Spatial analysis data for 'Lines in the sand: quantifying the cumulative development footprint in the world's largest remaining temperate woodland'

These datasets provide the data underlying the publication (see abstract below).
The datasets are:

data in csv format:
1. development footprint by sample area:

Information on the 24 ~490 km2 sample areas assessed in the study, including the different infrastructure types (roads, railways, mapped tracks, unmapped tracks which have been manually digitised in the study using aerial imagery and hub infrastructure such as mine pits and waste rock dumps, also manually digitised in the study). Also contains some key covariables assessed as potential explanatory variables for development footprint.

The region-wide modelling of development footprint found strong positive effects of mining project density and pastoralism, as well as a highly significant negative interaction between the two. At low mining project densities, development footprints are more extensive in pastoral areas, but at high mining project densities, pastoral areas are relatively less developed than non-pastoral areas, on average.

2. gww 20 km grid:

This dataset provides data for the 20x20 km grid placed over the whole Great Western Woodlands and used for the regional estimation of development footprint, linear infrastructure density, and linear infrastructure type based on the region-wide analysis. Data is for each cell in the grid and provides the total length of roads in that grid cell, MINEDEX mining projects, pastoral status, etc.

This dataset was used to project the data from the 24 study areas across the whole of the Great Western Woodlands and calculate region-wide estimates of development footprint and linear infrastructure lengths.

3. disturbance by patch:

This dataset provides the data for each patch for the analysis of patch-level drivers of development footprint, which was performed to gain further insights into the effects of other landscape variables that what could be gleaned from the region-wide analysis. For this analysis, we divided sample areas into polygonal ‘patch types’, each with a unique combination of the following categorical covariables: pastoral tenure, greenstone lithology, conservation tenure, ironstone formation, schedule 1 area clearing restrictions, environmentally sensitive area designation, vegetation formation, and sample area.

For each patch type (n=261), we calculated the following attributes: number of mining projects, number of dead mineral tenements, sum of duration of all live and dead tenements, type of tenements (exploration/prospecting tenement, mining and related activities tenement, none), primary target commodity (gold, nickel, iron-ore, other), distance to wheatbelt, and distance to nearest town.

4. mapped versus digitised tracks:

This dataset provides mapped and unmapped track widths, measured using high-resolution aerial imagery at at least 20 randomly-generated locations within each of 24 sample areas. Pastoral tenure and mining intensity for each sample area are included for analysis purposes.

This data was analysed as follows: we used a t-test to test for a difference between mapped and unmapped track width, conducted data exploration as per (Zuur et al. 2009), and modelled track widths using linear mixed models with ‘lme4’ package in R. We created a global model containing the following fixed variables: mapped/unmapped status; mining activity level for the relevant sample area, and pastoral status. Sample area identity was included as the random effect in all models after testing for its significance.

We used the &#145;dredge&#146; function in &#145;MuMin&#146; package to model all possible subsets of the global model and rank them based on AICc values. The optimal model included only mapped/unmapped status as a fixed effect, and the other top-ranking model also included a positive effect of pastoral tenure on track width. Mapped tracks were found to be on average ~1 m wider than unmapped tracks (p < 0.001) (Figure A2.1). Average widths of mapped and unmapped tracks were 6.06 m (s.e. 0.15 m) and 4.92 m (s.e. 0.10 m) respectively. No effect of mining activity was included in the top-ranking models.

5. edge effect scenarios:

Hypothetical edge effect zones were created, based on effect zones gleaned from the literature and arranged under three scenarios, to reflect potential risks of offsite impacts in areas adjacent to development footprints observed (see appendix 3 of article).

The calculated proportion of the entire GWW within edge effect zones varied from ~3% under the conservative scenario to ~35% under the maximal scenario. Within the range of development footprints observed in this study, the proportion of a landscape that lies within edge effect zones increases hyperbolically with the number of mining projects, and approaches 100% in the maximal scenario, 60% in the moderate scenario, and ~20% under the conservative scenario.

shapefiles:

6. Great Western Woodlands boundary

7. sample areas (layer file shows sample areas by category). We used stratified random sampling to distribute 24 circular sample areas, each 25 km in diameter, among the 8 mining and pastoral categories. We used circular sample areas to minimise the edge-to-area ratio of the sample areas and therefore maximise the extent to which the sample areas reflected the category represented rather than the adjacent landscape.

8. linear infrastructure extending beyond gww boundary by ~100 km. This is a dataset compiled from 23 different sources that represents the most comprehensive spatial dataset for the GWW available at the time of publication, to KR&#146;s knowledge. However, it does contain a number of different sources of error and should not be considered to necessarily reflect an updates, accurate dataset (note there is a more detailed metadata document inside this folder).

9. linear infrastructure footprints. Linear features buffered by average width of that linear infrastructure type for each sample area. Linear features include paved roads and railways, unpaved roads, mapped tracks, and unmapped tracks (digitized from aerial images in this study).

