This is a non-peer reviewed preprint submitted to EarthArXiv

Global water cycle shifts far beyond pre-industrial

conditions – planetary boundary for freshwater change
transgressed
Miina Porkka1,2*†, Vili Virkki1*†, Lan Wang-Erlandsson3,4,5, Dieter Gerten5,6, Tom
5 Gleeson7,8, Chinchu Mohan7,9, Ingo Fetzer3,4, Fernando Jaramillo4,10, Arie Staal11, Sofie te
Wierik12, Arne Tobian3,5, Ruud van der Ent13, Petra Döll14,15, Martina Flörke16, Simon N.
Gosling17, Naota Hanasaki18, Yusuke Satoh19,20, Hannes Müller Schmied14,15, Niko
Wanders21, Johan Rockström3,5, Matti Kummu1
1
Water and Development Research Group, Aalto University; Espoo, Finland.
10 2
Global Economic Dynamics and the Biosphere, Royal Swedish Academy of Sciences; Stockholm, Sweden.
3
Stockholm Resilience Centre, Stockholm University; Stockholm, Sweden.
4
Bolin Centre for Climate Research, Stockholm University; Stockholm, Sweden.
5
Potsdam Institute for Climate Impact Research (PIK), Member of the Leibniz Association; Potsdam, Germany.
6
Geography Department and IRI THESys, Humboldt-Universität zu Berlin; Berlin, Germany.
15 7
Department of Civil Engineering, University of Victoria; Victoria, British Columbia, Canada.
8
School of Earth and Ocean Sciences, University of Victoria; Victoria, British Columbia, Canada.
9
Global Institute for Water Security, University of Saskatchewan; Saskatoon, Saskatchewan, Canada.
10
Department of Physical Geography, Stockholm University; Stockholm, Sweden.
11
Copernicus Institute of Sustainable Development, Utrecht University; Utrecht, Netherlands.
20 12
Institute for Biodiversity and Ecosystem Dynamics, Governance and Inclusive Development, University of
Amsterdam; Amsterdam, Netherlands.
13
Department of Water Management, Delft University of Technology; Delft, Netherlands.
14
Institute of Physical Geography, Goethe University Frankfurt; Frankfurt, Germany.
15
Senckenberg Leibniz Biodiversity and Climate Research Centre Frankfurt (SBiK‐F); Frankfurt, Germany.
25 16
Institute of Engineering Hydrology and Water Resources Management, Ruhr-University Bochum; Bochum,
Germany.
17
School of Geography, University of Nottingham; Nottingham, NG7 2RD, United Kingdom.
18
Center for Climate Change Adaptation, National Institute for Environmental Studies; Tsukuba, Japan.
19
Moon Soul Graduate School of Future Strategy, Korea Advanced Institute of Science and
30 Technology; Daejeon, Korea.
20
International Institute for Applied Systems Analysis; Laxenburg, Austria.
21
Department of Physical Geography, Utrecht University; Utrecht, Netherlands.

*Corresponding author. Email: [email protected], [email protected]

35 †These authors contributed equally to this work.

1
 This is a non-peer reviewed preprint submitted to EarthArXiv

Abstract
Human actions compromise the many life-supporting functions provided by the freshwater
cycle. Yet, scientific understanding of anthropogenic freshwater change and its long-term
evolution is limited. Using a multi-model ensemble of global hydrological models, we estimate
40 how, over a 145-year industrial period, streamflow and soil moisture have deviated from pre-
industrial baseline conditions (defined by 5th–95th percentiles, at 0.5° grid level and monthly
time step). We find increased frequency of local deviations on ~45% of land area, mainly in
regions under heavy direct or indirect human pressures. To estimate humanity’s aggregate
impact on the freshwater cycle, we present the evolution of deviation occurrence at regional to
45 global scales. Currently, local streamflow and soil moisture deviations occur on 18.2% and
15.8% of global land area, respectively, which is 8.0 and 4.7 percentage points beyond the ~3
percentage point wide pre-industrial variability envelope. Finally, we discuss applying our
approach to define a planetary boundary for freshwater change. Our results indicate a
substantial shift from stable pre-industrial streamflow and soil moisture conditions to
50 persistently increasing change and a transgression of the freshwater change planetary
boundary, calling for urgent actions to reduce human disturbance of the freshwater cycle.

