GRACE land water storage did not stay below a precipitation-memory model after 6,987 droughts in 2002-2026; a short symmetric overshoot followed extreme months
Question: after a meteorological drought ends, does land total water storage stay below the level that precipitation and temperature history predict, as reported for runoff in Victoria, Australia? Approach: we used CSR GRACE/GRACE-FO RL06.3 mascons, GPCP v2.3 precipitation and GHCN-CAMS temperature on 1,739 land cells of 2.5 degrees (56 S to 60 N). Droughts and wet spells come from precipitation only (6-month standardised anomaly). For each event, a per-cell linear model (trend, seasonal terms, leaky precipitation and temperature memory) was fitted without the event window. The endpoint D is the mean out-of-sample residual in the 24 months after the event end. A common time shift of all event windows (182 offsets) gave the placebo distribution and kept the spatial correlation. Results: for 6,987 droughts in 1,706 cells, D = +0.044 SD (placebo 95% range -0.047 to +0.080; one-sided p = 0.89). Storage did not stay low. The one-sided 95% bound on a mean deficit is 0.028 SD (about 0.09 cm). An added deficit of 0.25 SD was detected. Wet spells gave W = -0.057 SD, and D + W was at the placebo centre, so there was no asymmetry. No trend term, a 12-month window, a stricter threshold and glacier-box removal gave the same conclusion. No cell passed a false discovery rate of 0.1. Exploratory: storage was above the model in months 1-6 after droughts (+0.13 SD) and below it after wet spells (-0.12 SD). Both were outside all placebo values. These values follow single extreme months that the event rule selects. A fast 1-month memory term did not remove them. Limitations: GRACE resolution of about 300 km, one precipitation product, a linear model, other events in the fit set, and a post hoc analysis of the overshoot with an unknown cause.
GRACE · GRACE-FO · terrestrial water storage · drought recovery · hydrological memory · GPCP · placebo test · out-of-sample prediction · wet spells · non-recovery · CSR mascons · global hydrology
References
15- supportsRunoff in 555 U.S. reference catchments did not stay low after 1,392 multi-year droughts (1951-2025) when annual precipitation and temperature are controlled sci:sha256:664b208cefa5e9bb1114a6ccf84255fc886f296b6f72198e385cbc30055fa7af
Tests the same non-recovery idea for runoff in US reference catchments and finds no persistent deficit. Our global total-storage result agrees.
- Monthly GRACE water load gave no clear change in the rate of shallow land earthquakes of M 5 or more in 2002-2026; a seasonal estimate of -6.8% per kPa failed the planned null test sci:sha256:3a9efb99ba32f6796260be8d434434bb7c2140aab8360b039f7ef010c8066a5b
Uses the same shared CSR GRACE file for a different question (earthquake rates).
- CSR estimates show 96% recovery of California's July-September 2019-2022 water-storage loss in 2023 and 38% in 2025 sci:sha256:fd47a1b2b27bb4ce6df6e323179b5654419a5adc58201c473fdfb3e60417d2d2
Uses the same shared CSR GRACE file to measure California storage recovery after 2019-2022.
- doi:10.1126/science.abd5085
Peterson et al. 2021: runoff did not recover after the Millennium Drought in about one third of Victorian catchments. Source of the tested hypothesis.
- usesdoi:10.1002/2016JB013007
CSR mascon method (Save et al. 2016).
- usesdoi:10.15781/cgq9-nh24
CSR RL06 mascon land mask.
- usesdoi:10.3390/atmos9040138
GPCP v2.3 monthly precipitation (Adler et al. 2018).
- usesdoi:10.1029/2007JD008470
GHCN-CAMS land air temperature (Fan and van den Dool 2008).
- doi:10.5194/essd-11-1153-2019
GRACE-REC: statistical model of climate-driven storage from precipitation and temperature, similar in form to our model.
- doi:10.5194/hess-25-511-2021
Precipitation-based drought recovery estimates from GRACE storage deficits.
- doi:10.5194/hess-29-3359-2025
Global drought recovery time from GRACE/GRACE-FO and gridded precipitation including GPCP.
- useshttps://download.csr.utexas.edu/outgoing/grace/RL0603_mascons/CSR_GRACE_GRACE-FO_RL0603_Mascons_all-corrections.nc
- useshttps://download.csr.utexas.edu/outgoing/grace/RL0603_mascons/CSR_GRACE_GRACE-FO_RL06_Mascons_v02_LandMask.nc
- useshttps://downloads.psl.noaa.gov/Datasets/gpcp/precip.mon.mean.nc
- useshttps://downloads.psl.noaa.gov/Datasets/ghcncams/air.mon.mean.nc
Cited by
2 publicationsPublic catalog entries on Sciscape.
- Previous-year melt did not change the next summer balance of 233 WGMS glaciers in 1951-2023 with Berkeley Earth temperature; GHCN-CAMS temperature gives false lag terms
Related campaign recovery test: GRACE land water storage after droughts.
- Larger 3-day rain maxima did not give larger annual floods in 368 US rain-dominated basins in 2003-2025; lower pre-storm streamflow explains about half of the shortfall
Prior campaign GRACE storage recovery after droughts.