基于MaxEnt生态位模型预测木灵藓科三属植物在新疆的潜在分布区

木灵藓科植物是干旱、半干旱环境中常见的藓类植物，对温度和降水极为敏感.木灵藓科在中国的物种丰富度由东向西增加，在新疆的丰富度较高.木灵藓属(Orthotrichum)、显孔藓属(Lewinskya)、多胞藓属(Nyholmiella)是三个在新疆分布较为广泛的属.预测当前以及未来气候情景下这三个属在新疆的潜在分布范围，将为木灵藓科区系和多样性研究提供一定的参考依据，也能预判未来气候变化对新疆地区木灵藓属的分布影响.该文根据木灵藓属、显孔藓属、多胞藓属在新疆的125个地理采集样点和19个气候因子，运用MaxEnt模型预测了它们适生区，并分析这些气候因素对其分布范围的影响.结果显示：1) AUC值高达0.945，说明该模型能够很好地预测了三个属的适生区范围.Jackknife检验表明，影响三个属分布的主要气候因子为最湿季平均气温以及最干季降雨量，贡献率分别为24.5%、23.9%.2) 该三属主要分布在阿勒泰、塔城、中部天山山脉、哈密、喀什和克州等地区，在和田和巴州也有一定的分布.在21世纪50年代的气候情景下，这些属在阿尔泰和塔城的分布范围将扩大，而在喀什和巴州的则缩小.总体而言，在21世纪70年代的气候情景下，这三个属在新疆的潜在分布区面积比现在的有所缩小.该文以气候因子预测了木灵藓科在新疆当前和未来的潜在分布范围，今后的工作中，将考虑影响苔藓植物生长的地形、植被的类型等，以进一步提高预测结果的准确性.

Potential distribution prediction of three genus of Orthotrichaceae in Xinjiang based on MaxEnt niche model

The normalized digital surface model is an important feature to characterize the height of ground objects and assist in the classification of remote sensing images， but its flaky features and unstable precision restrict the improvement of classification accuracy. Aiming at this problem， this paper proposes a dual-path input semantic segmentation network considering the local normalized height. On the one hand， a dual-path input structure is designed to extract the spectral features and geometric features of the ground objects， and connect them through skip connections to fully learn the multi-modal information of ground objects. On the other hand， a new method of ground object height representation is proposed. Considering that deep neural network can only process images in a small area due to the limitation of GPU memory， the height features are calculated within the local area of the digital surface model. Finally， by comparing the three network frameworks on the ISPRS standard data set， it is demonstrated that the overall accuracy of the proposed method is improved by 4.5%～4.7% compared to the method using only spectral images， and the classification accuracy， computational efficiency and degree of automation are better than method with normalized digital surface model.