2
 This is a non-peer reviewed preprint submitted to EarthArXiv

Introduction
Freshwater systems globally are under unprecedented pressure from human actions. Water
extraction and infrastructure, land use and land cover change, and climate change now
55 considerably modify the quantity and timing of atmospheric and terrestrial freshwater flows,
with crucial implications for Earth’s climate and ecosystems1. Remarkable signals of water
cycle changes include widespread and severe river flow regime alterations2, intensification3
and homogenisation4 of the global water cycle, and increases in the severity, frequency, and
duration of floods and droughts5.
60 Scientific understanding of the magnitude and nature of anthropogenic water cycle change that
matters for the hydroclimatic and hydroecological stability of the Earth system is yet limited.
Water cycle changes are typically analysed over relatively short time periods, typically
decades6–8. Furthermore, changes are often quantified against reference conditions that are
already affected by anthropogenic influences such as climate change (e.g. 9–11). Studies also
65 rarely provide global aggregation of water cycle changes that would be meaningful for
assessing the state and trajectory of the global water cycle as a whole. For example, assessing
change in single elements of the water cycle2,6,11–17 or aggregating freshwater flows and stocks
very broadly (whether in space7,18,19 or in time8,20) may disregard the Earth system relevance
of freshwater. Thus, overall, existing studies lack a long-term overarching perspective on
70 aggregated anthropogenic water cycle change in the Earth system.
Here, we assess humanity’s ever-increasing aggregate impact on the freshwater cycle,
quantifying blue (represented by streamflow) and green water (represented by root-zone soil
moisture – hereinafter simply soil moisture – that is the soil water available to plants) deviations
from pre-industrial baseline conditions, across scales from local to global. Jointly assessing
75 change in streamflow and soil moisture conditions, using coherent methodology, allows for
inferring changes in other water cycle elements and the varying drivers of freshwater change.
We define the pre-industrial period 1661–1860 as a baseline that we assume to be largely
undisturbed by human impacts and assess freshwater change over the following one and a half
centuries (1861–2005) against it – illustrating the trajectory of how human-driven freshwater
80 change has evolved. Widespread deviations from the ‘pristine’ state that are detected by our
approach can be considered to pose elevated risks to the Earth system functions of freshwater.
For example, terrestrial and freshwater ecosystems have adapted to specific quantities and
timing of water flows such that changes in soil moisture and streamflow can severely affect
ecosystem status and biodiversity21–23. Moreover, the wetness of landscapes regulates climate
85 from micro- to regional and global scales, such that changes in the magnitude and timing of
wetness can impact rainfall and consequently streamflow and soil moisture both locally and
remotely24.
Our approach refines the method proposed by Wang-Erlandsson et al.25 for quantifying a
planetary boundary (PB) for green water. PBs set limits to nine processes, including freshwater
90 use, that together regulate the state of the Earth system and thus delimit a safe operating space
for retaining a quasi-stable Holocene-like state26,27. Previous freshwater PB estimates, which
have focused on only blue water use, have been criticised for their limited capacity to capture
the interconnected direct and indirect anthropogenic pressures on the water cycle28. Advancing
the work of Wang-Erlandsson et al.25, we present a coherent, comparable, and spatially explicit

3
 This is a non-peer reviewed preprint submitted to EarthArXiv

95 assessment of blue (streamflow) and green water (soil moisture) change, utilising an ensemble
of consistently forced global hydrological models (Fig. 1, Methods). Our approach robustly
addresses criticism of previous freshwater PB definitions and quantifications. We therefore
follow Wang-Erlandsson et al.25 in proposing to use it to replace the current definition of a PB
for freshwater use with an integrated PB for freshwater change.

