Turkish Coastal City Could See Hotspots Explode Elevenfold by 2040, Satellite Study Warns

A small province on the southern shore of the Sea of Marmara has become an unlikely warning bell for coastal cities everywhere. Yalova, Türkiye, a region known for its thermal springs, orchards, and forested hillsides, has been transformed over the past two decades by some of the fastest urban expansion in the country. A new study published in Theoretical and Applied Climatology by Ahmet Salih Değermenci of Duzce University combines twenty years of satellite observations with machine-learning simulation to trace exactly how that expansion has reshaped the province’s surface, and to project what its climate could look like by 2040 if current development patterns continue unchecked.

The research rests on a straightforward but powerful premise: the way humans rearrange the land surface directly rewrites the local thermal environment. Using multi-temporal Landsat imagery from 2000, 2010, and 2020, Değermenci classified the province’s land use and land cover with a Maximum Likelihood Classification approach, a statistical technique that assigns each pixel to the land class it most probably belongs to based on its spectral signature. The classifications performed remarkably well, achieving overall accuracies of 87.8 percent, 90.0 percent, and 94.8 percent for the three dates respectively, with Cohen’s Kappa values ranging from 0.84 to 0.93, levels conventionally regarded as indicating strong to almost perfect agreement between the classified maps and ground reality.

What those maps reveal is a landscape in rapid transition. Between 2000 and 2020, Yalova’s built-up area grew by 138 percent, expanding from 47.30 square kilometers to 112.58 square kilometers. That growth did not occur on empty land. It came overwhelmingly at the expense of dense vegetation and agricultural fields, which together suffered a net loss of 121.02 square kilometers, a decline of 21.5 percent. In practical terms, more than two-fifths of the province’s green and cultivated fabric was converted into roads, rooftops, and other impervious surfaces within a single generation, a pace of land conversion that places Yalova among the more dramatically urbanizing coastal corridors of the eastern Mediterranean region.

To connect this transformation to temperature, the study derived Land Surface Temperature, or LST, from the same Landsat archive. LST is the radiative skin temperature of the ground as seen from orbit, and it responds almost immediately to changes in surface material: asphalt and concrete absorb solar radiation and re-emit it as heat, while leafy canopies cool their surroundings through evapotranspiration and shading. Because satellite-derived LST can be sensitive to retrieval method and sensor differences, Değermenci rigorously cross-validated the estimates against two independent references: MODIS Terra satellite temperature products, which correlated with the Landsat estimates at R² values of 0.77 to 0.86, and two-meter air temperature records from ground meteorological stations, which correlated at R² values of 0.79 to 0.88. This dual validation gives the thermal analysis a firm empirical footing.

The coupling between land cover and heat proved strikingly strong. Spatial regression analysis showed a robust positive association between the Normalized Difference Built-up Index, a spectral measure of built-up density abbreviated NDBI, and land surface temperature, with correlation coefficients between 0.78 and 0.88. In plain terms, the more densely built a neighborhood became, the hotter its surface ran. The mirror image held for vegetation: the Normalized Difference Vegetation Index, NDVI, showed significant negative correlations with LST, ranging from −0.70 to −0.84, confirming that vegetated corridors act as measurable thermal attenuators, actively suppressing surface temperatures across the landscape.

Having established the historical relationship between urbanization and heat, the study then turned to the future using a Cellular Automata–Artificial Neural Network, or CA–ANN, simulation framework. This hybrid modeling approach, widely used in land-change science, couples an artificial neural network, which learns the probabilistic relationships between land transitions and their driving variables such as slope, distance to roads, and proximity to existing development, with cellular automata, which enforce local spatial rules so that new development emerges in spatially realistic patterns rather than scattered randomly. Under a Business-as-Usual baseline scenario, the model projects that Yalova’s built-up surfaces will reach 148.42 square kilometers by 2040, a further increase of roughly a third over 2020 levels.

The thermal consequences of that projected growth are the study’s most eye-catching finding. The model predicts that the extreme thermal class, defined as surface temperatures between 35 and 40 degrees Celsius, will expand approximately elevenfold, from 9.52 square kilometers in 2020 to 104.16 square kilometers by 2040. In other words, land that today experiences only occasional extreme surface heating would become routinely, extensively hot, an area larger than the province’s entire current built-up footprint. Such a shift would intensify heat stress for residents, raise cooling energy demands, and place additional pressure on the ecosystems and agricultural lands that currently buffer the local microclimate.

The study is also notable for its honesty about uncertainty. Residual analysis, which compares predicted land cover against observed conditions, found that 10.8 percent of pixels deviated by more than 2.5 degrees Celsius in the thermal simulation, with errors concentrated primarily in rural–urban transition ecotones. These fringe zones, where farmland, forest fragments, and new subdivisions intermingle, are precisely where land-change models struggle most, because conversion decisions there are driven by volatile economic and policy factors that are difficult to capture in any statistical model. Flagging these zones explicitly turns the map of model error into a map of where planners should watch development pressure most closely.

Değermenci frames the outputs as a spatial early-warning diagnostic rather than a deterministic forecast. The value of the CA–ANN baseline lies in showing what happens if nothing changes: if zoning continues to permit dispersed growth along the coast and agricultural conversion proceeds at historical rates, the extreme-heat footprint will balloon. The same framework can, in principle, be rerun under alternative scenarios, compact-growth zoning that concentrates development, protected forest corridors that block conversion of vegetated land, or nature-based mitigation strategies such as expanded urban green infrastructure, allowing decision-makers to compare the thermal futures associated with different planning choices before committing concrete to the ground.

The implications extend well beyond one Turkish province. Coastal corridors around the Mediterranean, the Black Sea, and rapidly urbanizing coastlines across Asia and Africa face the same compound dynamic: population growth, tourism development, and agricultural land conversion converging on landscapes whose vegetation once kept them cool. The Yalova study demonstrates that with freely available Landsat data, careful validation against ground stations, and accessible machine-learning tools, even a single province can quantify its own heat trajectory with enough precision to inform policy. As extreme heat becomes one of the deadliest consequences of a warming world, the ability to see, decades in advance, exactly which neighborhoods will become heat traps, and which green corridors can still prevent them, may prove one of the most valuable instruments in the urban planning toolkit.

Subject of Research: Land use/land cover change and land surface temperature dynamics modeled with CA–ANN scenarios in Yalova, Türkiye

Article Title: Spatiotemporal analysis and CA–ANN scenario modeling of land use/land cover and surface thermal classes in Yalova, Türkiye

Article References: Değermenci, A. S. (2026). Spatiotemporal analysis and CA–ANN scenario modeling of land use/land cover and surface thermal classes in Yalova, Türkiye. Theoretical and Applied Climatology, 157(11), Article 705. https://doi.org/10.1007/s00704-026-06637-6

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06637-6

Keywords: land use land cover, land surface temperature, urban heat, CA–ANN modeling, remote sensing, Landsat, Yalova, Türkiye, urbanization, NDVI, coastal cities, climate adaptation