Abstract
Satellite gravimetry as realized with GRACE and GRACE‐FO provides a novel opportunity to
study extreme deviations from annually varying terrestrial water storage (TWS) in all continental areas of our
planet. By utilizing the generalized extreme value (GEV) distribution, we estimate return levels for events
that are expected to happen once every 10 (i.e., 1‐in‐10) years. With two GRACE‐like reconstructions
spanning over 40 and 114 years, respectively, we show that the currently available data record of 20 years is
already sufficiently long to derive robust estimates of those return levels. When contrasting the GRACE/‐FO
results to model experiments from the CMIP6 archive extending until the year 2100 by concatenating
historical runs and climate projections under the SSP5‐8.5 socioeconomic pathway, we find that (a) the
multi‐model median from CMIP6 has the overall best agreement with the satellite data, thereby nicely
confirming the validity of a central assumption of many climate‐related studies that heavily rely on ensemble
statistics. We also find that (b) CMIP6 model runs contain only modest deviations of 1‐in‐10 years return
levels from the beginning of the 20th century when compared to present‐day, but predict stronger changes
toward more extreme return levels by the end of the 21st century. On the other hand, we also find substantial
differences between satellite data and individual model experiments, which opens new opportunities to
inform, validate and/or calibrate numerical climate models with satellite gravimetry data from GRACE,
GRACE‐FO, and in future also GRACE‐C.
study extreme deviations from annually varying terrestrial water storage (TWS) in all continental areas of our
planet. By utilizing the generalized extreme value (GEV) distribution, we estimate return levels for events
that are expected to happen once every 10 (i.e., 1‐in‐10) years. With two GRACE‐like reconstructions
spanning over 40 and 114 years, respectively, we show that the currently available data record of 20 years is
already sufficiently long to derive robust estimates of those return levels. When contrasting the GRACE/‐FO
results to model experiments from the CMIP6 archive extending until the year 2100 by concatenating
historical runs and climate projections under the SSP5‐8.5 socioeconomic pathway, we find that (a) the
multi‐model median from CMIP6 has the overall best agreement with the satellite data, thereby nicely
confirming the validity of a central assumption of many climate‐related studies that heavily rely on ensemble
statistics. We also find that (b) CMIP6 model runs contain only modest deviations of 1‐in‐10 years return
levels from the beginning of the 20th century when compared to present‐day, but predict stronger changes
toward more extreme return levels by the end of the 21st century. On the other hand, we also find substantial
differences between satellite data and individual model experiments, which opens new opportunities to
inform, validate and/or calibrate numerical climate models with satellite gravimetry data from GRACE,
GRACE‐FO, and in future also GRACE‐C.
| Original language | American English |
|---|---|
| Article number | e2024JB031011 |
| Number of pages | 22 |
| Journal | Journal of Geophysical Research: Solid Earth |
| Volume | 130 |
| Issue number | 10 |
| DOIs | |
| Publication status | Published - 28 Sept 2025 |