Global fire cases in recent years and their vast damages are vivid reasons to study the wildfires more deeply. A 25-year period natural wildfire database and a wide array of environmental variables are used in this study to develop an artificial neural network model with the aim of predicting potential fire spots. This study focuses on non-human reasons of wildfires (natural) to compute global warming effects on wildfires. Among the environmental variables, this study shows the significance of temperature for predicting wildfire cases while other parameters are presented in a next study. The study area of this study includes all natural forest fire cases in United States from 1992 to 2015. The data of eight days including the day fire occurred and 7 previous days are used as input to the model to forecast fire occurrence probability of that day. The climatic inputs are extracted from ECMWF. The inputs of the model are temperature at 2 meter above surface, relative humidity, total pressure, evaporation, volumetric soil water layer, snow melt, Keetch–Byram drought index, total precipitation, wind speed, and NDVI. The results show there is a transient temperature span for each forest type which acts like a threshold to predict fire occurrence. In temperate forests, a 0.1-degree Celsius increase in temperature relative to 7-day average temperature before a fire occurrence results in prediction model output of greater than 0.8 for 4.75% of fire forest cases. In Boreal forests, the model output for temperature increase of less than 1 degree relative to past 7-day average temperature represents no chance of wildfire. But the non-zero fire forest starts at 2 degrees increase of temperature which ends to 2.62% of fire forest cases with model output of larger than 0.8. It is concluded that other variables except temperature are more determinant to predict wildfires in temperate forests rather than in boreal forests.