| Forest fire is a natural ecosystem process that has a great impact on environmental protection,ecosystem balance and climate change,which can change ecosystem composition and biodiversity,and has an important impact on atmospheric carbon balance and global climate change.The identification of forest fires has always relied mainly on manual patrol and satellite identification,but the manual operation has been gradually eliminated due to its slow transmission speed,small scope and single information.Satellite forest fire identification has the advantages of wide range,high time frequency and high accuracy.Previously,the polar orbit and geostationary satellites with low spatial resolution(≥1 km)were mainly used for forest fire identification,and the accuracy of forest fire identification was relatively low,which could not meet the current requirements of forest fire fighting.In this study,aiming at the problems of low spatial resolution and incomplete information acquisition dimension of satellite forest fire recognition,a forest fire recognition algorithm which can be used in Landsat 8 OLI images is proposed.The algorithm combines the traditional forest fire identification algorithm with the geographic object-based image analysis(GEOBIA)for forest fire identification.The algorithm is applied to a Landsat 8 OLI image in six areas where forest fires are active,including China,the United States,Canada,Australia,Indonesia and southern Africa.Comparing Landsat 8 OLI images with remote sensing images before fire,the misclassification error and missing error were evaluated.The main results are as follows:The Landsat 8 OLI images of 6 experimental sites all have wrong scores and missed scores by using this algorithm.The error rate ranged from 10.89%to 17.45%,with an average error rate of 14.58%.The leakage rate ranges from 3.12%to 6.32%,and the average leakage rate is 4.55%.The misclassification rate and misclassification rate of Landsat 8 OLI images at six experimental sites differ by 6.56%.The top three fire pixels identified by this algorithm are Canadian,American and Australian images,with 11,370 pixels,3,356 pixels and 3,312 pixels respectively.If the misclassification rate is not 0%,it means that some background pixels in non-fire areas are identified as fire spots by the algorithm.The average misclassification error of this algorithm in these three research areas is less than 5%,while the average missing error is less than 17%.In this paper,a detailed comparative study was made on the forest fires that raged continuously in Northwest Canada from June to August,2014,and the statistics on the recognition accuracy and effect of single fire point,small forest fires and large forest fires were completed.The results show that the background characteristics of different recognition areas have a great influence on the accuracy of the algorithm for geographic object recognition in this study.At the same time,it is found that the abnormal points in the algorithm recognition are caused by "supersaturation" of high temperature forest fire pixels.The algorithm can identify the forest fire with a single pixel(30m×30m).Compared with the current satellite forest fire identification technology with spatial resolution of 1km,it can detect forest fires with an area below lha and draw specific forest fire information(spatial location and area).The algorithm can be applied not only to Landsat 8 OLI remote sensing images,but also to other terrestrial satellite sensors(Sentinel-2,etc.).This study can not only provide reference for remote sensing identification of forest fires,but also provide technical support for quantification of forest fire characteristics. |
PDF Full Text Request