This document presents a new approach for identifying forest fires using land surface temperature satellite imagery. It involves segmenting the imagery into clusters using k-means clustering to locate regions of abnormal temperature distribution. The mean wavelength value is then computed for regions of interest using Haar wavelets. Experimental results on 312 images found that a mean wavelength exceeding 10.14 accurately identified forest fire locations according to historical records, with an average accuracy of 89.5%. This approach provides an improved method for early forest fire detection compared to existing techniques.