University of Tehran Press Journal of the Earth and Space Physics 2538-371X 50 3 2024 10 05 Using the multivariate linear regression method to model the 2-meter air temperature from MODIS sensor data Using the multivariate linear regression method to model the 2-meter air temperature from MODIS sensor data 803 821 98112 10.22059/jesphys.2024.376789.1007609 FA Mohammad Amin Mohammadi Ahoei Department of Surveying Engineering, Faculty of Civil Engineering, Babol Noshirvani University of Technology, Babol, Iran. 0009-0003-4009-6852 Ali Sam-Khaniani Department of Surveying Engineering, Faculty of Civil Engineering, Babol Noshirvani University of Technology, Babol, Iran. 0000-0002-3771-6761 Journal Article 2024 05 25 The air temperature near the earth's surface is one of the influential variables in various climate studies, hydrology and weather forecasting. In many areas, this parameter is usually measured with the help of weather stations located on the ground. Due to the lack of uniform spatial distribution of weather stations in areas with the different topography, in many inaccessible or unpopulated places, enough ground stations are not available to observe and record surface air temperature data. On the other hand, remote sensing satellite images are used as a potential alternative to describe temperature patterns with appropriate spatial details in large areas. Land Surface Temperature (LST) is prepared with the help of satellite observations. Although the LST product is related to the temperature of the air near surface (T2m), they have different behavior and characteristics. Therefore, many researchers, in order to overcome the limitation of ground-based air temperature data spatial resolution, try to establish a relationship between near surface air temperature and satellite LST.The provinces along the Caspian Sea, such as Gilan and Mazandaran, are very important from various climatic, economic and agricultural aspects. Due to the vastness and diverse topography of these areas, the number of synoptic stations available in these areas is limited. On the other hand, so far, 2m air temperature modeling using satellite data has not been done in this region.The main goal of this study is to create a suitable model for estimating this parameter using satellite LST data. For this purpose, using multivariate linear modeling method, a model was created between MODIS sensor LST and T2m air temperature in Gilan and Mazandaran region. The parameters used in this model include LST obtained from MODIS sensor, height, Normalized Vegetation Index (NDVI), slope and curvature. The data collected between 2000 and 2017 were used to estimate the coefficients of the model, and the data from 2018 to 2020 were used to evaluate the model. In each province, separate models were estimated for night and day data and two different height categories. The comparison of the provincial model with the single-station model in terms of error statistics showed that the provincial model has little difference with the model constructed for each station. The results showed that the RMSE values in the provincial model are on average in the range of 2.60-3.11 degrees Celsius. The correlation coefficient of the T2m values obtained from the model with real observations was estimated to be more than 90%. In addition, the data of different seasons were separated and a separate model was presented for each season. On average, the use of the seasonal model led to an improvement in the estimation of T2m data in Gilan and Mazandaran provinces with RMSE of 2.72 and 2.55 degrees Celsius, respectively. The air temperature near the earth's surface is one of the influential variables in various climate studies, hydrology and weather forecasting. In many areas, this parameter is usually measured with the help of weather stations located on the ground. Due to the lack of uniform spatial distribution of weather stations in areas with the different topography, in many inaccessible or unpopulated places, enough ground stations are not available to observe and record surface air temperature data. On the other hand, remote sensing satellite images are used as a potential alternative to describe temperature patterns with appropriate spatial details in large areas. Land Surface Temperature (LST) is prepared with the help of satellite observations. Although the LST product is related to the temperature of the air near surface (T2m), they have different behavior and characteristics. Therefore, many researchers, in order to overcome the limitation of ground-based air temperature data spatial resolution, try to establish a relationship between near surface air temperature and satellite LST.The provinces along the Caspian Sea, such as Gilan and Mazandaran, are very important from various climatic, economic and agricultural aspects. Due to the vastness and diverse topography of these areas, the number of synoptic stations available in these areas is limited. On the other hand, so far, 2m air temperature modeling using satellite data has not been done in this region.The main goal of this study is to create a suitable model for estimating this parameter using satellite LST data. For this purpose, using multivariate linear modeling method, a model was created between MODIS sensor LST and T2m air temperature in Gilan and Mazandaran region. The parameters used in this model include LST obtained from MODIS sensor, height, Normalized Vegetation Index (NDVI), slope and curvature. The data collected between 2000 and 2017 were used to estimate the coefficients of the model, and the data from 2018 to 2020 were used to evaluate the model. In each province, separate models were estimated for night and day data and two different height categories. The comparison of the provincial model with the single-station model in terms of error statistics showed that the provincial model has little difference with the model constructed for each station. The results showed that the RMSE values in the provincial model are on average in the range of 2.60-3.11 degrees Celsius. The correlation coefficient of the T2m values obtained from the model with real observations was estimated to be more than 90%. In addition, the data of different seasons were separated and a separate model was presented for each season. On average, the use of the seasonal model led to an improvement in the estimation of T2m data in Gilan and Mazandaran provinces with RMSE of 2.72 and 2.55 degrees Celsius, respectively. https://jesphys.ut.ac.ir/article_98112_8e24d3c12fc3419717a7954b2594e14d.pdf