. The vegetation and water indices reinforced the pattern: areas rich in greenery and surface moisture consistently corresponded to lower thermal loads. The researchers also observed a perfect negative correlation between forest greenness and agricultural or mixed land dominance, suggesting a kind of mutual exclusivity in how the landscape is partitioned—as one expands, the other is suppressed.

From a biophysical standpoint, these correlations reflect fundamental changes in how the land exchanges energy with the atmosphere. Vegetated surfaces and water bodies act as thermal regulators through evapotranspiration and their high heat capacity, absorbing solar energy as latent heat rather than raising surface temperatures. Concrete, asphalt and bare soil do the opposite: they store heat during the day and release it slowly at night, amplifying sensible heat flux and suppressing the cooling processes that natural surfaces provide. As Sylhet’s settlements and farmland replaced forests and wetlands, the division’s capacity to buffer heat eroded, and the surface energy balance tipped decisively toward warming. The study describes this as a feedback loop in which urban expansion intensifies thermal stress, which in turn influences further land conversion patterns.

The findings align with a growing body of evidence from across Bangladesh and South Asia. Previous studies in Dhaka, Chittagong, Rajshahi, Khulna and Rangpur have documented similar linkages between impervious surface expansion and rising land surface temperature, and comparable dynamics have been reported in Cox’s Bazar, Gazipur and the Sylhet Sadar region itself. What sets the new research apart is its division-wide scope and its integration of field validation, which the authors argue makes the framework transferable to other rapidly urbanizing and ecologically sensitive regions. They acknowledge certain limitations, including the coarse spatial resolution of Landsat imagery, minor seasonal effects from varying acquisition dates within the dry season, and the use of unsupervised rather than machine learning-based classification, which can blur spectrally similar land cover types such as bare land and fallow fields.

The policy implications are urgent for a region where settlements are encroaching on haors, wetlands and forested hillocks. The authors call for climate-sensitive zoning, the preservation of remaining wetlands, riparian corridors and forest patches as critical cooling infrastructure, and the expansion of urban green cover through tree planting and green corridors to offset rising temperatures. They also recommend future research incorporating higher-resolution data from Sentinel satellites or drones, machine learning classification techniques, and urban climate modeling coupled with socioeconomic variables to anticipate how heat islands will expand under different development scenarios. For now, the message from three decades of satellite observations is unambiguous: when a landscape loses its forests and water, it loses its thermostat, and Sylhet Division is feeling the heat.

Subject of Research: Land cover change, land surface temperature dynamics, and urban heat island formation in Sylhet Division, Bangladesh, assessed through geospatial analysis of Landsat imagery from 1994 to 2024.

Article Title: Evaluating the transformation of land cover, land surface temperature, and urban heat island in Sylhet Division through geospatial analysis

Article References: Joy, M. F. R., Rahim, M. A., Rahman, M. M., Preota, S. A., & Hossain, M. M. (2026). Evaluating the transformation of land cover, land surface temperature, and urban heat island in Sylhet Division through geospatial analysis. Discover Geoscience, 4(1), Article 306. https://doi.org/10.1007/s44288-026-00687-z

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00687-z

Keywords: urban heat island, land surface temperature, land use land cover change, Sylhet Division, Bangladesh, Landsat, remote sensing, NDVI, NDWI, deforestation, urbanization, ground validation

The research, published in the journal Discover Geoscience, was carried out by Dereje Ketema and Addisu Bekele of Bule Hora University together with Barsisa Tola of Salale University. Their starting point was a familiar problem in land-change science: most studies of land use and land cover in Ethiopia have been retrospective, describing what has already happened at regional or national scales, while the most intense transformations are unfolding in small and medium-sized towns that rarely receive dedicated analysis. Fiche, administratively divided into four kebeles and expanding rapidly under population pressure and infrastructure growth, offered precisely the kind of case where predictive evidence could shift planning from reactive to proactive.

To build the historical record, the team turned to the Landsat archive, the workhorse of long-term land-cover mapping. They acquired cloud-free scenes from Landsat 7 ETM+ for 2002 and Landsat 8 OLI/TIRS for 2013 and 2022, keeping cloud cover below ten percent to minimize atmospheric interference. After radiometric calibration to correct for sensor and illumination differences, the images were classified in ENVI 5.3 using a supervised Support Vector Machine approach, an algorithm favored for its ability to deliver high accuracy even with relatively modest training datasets. Following the AFRICOVER classification scheme widely used in East Africa, the landscape was divided into four classes: built-up area, farmland, open space, and shrubland, with training signatures drawn from 50 to 90 sites per class informed by ground control points, historical Google Earth imagery, and expert knowledge of the terrain.

