【Objective】 Studying the temporal and spatial characteristics of forest fires in Yunnan Province, help quantify the occurrence of forest fires, identify areas of high incidence and finally carry out risk zoning scientifically. 【Method】Based on data from Yunnan Province between 2001 and 2020 (the MODIS fire location/thermal anomaly product, MCD14DL), we studied the correlation and heterogeneity of fire locations to analyze spatial and temporal distribution by statistical analysis, center point, standard deviational ellipse, Ripley’s K function, kernel density and other methods. 【Result】(1)We observed large annual fluctuations of forest fires between 2001 and 2020, peaking in 2010. Forest fires in Yunnan Province concentrated in winter and particularly spring, with fire spots peaking in March. (2)The occurrence of forest fires varied greatly between the regions, with the highest incidence in the southwest of Yunnan Province, followed by the southeast and northwest. The northeast had the smallest number of forest fires. The median centre was Pu’er, while the northwest of Pu’er had the highest kernel density(0.43) followed by the central and western regions of Xishuang banna Prefecture, with a kernel density of 0.34, and the northwest of Dali Prefecture and northeast of Wenshan Prefecture, with a nuclear density of 0.26; these four regions were hotspots for forest fires. (3)In the past 20 years, the spatial distribution of forest fires in Yunnan Province has displayed obvious directionality; the center of gravity has gradually shifted from Pu’er to Dali-Chuxiong-Yuxi-Honghezhou. (4)The spatial distribution of forest fires in Yunnan Province fits an aggregation model (but not to a high degree) and is gradually dispersed.【Conclusion】Forest fires in Yunnan Province between 2001 and 2020 were not randomly distributed, but followed a spatial-temporal distribution law. This law can be applied to inform a regional fire management, strengthen the construction and publicity for the fire prevention in priority regions, scientifically organize fire prevention work, reduce the incidence of forest fires and realize the sustainable development of forests.
【Objective】 The China-Mongolia border is prone to grassland fires in Erenhot, the East Ujimqin Banner and the New Barag Right Banner, threatening livelihoods and the local environment. Studying the parameters of fire behavior for grassland fuel on the China-Mongolia border in windless conditions can provide a theoretical basis for the grassland fuel management and grassland fire prevention. 【Method】Based on field investigations and sample collection, a combustion bed and thermocouple were used to simulate and measure the behavior of combustibles, and the fire behavior indexes were simulated and analyzed. 【Result】Simulations for windless conditions in Erenhot, the East Ujimqin Banner and the New Barag Right Banner, respectively, showed a fire intensity of 780.44, 805.80 and 1 195.60 kW/m; a spread rate of 0.61, 0.45, 0.67 m/min; an average flame height of 30.30, 35.53,33.90 cm; average flame temperature of 481.97, 529.33 and 389.93 ℃; heat radiation, 0.45, 1.20 and 1.44 kW/m2, and blocking time of combustibles, 28, 15 and 22 s. 【Conclusion】 Long periods of temperature fluctuation are observed for grassland fires in Erenhot, and the risk of fire is high. In the East Ujimqin Banner, the smoldering heat radiation of combustibles is close to flame heat radiation and the holding power of fires is strong, making grassland fires likely.
【Objective】The fuel load is important for forest fire management. This study investigated the distribution of fuel load and analyzed the relationship between the per unit area fuel load components and volumes, within the tree layer among subtropical forest types, to provide scientific references for sustainable forest management. 【Method】Forest fire survey and supplementary investigation data were taken from Huzhou City, Zhejiang Province. Nine typical subtropical forest types were chosen as research objects. The fuel load per unit area of each component of different forest types was measured and calculated by quadratic investigation and drying methods. One-way ANOVA was used to test for differences of the fuel load per unit area among forest types and components. The linear or nonlinear correlation fitting was carried out on the fuel load per unit area of the components and the volume per unit area of the tree layer of typical forest types to analyze their correlation.【Result】The amount of the fuel load per unit area of tree layer in most forest types was significantly higher than that in the fallen dead wood layer (P< 0.05), and the fallen dead wood layer was significantly higher than other components such as the shrub layer (P< 0.05). However, the fuel load per unit area of the fallen dead wood layer in Phyllostachys edulis monoculture forests was significantly higher than that of other components (P< 0.05). The fuel load per unit area in the tree layer of Quercus spp. forests, Schima superba forests, and coniferous and broad-leaved mixed forests were higher than that of other forest types. The fuel load per unit area of the litter layer in coniferous and broad-leaved mixed forests, Schima superba forests, and other soft broad-leaved forests were higher than that of other forest types. The results of the regression analysis showed that for most forest types, there was a significant linear positive correlation between fuel load per unit area and volume per unit area of tree layer. Among them, Quercus spp. forests have the highest correlation (R2=0.89). Nevertheless, the amount of the fuel load per unit area in the shrub layer and herb layer in most forest types decreased with the increase of the amount of volume per unit area.【Conclusion】The management of subtropical forest fuel loads should take full account of the differences in the distribution characteristics of different forest types and the interrelationship among different components. The composition of tree species in the canopy layer has a significant impact on the distribution of forest surface fuels. The storage volume per unit area of tree layer is closely related to the fuel load per unit area of tree layer and is an important reference indicator for the analysis and prediction of fuel loads of forest components.