100 Main
To estimate freshwater change across scales, we used streamflow and soil moisture data
simulated by an ensemble of state-of-the-art gridded global hydrological models that were
forced with bias-adjusted CMIP5 climate models and dynamic socio-economic conditions
(Methods, Table S1). The data was given in two time periods: pre-industrial period (1661–
105 1860) and industrial period (1861–2005).
Based on the 200-year pre-industrial period, we set the local baseline range (Fig. 1a) for each
0.5° grid cell and month, respectively for streamflow and soil moisture. The local baseline
range is defined as the range between the 5th (dry bound) and the 95th (wet bound) percentiles
of pre-industrial monthly streamflow or soil moisture values (Fig. 1a). We then compared pre-
110 industrial streamflow and soil moisture data against the local baseline range to detect local
deviations (Fig. 1b). Months with values below the dry bound were marked as dry local
deviations, and months with values above the wet bound were marked as wet local deviations
(Fig. 1b). The areas of grid cells with local streamflow or soil moisture deviations were then
summed up globally (regionally). We divided this sum by total ice-free land area to yield the
115 percentage of land area with local deviations (Fig. 1c). Finally, to establish global (regional)
reference conditions, we defined pre-industrial variability (Fig. 1d) from the pre-industrial
percentage of land area with local deviations. Two key metrics characterise pre-industrial
variability: the median of pre-industrial variability and the upper end of pre-industrial
variability (Fig. 1d). The median corresponds to the 50th and the upper end to the 95th percentile
120 of the percentage of land area with local streamflow or soil moisture deviations during the pre-
industrial period.
After defining the pre-industrial reference conditions, we compared industrial streamflow and
soil moisture data against them. We repeated the detection of local deviations (Fig. 1b) and
computed the percentage of land area with local deviations (Fig. 1c), which allowed us to
125 compare changes in local deviation frequency (Fig. 1e) and in the percentage of land area with
local deviations (Fig. 1f).

4
 This is a non-peer reviewed preprint submitted to EarthArXiv

Fig. 1. Methodological outline: setting the local baseline range (a), identifying local deviations (b),
computing the percentage of land area with local deviations (c), defining pre-industrial variability (d), and
130 comparing the industrial period against the pre-industrial period (e–f). The steps a–d are described in detail
in Methods.

Global freshwater change

We found no substantial changes or trends in the global land area with local streamflow or soil
moisture deviations during the pre-industrial period, which indicates stable water cycle
135 conditions prior to industrialisation (Fig. 2). The median of pre-industrial variability (i.e. the
typical land area with freshwater deviations) was at 9.4% for streamflow and at 9.8% for soil
moisture, respectively (Fig. 2). This is in accordance with the expectation that dry or wet
deviations should each occur in 5% of the data points in each grid cell over the pre-industrial
period (Methods). Annually, the pre-industrial percentage of land area with local deviations
140 varied mostly within +/- 1.5 percentage points (pp) around the median of pre-industrial
variability, and no global trends could be observed before 1860 (Fig. 2). Therefore, these
relatively stable pre-industrial conditions can be considered a useful reference baseline that is
not influenced by major anthropogenic impacts on the water cycle.

5
 This is a non-peer reviewed preprint submitted to EarthArXiv

Fig. 2. Percentage of global ice-free land area with local streamflow (a) and soil moisture (b) deviations.
145 Shown is the annual percentage, which is computed as an average of monthly percentages. The annotated years
mark the 10-year moving (trailing) mean transgressing the upper end of pre-industrial variability (95th percentile
of the pre-industrial percentage of land area with local deviations; Fig. 1d). The ensemble median and interquartile
range are computed from n = 23 (streamflow) and n = 15 (soil moisture) ensemble members. Sensitivity analysis
using different definitions of the local baseline range (Fig. 1a) is presented in Extended Data Fig. 1. Values prior
150 to 1691 are excluded and the ensemble median line for 1861–1890 is shaded and dashed due to traces of model
spinups being common during these years (Extended Data Fig. 2–3). Model-wise percentages of land area with
local freshwater deviations are shown in Extended Data Fig. 4–5. Sensitivity analysis using different definitions
of the local baseline range (Fig. 1a) is presented in Extended Data Fig. 1.