The classification proved remarkably reliable. Validation against reference data, including 117, 147, and 100 randomly sampled testing points for 2002, 2013, and 2022 respectively, plus 55 field-collected ground control points for 2022, produced overall accuracies of 94.9 percent, 96.6 percent, and 98.7 percent, with Kappa coefficients of 0.932, 0.955, and 0.983. All values comfortably exceeded the 85 percent accuracy threshold conventionally required for dependable change detection, giving the researchers a solid foundation for the transition analysis that followed.

What those maps revealed is a landscape in the grip of an accelerating regime shift. In 2002, farmland dominated Fiche’s surroundings, covering 66.4 percent of the study area, or about 2,300 hectares, while built-up surfaces occupied just 10.4 percent, roughly 361 hectares. By 2013 the urban footprint had grown to 16.9 percent, and by 2022 it had surged to 25.8 percent, or 891.89 hectares, an increase of nearly 147 percent over two decades. The pace quickened dramatically in the second decade: built-up expansion ran at 20.4 hectares per year between 2002 and 2013, then jumped to 33.9 hectares per year between 2013 and 2022. Farmland loss followed the same pattern, doubling from 36.1 hectares per year in the first interval to 71.2 hectares per year in the second, culminating in the disappearance of more than 640 hectares of cropland in under ten years. Shrubland fared even worse in relative terms, collapsing by more than half, from 100 hectares to just 44.79.

Transition matrices generated with the MOLUSCE plugin in QGIS added a probabilistic layer to the story, quantifying the likelihood of pixels shifting from one class to another across each interval. Farmland proved to be the primary donor for urban growth: over the full 2002 to 2022 period, its persistence probability fell to just 0.452, with substantial contributions flowing to open space and built-up areas. Built-up land, by contrast, showed deep stability at 0.833, meaning that once land is urbanized in Fiche, it almost never reverts. Shrubland was the most volatile category of all, with a persistence of only 0.180, its area dispersed across farmland, built-up surfaces, and open space. The spatial pattern of conversion followed a proximity-based logic, with peripheral agricultural belts progressively absorbed by the advancing urban fringe rather than being lost at random.

The predictive engine at the heart of the study is a hybrid Cellular Automata-Artificial Neural Network framework, implemented through MOLUSCE’s multilayer perceptron option. The researchers selected the ANN approach over alternatives such as logistic regression and weights of evidence because of its strength in capturing complex, non-linear relationships between land change and its drivers. Those drivers included a digital elevation model, slope, aspect, and distance from roads and existing built-up areas, all standardized to a uniform 30-meter resolution in the UTM Zone 37N coordinate system and screened for multicollinearity using Pearson’s correlation, Cramér’s V, and joint information uncertainty. The network was trained on the 2002 to 2013 transition dynamics with 1,000 iterations, ten hidden layers, a momentum value of 0.06, and a learning rate of 0.001, then coupled with cellular automata allocation rules to simulate the 2022 landscape.

That simulation passed a demanding test. When compared against the actual satellite-derived 2022 classification, the modeled map achieved an overall Kappa of 0.83 and a correctness rate of 88.8 percent, with component Kappa statistics as high as 0.93 for the spatial allocation of change. Confident in the model, the team ran it forward to 2030, 2040, and 2050 under a business-as-usual assumption that historical dynamics and driving factors will persist. The projections show built-up area climbing from 891.89 hectares in 2022 to 1,043.6 hectares by 2030, 1,140.5 by 2040, and 1,268.7 by 2050, a net gain of 376.81 hectares, or 42.2 percent, over 28 years. Farmland, meanwhile, is forecast to fall from 1,261.2 hectares to 762.2 hectares by 2050, a loss of 499 hectares, or 39.6 percent. In proportional terms, built-up land would reach 29.76 percent of the study area in 2030, 32.31 percent in 2040, and 35.74 percent in 2050, while agriculture slides to 29.74, 26.56, and finally 22.01 percent. Open space is projected to keep growing as well, a signal of ongoing land clearing and fragmentation, while the last remnants of shrubland thin further.

