Global soil organic carbon in tidal marshes version 1

This dataset is the first version of the predictions, expected model error, and area of applicability of the global soil organic carbon in tidal marshes at a 30 m resolution. All methods are provided in detail in the accompanying bioRxiv preprint, Maxwell et al. (2024) Soil carbon in the world's tidal marshes. Tidal marsh extent map Worthington et al. (2023) The distribution of global tidal marshes from earth observation data. bioRxiv. Training data Maxwell et al. (2023) Global dataset of soil organic carbon in tidal marshes. Scientific Data. Holmquist et al. (2024) The Coastal Carbon Library and Atlas: Open source soil data and tools supporting blue carbon research and policy. Global Change Biology. Citations for the training data from the above-mentioned syntheses are available here. Model Code available on Github. 3D soil modelling approach: Hengl & MacMillan (2019). Predictive Soil Mapping with R. Random forest model: Kuhn (2008). Building Predictive Models in R Using the caret Package. J. Stat. Softw. k-NNDM spatial cross validation: Meyer, Milà & Ludwig (2022). CAST: ‘caret’ Applications for Spatial-Temporal Models. Area of applicability: Meyer & Pebesma (2022). Machine learning-based global maps of ecological variables and the challenge of assessing them. Nature Communications. Description of files GRID.zip: shapefile with the location of each tile in the zipped folders below Final_predicted_SOC_both_layers.png: final predicted tidal marsh soil organic carbon (SOC) for a) the 0-30 cm soil layer and b) the 30-100 cm soil layer (aggregated per 2° cell). Area of applicability aoa0.zip: the area of applicability (AOA) mask for the 0-30 cm layer. Pixels with an AOA value of 0 or 0.5 are considered outside the AOA; with an AOA value of 1 are considered inside the AOA. aoa30.zip: the area of applicability (AOA) mask for the 30-100 cm layer. Pixels with an AOA value of 0 or 0.5 are considered outside the AOA; with an AOA value of 1 are considered inside the AOA. Final predictions and expected error pred0_aoa.zip: predicted soil organic carbon for the 0-30 cm layer (Mg C ha-1), masked by the area of applicability. pred30_aoa.zip: predicted soil organic carbon for the 30-100 cm layer (Mg C ha-1), masked by the area of applicability. err0_aoa.zip: expected model error for the 0-30 cm layer (Mg C ha-1), masked by the area of applicability. err30_aoa.zip: expected model error for the 30-100 cm layer (Mg C ha-1), masked by the area of applicability. Initial predictions and expected error pred0.zip: predicted soil organic carbon for the 0-30 cm layer (Mg C ha-1). pred30.zip: predicted soil organic carbon for the 30-100 cm layer (Mg C ha-1). err0.zip: expected model error for the 0-30 cm layer for all tidal marsh extent pixels (Mg C ha-1). err30.zip: expected model error for the 30-100 cm layer for all tidal marsh extent pixels (Mg C ha-1).

本数据集为全球潮汐沼泽土壤有机碳（Soil Organic Carbon, SOC）的预测值、模型预估误差及适用范围（Area of Applicability, AOA）的首个版本，空间分辨率为30米。所有方法的详细说明见伴随发布的bioRxiv预印本：Maxwell等人(2024)《全球潮汐沼泽土壤碳》。潮汐沼泽范围图来源：Worthington等人(2023)《基于地球观测数据的全球潮汐沼泽分布》，发表于bioRxiv。训练数据集来源包括：Maxwell等人(2023)《全球潮汐沼泽土壤有机碳数据集》，发表于《Scientific Data》；Holmquist等人(2024)《海岸碳库与图集：支持蓝碳研究与政策的开源土壤数据与工具》，发表于《Global Change Biology》。上述综合研究所用训练数据的引用信息可在此处获取。模型代码已上传至Github。3D土壤建模方法：Hengl与MacMillan(2019)《基于R语言的预测性土壤制图》（Predictive Soil Mapping with R）。随机森林模型：Kuhn(2008)《使用R语言caret包构建预测模型》，发表于《Journal of Statistical Software》。k-NNDM空间交叉验证：Meyer、Milà与Ludwig(2022)《CAST：面向时空模型的caret应用工具包》。适用范围分析方法：Meyer与Pebesma(2022)《基于机器学习的生态变量全球制图及评估挑战》，发表于《Nature Communications》。

文件说明如下：
GRID.zip：包含以下压缩文件夹中各瓦片位置的形状文件（Shapefile）。
Final_predicted_SOC_both_layers.png：最终预测的潮汐沼泽土壤有机碳（SOC）结果，其中(a)为0-30厘米土层，(b)为30-100厘米土层（按2°栅格单元聚合）。
aoa0.zip：0-30厘米土层的适用范围（AOA）掩膜文件。AOA值为0或0.5的像素被视为处于适用范围外；AOA值为1的像素被视为处于适用范围内。
aoa30.zip：30-100厘米土层的适用范围（AOA）掩膜文件。AOA值为0或0.5的像素被视为处于适用范围外；AOA值为1的像素被视为处于适用范围内。

经适用范围掩膜的预测值与预估误差：
pred0_aoa.zip：经适用范围掩膜的0-30厘米土层土壤有机碳预测值（单位：Mg C ha⁻¹）。
pred30_aoa.zip：经适用范围掩膜的30-100厘米土层土壤有机碳预测值（单位：Mg C ha⁻¹）。
err0_aoa.zip：经适用范围掩膜的0-30厘米土层模型预估误差（单位：Mg C ha⁻¹）。
err30_aoa.zip：经适用范围掩膜的30-100厘米土层模型预估误差（单位：Mg C ha⁻¹）。

初始预测值与预估误差（未掩膜）：
pred0.zip：0-30厘米土层的土壤有机碳预测值（单位：Mg C ha⁻¹）。
pred30.zip：30-100厘米土层的土壤有机碳预测值（单位：Mg C ha⁻¹）。
err0.zip：覆盖所有潮汐沼泽范围像素的0-30厘米土层模型预估误差（单位：Mg C ha⁻¹）。
err30.zip：覆盖所有潮汐沼泽范围像素的30-100厘米土层模型预估误差（单位：Mg C ha⁻¹）。