Iran is a fire-prone country because of its climatic conditions. An extensive area of the forests and pastures of this country has been burned by wildfires in recent decades. Golestan province in northeastern Iran is the most fire-affected region in Iran in terms of fire frequency and extent. Therefore, the current study was done to evaluate the spatio-temporal relationships among meteorological factors and fires in this region in recent years. For this purpose, regression analysis and machine learning models were applied. The fire regime statistics were frequency and extent of fires, while meteorological variables were mean temperature, maximum temperature, absolute maximum temperature, mean precipitation, mean relative humidity, mean wind speed, and maximum wind speed in fire season for a 26-year period. For analyzing the temporal relationships between fire statistics and meteorological variables, Pearson's correlation was applied. For those significant correlations between fire statistics and meteorological variables, the regression relationships were calculated. For spatial relationships between fire statistics and meteorological variables, machine learning models were used. First, a fire map of Golestan province was created by fire data available in Golestan Natural Resources and Watershed Administration (GNRWA) and Moderate-Resolution Imaging Spectroradiometer (MODIS) sensor. The maps of meteorological variables were also prepared by Inverse Distance Weighting (IDW) procedure in GIS. The spatial role of meteorological variables on fire susceptibility was determined by Mean Decrease Gini (MDG) and Mean Decrease Accuracy (MDA) parameters. Four machine learning methods (support vector machine, random forest, logistic regression, and SVM-RF methods) and 70% of fire occurrences were applied for spatial modeling and mapping of fire susceptibility in the study area. Area Under the Curve (AUC) and 30% of fire occurrences were used for evaluating the model's performance. The findings demonstrated that 4466 fires have been occurred in the study area, burning 14,907.09 hectares of natural areas in Golestan province for 26 years. The findings of temporal relationships revealed the significant relationships between frequency of fires and mean temperature, maximum temperature, absolute maximum temperature, mean precipitation, mean wind speed, and maximum wind speed in fire season during 26 years. In addition, significant relationships were observed among extent of fires and mean precipitation, and mean relative humidity in fire season. The results of spatial relationships showed that absolute maximum temperature, mean precipitation, and mean relative humidity had the most importances in fire susceptibility in the study area. Validation of fire susceptibility maps demonstrated that RF and SVM-RF methods (AUC: 0.83) were the most precise models for fire susceptibility mapping in the study area. Therefore, forecasting the future fires in Golestan province is possible based on these maps. Results of this study depict the effective role of long-term climate change (mainly increasing temperature and decreasing precipitation) on fire occurrence, fire frequency, and fire extent in natural areas of Golestan province in northeastern Iran. Among all climatic factors, temperature was the most important parameter in fire regime in both temporal and spatial scales. Increasing temperature due to climate change and global warming can change the patterns of fire regime in the study area over time and space. Therefore, protective management of forests and pastures of Golestan province against fire is necessary especially in the warmer locations during fire season at spatio-temporal scale.