Urban heat island classification through alternative normalized difference vegetation index

BACKGROUND AND OBJECTIVES: Urban heat island is characterized by higher temperatures in urban areas compared to their surroundings. Vegetation, quantified by the normalized difference vegetation index, is key in mitigating urban heat island effects and influencing land surface temperature. With the rise of Machine Learning techniques, particularly random forest, land surface temperature predictions have become more accurate. This study explores alternative normalized difference vegetation index adjustments to understand their impact on urban heat island classification in Chiang Mai, Thailand. It investigates how changes to the normalized difference vegetation index can help to be part of practical urban planning measures, such as prioritizing vegetation type and location for cooling strategies in urban areas. Furthermore, the study aims to highlight the importance of vegetation as a sustainable solution for mitigating the adverse effects of urban heat island and enhancing urban livability.METHODS: Satellite data from Sentinel-۲ and Landsat ۸ for ۲۰۱۶–۲۰۲۲ were used to develop a ۲۰-meters grid resolution dataset, resulting in approximately ۲ million points. Random Forest was employed to predict land surface temperature, followed by systematically adjusting normalized difference vegetation index values from -۱۰۰ percent to +۱۰۰ percent in ۱۰ percent increments. Urban heat island was classified based on standard deviation thresholds. The results were analyzed and compared visually using geographic information system, incorporating spatial variations and heat intensity patterns to better understand the urban landscape.FINDINGS: Adjusting normalized difference vegetation index values showed a nonlinear relationship with Land Surface Temperature predictions, where certain thresholds caused unexpected decreases in Land Surface Temperature. Urban heat island classifications identified distinct urban regions with varying heat intensities. The visual comparison highlighted significant differences between the base case and alternative scenarios, revealing the sensitivity of land surface temperature to vegetation density. the results also emphasized the role of high normalized difference vegetation index values in cooling urban regions and reducing urban heat island intensity, while extreme reductions in vegetation led to potential misclassification of water bodies, creating anomalies in cooling patterns. The results of this research provide information on important variables that affect the changes in the urban heat islands, focusing on changes in vegetation, which can be a part of decision-making to improve urban planning in the future.CONCLUSION: The study demonstrates the influence of normalized difference vegetation index on urban heat island classification and its potential in urban planning strategies. By highlighting nonlinear trends, the research underscores further the need to explore vegetation dynamics in land surface temperature predictions. The findings contribute to a deeper understanding of urban heat island effects and provide a basis for enhancing machine Learning models and urban planning frameworks. Future studies could expand to other urban areas, incorporate additional variables, and refine predictive algorithms for broader applications. This study will serve as a foundation for the development of future real-time monitoring tools that will enable proactive and sustainable solutions to UHI problems.

کلیدواژه ها:

Land surface temperature ، Machine Learning (ML) ، Random Forest ، Remote Sensing ، Urban Heat Islands (UHI)