However, occurrence of streamflow and soil moisture deviations started to steadily increase
155 after the end of the pre-industrial period, with both global indicators surpassing the upper end
of pre-industrial variability in the early 20th century (Fig. 2). Global land area with local
streamflow deviations transgressed the upper end of pre-industrial variability (10.2%) already
in 1905. The degree of transgression continued to increase apart from two dips around 1940
and 1970. By the end of our study period (mean of 1996–2005), areas with local streamflow
160 deviations covered 18.2% (~24 million km2) of global ice-free land area (Fig. 2a). This
corresponds to a 78% (8.0 pp) increase compared to the upper end and a 94% (8.8 pp) increase
compared to the median of pre-industrial variability. The development of streamflow deviation
occurrence is mainly due to dry deviations, which were still increasing in spatial coverage by
the end of our study period, while the increase in wet deviations had plateaued decades earlier
165 (Extended Data Fig. 6b–c).
Soil moisture exhibits a similar, though slightly less pronounced, global trajectory and pattern:
the upper end of pre-industrial variability (11.1%) was transgressed in 1929, after which the
land area with local soil moisture deviations increased consistently, apart from two dips (like

6
 This is a non-peer reviewed preprint submitted to EarthArXiv

streamflow) in the latter half of the 20th century (Fig 2b). During the last decade (1996–2005)
170 of our analysis period, local soil moisture deviations occurred on 15.8% (~ 20 million km2) of
global ice-free land area, which is 42% (4.7 pp) above the upper end of pre-industrial variability
and 61% (6.0 pp) above its median. In contrast to streamflow, by the end of our study period,
occurrence of dry soil moisture deviations was declining slowly, yet remaining at considerably
higher levels than during the pre-industrial period, while occurrence of wet deviations
175 continued to increase towards the end of the 20th century (Extended Data Fig. 6e–f).
These global patterns are robust against key methodological sensitivity. When widening the
local baseline range, relative transgressions of the upper end of pre-industrial variability
increase (indicating that the most extreme conditions increased more than less extreme), but,
importantly, pre-industrial stability and the temporal evolution during the industrial period
180 remain similar (Extended Data Fig. 1). In addition, though the model spread is relatively large
especially towards the present day and in the case of dry streamflow deviations (Extended Data
Fig. 4–6), the entire ensemble interquartile range surpasses the upper end of pre-industrial
variability in the latter half of the 20th century (Fig. 2). This means that the hydrological models
used here mostly agree on the evolution of the percentage of global land area with local
185 deviations. Hence, there has been – and is still ongoing – a robust and remarkable drift away
from the quasi-stable pre-industrial soil moisture and streamflow conditions (characterised by
a narrow, ~3 pp wide variability range) to persistently increasing change in the freshwater cycle
(currently 4.7–8.0 pp beyond pre-industrial variability).
Local and regional freshwater change
190 Mapping changes in the frequency of local deviations reveals a general pattern of more frequent
dry deviations of both streamflow and soil moisture in much of the tropics and subtropics, with
wet deviations becoming more common in temperate and subpolar regions, and in many
highland areas (Fig. 3, Extended Data Fig. 7). For streamflow, increases in the frequency of
dry local deviations are more common, while for soil moisture, neither dry nor wet deviation
195 frequency increase dominates over the other. Comparing local deviation frequency in 1976–
2005 against 1691–1860, 39.2% of land area shows a statistically significant increase in dry
streamflow deviation frequency, while 9.2% of land area exhibits a statistically significant
increase in wet streamflow deviation frequency. For soil moisture, 26.9% of land area shows
an increase in dry and 19.2% an increase in wet deviation frequency. Areas in which the
200 frequency of both dry and wet local deviations increased significantly cover 2.0% (streamflow)
and 1.0% (soil moisture) of land area. Therefore, 46.4% of land area has experienced a
statistically significant increase in streamflow deviation frequency and 45.1% in soil moisture
deviation frequency. Increases in local deviation frequency were major (> 5 pp increase, i.e.
the frequency of deviations has more than doubled compared to the expected 5% frequency) in
205 approximately half of the area with increased frequency of dry streamflow deviations, a third
of the area with increased frequency of soil moisture deviations (both dry and wet), and a
quarter of the area with increased frequency of wet streamflow deviations (Fig. 3).