10. digitised tracks All linear infrastructure that hadn&#146;t already been mapped in #8 above. Manually digitised from high-resolution aerial images in this study.

11. digitised hub infrastructure Development footprints of all non-linear (i.e. polygonal) anthropogenic disturbance, including mine pits, waste rock dumps, mining camps and accommodation villages, dams, and other cleared areas, manually digitised from high-resolution aerial imagery in this study.

12. edge effect zones Polygons created by creating buffers around the development footprint as described in Appendix 3 of the article. These zones around the direct development footprint represent offsite impact risk for each type of infrastructure, using a hypothesized set of risk buffers. These were based on edge effect distances reported in the literature for species and processes from around the world. Three scenarios are represented: a conservative, moderate, and maximal scenario.

The abstract for the publication is as follows:

Context
The acceleration of infrastructure development presents many challenges for the mitigation of ecological impacts. The type, extent, and cumulative effects of multiple developments must be quantified to enable mitigation.

Objectives
We quantified anthropogenic development footprints in a globally significant and relatively intact region. We identified the proportion accounted for by linear infrastructure (e.g. roads) including infrastructure that is currently unmapped; investigated the importance of key landscape drivers; and explored potential ramifications of offsite impacts (edge effects).

Methods
We quantified direct development footprints of linear and 'hub' infrastructure in the Great Western Woodlands (GWW) in south-western Australia, using digitisation and extrapolation from a stratified random sample of aerial imagery. We used spatial datasets and literature resources to identify predictors of development footprint extent and calculate hypothetical &#145;edge effect zones&#146;.

Results
Unmapped linear infrastructure, only detectable through manual digitisation, accounts for the greatest proportion of the direct development footprint. Across the 160,000 km2 GWW, the estimated development footprint is 690 km2, of which 67% consists of linear infrastructure and the remainder is &#145;hub&#146; infrastructure. An estimated 150,000 km of linear infrastructure exists in the study area, equating to an average of ~1 km per km2. Beyond the direct footprint, a further 4,000&#150;55,000 km2 (3&#150;35% of the region) lies within edge effect zones.

Conclusions
This study highlights the pervasiveness of linear infrastructure and hence the importance of managing its cumulative impacts as a key component of landscape conservation. Our methodology can be applied to other relatively intact landscapes worldwide.

本数据集支撑下述论文的相关数据（论文摘要见下文）。本数据集包含以下内容：

### CSV格式数据集
1. 样区开发足迹数据
本数据集涵盖本研究中评估的24个面积约490 km²的样区相关信息，包含各类基础设施类型（道路、铁路、已制图轨道、本研究通过航空影像手动数字化的未制图轨道，以及矿坑、废石堆等枢纽基础设施（hub infrastructure），后者亦由本研究手动数字化完成）。同时包含部分被评估为开发足迹潜在解释变量的关键协变量。

针对开发足迹的全域建模结果显示，矿业项目密度与放牧活动均对开发足迹存在显著正向影响，且二者间存在高度显著的负向交互效应。当矿业项目密度较低时，放牧区域的开发足迹范围更广；而当矿业项目密度较高时，放牧区域的平均开发程度相对低于非放牧区域。

2. 大西部林地（Great Western Woodlands, GWW）20 km网格数据
本数据集覆盖整个大西部林地（Great Western Woodlands, GWW）的20×20 km网格数据，用于基于全域分析开展开发足迹、线性基础设施（linear infrastructure）密度及线性基础设施类型的区域估算。数据涵盖网格内每个单元格的道路总长度、MINDEX矿业项目信息、放牧状态等属性。

本数据集用于将24个研究样区的数据外推至整个大西部林地，并计算开发足迹与线性基础设施长度的全域估算值。

3. 斑块扰动数据
本数据集包含用于开展开发足迹斑块级驱动因子分析的各斑块数据，旨在进一步挖掘全域分析中无法获取的其他景观变量的影响效应。本研究将样区划分为多边形‘斑块类型’，每类斑块由以下分类协变量的唯一组合构成：放牧权属、绿岩岩性、保护地权属、铁质岩组、Schedule 1区域清理限制区、环境敏感区标识、植被群系及样区。

针对每类斑块类型（共261类），本研究计算了以下属性：矿业项目数量、失效矿业租约数量、有效与失效租约的总存续时长、租约类型（勘探/探矿租约、采矿及相关活动租约、无租约）、主要目标矿种（金、镍、铁矿石、其他）、至小麦带的距离以及至最近城镇的距离。

4. 已制图轨道与数字化轨道对比数据
本数据集包含24个样区内各至少20个随机采样点的已制图与未制图轨道宽度数据，数据通过高分辨率航空影像测量获取。同时包含每个样区的放牧权属与矿业强度数据，用于后续分析。

