基于地学信息图谱的重庆岩溶石漠化植被恢复演替研究

重庆岩溶石漠化区的植被恢复演替动态变化研究对于该地区的石漠化治理和生态恢复具有十分重要的指导意义。而多时相的遥感数据和地学图谱分析法为植被恢复的研究提供了一种动态性和综合性的研究方法。以重庆市中梁山的典型植被恢复区为例,在基于1996、2001、2007和2013年4期遥感影像解译分类的基础上,采用"空间代替时间"的生态学植被演替研究方法,建立重庆市中梁山区退耕还林前后的植被恢复演替图谱,并结合地学图谱的相关分析方法,得到该区的植被恢复演替动态格局演变规律,体现了空间信息科学技术、生态学方法和地学信息图谱分析法在植被恢复演替研究中的有效结合。结果表明:(1)运用BP神经网络和BP算法进行分类,分类精度达到87.42%,比传统监督分类提高了5.57%。(2)自2002年全国范围内的"退耕还林(草)"工程全面启动后,该区域植被恢复演变特征明显,耕地面积明显减少而植被面积明显增加。(3)从2001—2013年,植被演替在该时期内依然存在着进展演替和逆向演替两个方向。虽然逆向演替比例仅占到18.63%,但它却使该区的演替研究变得复杂。(4)质心反映了各植被类型在恢复演替过程中的聚散与迁移,1996—2013年,马尾松群落和落叶阔叶林群落的质心变化较小,其他植被群落的质心都有很明显的变化。

Quantifying vegetation restoration in a karst rocky desertification area in Chongqing based on Geo-informatic Tupu

Karst rocky desertification (KRD) has become one of the most important ecological and environmental problems in China, and the control of rocky desertification has been listed as a goal of both social development and national environmental managment projects. However, patterns of plant succession in the process of KRD reversal activities are still unclear. Understanding plant dynamics is important for both the theory and practice for successful ecosystem restoration. We used multi-temporal, remotely sensed images and a Geo-informatic Tupu method to investigate the succession patterns of vegetation restoration at Zhongliang Mountain in Chongqing, Southwest China. This region is a typical KRD vegetation restoration area, with a rich diversity of regional vegetation types. In this study, remotely sensed images for four different time periods (1996, 2001, 2007, and 2013), representing four different stages in vegetation succession, were selected and analyzed using back-propagation (BP) neural network models for interpretation and classification. This resulted in maps of vegetation restoration based spatial structure rather than time series images, captured before and after the Grain for Green project, and thus, established information about principles for vegetation restoration succession in the region. Thereafter, the maps were analyzed using Geo-informatic Tupu to identify the dynamic patterns of vegetation restoration succession in the region. Our results indicate the following. (1) The BP neural network model provides an efficient vegetation classification method in the Zhongliang Mountain region. The overall accuracy of the (BP) neural network classification was 87.42%, which was 5.57% higher than traditional supervised methods. (2) Since 2002, a series of ecological restoration projects, including the Grain for Green project (the conversion of cropland into forest or pasture), have been implemented in this region, leading to a reduction in the area of farmland and an increase in the area of natural vegetation. The positive trends observed in the study site are interpreted as being the result of human-induced restoration. Comparing vegetation change in the different sub-regions of the study site, the most significant vegetation changes occurred on farmland that was located in the valley and foothills of Zhongliang Mountain. In contrast, regions with moderate change included the acid and the alkaline soil areas at higher elevations of Zhongliang Mountain. Here, the Masson pine community and kashiwagi community, respectively, are dominant. These communities have effective energy and nutrient conversion capabilities and form stable ecosystems that moderate changes in the natural succession of vegetation. (3) The Zhongliang Mountain communities were classified into two different stages, forward succession and reverse succession, during the period 2001-2013. In 18.63% of the region, reverse succession occurred, resulting in complex patterns of vegetation change. (4) Analysis of changes in vegetation structure was performed based on the centroid shifting method. From 1996 to 2013, the centroid of the Masson pine community and deciduous broad-leaved forest communities changed only slightly, while changes observed in other vegetation types were more marked. The Tupu method quantified the spatial pattern, distribution, and change processes of vegetation in different successional stages. It appears that the Tupu method can identify the state of each image element in each sampling time and achieve an integration of Space-Property-Process, providing a strong scientific basis and technical means for vegetation restoration in karst rocky desertification regions.