7
 This is a non-peer reviewed preprint submitted to EarthArXiv

Fig. 3. Statistically significant increases in dry and wet local deviation frequency for streamflow (a) and for
soil moisture (b). Changes in the frequency of local deviations are computed by comparing the ensemble median
210 frequency of local deviations during 1976–2005 against 1691–1860, and significance of change is tested at a 95%
confidence level (p = 0.05) with R package stats function prop.test29. Colours denoted with + indicate statistically
significant increases with magnitude ≤5 pp (minor), whereas colours denoted with ++ indicate statistically
significant increases with magnitude >5 pp (major). Colours denoted with +(+) pool together any statistically
significant increase (minor or major).

215 Persistent freshwater changes and their timing are spread out unevenly (see Fig. 4, in which
local freshwater deviations are aggregated at river basin scales instead of the global scale). In
many regions – especially around mid-latitudes – land area with local deviations persistently
transgressed the region-specific upper end of pre-industrial variability already before 1940,
while in other areas – particularly in humid regions – a transgression has not occurred yet. The
220 Mississippi, Indus, and Nile basins, for instance, were among the first regions where persistent
transgression happened in case of streamflow (in the Nile, also for soil moisture) (Fig. 4a–b).
For soil moisture, persistent transgression has occurred in fewer regions and often later than in
the case of streamflow (Fig. 4c–d). This is visible particularly in Siberian basins, the Congo
basin, and South/Southeast Asia, where the persistent transgression year for soil moisture
225 followed 10–20 years after that of streamflow – or the transgression has not occurred yet (Fig.
4).

8
 This is a non-peer reviewed preprint submitted to EarthArXiv

Fig. 4. Timing of persistent transgression of the region-specific upper end of pre-industrial variability, for
streamflow (a–b) and soil moisture (c–d). Here, local deviations are first aggregated regionally to get pre-
industrial percentage of land area with local deviations (Fig. 1c), then the upper end of pre-industrial variability
230 is set to the 95th percentile of the percentage of land area with local deviations (Fig. 1d), and finally the percentage
of land area with local deviations is calculated for the industrial period and compared against pre-industrial
variability (Fig. 1f). The persistent transgression year is defined as the first year, during which the 10-year moving
(trailing) mean of percentage of land area with local deviations has exceeded the region-specific upper end of pre-
industrial variability for ten consecutive years, without returning below this limit after this year. The regions
235 shown here depict basins delineated by the HydroBASINS data set level 2 (a, c; n = 60, mean area = 2,247,000
km2, median area = 2,045,000 km2) and level 3 (b, d; n = 264, mean area = 511,000 km2 median area = 313,000
km2)30.

Drivers and impacts of freshwater change

Assessing the changes in local streamflow and soil moisture deviations in unison provides more
240 detailed insights into their causes and potential impacts. The general pattern of increasing wet
deviation frequency of both streamflow and soil moisture in the Northern Hemisphere likely
arises from the precipitation change due to ~1°C of mean global warming5, which affects both
streamflow and soil moisture similarly if major human impacts on land surface, such as land
cover change and soil degradation, do not change precipitation partitioning. Another distinctive
245 example of climate-induced deviations is the Sahel region (stretching across the African
continent south of the Sahara; partially covered by the Niger river basin in Fig. 5). Drier
conditions (relative to pre-industrial climate) dominated in the region already in the first half
of the 20th century31 – which agrees with our analysis showing increased dry deviation
frequency already when comparing 1931–1960 against 1691–1860 (Extended Data Fig. 8) –
250 and intensified from the 1970s onwards. Recent decades have additionally witnessed
widespread wet local deviations (and different combinations of dry and wet), which could be
explained by changes in temporal dynamics, such as intensity of precipitation, number of wet
days, occurrence of dry spells and timing and length of the rainy season31,32.