本数据集的分析流程如下：采用t检验对比已制图与未制图轨道的宽度差异，参考Zuur et al. 2009的方法开展数据探索，并通过R语言的`lme4`包构建线性混合模型以拟合轨道宽度。本研究构建的全局模型包含以下固定效应变量：轨道是否已制图状态、对应样区的矿业活动水平以及放牧状态。在检验显著性后，将样区标识作为所有模型的随机效应。

使用`MuMin`包的`dredge`函数对全局模型的所有可能子集进行建模，并基于AICc值对模型进行排序。最优模型仅包含轨道是否已制图状态作为固定效应；次优模型则额外包含放牧权属对轨道宽度的正向影响。结果显示，已制图轨道的平均宽度较未制图轨道高出约1 m（p < 0.001）（见图A2.1）。已制图与未制图轨道的平均宽度分别为6.06 m（标准误0.15 m）与4.92 m（标准误0.10 m）。顶级排序模型中未体现矿业活动的影响效应。

5. 边缘效应情景数据
本数据集基于文献报道的效应区构建了假想边缘效应区（edge effect zones），并设置三种情景，用于反映观测到的开发足迹周边区域的潜在场外影响风险（详见论文附录3）。

经计算，整个大西部林地处于边缘效应区内的面积占比从保守情景下的约3%变化至极端情景下的约35%。在本研究观测到的开发足迹范围内，景观中处于边缘效应区的比例随矿业项目数量呈双曲线增长；在极端情景下该比例接近100%，中度情景下约为60%，保守情景下约为20%。

### Shapefile格式矢量数据集
6. 大西部林地边界矢量文件
7. 样区矢量文件：本研究采用分层随机抽样方法，在8个矿业与放牧类别中布设24个直径为25 km的圆形样区。选用圆形样区旨在最小化样区的边缘面积比，从而最大化样区对所属类别的代表性，减少邻接景观的干扰。
8. 延伸至大西部林地边界外约100 km的线性基础设施数据集：本数据集整合自23个不同来源，据作者KR所知，为本论文发表时可获取的最全面的大西部林地空间数据集。但该数据集存在多种误差来源，不应被视为更新后的精准数据集（注：本文件夹内包含更详细的元数据文档）。
9. 线性基础设施足迹矢量文件：将线性要素按照对应样区中该类线性基础设施的平均宽度进行缓冲区分析得到的数据集。线性要素包括铺装道路与铁路、非铺装道路、已制图轨道以及本研究通过航空影像数字化的未制图轨道。
10. 数字化轨道矢量文件：本研究通过高分辨率航空影像手动数字化的、未包含在第8项数据中的线性基础设施。
11. 数字化枢纽基础设施矢量文件：本研究通过高分辨率航空影像手动数字化的所有非线形（即多边形）人为扰动开发足迹，包括矿坑、废石堆、采矿营地与住宿村落、水坝及其他清理区域。
12. 边缘效应区矢量文件：基于论文附录3所述方法，围绕开发足迹构建缓冲区得到的多边形数据集。这些围绕直接开发足迹的区域代表各类基础设施的场外影响风险，采用基于全球范围内物种与生态过程的文献报道的边缘效应距离构建的假想风险缓冲区。本数据集包含三种情景：保守情景、中度情景与极端情景。

---

### 论文摘要
#### 研究背景
基础设施开发的加速为生态影响减缓带来诸多挑战。需量化多开发项目的类型、范围与累积效应，方能开展有效的减缓措施。

#### 研究目标
本研究对全球重要且相对完整的区域开展人为开发足迹量化工作，识别其中线性基础设施（linear infrastructure，如道路）的占比，包括当前未制图的基础设施；探究关键景观驱动因子的重要性；并探讨场外影响（边缘效应）的潜在后果。

#### 研究方法
本研究通过对航空影像开展分层随机采样并进行数字化与外推，量化澳大利亚西南部大西部林地（Great Western Woodlands, GWW）内线性与‘枢纽’基础设施（hub infrastructure）的直接开发足迹。采用空间数据集与文献资源确定开发足迹范围的预测因子，并计算假想‘边缘效应区’。

#### 研究结果
仅能通过手动数字化识别的未制图线性基础设施，占直接开发足迹的最大比例。在面积达160,000 km²的大西部林地中，估算的开发足迹总面积为690 km²，其中67%为线性基础设施，剩余为‘枢纽’基础设施。研究区内估算的线性基础设施总长度达150,000 km，平均约1 km/km²。除直接开发足迹外，另有4,000~55,000 km²（占区域总面积的3%~35%）的区域处于边缘效应区内。

#### 研究结论
本研究凸显了线性基础设施的普遍性，因此将其累积影响管理作为景观保护的核心内容至关重要。本研究方法可推广应用至全球其他相对完整的景观区域。