Landsat-8 and Sentinel-2 based Forest fire burn area mapping using machine learning algorithms on GEE cloud platform over Uttarakhand, Western Himalaya

The accurate quantitative and qualitative estimation of burn-area are crucial to analyze the impact of fire on forest. The medium resolution optical-satellite imagery of Landsat-8 and Sentinel-2 are employed covering the period 2016 to 2019 for forest fire patches identification on Google Earth Engine (GEE). The most indispensable season of Forest Fire (FF) is pre-monsoon in Uttarakhand, western Himalaya, India. Bi-temporal (pre and post fire) reflectance contrast of burn-sensitive spectral bands was used to compute differential spectral indices, namely, Normalized Burn Ratio (dNBR), Normalized Difference Vegetation Index (dNDVI), Normalized Difference Water Index (dNDWI), and Short-Wave Infrared (dSWIR). The differential spectral-indices composite is further used as an input to unsupervised Weka clustering algorithms for capturing the shape and pattern of fire patches. Sample training-data of burn and unburn classes were collected with reference to thermal and optical spectral principle. Classification Regression Tree (CART), Random Forest (RF), and Support Vector Machine (SVM) algorithms have been employed to identify FF. The key findings revealed that CART and RF algorithms displayed similar forest fire patches with an overall accuracy of 97–100%. The classification accuracy is slightly lower in SVM and its underestimating forest fire patches detections. Landsat-8 OLI derived burn area was fitted better with fire product of Climate Change Initiative (Fire-CCI of ESA) and MCD64A1 of MODIS burn area product with R-square of 0.71–0.93 and 0.62–0.91, respectively which attributed to better spectral bands of Landsat-8 than the Sentinel-2. However, Sentinel-2 bands have the potential to capture fire patches during post-fire events. This study has demonstrated the potential utilities of combined effort of unsupervised and supervised algorithms on Landsat-8 and Sentinel-2 on GEE to identify fire patches.