Abstract: [Objective] In this study, the influences of landscape pattern on forest fires in Yunnan Province were studied. The landscape index was calculated and the correlation analysis was conducted to examine the influences of different land use types on forest fires, as well as the influence strength of the increasing fire burned area. This study aimed to lay a theoretical foundation for future in-depth study on the impact of forest landscape on forest fires.[Method] Forest fires were classified based on the Forest Protection Regulations combined with the actual situation in the study area in Yunnan Province. The remote sensing images were pretreated and reclassified, and the land use classification diagram of Yunnan Province was granted. The ArcGIS grid method was used to grid and seven kinds of landscape indexes were chosen[number of plaques (NP), mean patch size(MPS), patch density(PD), contagion index(CONTAG), patch richness(PR), patch richness density(PRD),Shannon's diversity index (SHDI)] and calculated. The correlation analysis and the Bootstrap method were applied to study the relationship between forest fires and the landscape index, furthermore to investigate the influence of landscape pattern on forest fires.[Result] A total of 1457 fires occurred during 2005 to 2008 in Yunnan Province, and among which the 98 were major fires. Within the grid with the side length of 10 km, the chance of high-grade forest fire occurrence was increased with the increase of the average plaque area of coniferous forest, broadleaf forest, grass and shrubs, but it decreased with the increase of mixed coniferous and broad-leaved forest, river and wetland. In more than 100 hectares of forest fires, the correlation coefficient of contagion, Shannon diversity index, landscape shape index, patch density, the number of plaques and forest fire area were -0.022, 0.300, 0.253,-0.333, -0.360 and -0.302, respectively. All correlations except for that with landscape shape index were significant. In 0-1 hm² of forest fires, the correlation between the contagion index and the forest fire area was believable at the 0.01 level. For forest fires with burned area of 50-100 hm², Bootstrap significance of CONTAG and SHDI was credible at 0.01 and 0.05 levels. When forest fire area was more than 100 hm², burned area was negatively correlated with SHDI with the negative correlation coefficient of -0.333, and the correlation was significant at the 0.01 level.[Conclusion] The increase of the average patch area of coniferous forest, broadleaf forest, grassland and shrub promoted the spread of forest fires, while the increase of the average patch area of the mixed broadleaved-conifer forest and wetland inhibited the spread of forest fires. Through the Bootstrap test of significance, the influences of Shannon diversity index, landscape abundance and spread index on burned area dropped in the range of 0-1 hm² fire burned area. However, in the range of 50-100 hm² fire burned area, the influences of spread index, Shannon diversity index and plaque quantity on burned area decreased. The correlations of forest fire and Shannon diversity index, the patch number and the patch density were changed to more negative with the increase of fire rating, and the inhibiting effect on forest fire gradually increased. On the contrary, the correlation of forest fire and spread index was changed to more positive with the increase of fire rating, and the promoting effect gradually increased. The impacts of landscape indices on fire burned area were obvious for major forest fires.

Key words: landscape pattern, landscape index, forest fires