Fire activity has a huge impact on human lives. Different models have been proposed to predict fire activity, which can be classified into global and regional ones. Global fire models focus on longer timescale simulations and can be very complex. Regional fire models concentrate on seasonal forecasting but usually require inputs that are not available in many places. Motivated by the possibility of having a simple, fast, and general model, we propose a seasonal fire prediction methodology based on time series forecasting methods. It consists of dividing the studied area into grid cells and extracting time series of fire counts to fit the forecasting models. We apply these models to estimate the fire season severity (FSS) from each cell, here defined as the sum of the fire counts detected in a season. Experimental results using a global fire detection data set show that the proposed approach can predict FSS with a relatively low error in many regions. The proposed approach is reasonably fast and can be applied on a global scale.