Global ecosystem restoration has unexpectedly low potential to mitigate climate change

Ecosystem restoration is increasingly recognized as a tool of climate
change mitigation. Former global-scale studies predicted that ecosystem
restoration can nearly offset human carbon emissions since the Industrial
Revolution, but these were heavily criticized for their tree-centric view
of global ecosystems or questionable modeling approaches, which hinders
planning large-scale, long-term, ecologically appropriate restoration
strategies. Here, we developed a model for the carbon capture potential of
ecosystem restoration until 2100, and show that the maximum amount is 96.9
Gt of carbon, which is 17.6% of the 640 Gt emitted since 1750, and only
3.7–12.0% if taking into account future emissions until 2100. Thus,
ecosystem restoration can play a minor role in climate change mitigation
even if orchestrated with a pervasive shift toward sustainable,
low-emission economies globally. In addition, if restoration targets are
planned to match future climatic conditions and consider state transitions
of current natural ecosystems due to climate change, the mitigation
potential of ecosystem restoration is close to zero. Therefore, we
recommend ensuring that ecosystem restoration is pursued primarily for
restoring biodiversity, supporting livelihoods, and resilience of
ecosystem services (apart from carbon sequestration), as the climate
mitigation potential of restoration is relatively minor and will vary
depending on the actual state transitions that occur between vegetation
types and feedbacks. 

# Global ecosystem restoration has unexpectedly low potential to mitigate
climate change
[https://doi.org/10.5061/dryad.ksn02v7g4](https://doi.org/10.5061/dryad.ksn02v7g4) ## Description of the data and file structure Data were collected from online open-access databases.  ### Files and variables #### File: Global\_model\_of\_ecosystem\_restoration.zip **Description:**  Scripts, input data, and output data of the paper "Global ecosystem restoration has unexpectedly low potential to mitigate climate change" by Tölgyesi et al., Nature Geoscience # Scripts ## 01\_prepare\_modeling.r R script that prepares the modeling. It creates a list of pieces of information (variable names, modeling parameters, etc.), loads the environmental dataset, and convert the environmental dataset to simple features (sf). ## 02\_train\_and\_evaluate\_models.r R script that trains and evaluates Random Forest models for the four studied ecosystem types, using 5-fold cross-validation. It iterates through the four ecosystem types, and repeats the following steps in each iteration: (1) creates five folds in a balanced way for cross-validation, (2) display the folds in a map for visual checking, (3) train five submodels (one submodel for one cross-validation fold), (4) makes prediction with each submodels for uncertainty calculation and model evaluation, (5) calculates prediction uncertainty, and (6) evaluates the model. ## 03\_display\_boxplot.r R script that displays the distribution of the predicted cover of the studied ecosystem types in boxplot grouped by 10% classes of observed cover. ## 04\_predict\_reference\_period.r R script that predicts the potential cover of the studied ecosystem types for the reference period. It iterates through the four ecosystem types, and repeats the following steps in each iteration: (1) makes prediction with each of the five submodels, (2) display the predictions in a map for visual checking, (3) calculates the ensemble mean of the five predictions, and (4) display the ensemble prediction in a map for visual checking. ## 05\_predict\_future\_period.r R script that predicts the potential cover of the studied ecosystem types for the future period. It iterates through the four scenarios, and repeats the following steps in each iteration: (1) if not yet downloaded, downloads the bioclimatic raster of the future period from WorldClim, (2) spatially align (reproject) the bioclimatic raster to the raster of the reference period, (3) merge the static variables from the raster of the reference period with the dynamic (bioclimatic) variables of the raster of the future period, and (4) iterates through the four ecosystem types, and repeats the following steps in each iteration: (i) makes prediction with each of the five submodels, (ii) display the predictions in a map for visual checking, (iii) calculates the ensemble mean of the five predictions, and (iv) display the ensemble prediction in a map for visual checking. # Input ## Supplementary Table 1.xlsx Excel spreadsheet containing (1) the coordinates (in WGS-84 reference system), (2) the observed cover of the four vegetation types, and (3) the environmental predictors for the 78,850 and 80,000 points used by Bastin et al. (2019) and