Forest Fire Damage Assessment and Biomass loss using Sentinel-2 Satellite Imagery in Doi Inthanon National Park in Chiangmai Province, Thailand

Abstract

Forest fire is a significant disturbing factor in ecosystems that influence land cover alteration. Now, it is generally realized that the global forest is greatly affected by fire. This research is the application of geoinformatics and remote sensing techniques in the analysis of the forest fire damage area in the Doi Inthanon national park in Chiangmai Province. The Sentinel-2 Satellites were used to analyze the burned area and forest monitoring due to Sentinel-2 satellite image data has a medium and high spatial resolution and temporal resolution of five days. From Thailand forest fires report, it is found that Thailand has problems due to forest fires every year and the impacts are quite devastating, particularly in northern Thailand. Forest fires tends to increase over recent decades, with most of the increase in dry season beginning from January to May with its peak in March. According to the statistics of forest fires occurrences in Thailand in 2016-2019, it shows the results of more than 7 million rai or approximately 11,538.4272 square kilometers of burned areas (DNP, 2019; GISTDA, 2019). Therefore, the solution to solve this problem should be significantly taken into account. This study aims to detect forest fire burned scars and the classification of burn severity in Doi Inthanon National Park in 2019. We found that Forest fires in the studied area occurred during the period from January 1 to May 31, 2019 and the highest is in March. in the study area, we found 553 of active fire hotspots and most of these hotspots are in the area of deciduous forests, these hotspots have been conducted by the Fire Information for Resource Management System (FIRMS, 2019). The analysis of the damaged areas of forest fires were calculated using Differenced Normalized Burn Ratio (dNBR), Relativized Burn Ratio (RBR) and Burned Area Index for Sentinel-2(BAIS2) and compared with the burn area published by the Geo-informatics Technology Development Agency (GISTDA) of Thailand. The result showed that RBR, BAIS2 and dNBR are good performance and consistent with the actual area of the forest fire. The overall accuracy of the detection of burned area of three indices were 90.00%, 87.14%, 81.43% (Kappa coefficient = 0.81, 0.75, 0.64), respectively.

Moreover, when considered the fire problems mentioned above, it can be said that forest is the area that is profoundly affected by the fire. Biomass and ecological systems were destroyed by fire. The loss of the biomass causes various impacts on forests and environment. Forests also play a major role in absorbing carbon dioxide (CO2) throughout the process of photosynthesis. This stage produces organic substances; carbon-based components that are stored in various parts of the tree, known as Aboveground biomass. For this reason, the studies on evaluation of biomass and carbon storage are very important. Remote sensing has been applied in the field of forest surveys for decades and has become a quality method for estimating aboveground biomass (AGB) and carbon stocks of trees. Lately, Random Forest (RF) and Support Vector Machine (SVM) in machine-learning model were used to improve the accuracy of satellite image analysis with a more complex algorithm. In this section, the burned area of the RBR index is used to assess the loss of above-ground biomass, and the result found that most of the burned areas are the deciduous forests. This study aims to evaluate the ability of Sentinel-2 images using vegetation indices and AGB obtained from field measurements, to compare and evaluate the accuracy of the AGB prediction models and to create the map of AGB loss in the burned area by optimal models. The Vegetation indices (Vis) combine and forest inventory parameters were used to estimate above-ground biomass loss in damage area from forest fires and compare the accuracy of the biomass model using Machine Learning method (RF, SVM). 47 sample plots and 5 vegetation indices in dry dipterocarp forest and mixed deciduous forest area were considered to calculate the loss of aboveground biomass. Evaluate the efficiency of the model using The 6-fold cross-validation. The results show that the RF model is the lowest root mean square error (RMSE= 6.04 Mg/ha) from predicted and observed AGB (Ws+Wl+Wb) and highest coefficient of determination (R2 = 0.98). The average predicted AGB was 108.815 Mg/ha. Mean absolute error (MAE) of RF was 4.17 Mg/ha. Consequently, The results show that the RF model AGB of leaves (R2 = 0.93, RMSE= 0.277 Mg/ha). The average predicted AGB loss of leaves was 2.91 Mg ha. Thus, the RF model was the most accurate model for estimating AGB in this study area. These results demonstrated that Sentinel-2 satellite is a source of valuable information for the monitor after the fire event. It is also suitable for assessing the changes in vegetation, forestry, agriculture and other areas. Furthermore, the application of satellite data with machine learning algorithms indicates the effective potential to assess aboveground biomass.

-