Spectral index and machine learning-based evaluation of coastal urban thermal dynamics using google earth engine
Abstract
This present study introduces a cloud-based geospatial and machine learning framework for assessing coastal urban heat island (UHI) dynamics in Thoothukudi coastal region, Southern India, using 12 seasonal Landsat composites (2010–2025) processed through Google Earth Engine (GEE). Seasonal classification followed Indian Meteorological Department definitions: Summer (March-June), Northeast Monsoon (October-November) and Winter (December-February). Multi-sensor Landsat Collection 2 Tier 1 Level-2 data included Landsat 5/7 (2010), Landsat 8 (2015) and Landsat 8/9 (2020, 2025). Land surface temperature (LST) was retrieved using USGS Collection 2 scale-factor protocol, capped to 0–65 °C to suppress cloud-contaminated pixels. UHI intensity (UHI = LST − P₂₅(LST)) and Urban Thermal Field Variance Index (UTFVI) were computed alongside four spectral indices (NDVI, NDBI, NDWI, NDMI). Summer was the hottest season in three of four observation years (mean LST: 38.69–48.67 °C). The 2020 Northeast Monsoon composite was excluded due to cloud contamination. Mean UHI intensity ranged from 2.03 to 5.17 °C. Pearson correlation confirmed NDBI as the strongest positive LST predictor ( r = 0.30–0.60), while NDMI mirrors this inversely. Random Forest and Gradient Tree Boosting models were trained on seven predictors using dual validation (70/30 split and 2 km spatial block cross-validation). Model performance yielded R² of 0.301–0.476 and RMSE of 2.74–6.23 °C with physically realistic values. GTB permutation feature importance identified NDBI and SWIR1 as dominant predictors. Ensemble scenario illustrations for 2030 and 2050, presented as linear extrapolations rather than climate-model forecasts, suggest sustained thermal stress in industrial and port zones with explicit uncertainty quantification (± 2.75 to ± 8.61 °C). The study demonstrates that rigorously validated Landsat time-series analysis provides actionable coastal urban thermal intelligence for heat adaptation planning aligned with SDGs 11, 13 and 3. The framework is applied here as the first four-epoch, IMD-season-resolved thermal characterisation of the Thoothukudi coast and the 2030 and 2050 outputs are intended to support the spatial targeting of persistent thermal hotspots rather than the prediction of absolute future temperatures.
Related
- Uncertainty analysis of a machine learning-aided ensemble forecast for coastal flood maps
- AQUAVERSE-AC: a machine learning-based atmospheric correction framework for inland and coastal waters
- Interpretable machine learning quantifies composition and size influences on aerosol spectral absorption
- Optical Remote Sensing and Machine Learning Enable Scalable Carbon Mapping in Himalayan Forests Despite Spectral Saturation Challenges
- A Station-Based Evaluation of Machine Learning-based Weather Forecasting Models in Northern Norway
BibTeX
@article{roy2026spectral,
title = {{Spectral index and machine learning-based evaluation of coastal urban thermal dynamics using google earth engine}},
author = {A. Ernest Amita Roy and John Prince Soundaranayagam and A. Antony Alosanai Promilton and Neelam Sidhu and V. Stephen Pitchaimani and Hazratullah Paktin and Abhishek Sharma},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-73768-1},
url = {https://www.nature.com/articles/s41598-026-73768-1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Spectral index and machine learning-based evaluation of coastal urban thermal dynamics using google earth engine
AU - A. Ernest Amita Roy
AU - John Prince Soundaranayagam
AU - A. Antony Alosanai Promilton
AU - Neelam Sidhu
AU - V. Stephen Pitchaimani
AU - Hazratullah Paktin
AU - Abhishek Sharma
PY - 2026
DA - 2026/10/03
JO - Scientific Reports
DO - 10.1038/s41598-026-73768-1
UR - https://www.nature.com/articles/s41598-026-73768-1
AB - This present study introduces a cloud-based geospatial and machine learning framework for assessing coastal urban heat island (UHI) dynamics in Thoothukudi coastal region, Southern India, using 12 seasonal Landsat composites (2010–2025) processed through Google Earth Engine (GEE). Seasonal classification followed Indian Meteorological Department definitions: Summer (March-June), Northeast Monsoon (October-November) and Winter (December-February). Multi-sensor Landsat Collection 2 Tier 1 Level-2 data included Landsat 5/7 (2010), Landsat 8 (2015) and Landsat 8/9 (2020, 2025). Land surface temperature (LST) was retrieved using USGS Collection 2 scale-factor protocol, capped to 0–65 °C to suppress cloud-contaminated pixels. UHI intensity (UHI = LST − P₂₅(LST)) and Urban Thermal Field Variance Index (UTFVI) were computed alongside four spectral indices (NDVI, NDBI, NDWI, NDMI). Summer was the hottest season in three of four observation years (mean LST: 38.69–48.67 °C). The 2020 Northeast Monsoon composite was excluded due to cloud contamination. Mean UHI intensity ranged from 2.03 to 5.17 °C. Pearson correlation confirmed NDBI as the strongest positive LST predictor ( r = 0.30–0.60), while NDMI mirrors this inversely. Random Forest and Gradient Tree Boosting models were trained on seven predictors using dual validation (70/30 split and 2 km spatial block cross-validation). Model performance yielded R² of 0.301–0.476 and RMSE of 2.74–6.23 °C with physically realistic values. GTB permutation feature importance identified NDBI and SWIR1 as dominant predictors. Ensemble scenario illustrations for 2030 and 2050, presented as linear extrapolations rather than climate-model forecasts, suggest sustained thermal stress in industrial and port zones with explicit uncertainty quantification (± 2.75 to ± 8.61 °C). The study demonstrates that rigorously validated Landsat time-series analysis provides actionable coastal urban thermal intelligence for heat adaptation planning aligned with SDGs 11, 13 and 3. The framework is applied here as the first four-epoch, IMD-season-resolved thermal characterisation of the Thoothukudi coast and the 2030 and 2050 outputs are intended to support the spatial targeting of persistent thermal hotspots rather than the prediction of absolute future temperatures.
SN - 2045-2322
ER -