ABSTRACT The rehabilitation of degraded subtropical natural forests is a global concern. A detailed assessment of their structure is a challenging and costly prerequisite because diverse structures exist depending on the cause and degree of degradation. Recent remote sensing concepts and technologies provide a detailed picture of actual forest structure, even in difficult terrain. When it comes to planning and implementing rehabilitation measures on the ground, however, meaningful forest management units (FMUs) must be created that are large enough to allow technical implementation, but which are also homogenous in structure. To date, the delineation of FMUs has, in most cases, been achieved qualitatively based on expert knowledge. The aim of this contribution is to develop and demonstrate a method for creating and delineating meaningful FMUs based on quantitative information acquired from remote sensing and spatial statistics. Therefore, a case study was conducted in a 3940-ha fire-degraded forest area in the Argentinean cloud forest of Yungas Pedemontana. A plot-based field inventory and an aerial survey with an unmanned aerial vehicle were conducted. The Adjusted Canopy Coverage Index (ACCI), as a metric for stand structure, was formulated to predict basal area from canopy height models. A SPOT6 image of the area was object-based segmented and classified into four fire-severity strata by training it with the ACCI values. The resulting classification presented a mosaic pattern in which the stands are homogenous but far too small (average 3129 m2) for planning adaptive management. Therefore, features in close proximity with similar structure (i.e. ACCI values) were aggregated using the Hot Spot Analysis (Getis-Ord Gi*) tool from the Arc geographic information system environment to create FMUs. Clusters were calculated at four scales: 10, 20, 30 and 40 ha (resulting in threshold radii of 178, 252, 309 and 357 m, respectively), using ACCI values as the variable of aggregation. As a result, average cluster areas were obtained of 33.9 ha for the shortest threshold distance of analysis and 138.5 ha for the greatest threshold distance. The tool significantly aggregated between 30.7% and 60.8% of the area into either coldspots or hotspots of ACCI, facilitating the delineation of FMUs for the planning of adaptive rehabilitation measures. There is a trade-off, however, between the gain in area of the FMUs and the loss of homogeneity: for a 357 m distance threshold, 12% more of the area was misclassified, compared with a 178 m threshold.

中文翻译：

---

使用热点分析 (Getis-Ord Gi*) 在受混合严重程度火灾影响的森林上创建森林管理单元

摘要 退化的亚热带天然林的恢复是全球关注的问题。对其结构的详细评估是一项具有挑战性且成本高昂的先决条件，因为根据退化的原因和程度存在不同的结构。最近的遥感概念和技术提供了实际森林结构的详细图片，即使是在困难的地形中。然而，当涉及到实地规划和实施恢复措施时，必须创建有意义的森林管理单位 (FMU)，其规模足够大以允许技术实施，但在结构上也是同质的。迄今为止，在大多数情况下，FMU 的描述是基于专家知识定性地实现的。这项贡献的目的是开发和演示一种基于从遥感和空间统计中获得的定量信息来创建和描绘有意义的 FMU 的方法。因此，在阿根廷云加斯佩德蒙塔纳云林的 3940 公顷火灾退化林区进行了案例研究。进行了基于地块的实地清查和无人机航测。调整后的冠层覆盖指数 (ACCI) 作为林分结构的度量标准，用于根据冠层高度模型预测基底面积。通过使用 ACCI 值对其进行训练，该区域的 SPOT6 图像被基于对象分割并分为四个火灾严重程度层。由此产生的分类呈现出一种镶嵌模式，其中林分是同质的，但对于规划适应性管理来说太小了（平均 3129 平方米）。因此，使用来自 Arc 地理信息系统环境的热点分析 (Getis-Ord Gi*) 工具聚合具有相似结构（即 ACCI 值）的邻近特征以创建 FMU。在四个尺度上计算集群：10、20、30 和 40 公顷（导致阈值半径分别为 178、252、309 和 357 m），使用 ACCI 值作为聚合变量。结果，分析的最短阈值距离为 33.9 公顷，最大阈值距离为 138.5 公顷。该工具将 30.7% 到 60.8% 的区域显着聚合到 ACCI 的冷点或热点中，促进划定 FMUs 以规划适应性康复措施。然而，在 FMU 面积增加和同质性损失之间存在权衡：对于 357 m 的距离阈值，与 178 m 的阈值相比，错误分类的区域多出 12%。