The authors frame these numbers as more than a cartographic exercise. Ethiopia, Africa’s second most populous country with over 120 million people, remains one of the world’s least urbanized nations but is urbanizing at unprecedented speed, with the national urban population projected to nearly triple from 15.2 million in 2012 to 42.3 million by 2037. In peri-urban zones like Fiche’s surroundings, that demographic surge translates directly into competition between farms and houses, threatening local food security and the livelihoods of farming communities that the urban expansion progressively marginalizes. The conversion of productive land into impervious surfaces also degrades ecosystem services, alters hydrological processes, and reshapes microclimates, compounding the socioeconomic stakes.

The study’s prescription is correspondingly practical. Ketema and his colleagues argue that Fiche’s trajectory can be bent only through deliberate intervention: strict zoning to protect high-value agricultural land, compact and vertical development strategies that discourage horizontal sprawl, green infrastructure, improved urban mobility, and continuous geospatial monitoring embedded in municipal governance. They also point to the next generation of models, suggesting that integrating higher-resolution Sentinel imagery and socioeconomic variables such as population growth and policy shifts would sharpen future projections beyond what purely spatial drivers can explain. For now, the message from the highlands of Central Ethiopia is unambiguous: without timely action, the satellite record and the neural network agree that by 2050 the town’s fields will have been reduced to a fragile remnant, and the landscape fragmentation that follows may prove irreversible.

Subject of Research: Spatiotemporal analysis and neural network modeling of land use and land cover change and urban expansion in Fiche Town, Central Ethiopia

Article Title: Spatiotemporal analysis and modeling of land use and land cover change in Fiche Town, Central Ethiopia using geospatial techniques and MOLUSCE

Article References: Ketema, D., Bekele, A., & Tola, B. (2026). Spatiotemporal analysis and modeling of land use and land cover change in Fiche Town, Central Ethiopia using geospatial techniques and MOLUSCE. Discover Geoscience, 4(1), Article 309. https://doi.org/10.1007/s44288-026-00689-x

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00689-x

Keywords: land use land cover change, urban expansion, Ethiopia, remote sensing, Landsat, MOLUSCE, artificial neural network, cellular automata, support vector machine, GIS, peri-urban agriculture, food security

The study, published in the journal Theoretical and Applied Climatology, was led by Kajesh Gadekar and Aneesh Mathew, who combined satellite imagery, spectral indices, and machine learning into a single analytical framework. To track how the city’s surface changed, the team classified multi-temporal Landsat imagery into land use and land cover categories, revealing the steady march of construction across Chennai’s fringes. To quantify the heat island itself, they turned to nighttime land surface temperature data from the Moderate Resolution Imaging Spectroradiometer, or MODIS, instrument, which allowed them to compute annual and seasonal heat island intensities consistently across the entire two-decade window. Nighttime measurements are particularly revealing for urban climate science because that is when the thermal contrast between dense cities and their rural surroundings is most pronounced, as built-up materials continue radiating stored heat long after sunset while rural soils and vegetation cool quickly.

One of the most striking findings is seasonal. Counterintuitively, the urban heat island in Chennai was strongest in winter, not summer, peaking at 2.21 degrees Celsius in 2018. The explanation lies in the physics of the surrounding landscape. During the monsoon and post-monsoon months, rural areas around the city are moist and vegetated, so a large fraction of incoming solar energy is consumed by evapotranspiration rather than heating the ground. The city, by contrast, remains dry and impervious year-round, so its surfaces warm efficiently and hold that warmth into the night. The result is a large urban-rural temperature differential precisely when residents might least expect it. Summer heat islands, while lower in intensity, still showed significant peaks in the urban core, underscoring the role of construction materials such as concrete and bitumen in retaining heat even during the hottest months of the year.

To understand what drives these temperature patterns at the level of individual pixels, the researchers computed a suite of environmental predictors from satellite data. These included the Enhanced Vegetation Index, a measure of green biomass; the Normalized Difference Built-up Index, which flags impervious construction; the Modified Normalized Difference Water Index, which captures surface moisture and open water; the Modified Bare Soil Index, which identifies exposed earth; and Aerosol Optical Depth, a proxy for atmospheric particulate loading. They added elevation, latitude, and longitude to account for terrain and geographic position. Correlation analysis between land surface temperature and these parameters revealed a coherent physical story. Vegetation and elevation were negatively correlated with temperature, meaning greener and higher ground runs cooler. Built-up index and aerosol loading were positively correlated, confirming that denser construction and hazier air accompany hotter surfaces.