【Objective】 Forest fire risk maps are necessary to improve effective patrols and optimize the scientific layout of limited fire prevention resources. This study uses machine learning algorithms to construct forest fire occurrences based on terrain, human activities, vegetation, and meteorological factor data.【Method】With Jiushan Mountain in Chuzhou city in Anhui province as the research object, we extracted the following potential driving factors: slope, elevation, aspect, distance to settlement, distance to road, topographic wetness index, normalized difference vegetation index and temperature of the study area, evaluated the driving factors of fire occurrence, and then divided the potential driving factors into topography, human activities, vegetation, meteorological factors, and other four categories. Historical fire points in the forest area were extracted from the sentinel fire products. A prediction model for forest fire occurrence was then constructed using a machine learning algorithm. Finally, the accuracy of the models was evaluated using a confusion matrix and receiver operating characteristic curves. 【Result】We found that vegetation, temperature and distance to the road are the main driving factors of forest fires in the study area. The ROC curves of the two models showed that the Logistic regression prediction model had an accuracy of 71.07%, where the area under the curve was 0.717 2. Meanwhile, the random forest model had a better accuracy, with an accuracy of 84.91% and an area under the curve of 0.850 1.【Conclusion】The random forest model exhibits a better predictive ability than that does the logistic regression model. Furthermore, the generated forest fire risk map shows that 11.91% (29.36 km2) of the study area is at a high or extremely a high risk. If utilized, this forest fire risk map can effectively help forest fire protection managers implement appropriate measures to protect forest resources in Jiushan Mountain.
【Objective】The study aims to evaluate the response of acoustic indices to the dynamic changes of animal sound diversity, further to explore the characteristics of the variation of animal sound diversity in Shennongjia National Park, China, in order to provide a quantitative basis for the local ecological protection. 【Method】We deployed nine sound recording equipments in nine sampling sites in Shennongjia National Park, and sound recording data from May to July 2021 were obtained. A time series of eco-acosutic indices including acoustic complexity index (ACI), bioacoustic index (BI), normalized difference soundscape index (NDSI) were extracted from the recording data after noise reduction processing. Further the summer dynamic characteristics of these three acoustic diversity indices were analyzed.【Result】Results showed that the variation of ACI was not obvious during the recording period, while BI showed a double peak variation, NDSI index showed a triple or four peak variation. Results of the Mann-Kendall mutagenicity test showed that ACI only had a mutations at a few sampling sites, while BI and NDSI had mutation at most of the sampling sites. And the analysis of daily variations on the date of the index mutation suggested that weather and human disturbance are the main causes of the index mutation. Results of daily changes of acoustic indices showed that the variation of ACI was not obvious also, while BI and NDSI had a obvious diurnal variation trend, which was consistent with the species of dawn/dusk chorus. The spatial variation of acoustic indexes with altitude gradient indicated that ACI and BI have the maximum value in the middle altitude area, and ACI has a strong correlation with the altitude, while the NDSI has no obvious change.【Conclusion】Our results demonstrated that BI and NDSI could better reflect the dynamic changes of animal sound diversity over time, and the changes of animal sound diversity in Shennongjia National Park showed a multi-peak variation during the recording period, and weather and human activities would cause the abrupt change. The ACI can well explain the spatial variation of animal sound diversity, and the maximum diversity was found in the middle altitude area.
【Objective】Terrestrial laser scanning involves collecting dense laser point clouds of plants to finely characterize the structural parameters of a forest, such as tree skeletons and true leaf area. True leaf area is an important index for phenotypic studies in forestry and botany. At present, there are no well-evidenced methods for measuring phenotypic traits in plant science. Here we develop a novel approach for true leaf area retrieval from terrestrial laser scanned points to appraise key phenotypic parameters. 【Method】 First we designed an individual leaf segmentation algorithm based on the small plane locating and region growing for plant point clouds, to achieve an accurate single-leaf point cloud segmentation. Second, we input three parameters: the angle between the normal vector of a single leaf and the incident laser beam from the scanner, the distance between the scanner and leaf, and the number of point clouds of a single leaf. Training samples combined L1 and L2 regularized multiple regression methods to realize inverse calculations of the total leaf area of all leaf elements in a tree canopy. Finally, we chose four individual trees on our campus(crape myrtle, cherry, ginkgo and camphor) to verify the effectiveness of our results by comparison with field measurements. 【Result】Leaf area retrieval results indicated the superiority of our approach over existing least-square fitting methods. Compared with field measurements, we saw better performance for two small trees: the crape myrtle [coefficient of determination (R2) was 0.95 and root mean square error (RMSE) was 0.42 cm2] and the cherry (R2 = 0.9 and RMSE was 0.42 cm2). Appreciable results were achieved for the ginkgo (R2 = 0.83 and RMSE was 1.24 cm2) and camphor trees (R2 = 0.86; RMSE was 1.10 cm2); these are larger trees with extended crowns and more vegetative elements in the canopy. 【Conclusion】 This method synergistically employed a computer vision and machine learning to accurately calculate the leaf area of canopies using scanned points, yielding novel perspectives for assessing the true leaf area of canopies.
