Validation dataset for the article 'Rubber planting and deforestation'

This repository contains the reference dataset used for validating the rubber layer published in Wang et al.’s paper "High-resolution maps show that rubber causes substantial deforestation".

The repository is organized as follows:

The folder “Code_GEE” contains Google Earth Engine code for:

Extraction of country-wide areas for rubber, rubber deforestation, overlapping area with oil palm and pulpwood, and related statistics (“Code_GEE/Export_areas.js”)

Figure showing the validation datasets from Wang et al. and from this repository and the associated errors for class ‘Rubber’ (“Code_GEE/Figure_validationDatasets”)

Visualization of validation points, high-resolution images from Google Maps, and Sentinel-2 NBR time series (“Code_GEE/Visualize_validationPoints”)

The folder “Screenshots_reference_dataset” contains screenshots of high-resolution images from Google Maps and Sentinel-2 NBR time series in the validation points. The screenshot images are sorted by:

Map class according to the rubber layer in Wang et. al, 2023 (subfolder “Screenshots_reference_dataset/Sorted_by_MapClass”)

Truth class assigned by non-automated visual interpretation of high-resolution images and NBR time series (subfolder “Screenshots_reference_dataset/Sorted_by_TruthClass”)

The file “ReferenceDataset_MattersArising_rubber.xlsx” contains the coordinates of the validation points (‘lat’ and ‘lon’) along with a unique identifier (‘ID’), the mapped and truth classes (‘MapClass’ and ‘TruthClass’) and a boolean specifying whether the points were included in the validation of the entire study area (‘subset_global’) or only the primary forest area (‘subset_primaryForest’). Truth classes are the following:

Class 0 (Non-forest): Areas with permanent non-tree cover. This class includes non-tree covers such as grasslands, annual crops, and water and impervious surfaces.

Class 1 (Forest): Areas with permanent tree cover. This class includes any area with presence of shrublands or trees. Individual trees, clear-cut areas, and tree plantations (except rubber plantations) are included in this class.  

Class 2 (Monoculture rubber): Monoculture rubber plantations, either industrial or smallholder owned.

Class 3 (Agroforest rubber): Jungle rubber (mixed species multi-strata agroforests). 

The file “TotalAreas_Rubber-PrimaryForest-and-Overlap-OilPalm-Pulpwood.xlsx” is the output file generated from the code “Code_GEE/Export_areas.js”.

The file “Validation_RubberLayer-Wang2023.xlsx” contains the calculations for post-stratified accuracy metrics and area estimates. The Excel file includes the following worksheets:

‘ALL_mono’: Validation of the rubber layer for the entire study area (eight countries in Southeast Asia). Class ‘Rubber’ only includes Class 2 ‘Monoculture rubber’. Class ‘Agroforest rubber’ is included in Class 1 ‘Forest’.

‘ALL_mono&agro’: Validation of the rubber layer for the entire study area. Class ‘Rubber’ includes both Class 2 ‘Monoculture rubber’ and Class 3 ‘Agroforest rubber’.

‘DEF_mono’: Validation of the rubber layer only for the primary forest in the study area. Class ‘Rubber’ only includes Class 2 ‘Monoculture rubber’. Class 3 ‘Agroforest rubber’ is included in Class 1 ‘Forest’.

‘DEF_mono&agro’: Validation of the rubber layer only for the primary forest in the study area. Class ‘Rubber’ includes both Class 2 ‘Monoculture rubber’ and Class 3 ‘Agroforest rubber’.

本仓库存储了用于验证Wang等人发表于论文《高分辨率地图显示橡胶种植引发大规模森林砍伐》（High-resolution maps show that rubber causes substantial deforestation）中所提出的橡胶图层的参考数据集。

本仓库的组织形式如下：

### 文件夹`Code_GEE`
该文件夹包含谷歌地球引擎（Google Earth Engine）代码，用于实现以下功能：
1. 提取全国范围的橡胶、橡胶砍伐、与油棕和纸浆林的重叠区域及相关统计数据（对应脚本`Code_GEE/Export_areas.js`）；
2. 展示Wang等人的验证数据集与本仓库的验证数据集，以及"橡胶"类别的相关误差（对应脚本`Code_GEE/Figure_validationDatasets`）；
3. 可视化验证点、谷歌地图高分辨率影像以及哨兵2号（Sentinel-2）归一化燃烧比（Normalized Burn Ratio，NBR）时间序列（对应脚本`Code_GEE/Visualize_validationPoints`）。

### 文件夹`Screenshots_reference_dataset`
该文件夹包含验证点处谷歌地图高分辨率影像和哨兵2号NBR时间序列的截图。截图按以下两种方式分类存储：
1. 按照Wang等人2023年研究中橡胶图层的地图类别"MapClass"排序，存储于子文件夹`Screenshots_reference_dataset/Sorted_by_MapClass`；
2. 按照通过人工目视解译高分辨率影像和NBR时间序列所赋予的真实类别"TruthClass"排序，存储于子文件夹`Screenshots_reference_dataset/Sorted_by_TruthClass`。

### 文件`ReferenceDataset_MattersArising_rubber.xlsx`
该文件包含验证点的坐标（"lat"为纬度、"lon"为经度）、唯一标识符"ID"、映射类别"MapClass"与真实类别"TruthClass"，以及一个布尔值，用于标识该验证点是用于全研究区验证（"subset_global"）还是仅用于原生森林区域验证（"subset_primaryForest"）。真实类别定义如下：
- 类别0（非森林）：具有永久性非树木覆盖的区域，包括草地、一年生作物、水体以及不透水地表等非树木覆盖类型。
- 类别1（森林）：具有永久性树木覆盖的区域，涵盖任何存在灌丛或树木的区域，其中单个树木、皆伐区域以及人工林（橡胶种植园除外）均归入此类。
- 类别2（单一栽培橡胶）：单一栽培橡胶种植园，包括工业种植园或小农经营种植园。
- 类别3（农林复合橡胶）：丛林橡胶（混交多层物种农林复合系统）。

### 文件`TotalAreas_Rubber-PrimaryForest-and-Overlap-OilPalm-Pulpwood.xlsx`
该文件为脚本`Code_GEE/Export_areas.js`生成的输出文件。

### 文件`Validation_RubberLayer-Wang2023.xlsx`
该文件包含后分层精度评估指标和面积估算的计算结果。该Excel文件包含以下工作表：
1. `ALL_mono`：针对全研究区（东南亚八个国家）的橡胶图层验证，其中"橡胶"类别仅包含类别2"单一栽培橡胶"，"农林复合橡胶"归入类别1"森林"。
2. `ALL_mono&agro`：针对全研究区的橡胶图层验证，其中"橡胶"类别同时包含类别2"单一栽培橡胶"与类别3"农林复合橡胶"。
3. `DEF_mono`：仅针对研究区内原生森林区域的橡胶图层验证，其中"橡胶"类别仅包含类别2"单一栽培橡胶"，"农林复合橡胶"归入类别1"森林"。
4. `DEF_mono&agro`：仅针对研究区内原生森林区域的橡胶图层验证，其中"橡胶"类别同时包含类别2"单一栽培橡胶"与类别3"农林复合橡胶"。