| The biggest issues for our forestry management and research department are how to forecast the forest fire effectively and scientifically and limited the loss those caused by forest fire and decreased the number of forest fire greatly. There was a big disparity comparing with some developed country because of our study started relative late, so this paper made a deep research on the distribution of forest fire and modeling the relationship between forest fire and weather and some spatial factors in Daxing'an mountain in Heilongjiang Province of china basis on some statistic and spatial analysis software, for instance, ArcGIS, S-Plus, SAS, R, GWR2 and so on. The results of this paper are very meaningful and can provide a scientific thesis foundation for forest fire's forecast.The first step for this paper is make a simple analysis on forest fire data of Daxing'an Mountain in Heilongjiang Province, the results show that the number of spring forest fire was decreased after 1987 year and it was increased for summer fire. The change trend of climate for the study area also be studied, which include temperature, precipitation, relative humid.Computing and judging the types of distribution of lighting fire and human-caused fire using K-function and L-function meanwhile combine with S-plus software. The results show that lighting fire is cluster distribution and the peak in 200Km scale. It shows random distribution over that scale in the period of 1973-1975. The distribution type of lighting fire during 1976-1981 is similar with 1973-1975. The main distribution type of lighting fire is cluster and no obvious peak during 1982-1984. It is cluster distribution but the degree is not high in the period of 1985-1987. Except the random distribution of 1991-1993, the other periods are cluster during 1988-2005. When the distance scale is 350Km, it shows cluster peak during 1988-1990 and when the scale less than 450Km the lighting fire is cluster, greater than 450Km are uniform distribution in the period of 1994-1996. It is cluster distribution when the scale less than 300Km and it is uniform distribution when the scale greater than 300Km during 1997-1999 and the results are similar with the period of 1999-2005.the main distribution type of lighting fire is cluster in the period of 1973-2005, the cluster peak is in 260Km scale, namely the lighting fire are cluster distribution under small distance scale.The distribution of human-caused fire during 1972-1976 are cluster, the peak appears at 200Km scale. It is cluster distribution under 400Km scale during 1977-1981, and it also happens in the period of 1982-1991 but it is random distribution tendency of human-caused fire during 1992-1996. It is cluster distribution when the scale less than 300 Km and if scale is greater than 300 Km, it would be uniform distribution,1997-2001.The human-caused fire are cluster distribution in all different scales during 2002-2005.Computing and modeling the spatial density of lighting and human-caused fire occurrence by the use of Kernel spatial function and the results show that there are several hot points and lower hot points in Daxing'an mountain area during 1973-1977, the coordinates of cores of hot and lower hot points are 123°23'E,52°12'N和123°58'E,51°11'N,separately, and located in north HuZhong forestry bureau and west of SongLing forestry bureau. In the period of 1978-1982, the lighting fire exist two obvious hot points and the coordinates of cores are located in 122°26'E,52°38'N and 124°08'E,53°17'N, located in much area of southwest of TuQiang forestry bureau and north of Tahe bureau. Two hot points exist in Daxing'an mountain area during 1983-1987. One is in the junction of HuZhong and north of XinLin forestry bureau and the other is in the junction of north of TaHe and east of HuZhong bureau, the coordinates of cores are located in 124°06'E,52°04'N and 124°04'E,52°23'N, respectively. duringl988-2005, the coordinates of cores are 123°06'E,52°20'N,123°41'E,51°34'N and 124°08'E, 50°48'N. The hot points of human-caused fire in Daxing'an mountain area are similar each other among the period of 1972-1976,1977-1981,1982-1986,1987-1991, all the hot points are located in 124°10'E,50°23'N, inside JIaGedaqi forestry bureau. There are no hot points of human-caused fire in Daxing'an mountain during 1992-1996. It is different comparing with previous periods, the coordinates of cores of human-caused fire during 1997-2001 are 52°09'N,125°55'E, located in north of Hanjiayuan bureau. The number of human-caused fire occurrence during 2002-2005 is bigger than other periods, the fire happens mainly focus on south of Daxing'an mountain, the coordinates of cores are 51°35'N,125°25'E; 51°13'N, 125°27'E; 51°07'N,124°38'E; 50°34'N,125°26'E; 51°16'N,123°41'E and located in the north junction of HuMa and SongLing bureau, east junction of SongLing and JiaGedaqi bureau, in the middle of SongLing area and the junction of east of JiaGedaqi, HuZhong and SongLing bureau.Besides traditional OLS model, we choose Poisson, Zero-inflated Poisson (ZIP), negative binomial (NB) and Zero-inflated negative binomial (ZINB) models to model and analysis the relationship between lighting fire, human-caused fire and weather factors, furthermore, comparing those five models by the use of AIC and Voung methods and find out the best model. The results show that ZINB is the best model for fitting lighting fire data and weather factors data, the prediction model as following: log(λ= E(Y))=-3.637929+0.124678MAT+0.010395 MAE and log it(p)=-72.175944+0.962976 MARHMeanwhile the ZINB model are also the best model for modeling and analyzing the relationship between human-caused fire and weather factors, the prediction models we get are log(λ=E(Y))=-0.6534+0.0056MAE and logit(p)=-20.4509+0.0175MAP-0.2615MAT+0.3183MARH.By the use of ArcGIS, Arcmap, GWR2 and Moran software to analyze the relationship between forest fire occurrence and spatial relative factors. This thesis to analyze compute and compare the models basis on different spatial scale which included 5KmX5Km grid and random choosing samples two scales. The models we choice were global logistic and GWR model. The result shows that the spatial factors, distance to street and distance to railway, have a significant correlation with fire occurrence, and the Global Logistic model we get is where x1is the distance to street; x2 is the distance to railway. The probability function we get from the model result of GWR is a matrix function, the intercept coefficient (IC) have five levels, they are IC≤-0.4,-0.4 |
PDF Full Text Request