ML Predicts Urban Heat from Land Use Changes

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Aug 22 2024

In a paper published in the journal Scientific Reports, researchers analyzed land use and land cover (LULC) changes in Kabul province from 1998 to 2022 using Landsat images and support vector machine (SVM) classification. They observed an increase in built-up areas and a corresponding rise in land surface temperature (LST).

Future predictions using the cellular automata-logistic regression (CA-LR) model suggest continued expansion of built-up areas and a significant increase in high LST areas over the coming decades.

Related Work

Past work has highlighted the serious environmental issues of LULC changes and their impact on LST, driven by urbanization that replaces natural and porous surfaces with heat-absorbing built-up areas. It has increased LST and exacerbated climate change effects, including higher energy demands and CO₂ emissions. Studies have shown significant LST increases in various regions due to LULC changes, with Kabul, Afghanistan, experiencing rapid urban expansion and unplanned growth contributing to these trends.

Kabul Urban Dynamics

This study focuses on Kabul, Afghanistan's largest urban province, situated at 34°31′31′′ north latitude and 69°10′42′′ east longitude. With a population of approximately 6 million and a land area of 4,523 km², Kabul is predominantly mountainous, with only about 38% flat terrain.

The city's weather varies greatly from minus 11 °C in the winter to about 40 °C in the summer. Its climate is semi-arid to dry. Because of the area's high immigration rates, fast urbanization, and unchecked growth—which has resulted in multiple informal settlements—the land cover and use have undergone significant changes.

Remote sensing data from Landsat satellites assessed LULC and LST changes from 1998 to 2022. Images were specifically acquired in May to minimize seasonal variations and cloud cover. Researchers corrected the data for radiometric and atmospheric factors. The team employed the SVM algorithm for LULC classification and used thermal bands from Landsat satellites to generate LST maps. The CA-LR model was applied to project future LULC and LST patterns for 2034 and 2046.

Built-up regions, bare soil, vegetation, and water bodies were the four categories into which LULC divided the Landsat images. SVM was used for supervised classification, and transition matrices were created to detect changes over time. LST was calculated from thermal bands, standardized for seasonal and topographic variations, and linked with normalized difference vegetation index (NDVI), normalized difference built-up index (NDBI), and normalized difference water index (NDWI) indices to explore the relationship between land cover types and LST.

The model incorporated LULC change maps, LST distributions, and various spatial variables. Accuracy was evaluated using the Kappa coefficient, with training samples and ground truth data ensuring reliable classification results. The study calculated overall, producer, and user accuracy to validate the predictions and classifications.

LULC and LST Changes

The analysis of LULC changes in Kabul from 1998 to 2022, utilizing SVM algorithms, revealed significant shifts. Built-up areas expanded while vegetation and bare soil decreased. The overall categorization accuracy for 1998, 2010, and 2022 varied between 93.26% and 93.78%. Specifically, built-up areas grew by approximately 481 km², whereas bare soil and vegetation declined by 395 km² and 89 km², respectively. The conversion of these land cover types into built-up areas is evident, impacting local climate and contributing to urban heat islands.

Past LST trends show a marked increase in high-temperature areas (>32 °C) from 1998 to 2022, with lower LST areas shrinking. Between 1998 and 2010 and 2010 and 2022, the yearly mean LST increased by 1.03 °C and 1.10 °C, respectively. Notably, the distribution of LST classes shifted, with higher temperature classes expanding and lower ones contracting. The temperature increases across all LULC classes, particularly in built-up areas, underscore the influence of urbanization on local climate dynamics.

Using the CA-LR model, future projections indicate further LULC and LST changes. Similarly, the percentage of land with temperatures above 32 °C is anticipated to rise to 16.32% in 2034 and 16.47% in 2046. These changes will likely exacerbate environmental and climatic impacts, affecting human and ecological systems.

Conclusion

To sum up, this study analyzed the impact of LULC changes on seasonal and annual LST in Kabul province, Afghanistan, from 1998 to 2022. Using the SVM algorithm to classify Landsat satellite images, the study identified four distinct LULC classes: built-up areas, bare soil, vegetation, and water bodies. The analysis revealed a 9.37% increase in built-up areas, corresponding decreases of 7.20% and 2.35% in bare soil and vegetation, respectively.

Built-up areas consistently exhibited the highest annual mean LST, reaching 27.53 °C in 2022, while water bodies had the lowest at 17.83 °C. Seasonal maximum temperatures rose from 40.91 °C in 1998 to 44.89 °C in 2022. Future projections show built-up areas in Kabul rising to 17.08% and 23.10% by 2034 and 2046, respectively, with temperatures ≥ 32 °C also significantly increasing.