Abstract: Since forest fire is too complicated to search, extinguish and rescue, the wireless sensor network (WSN) is applied to forest fire detection and an exact model is developed to estimate the location and intensity of fire. The approach to adaptive sampling by use of sensor nodes is discussed with a multi-peak Gaussian model proposed. Combining the expectation maximum (EM) with multiple resolution (MR) algorithm, an advanced EM-MR algorithm is proposed to estimate the parameters in the forest fire model. Based on the results of comparative simulation, this model can depict the relative real site of forest fire and provide the satisfactory estimation accuracy with computation complexity significantly reduced.