【Objective】 Research into virtual forest collision is hampered by redundant detection objects and simple collision response modes. Our study addresses the technical bottlenecks of high time-cost collision-detection algorithms (aiming to achieve a rapid detection of collisions) and a lack of interaction between the response mode and environmental factors (in favor of real response in virtual forest scenes). 【Method】 We studied a pure-planted Chinese fir (Cunninghamia lanceolata) forest on Shanxia Farm, based at the Experimental Center of Subtropical Forestry in the Chinese Academy of Forestry. Our study compared the efficiency of three methods for the collision detection: an axis-aligned bounding box (AABB) algorithm;a mixed bounding volume hierarchy tree (MBVT) algorithm and a MBVT algorithm optimized using a ‘Finding-the-Four-Closest-Trees’ method, to understand the effect of population size and plant density on collision-detection efficiency and to explore the feasibility of collision response under the strategy proposed in our study.【Result】The total consumption time, t1 for our optimized ‘Finding the Four Closest Trees’ method was approximately 29% that of the MBVT algorithm: 13.75 ms shorter. Compared with the BVH, the cross-test time consumption, t2 was effectively reduced, as was the construction time consumption(t3) and updating time(t4). Both the MBVT and optimized MBVT algorithms reduced the total consumption time, t1 was approximately 73% and 81% that of a single BVH tree AABB by 124.93 and 138.68 ms, respectively. Population was positively correlated with t1, t2 and t3 of the BVH tree: total consumption time, intersection-test time, and construction time. There was a negative correlation between the density of different plants and the total consumption time(t1) and intersection-test time(t2) and no significant correlation with construction time(t3). Conversely, an increase in population and decrease in plant density saw an increase in the total consumption time(t1) and intersection-test time(t2), while there was no significant difference in construction time(t3). Compared with traditional collision response models, the proposed collision response algorithm took into account phototaxis and more realistically simulated a virtual scene of Chinese fir. The virtual scene had a frame rate of 8.6 frames per second and an accuracy of 100%. 【Conclusion】 Improving a MBVT algorithm using our ‘Finding the Four Closest Trees’ can optimize the number of collision detection objects in mixed bounding box hierarchy trees, reducing the consumption time of BVH cross tests and construction to improve collision detection efficiency in a virtual forest of Chinese fir. A collision response strategy for adjacent trees, which accounts for phototaxis, can solve the problem of collision response without interacting with environmental factors in virtual forest scenes, improving the realism of a Chinese fir virtual forest with the Lambert model and collision response function.
【Objective】We used the enhanced Frost local filtering and single-tree distance map reconstruction marking technology to segment a canopy height model (CHM), improving the accuracy and efficiency of unpiloted aerial system (UAV)-light detection and ranging (LiDAR) segmentation in single-tree crowns.【Method】 We selected three forest types - coniferous mixed, coniferous-broad mixed, and broad-leaved mixed - in the Shanxia Experimental Forest Farm of Fenyi, Jiangxi Province. We then used UAV-LiDAR data to construct the CHM. To combat the increased pores in the crown area of the high-resolution CHM, we used the enhanced Frost local filtering to optimize the CHM and results were compared with different filtering methods. Next we applied the distance map reconstruction marker segmentation technology to segment and analyze the CHM-with resolutions of 0.1, 0.2, 0.5 and 1.0 m after optimization of the enhanced Froest local filter. Finally, we determined the CHM with the optimal resolution, and compared segmentation results with that of a watershed algorithm with the same resolution and mean-shift segmentation algorithm. 【Result】 Applying the enhanced Frost local filter indeed optimized the CHM-preserving image details while suppressing phase crown noise. A resolution of 0.2 m performed best for the CHM segmentation. An overall accuracy of 0.96, 0.84 and 0.75 was observed for coniferous mixed, coniferous-broad mixed, and broad-leaved mixed forests, respectively. The crown width of a single tree was calculated according to crown segmentation results, and the R2 estimated at 0.83, 0.82 and 0.71, respectively. 【Conclusion】Through the enhanced Frost local filtering and distance map reconstruction marking technology, the single-tree segmentation and crown estimation of laser point cloud CHMs can be realized, meeting key requirements of forest surveys and monitoring.
