Water retention is important in forest ecosystem services. The heterogeneity analysis of water-retention capacity and its influencing factors is of great significance for the construction of water-retention functional...Water retention is important in forest ecosystem services. The heterogeneity analysis of water-retention capacity and its influencing factors is of great significance for the construction of water-retention functional areas, restoration of vegetation, and the protection of forest ecosystems in the Beijing-Tianjin-Hebei region. A total of 1366 records concerning water-retention capacity in the canopy layer, litter layer, and soil layer of forest ecosystem in this region were obtained from 193 literature published from 1980 to 2017. The influencing factors of water-retention capacity in each layer were analyzed, and path analysis was used to investigate the contribution of the factors to the water-retention capacity of the three layers. The results showed that mixed forests had the highest water-retention capacity, followed by broad-leaved forests, coniferous forests, and shrub forests. In addition, no matter the forest type, the ranking of the water-retention capacity was soil layer, canopy layer, and litter layer from high to low. The main influencing factors of water-retention capacity in forest canopy were leaf area index and maximum daily precipitation(R2=0.49), and the influencing coefficients were 0.34 and 0.30, respectively. The main influencing factors of water-retention capacity in the litter layer were semi-decomposed litter(R2=0.51), and the influencing coefficient was 0.51. The main influencing factors of water-retention capacity in the soil layer were non-capillary porosity and soil depth(R2=0.61), the influencing coefficients were 0.60 and 0.38, respectively. This study verifies the simulation of the water balance model or inversion of remote sensing of the water-retention capacity at the site scale, and provides scientific basis for further study of the impact of global change on water retention.展开更多
异质图是由不同类型节点及边构成的图,可建模现实世界中各种类型对象及其关系。异质图嵌入旨在捕捉图中丰富的属性、结构和语义等信息,学习节点嵌入向量,用于节点分类、链接预测等任务,进而实现用户识别、商品推荐等应用。在异质图嵌入...异质图是由不同类型节点及边构成的图,可建模现实世界中各种类型对象及其关系。异质图嵌入旨在捕捉图中丰富的属性、结构和语义等信息,学习节点嵌入向量,用于节点分类、链接预测等任务,进而实现用户识别、商品推荐等应用。在异质图嵌入方法中,元路径通常被用来获取节点间的高阶结构和语义信息,然而现有方法忽略了元路径实例中不同类型节点或异质图中不同类型邻居节点的差异,导致信息丢失,进而影响节点嵌入质量。针对上述问题,提出基于数据增强的异质图注意力网络(Heterogeneous graph Attention Network based on Data Augmentation,HANDA),以更好地学习节点嵌入向量。首先,提出基于元路径邻居的边增强。该方法基于元路径获取节点的元路径邻居,用节点及其元路径邻居形成的语义边增强异质图。这些增强边不仅蕴含了节点间的高阶结构和语义,还缓解了异质图的稀疏性。其次,提出融入节点类型注意力的节点嵌入。该方法采用多头注意力从多个角度学习不同直接边邻居及增强边邻居的重要性并在注意力中融入节点的类型信息,进而通过消息传递、直接边邻居及增强边邻居同时获取节点的属性、高阶结构和语义信息,提升了节点嵌入质量。在真实数据集上的实验验证了HANDA模型在节点分类、链接预测任务上的效果优于基准模型。展开更多
基金National Key R&D Program of China,No.2017YFA0604703National Natural Science Foundation of China,No.41771111+4 种基金Hebei Natural Science Foundation,No.D2019205123Youth Innovation Promotion Association,No.2018071Research Fund Project of Hebei Normal University,No.L052018Z09Key Subject of Physical Geography of Hebei ProvinceInvestigation and Monitoring Project of Ministry of Natural Resources,No.JCQQ191504-06。
文摘Water retention is important in forest ecosystem services. The heterogeneity analysis of water-retention capacity and its influencing factors is of great significance for the construction of water-retention functional areas, restoration of vegetation, and the protection of forest ecosystems in the Beijing-Tianjin-Hebei region. A total of 1366 records concerning water-retention capacity in the canopy layer, litter layer, and soil layer of forest ecosystem in this region were obtained from 193 literature published from 1980 to 2017. The influencing factors of water-retention capacity in each layer were analyzed, and path analysis was used to investigate the contribution of the factors to the water-retention capacity of the three layers. The results showed that mixed forests had the highest water-retention capacity, followed by broad-leaved forests, coniferous forests, and shrub forests. In addition, no matter the forest type, the ranking of the water-retention capacity was soil layer, canopy layer, and litter layer from high to low. The main influencing factors of water-retention capacity in forest canopy were leaf area index and maximum daily precipitation(R2=0.49), and the influencing coefficients were 0.34 and 0.30, respectively. The main influencing factors of water-retention capacity in the litter layer were semi-decomposed litter(R2=0.51), and the influencing coefficient was 0.51. The main influencing factors of water-retention capacity in the soil layer were non-capillary porosity and soil depth(R2=0.61), the influencing coefficients were 0.60 and 0.38, respectively. This study verifies the simulation of the water balance model or inversion of remote sensing of the water-retention capacity at the site scale, and provides scientific basis for further study of the impact of global change on water retention.
文摘异质图是由不同类型节点及边构成的图,可建模现实世界中各种类型对象及其关系。异质图嵌入旨在捕捉图中丰富的属性、结构和语义等信息,学习节点嵌入向量,用于节点分类、链接预测等任务,进而实现用户识别、商品推荐等应用。在异质图嵌入方法中,元路径通常被用来获取节点间的高阶结构和语义信息,然而现有方法忽略了元路径实例中不同类型节点或异质图中不同类型邻居节点的差异,导致信息丢失,进而影响节点嵌入质量。针对上述问题,提出基于数据增强的异质图注意力网络(Heterogeneous graph Attention Network based on Data Augmentation,HANDA),以更好地学习节点嵌入向量。首先,提出基于元路径邻居的边增强。该方法基于元路径获取节点的元路径邻居,用节点及其元路径邻居形成的语义边增强异质图。这些增强边不仅蕴含了节点间的高阶结构和语义,还缓解了异质图的稀疏性。其次,提出融入节点类型注意力的节点嵌入。该方法采用多头注意力从多个角度学习不同直接边邻居及增强边邻居的重要性并在注意力中融入节点的类型信息,进而通过消息传递、直接边邻居及增强边邻居同时获取节点的属性、高阶结构和语义信息,提升了节点嵌入质量。在真实数据集上的实验验证了HANDA模型在节点分类、链接预测任务上的效果优于基准模型。
