Water and Land Resource Centre, P.O. Box 3880, Addis Ababa, Ethiopia, and private P.O.Box, 7985, Addis Ababa, Ethiopia.
Department of Geography & Environmental Studies, Addis Ababa University, P.O.Box, 1176, AAU, Ethiopia.
Sci Rep. 2019 Feb 7;9(1):1576. doi: 10.1038/s41598-018-36587-7.
The development of spatially differentiated management strategies against invasive alien plant species requires a detailed understanding of their current distribution and of the level of invasion across the invaded range. The objectives of this study were to estimate the current fractional cover gradient of invasive trees of the genus Prosopis in the Afar Region, Ethiopia, and to identify drivers of its invasion. We used seventeen explanatory variables describing Landsat 8 image reflectance, topography, climate and landscape structures to model the current cover of Prosopis across the invaded range using the random forest (RF) algorithm. Validation of the RF algorithm confirmed high model performance with an accuracy of 92% and a Kappa-coefficient of 0.8. We found that, within 35 years after its introduction, Prosopis has invaded approximately 1.17 million ha at different cover levels in the Afar Region (12.3% of the surface). Normalized difference vegetation index (NDVI) and elevation showed the highest explanatory power among the 17 variables, in terms of both the invader's overall distribution as well as areas with high cover. Villages and linear landscape structures (rivers and roads) were found to be more important drivers of future Prosopis invasion than environmental variables, such as climate and topography, suggesting that Prosopis is likely to continue spreading and increasing in abundance in the case study area if left uncontrolled. We discuss how information on the fractional cover and the drivers of invasion can help in developing spatially-explicit management recommendations against a target invasive plant species.
制定针对入侵外来植物物种的空间差异化管理策略需要详细了解它们当前的分布范围和入侵范围的入侵程度。本研究的目的是估计埃塞俄比亚阿法尔地区入侵树种金合欢属的当前分数覆盖梯度，并确定其入侵的驱动因素。我们使用了 17 个解释变量来描述 Landsat 8 图像反射率、地形、气候和景观结构，使用随机森林（RF）算法来模拟入侵范围内金合欢属的当前覆盖情况。RF 算法的验证证实了模型具有很高的性能，准确率为 92%，Kappa 系数为 0.8。我们发现，在引入后的 35 年内，金合欢属已在阿法尔地区不同覆盖水平的约 117 万公顷范围内入侵（占该地区面积的 12.3%）。归一化植被指数（NDVI）和海拔高度在 17 个变量中表现出最高的解释能力，无论是在入侵者的整体分布还是在高覆盖区域方面都是如此。与气候和地形等环境变量相比，村庄和线性景观结构（河流和道路）被发现是金合欢属未来入侵的更重要驱动因素，这表明如果不加控制，金合欢属很可能继续在研究区扩散并增加丰度。我们讨论了关于分数覆盖和入侵驱动因素的信息如何有助于制定针对目标入侵植物物种的空间明确管理建议。
