Introduction

Currently, many scientific evidences indicate a strong correlation between the earth's surface and its functional features and coverage in cells. Various patterns of settlement development have a direct impact on the formation of urban heat islands, and the use of spatial models plays a crucial role in enhancing the understanding of these effects. Therefore, in the present study, a cellular automata model was employed to predict the future development of residential areas in northern Iran (Guilan plain) for the years 2035 and 2050 under three urban growth scenarios: Business as Usual (BAU), Environmental Protection (ENV), and Compact Growth (COM).



Methodology

In this research, a combined modeling approach was utilized to extract historical patterns of residential expansion and land surface temperature for the years 2002, 2012, and 2022, using cloud-free Landsat images. Subsequently, the cellular automata model was employed to predict the future development of residential areas for the years 2035 and 2050 under three different scenarios. The relationship between land surface temperature and residential areas was examined to identify the most effective regression models explaining the impact of residential expansion on the average land surface temperature of the region.

Urban growth modeling was executed using the Cellular Automata-Markov model. For this purpose, settlement layers from 2002 and 2012 were employed to train the Markov model, and the transformed land-use ratio was estimated for constructed areas until 2022. Additionally, layers from 2002 and 2022 were utilized to determine the area and probability of land transformation until 2035 and 2050. A multi-criteria evaluation method was applied to prioritize land transformations and formulate various urban growth scenarios. The factors for the multi-criteria evaluation included soil quality, erosion, fault lines, proximity to roads, provincial centers, various sizes of urban areas, as well as forest and agricultural categories. Constraints, scenario-dependent, considered river basins (and their 200-meter boundaries), water bodies, impermeable surfaces, and forested areas. Ultimately, a weighted linear combination method was employed to integrate layers of factors and constraints, creating suitable layers for each scenario, including the Business as Usual (BAU) economic growth scenario with the aim of simulating growth, Environmental Protection (ENV) scenario focusing on environmental conservation, and Compact Growth (COM) scenario aiming to prevent the fragmentation of agricultural lands. For the integration of layers, a hierarchical analysis process was utilized, and for the transformation of values for each factor and constraint layer into dimensionless comparative scales, fuzzy membership functions and Boolean logic were applied. For modeling land surface temperature, two spatial feature-centric approaches were employed, namely the patch-based and segment-based perspectives.



Results and discussion



The land surface temperature layers of residential areas were retrieved from three Landsat satellite images for the years 2002, 2012, and 2022, with average values of 33.14, 36.38, and 34.78 degrees Celsius, respectively. Statistical analysis results indicated that spectral segments extracted from object-based image analysis provide more accurate predictions of the average land surface temperature. Using the area of spectral segments and their common border percentage with neighboring segments, a regression model was constructed to predict the land surface temperature of residential areas for the years 2035 and 2050 (R2 = 0.617). The highest average land surface temperature was obtained under the COM scenario (33.88 degrees Celsius) in 2050, while the lowest was observed under the ENV scenario in 2050 (31.23 degrees Celsius).

The results of this study indicate that this correlation will be meaningful at smaller structural scales. In other words, urban blocks act as distinct residential units, each having its specific land surface temperature. Based on these findings, it seems that studying land surface temperature at the block scale, especially in larger and interconnected residential areas, offers greater capability in estimating land surface temperature. Despite this approach, structural criteria of segments could not establish a significant correlation with their average land surface temperature.

Conclusion



The results illustrate that the land surface temperature of residential areas may not accurately reflect the historical trend of temperature changes due to local land conditions. Therefore, models based on the relationships between temperature and land surface features were used for modeling in three different time periods. The extent of residential segments demonstrated a strong statistical association with the average land surface temperature, while structural features, such as neighborhood parameters, could not predict changes in residential land surface temperature. Based on the findings of this study, the average land surface temperature of residential areas is dependent on both the spatial configuration of residential spectral segments at the local scale and the patterns of residential area expansion (scenarios) at the regional level.