The predictive heart of the study lies in its comparison of two machine learning algorithms trained to estimate land surface temperature from those environmental predictors. The team tested Random Forest, an ensemble method that averages the outputs of many decision trees, against Extreme Gradient Boosting, known as XGBoost, which builds trees sequentially so that each new one corrects the errors of its predecessors. XGBoost emerged as the clear winner, achieving a testing coefficient of determination of 0.8417, meaning it explained more than 84 percent of the variance in observed temperatures. Its error metrics were correspondingly tight, with a mean absolute error of 0.5796 degrees Celsius, a mean squared error of 0.4275 degrees Celsius squared, and a root mean squared error of 0.6538 degrees Celsius. Random Forest performed respectably but trailed on every metric, and the gap confirms that gradient boosting’s iterative error correction is particularly well suited to the nonlinear relationships linking land cover, moisture, aerosols, and surface temperature.

The seasonal breakdown of model performance adds further confidence. XGBoost achieved its highest coefficient of determination in winter, at 0.8805, and a strong 0.799 in summer, with error metrics remaining within acceptable ranges in both seasons. That robustness matters because a model that only works in one season would be of limited use for city planners trying to anticipate thermal stress throughout the year. Spatial validation reinforced the picture: predicted temperatures deviated minimally from observations, especially in dense urban areas where the Enhanced Vegetation Index remains relatively stable and the thermal signal is dominated by built surfaces. In rural zones, prediction errors were larger, a consequence of dynamic vegetation phenology and more varied terrain, where crop cycles and soil moisture swings introduce variability that static predictors struggle to capture.

What makes this study resonate far beyond Chennai is its methodological template. By fusing remote sensing, geospatial analysis, and machine learning into a single pipeline, the researchers demonstrated a repeatable way to diagnose how land cover transformation reshapes a city’s thermal climate. The approach requires no dense network of ground weather stations, which many rapidly urbanizing cities in the Global South lack. Instead, freely available satellite data from Landsat and MODIS, combined with open-source machine learning tools, can produce spatially explicit temperature maps and intensity estimates at annual and seasonal resolution. For the dozens of coastal megacities expected to absorb enormous population growth in the coming decades, that is a powerful and accessible diagnostic instrument.

The implications for planning are concrete. The study’s findings provide quantitative evidence for prioritizing vegetation conservation, blue-green infrastructure, and thermally sensitive land-use planning in rapidly urbanizing coastal cities. In practice, that means protecting remaining green corridors and wetlands before they are paved, integrating parks, lakes, and vegetated buffers into new development, and treating surface materials as a design variable rather than an afterthought. Because the analysis shows vegetation-related indicators contribute directly to cooling while built-up expansion intensifies the heat island, the model can be used prospectively: planners can simulate how a proposed land use scenario would alter surface temperatures before a single foundation is poured. The XGBoost model’s accuracy makes such scenario testing credible at the neighborhood scale.

Chennai’s two-decade record is, in effect, a controlled experiment in what happens when a humid coastal city converts its cooling infrastructure into heat-storing mass. The doubling of annual heat island intensity, the winter peaks above two degrees Celsius, and the near-tripling of built-up area together trace a clear causal chain from land use decision to thermal outcome. As climate change raises baseline temperatures across South Asia, the additional heat generated by urban form will compound the burden on residents, particularly those in dense, low-income neighborhoods with little tree cover. Studies of this kind turn that burden from an invisible background condition into a measurable, modelable, and therefore manageable quantity. The heat island, the research makes clear, is not an inevitable byproduct of urban growth. It is the predictable consequence of specific land cover choices, and it can be reversed the same way it was created, one surface at a time.

Subject of Research: Urban heat island dynamics, land use change, and machine learning-based land surface temperature prediction in Chennai, India

Article Title: Spatiotemporal Dynamics of Urban Heat Islands in Chennai: A Two-Decade Study on Land Use Changes and Predictive Modeling

Article References: Gadekar, K., & Mathew, A. (2026). Spatiotemporal Dynamics of Urban Heat Islands in Chennai: A Two-Decade Study on Land Use Changes and Predictive Modeling. Theoretical and Applied Climatology, 157(10), Article 611. https://doi.org/10.1007/s00704-026-06496-1

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06496-1

Keywords: urban heat island, Chennai, land surface temperature, land use land cover change, XGBoost, Random Forest, MODIS, Landsat, remote sensing, machine learning, urbanization, blue-green infrastructure