基于深度学习的内蒙古大兴安岭林区火灾预测建模研究

[目的]对内蒙古大兴安岭地区的森林火灾进行预测,为森林防火工作的开展提供重要支持。[方法]以内蒙古大兴安岭林区为研究对象,结合MCD64 A1月度火点产品、地形、气候等数据,构建森林火灾潜在影响因子数据集,分别利用卷积神经网络、随机森林、支持向量机模型对研究区森林火灾的发生概率进行预测与可视化,在此基础上对模型效果进行评价并分析森林火灾空间分布特征。[结果]大兴安岭的主要林火驱动因子按重要性值由高到低排序为海拔、平均气温、总降水量、与水域的距离等;CNN、RF、SVM预测森林火灾发生概率的AUC值分别为0.838、0.794、0.788,CNN的精度最高;CNN能够有效划分出森林火灾易感性极高、极低的区域,有利于划分森林火灾的警示区。[结论]CNN模型比RF、SVM模型更适用于大兴安岭林火发生概率的预测;大兴安岭林火风险的空间分布有明显的区域性,主要发生在东南地区。

Long Term Evaluation of Green Vegetation Cover Dynamic in the Atacora Mountain Chain(Togo) and its Relation to Carbon Sequestration in West Africa

The research was done in the Atacora Mountain chain in Togo which tended to assess the change of vegetation cover during a 24-year period.It also aims to evaluate the dynamic of the net primary productivity(NPP) of the living plants over the same period.The Landsat imagery covering three different periods(1987, 2000, and 2011) was pre-processed to correct atmospheric and radiometric parameters as well as gapfilling the 2011 SCL-off images.Then, the vegetation indices such as NDVI(normalized difference vegetation index), SR(simple ratiovegetation index), SAVI(soil-adjusted vegetation index), and CASA(carnegie- ames- stanford approach)model for NPP were applied on these images after masking the study area.The results showed a quiet decrease in the vegetation cover.The vegetation loss was more significant from 2000 to 2011 than from1987 to 2000, and anthropogenic activities can be deemed as the main cause of the vegetation loss.The biomass assessment by NPP computation also showed a decrease over the time.Similar to the change of the vegetation cover, the ecosystem net productivity was very low in 2011 compared to 2000 and 1987.It seems that the general health condition of thevegetation, including its potentiality in carbon sinking,was negatively affected in this area, which has already been under threatened.A perpetual monitoring of these ecosystems by means of efficient techniques could enhance the sustainable management tools of in the framework of reducing emissions from deforestation and forest degradation(REDD).

利用机载激光雷达技术估测东北林区典型针叶林的蓄积量

为了推广激光雷达技术在森林蓄积量估测计量方面的应用,本研究以东北林区云冷杉林、落叶松林、红松林和樟子松林4种典型针叶林为对象,基于机载激光雷达获取的点云数据提取特征变量,结合800块地面样地数据,采用逐步回归方法和偏最小二乘方法,建立4种针叶林的蓄积量模型。结果表明:偏最小二乘法建立的模型精度优于逐步回归方法(ΔR^(2)=0.05~0.15,ΔRRMSE=2.6%~4.2%);在参与建模的3类点云特征变量中,贡献最大的是点云高度变量(被选择26次),其他变量有一定的辅助作用(分别被选择12次和11次);使用偏最小二乘方法建立的林分蓄积量模型中,红松林(R^(2)=0.79,RMSE=60.92,RRMSE=22.9%)和落叶松林(R^(2)=0.76,RMSE=28.39,RRMSE=25.8%)的精度最高,云冷杉林(R^(2)=0.81,RMSE=46.96,RRMSE=27.7%)次之,樟子松林(R^(2)=0.50,RMSE=55.49,RRMSE=30.4%)的精度稍低。研究结果为东北林区4种典型针叶林蓄积量估测提供了一种有效的方法。