https://doi.org/10.5061/dryad.p2ngf1w1f

Description of the data and file structure

In this study, we developed a global dataset and predictive models to explore the distribution, density, and biomass of arbuscular mycorrhizal (AM) fungal networks across Earth's underground ecosystems, with a focus on their role in nutrient cycling and carbon storage.

Database Creation:
To construct a database on mycelial network density, we compiled over 4,000 field measurements of arbuscular mycorrhizal (AM) fungi from both published sources using a literature search with Google Scholar in multiple languages. Each entry was georeferenced, including standardized measurements of hyphal length and density, adjusted for sampling depth, and if not georeferenced this was noted. We enriched the dataset with additional metadata, including soil chemistry, climate variables, and vegetation types, to enable comparisons across diverse ecosystems by sampling global rasters using Google Earth Engine. This harmonized database served as the foundation for training predictive models and conducting analyses on a global scale. We also used this database to test the effects of hosts and symbionts on mycelial network densities, and create a predictive map of AM fungal network density globally. 

Geospatial Modeling:
We leveraged the geolocated field samples in our database to train Random Forest models using Google Earth Engine, predicting AM fungal network densities at a 1 km² resolution in global soils. The models incorporated key environmental predictors, such as climate, soil properties, and land cover, to create spatially explicit predictions of fungal network density. We also performed uncertainty quantification at the pixel level. Multiple cross-validation strategies were used.

Experimental Approach:
In parallel, we conducted high-throughput imaging of over 280,000 hyphae from five AM fungal species using a custom-built robotic platform. This enabled precise measurements of hyphal radii and the calibration of biomass models based on network architecture. We used these data to calculate global fungal biomass. This relied on parameterizing the equation for the volume of hyphae for the AM fungal cell. 

ALL CONTENTS IN "Dryad_StewartBisot_etal.zip":

DIRECTORY AND FILE DESCRIPTIONS 

File "ReadMe_Metadata_TabularData" has all information on column names for all tabular files. 

Code

Agriculture_observed.R: Statistical analysis comparing observed hyphal density values between cultivated and non cultivated systems.

Agriculture_predicted.R: Analysis of model predicted hyphal density differences across agricultural land use.

Biomass turnover.R: Calculations related to hyphal biomass turnover rates and carbon flux estimates.

Deepsoil.R: Analysis of hyphal density patterns at deeper soil layers where available.

Global_Carbon_Prop_Uncertainty.R: Propagation of uncertainty for global carbon stock estimates derived from hyphal biomass.

HostEffects.R: Analysis testing the influence of plant host identity on hyphal density.

Main_pipeline.ipynb: Primary modeling workflow. Includes data preprocessing, model training, prediction, and raster export.

Model_uncertainty_map.R: Generation and mapping of spatial prediction uncertainty.

PotField.R: Comparisons between pot experiments and field measurements.

Sample_maps.R: Map exports.

SHAP analyses.R: Model interpretation using SHAP values to quantify environmental predictor importance.

Spatial autocorrelation.R: Assessment of spatial structure in residuals using Moran’s I and related statistics.

SymbiontEffects.R: Analysis testing fungal symbiont identity effects on hyphal density.

Zonal statistics.R: Extraction of raster statistics summarized by biome and ecoregion.

Databases

hyphal_density_m_cm3.csv: Core database of hyphal length density measurements standardized to meters per cubic centimeter.

hyphal_density_full_metadata.csv: Expanded database including geographic coordinates, sampling depth, plant host, ecosystem type, and study level metadata.

hyphal_density_m_cm3_GEE_processed.csv Hyphal database with environmental covariates extracted from Google Earth Engine.

HyphaePlantList.csv: List of plant species associated with hyphal measurements. Note that empty cells are included for code to run.

S_pironon_utilised_plants_species_list.csv: Reference list of cultivated plant species used in land use analyses.

Radius_effective_per_plate_avg.csv: Effective hyphal radius values derived from imaging. Used for converting hyphal length to biomass.

raw_data.zip: Includes folder "hyphal_density_data" with raw tif data in "tifs" subfolder on hyphal width. Also includes summaries of hyphal width in file "Radius_effective_per_plate_avg".

Rasters

Ecoregion_GEE.tif: Rasterized terrestrial ecoregions layer used for zonal summaries.

ESA_WorldCover_v2.tif: Global land cover classification layer.

hyphal_density_m_cm3_Classified_mean.tif: Global predicted mean hyphal density.

hyphal_density_m_cm3_Classified_sd.tif: Standard deviation of predicted hyphal density representing model uncertainty.

ResolveBiome.tif: Biome classification raster used for biome level summaries.

sample_intensity_hyphae_scaled_global.tif: scaled raster representing sampling intensity across the globe.

Shapefile: Ecoregions2017.shp 

Table

Biome_Hyphal_Summary.csv: Summary statistics of hyphal density and biomass aggregated by biome.

Ecoregion_Hyphal_Summary_update.csv: Hyphal density and biomass summaries aggregated by ecoregion.

plant_genus_sample_size.csv: Number of observations per plant genus.

DataSources.xlsx: Documentation of primary literature sources contributing to the hyphal database.

Software Requirements:

R version 4.6 with packages including: terra, sf, ranger, fastshap, tidyverse, ggplot2

Python 3.1

--------------------------------------------------------------------

Contact:

For questions regarding the dataset or code, contact: Justin Stewart, justin@spun.earth