In groundwater depletion (GWD) regions, negative trends in groundwater storage (GWS) are problematic for groundwater drought detection, since they mask climate-induced drought signals. As this is not yet considered in any large-scale drought early warning system (LDEWS), we used GWS from the global hydrological model WaterGAP 2.2e to investigate, for the first time at the global scale, how groundwater drought can best be quantified in GWD regions. We analyzed two methods: 1) Linear detrending of monthly GWS time series and 2) analysis of naturalized GWS computed by assuming no human water use. We found that linear detrending is unsuitable for global-scale groundwater drought monitoring and forecasting as even small deviations from a pronounced linear trend can lead to a systematic over- and underestimation of the drought hazard. In contrast, indicators from naturalized GWS can identify climate-induced GWS anomalies.
Here, the reader can download 1) monthly time series of GWS_nat during 1980-2019, 2) groundwater drought hazard indicators assessed in the study, 3) the R scripts for computing the indicators and other data (including required input data), and 4) WaterGAP-related data (e.g., landmask, big cities), and other meta data (e.g., GWD grid cells and LTC grid cells). WaterGAP 2.2e model output from an anthropogenic model run is available at https://doi.org/10.25716/GUDE.0TNY-KJPG.

General remarks:
"-99" or "-999" in the data refer to NA (not applicable, not computable).

1) GWS_nat 1980-2019: Monthly time series of GWS_nat from a naturalized WG22e model run.

2) Groundwater drought hazard indicators as computed by WaterGAP 2.2e (climate data GSWP3-W5E5) for the whole globe except Antarctica, spatial resolution: 0.5°, monthly data for the reference period 1980-2009 and the evaluation period 2010-2019:

- EP1_ant, EP1_nat, EP1_ltc, EP12_prec (unitless; values need to be multiplied by 100 to be transformed into percent)
- D1_ant,D1_nat (in months)
- RP1_ant, RP1_nat (in years)

3) Nine R scripts:
- Eight out of the nine scripts are numbered and should be used subsequently as indicated.

4) GWS text files as input for R scripts to compute the EP1 variants:
- GWS data (G_GROUND_WATER_STORAGE_mm_gswp3w5e5_[...].txt) compiled from WG22e output (ant and nat variants) as well as GWS_ltc computed based on GWS_ant for 1980-2009 and 1980-2019.

5) Other data:
- watergap_arcid_lon_lat_continentalarea.txt:
WaterGAP grid cell IDs ("arcid") with longitude, latitude, and continental area per grid cells in km².

- big_cities_watergap_arcid_lon_lat.txt:
Grid cells with big cities were excluded from the assessment.

- gwd_and_ltc_cells_watergap_arcid_lon_lat.txt:
Groundwater depletion (GWD) grid cells and grid cells selected for linear trend correction (LTC).

- linear_GWS_ant_trend_Theil_Sen_1980_2009_and_2010_2019_in_mm_per_year_arcid_lon_lat.txt:
Computed linear GWS_ant trend (slope) using STL and the Theil Sen estimator.

- mean_annual_actual_NAg_and_GWR_mm_yr_1980_2009_arcid_lon_lat.txt:
Mean annual net abstractions from groundwater (NAg) in mm/yr during 1980-2009.
Mean annual groundwater recharge (GWR) in mm/yr during 1980-2009.

- number_of_drought_events_1980_2009_ant_nat_ltc_arcid_lon_lat.txt:
The number of drought events is required for the computation of the severity indicators (RP1 variants).

- variance_fraction_of_linear_trend_monthly_GWS_ant_mm_1980_2009_arcid_lon_lat.txt:
Variance ratio VR_lin [-] computed as the variance of GWS_lin divided by the variance of GWS_trend during the reference period 1980-2009.
The values are unitless and need to be multiplied by 100 to be transformed into percent.