ForestAge-Constrained Eddy-Covariance Gridded NEP Product

Description

This repository holds global spatial estimates of the Net Ecosystem Productivity of forests (NEP), circa 2010, for a grid spacing of 0.5° by 0.5º pixel size. Three different approaches were used to create the maps.

Model M1 (Regional Age–NEP Relationships Per Biome): This model scales site-level NEP observations to a global gridded field using biome-specific NEP-age curves and site-level anomalies. The random forest model (RF1) is trained on forest age, GPP, temperature, and nitrogen deposition, producing NEP anomalies that reflect site-specific deviations from biome-wide trends. Gridded predictor fields of forest age, GPP, temperature (MAT), and nitrogen deposition are used to create 0.5° by 0.5° NEP grids, with uncertainties estimated using an ensemble of 180 members. The data from Model M1 can be investigated from the ForestAge_EC_NEP_M1_v1.0.nc file.

Model M2 (Global Age–NEP Relationship): This model uses a random forest algorithm (RF2) to upscale NEP observations but applies a global NEP-age relationship across all sites. It uses the same gridded predictor fields as M1—forest age, GPP, MAT, and nitrogen deposition—but the age–NEP relationship is determined globally. Uncertainty is calculated similarly to M1, using ensembles of model parameters and predictor fields. The data from Model M2 can be investigated from the ForestAge_EC_NEP_M2_v1.0.nc file.

Model M3 (Without Age Consideration): This model predicts NEP solely based on GPP, MAT, and nitrogen deposition without accounting for forest age. It follows a similar approach to RF3 models from previous work and uses the same gridded predictors and uncertainty estimation methods as M1 and M2. The data from Model M3 can be investigated from the ForestAge_EC_NEP_M3_v1.0.nc file.

The variation across each model's members can assess the uncertainty in each model, which represents uncertainty caused by input variables and the k-fold cross-validation approach. 

More details about the methodologies behind the three approaches can be found in Ciais, P., Yao, Y. Besnard, S. et al. (2024) (see reference below).

Data structure

The datasets are stored in NetCDF format with a structure consistent across the different models (M1, M2, M3). Each file contains multiple variables representing components of the Net Ecosystem Production (NEP) estimates, such as the mean NEP and its quantiles. The primary variables are:

NEP_MX_mean: The mean estimate of NEP for each model (M1, M2, M3), with units of grams of carbon per square meter per year (gC m⁻² year⁻¹).

NEP_MX_quantiles: Estimates of NEP at different quantiles, providing uncertainty ranges. The quantiles represented in the data are: [0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75]

Members dimension: Each model includes a members dimension, representing several NEP estimates generated using different ensemble members. These members capture uncertainty from input variables such as GPP, temperature, nitrogen deposition, and forest age. The members dimension provides users with multiple realizations of NEP estimates, reflecting the variability these factors introduce.

Coordinates include latitude and longitude with CRS information (EPSG:4326). Missing data values are represented by -9999.

Citation

When using the maps, please cite the dataset, including the version number and the following paper: Ciais, P.,  Yao, Y. Besnard, S. et al. (2024) The global carbon balance of forests based on flux towers and forest age data, submitted. 

Version History

1.0 - Initial version, covering 2010