| In recent decades, forest fire happen more and more frequently. Most of the forest fires are out of control burning under the synergy of human and natural factors. They not only ruthlessly ruin various species in forest, destroy the balance and stability of terrestrial ecosystem, generate a large amont of smoke particles which lead to the atmospheric and environmental pollution and negative effect of local climate, or even seriously threaten the living environment for human, but also cause significant loss of life and property of country and people. As a result, taking necessary actions and measures for predicting and finding forest fire, and then suppressing the forest fire or fighting it at the beginning phase is an arduous and fundamental mission for carrying out the work of forest fire prevention.In this paper, the topic of our research is carrying out the prediction, detection and assessment of forest fire in pre-fire, fire occurrence and post-fire phases respectively by using satellite remote sensing techniques. The aim is to build the fire danger dynamic model and the smoke detection algorithm that are able to monitor the status of the study area in time, and then conduct the analysis and assessment of vegetation in post-fire, which can provide abundant technological support for the work of forest fire prevention, improve the accuracy of fire prediction and detection and minimize the losses in the end. The basic idea is to build the fire dynamic danger model and smoke detection algorithm for fire prediction and monitoring and then conduct the statistics analysis of vegetation in post-fire based on the differences among various cover type’s spectral response in different ranges of electromagnetic spectrum to find the fire-related parameters.In order to be good knowledge and understanding of the application of remote sensing technique to forest fire prediction and dection, in Chapter 2, we introduced the definition of satellite remote sensing as well as its foundation of physics, and the current platform of satellite remote sensing by means of literature investigation and data collection. We also analyzed the main satellite remote sensing data fit for fire detection for further discussing the data (EOS/MODIS) source and building the satellite remote sensing platform used in this study. Finally, we introduced the typical application of fire detection by means of satellite remote sensing to deepen the acquaintance of remote sensing of forest fire.In order to define the high fire danger zones, that is to say, conduct accurate fire danger assessment in the study area, in Chapter 3, the fire dynamic factors, such as the fuel moisture content (FMC), land surface temperature (LST) and relative greenness index (RG) were used to integrate the fire danger dynamic model based on the previous models. In order to validate and assess the accuracy and feasible of the model used in the study area, the recevicer operating curve technique were used to analyzed the MODIS data acquired over the northeastern China on 26th,27th and 28th June,2010. Results pointed out that the predicted accuracy of fire danger level (the hot-spots were defined as high fire danger zones) in these three days were 76.338%、 88.853% and 80.910%. Thus, the forest fire danger dynamic model proposed in this chapter was considered to be applicable in the study area.In order to actomatically discriminate the smoke pixels from the background pixels, in Chapter 4, we proposed an identification algorithm based on the spectral analysis among the smoke, cloud and underlying surface pixels. In order to get satisfactory results, multi-threshold method (is used for extracting training sample sets) and back-propagation neural network classification were used for merging the smoke detection algorithm. The MODIS data from three forest fires were used to develop the algorithm and get parameter values. These fires occurred in 1) northeastern China on 16th Oct,2004 (autumn); 2) northeastern Asia on 29th April,2009 (spring) and 3) Russia on 29th July,2010 (summer) in different seasons. Then the data from four other fires were used to validate the applicability and robustness of the algorithm. Results indicated that the algorithm captured both thick smoke and thin dispersed smoke over land, as well as the mixed pixels of smoke over the ocean. These results could provide valuable information concerning forest fire location, fire spreading and so on.In order to analyze and assess the influence of fires on vegetation, in Chapter 5, the empirical analysis and normalized burned ratio (NBR) were integrated to map the burned area and extract the fire-affected pixels. And then the MODIS-NDVI time series data (2002-2014) of fire-affected and fire-unaffected pixels in the study area were alalyzed based on two statistics method:detrended fluctuation analysis (DFA) and Fisher-Shannon (FS) method. Our findings show that 1) the results obtained by jointly using the two methods are consistent, enabling the characterization and discrimination between fire-affected and fire-unaffected areas; 2) in particular, among the investigated indices, FIM (Fisher Information Measure) is the most significant estimators in discriminating burned and unburned sites; 3) the mean value of FIM for burned sites is about 2.5 that is significantly larger than that obtained for unburned sites (～1.3); 4) FIM is also associated the larger effectiveness in discriminating two classes of sites based on the ROC analysis of its performance; 5) the scaling exponents estimated from DFA of fire affected pixels are averagely higher than those of fire-unaffected ones, which means the fire affected pixels are characterized by higher organization and lower disorder degree. In conclusion, both of the two methods could contribute to identify the impact of fires on vegetation.In the end, in Chapter 6, we looked back and summarized our work and presented the main conclusions, meanwhile pointed out further research plan based on the limitation of this work. |
PDF Full Text Request