Spatio-temporal analysis of the relationship of land use/land cover and land surface temperatures and its modeling for the future

In this study, it is aimed to examine the effects of Land Use/Land Cover (LU/LC) changes on Land Surface Temperature (LST) values and regional climate. For this purpose, the main stages of the study are (i) determining the LU/LC areas between 1984-2022, examining the temporal changes of these areas and analyzing the LST values to reveal the current situation (ii) modelling the LU/LC areas for the years 2030, 2040 and 2050, simulating the LST values for the years between 2023-2050 and (iii) analyze the change of the Urban Heat Island (UHI ) effect in the region selected as the study area during the study period. While the Mediterranean Region is considered among the most sensitive regions to climate change and annual temperatures are predicted to increase by 1-5°C by the year 2100, with the additional heat load caused by the UHI effect, Mediterranean cities may be exposed to greater heat stress and negative health effects than other regions. Therefore, UHI, LST and climate studies carried out in this region are of great importance. For this reason, Antalya Basin was determined as the study area. In this context, firstly using the Google Earth Engine (GEE) platform annual multispectral bands, land cover indices and LST data were created using all summer months images (940 images) that have cloudiness lower than 5% in the period from 1984 to 2022, when access to Landsat thermal data began. Using multispectral bands, land cover indices, ASTER Digital Elevation Model and nighttime lights data 5 different data sets were created. Classification was applied to the data sets using the Random Forest (RF) machine learning algorithm for the years 1985, 1990, 1995, 2000, 2005, 2010, 2015 and 2020, and LU/LC thematic maps were created from the data set with the highest accuracy. By analyzing the created LU/LC thematic maps and the produced LST values together, the relationship between land classes and LST values in the region between 1984-2022 was revealed and UHI density classes were determined. In the next stage of the study, the identified land classes were modelled for the years 2030, 2040 and 2050 with Geographic Information Systems (GIS) data and Cellular Automata-Artificial Neural Networks (CA-ANN) algorithms. LST values were estimated for the period between 2023 and 2050 using different methods. For this process, firstly, predictions were made with the simple Linear Regression method, then, LST predictions were made for the years 2030, 2040 and 2050 using the modelled LU/LC classes and the modelled land cover indices and the RF regression method. Finally, UHI density classes were simulated for the years 2030, 2040 and 2050 with the CA-ANN algorithm. According to the results, the classification process has been successfully implemented with an overall accuracy value of 85% and above, and the auxiliary data used increased the classification accuracy. LU/LC thematic maps show that urban, industrial, agricultural and greenhouse areas increased in the studied period, while green areas and other areas decreased. The calculated LST values were compared with MODIS LST values and meteorological station measurements, and it was seen that there was a very high correlation of 0.96 and above between MODIS LST and Landsat LST values. Additionally, it was determined that there was a significant correlation between Landsat LST values and temperatures measured by meteorological stations. LST values monitored in the period 1984-2022 in all LU/LC classes tend to be increase. However, the highest LST values were observed in urban, industrial, and dry agriculture areas, and their upward trends were higher than other areas. In addition, with the LU/LC simulations made for 2010 and 2020, it was seen that the simulation process could be applied with approximately 90% accuracy with the CA-ANN algorithm. According to the simulation results, it is predicted that the city will grow, industrial and dry agricultural areas will increase, and green areas will decrease in 2030, 2040 and 2050. Future LST values have been determined by the RF regression method with an Out of Bag (OOB) error around 2°C and an r2 value over 0.80, and according to the LST values simulated by all 3 methods, temperatures will rise in the 2023-2050 period and in addition UHI intensity will increase.