Groundwater-dependent ecosystem map exposes global dryland protection needs

Abstract

Groundwater is the most ubiquitous source of liquid freshwater globally, yet its role in supporting diverse ecosystems is rarely acknowledged^1,2. However, the location and extent of groundwater-dependent ecosystems (GDEs) are unknown in many geographies, and protection measures are lacking^1,3. Here, we map GDEs at high-resolution (roughly 30 m) and find them present on more than one-third of global drylands analysed, including important global biodiversity hotspots⁴. GDEs are more extensive and contiguous in landscapes dominated by pastoralism with lower rates of groundwater depletion, suggesting that many GDEs are likely to have already been lost due to water and land use practices. Nevertheless, 53% of GDEs exist within regions showing declining groundwater trends, which highlights the urgent need to protect GDEs from the threat of groundwater depletion. However, we found that only 21% of GDEs exist on protected lands or in jurisdictions with sustainable groundwater management policies, invoking a call to action to protect these vital ecosystems. Furthermore, we examine the linkage of GDEs with cultural and socio-economic factors in the Greater Sahel region, where GDEs play an essential role in supporting biodiversity and rural livelihoods, to explore other means for protection of GDEs in politically unstable regions. Our GDE map provides critical information for prioritizing and developing policies and protection mechanisms across various local, regional or international scales to safeguard these important ecosystems and the societies dependent on them.

Fig. 1. Global GDE map.

a, Global map shows GDE area density at 30 arcsecond resolution (roughly 1 km grids). Call-out circles show binary GDE classification at the full 1 arcsecond resolution (roughly 30 m grids). Bar plot (bottom left) shows GDE surface area distribution across continents. AS, Asia; AF, Africa; OC, Oceania; SA, South America; NA, North America; EU, Europe. b–g, Regional maps shown at the full 1 arc second resolution for the western USA (b), central Argentina and Chile (c), the central Sahel region (d), southern Africa (e), central Asia (f) and eastern Australia (g). The global map is shown in the Robinson projection whereas all panel insets are shown in geographic projection (latitude and longitude) referenced to the World Geodetic System (WGS) 1984 datum. An interactive version of the full resolution map is available at https://codefornature.projects.earthengine.app/view/global-gde.

Fig. 2. GDE area density and regional groundwater storage trends.

Global relationship between GDE area density and groundwater storage trends at 30 arcminute resolution (roughly 50 km grids). Annotated numbers inside the legend correspond to the area-weighted average values per freshwater ecoregion (Methods) highlighted in the map.

Fig. 3. The fragility of GDEs and associated communities within the Greater Sahel.

a, GDE area density at 5 arcminute resolution. b, Pastoral land area density at 5 arcminute resolution. c, District-level food insecurity classes as of October 2021. Food insecurity classes are Min., Minimal; Stress., Stressed; Crisis; Emerg., emergency. d, Armed conflict location and event data (all events between January 1997 and February 2021) summarized for GADM level 1 administrative areas. Data sources from b–d are provided in Supplementary Table 6.

Fig. 4. Protection status of GDEs.

The proportion of mapped GDEs with no protection (red) is 79%, with the remaining 21% having some degree of protection (blue and purple). GDEs shown in purple exist on protected areas or in jurisdictions with sustainable groundwater management policies. GDEs shown in blue are protected by both measures (protected area and sustainable groundwater management policy). GDE area density is shown in this figure at 30 arcsecond resolution (roughly 1 km grids).

Extended Data Fig. 1. Groundwater-dependent ecosystem (GDE) training and validation data (n = 34,454 points).

(a) Data sources: LANDFIRE vegetation (n = 6,652 points), Australian GDE Atlas data (n = 19,111 points), ESRI 10 m land use land cover bare ground data (n = 4,075 points), and sPLOT vegetation (n = 4,616 points). (b) GDE (n = 16,805 points) and non-GDE (n = 17,649 points) classifications.

Extended Data Fig. 2. Köppen-Geiger climate classifications used to designate dryland regions for GDE mapping.

Data Source: Beck et al., 2018.

Extended Data Fig. 3. Global depth to groundwater (≥30 meters).

Masked areas where depth to groundwater >30 meters is denoted in dark blue and were not included in the model. Data Source: Fan et al. 2017.

Extended Data Fig. 4. Random forest variable importance plot.

Random forest variable importance plot ranked highest to lowest: ratio of the annual sum of plant transpiration and precipitation (ETaP), ambient land surface temperature spatial anomaly (LST), Landsat modified soil adjusted vegetation index annual average (MSAVI: annual), normalized difference moisture index annual average (NDMI: annual), normalized difference vegetation index annual average (NDVI: annual), compound topographic index (CTI), normalized difference water index annual average (NDWI: annual), normalized difference moisture index inter-annual variability (NDMI: multiyear), normalized difference vegetation index inter-annual variability (NDVI: multiyear), Landsat modified soil adjusted vegetation index inter-annual variability (MSAVI: multiyear), and normalized difference water index inter-annual variability (NDWI: multiyear).

Extended Data Fig. 5. Predictor variable distribution plots comparing predictor values for the main model’s training and validation data (n = 34,454 points) and random global points within the model extent (n = 32,954 points).

The numbers in the upper left of each plot indicate the degree of overlap between the global and training point distributions, with zero indicating no overlap and one indicating complete overlap.

Extended Data Fig. 6. Groundwater-dependent ecosystem (GDE) probability map, indicating how likely a pixel is a GDE (100%) or non-GDE (0%) within the model extent at 1 arcsecond (~30 m) resolution.

Probability data are provided in 5 degree x 5 degree tiles (see Data availability). Areas outside of these tiles extent are shaded in gray. The GDE probability data map is also available as an interactive web map (https://codefornature.projects.earthengine.app/view/global-gde).

Extended Data Fig. 7. Groundwater-dependent ecosystems (GDE) area density intersected with pastoral lands area density.

Globally, our map indicates 59% of GDEs (4.9 million km²) overlap lands with >25% pastoral land density at 5 arcminute resolution.

Extended Data Fig. 8. Hyperparameter Tuning for the main Global GDE model.

Selected hyperparameter values are indicated by the red vertical dashed lines. The red horizontal dashed lines represent the highest accuracy of parameter values.