added by the current research, respectively. This dataset is used for model training and evaluation. ## environmental\_raster.tif 30'' resolution geoTiff raster containing the environmental predictors. This dataset is used for making predictions for the reference period and the static predictors from this dataset are used (along with the dynamic predictors obtained from WorldClim) for making predictions for the future period. Layer names are described in the table below. | layer name | predictor name | unit of measurement | | ---------- | ---------------------------------------------------- | ------------------- | | bio1 | Annual mean temperature | °C × 10 | | bio8 | Mean temperature of wettest quarter | °C × 10 | | bio12 | Annual precipitation | mm | | bio15 | Precipitation seasonality (coefficient of variation) | - | | bio17 | Precipitation of the driest quarter of the year | mm | | gmtedhill | Slope | degree | | sndppt | Sand content in the top 15 cm of the soil | % | | ocstha | Soil organic carbon stock | g m-2 | | bdricm | Depth to bedrock | m | | mn75grd | Elevation above sea level | m | # Output ## 01/information.RData A list of pieces of information (variable names, modeling parameters, etc.). ## 01/dataset.RData A simple features (sf) dataset containing the observed cover of the ecosystem types and the environmental predictors in WGS-84 projection. ## 02/models.RData A four-elements list (one list element for each ecosystem type) of five-elements lists (one list element for each cross-validation fold) of trained Random Forest model objects. ## 02/predictions\_for\_evaluation.RData A four-elements list (one list element for each ecosystem type) of numeric vectors. Each vector contains the predicted cover values for the evaluation folds. ## 02/uncertainties.RData A four-elements list (one list element for each ecosystem type) of numeric vectors. Each vector contains the uncertainty of the predictions, i.e., the standard deviation of the five predicted cover values according to the five submodels. ## 02/evaluation\_dataset.RData A four-elements list (one list element for each ecosystem type) of simple features (sf) datasets used for model evaluation. The columns contain the predicted cover, the observed cover, and the observed cover reclassified to 11 categories (used for the boxplot visualization). ## 02/evaluation\_results.RData A data.frame containing the intercept, the slope and the R2 value of the linear model (used for model evaluation) for each ecosystem types. ## 02/evaluation\_results.xlsx An Excel spreadsheet containing the intercept, the slope and the R2 value of the linear model (used for model evaluation) for each ecosystem types. ## 02/*\[ecosystem type]*/folds.RData A numeric vector containing fold identifiers (i.e., integer values from 1 to 5) for each location where the cover value and the predictor values are known. Used for cross-validation. ## 02/*\[ecosystem type]*/folds.png A map visualization of the spatial distribution of the cross-validation folds. ## 02/*\[ecosystem type]*/mask\_of\_known.RData A logical mask indicating whether the observed cover value is known for each location where the predictor values are known. ## 03/boxplots.tif The predicted cover values in different observed cover intervals visualized in boxplots. The four vegetation types are displayed in separate subfigures. ## 03/percentages.RData A four-elements list (one list element for each ecosystem type) of data.frames. Each data.frame contains four columns: "observed_class", "count", "percentage", and "percentage_label". ## 04/*\[ecosystem type]*.tif 30'' resolution geoTiff raster containing the predicted cover of the studied ecosystem type in the reference period according to the ensemble model, i.e., the mean of the prediction of the five submodels. ## 05/*\[SSP scenario]*/ wc2.1\_30s\_bioc\_EC-Earth3-Veg\_*\[SSP scenario]*\_2061-2080.tif 30'' resolution geoTiff raster containing the bioclimatic variables of the future period (based on the studied SSP scenario) obtained from WorldClim. The bioclimatic variables are the dynamic predictors used (along with the static predictors originated from environmental_raster.tif) for making predictions for the future period. ## Access information Other publicly accessible locations of the data: * n/a Data was derived from the following sources: * Fick, S. E. & Hijmans, R. J. WorldClim 2: new 1-km spatial resolution climate surfaces for global land areas. *International Journal of Climatology* **37**, 4302–4315 (2017). * Hengl, T. et al. SoilGrids250m: Global gridded soil information based on machine learning. *Plos One* **12**, e0169748 (2017). * Danielson, J. J. & Gesch, D. B. *Global Multi-Resolution Terrain Elevation Data 2010 (GMTED2010).* Report No. OFR 2011-1073 (US Geological Survey, 2011).