9
 This is a non-peer reviewed preprint submitted to EarthArXiv

As an example of direct human drivers, increased frequency of dry local streamflow deviations
255 in combination with wet local soil moisture deviations in a given area most likely indicates an
effect of irrigation expansion. Indeed, these cases are observed in many heavily irrigated
regions33, such as South Asia, eastern China, Western USA, and the Nile delta (Fig. 5, pink
colour). These regions are also the likely early drivers of globally aggregated deviations, as
their irrigation extent was relatively large already in the early 20th century33, and local
260 deviations were prevalent already then (Extended Data Fig. 8). Increases in the frequency of
both wet and dry streamflow deviations, as well as wet soil moisture deviations, (Fig. 5, dark
blue) could imply a situation in which the natural river flow regime has been altered by dam
operation (decreased flood peaks and increased dry season streamflow) in combination with
extensive irrigation (decreased streamflow, increased soil moisture). These cases can be found
265 in many heavily modified river basins34, such as the Nile, the Aral Sea, and parts of India and
Thailand (Fig. 5).

Fig. 5. Combined increases in local streamflow and soil moisture deviation frequency along with regional
percentages of land area with dry or wet local deviations in selected regions. The classification is based on
local deviation frequency increases presented in Fig. 3, and it pools together both minor and major increases
270 (represented by +(+) in the legend). Computing the percentage of land area with local deviations (Fig. 1c) was
performed within each region, and the median and upper end of pre-industrial variability (Fig. 1d) were also
computed regionally.

Our findings of increased streamflow and soil moisture deviation frequency are consistent with
many regional to global freshwater-mediated impacts that have been reported independently.
275 One of the most dramatic examples of blue water change is the Aral Sea, where the
anthropogenic overuse of water for irrigation resulted in a massive lake depletion, consequent
ecological degradation, and regional climate change35. In our results, the region’s two large
river basins, Amu Darya and Syr Darya, exhibit a steep increase in dry streamflow and a

10
 This is a non-peer reviewed preprint submitted to EarthArXiv

moderate increase in wet soil moisture deviations, starting in the late 1960s (Fig. 5) – in
280 accordance with the substantial irrigation expansion that started in the region at that time35.
Soil moisture changes have been associated with productivity loss, as exemplified by drying-
induced forest dieback36 in many of the regions where we find a major frequency increase in
dry soil moisture deviations, such as the Mediterranean basin, central North America and West
Africa (e.g. Niger river basin in Fig. 5). Productivity shocks due to dry and wet soil moisture
285 deviations (droughts and floods) have also been reported on many cultivated lands, particularly
in South and East Asia, Australia, and North Africa37, where we find both drying and wetting
(Fig. 3b, India in Fig. 5). Other examples of ecological and climatic impacts of water surpluses
are habitat loss in the Central Amazon floodplains due to anthropogenic flood pulse disturbance
by dams38, and increased greenhouse gas emissions from reservoirs39, though these freshwater
290 alterations are relatively small-scale and thus not clearly visible in our maps.
Introducing the new planetary boundary for freshwater change
The existing freshwater planetary boundary estimates have fallen short in capturing freshwater
change due to e.g. land use and land cover change, and anthropogenic greenhouse gas and
aerosol emissions, which alter evaporation, soil moisture, precipitation and runoff patterns28,40–
44
295 . Existing approaches have also been criticised for aggregating freshwater fluxes too broadly,
e.g. by operating on an annual scale only or simply summing global water use and availability
despite diverse local impacts45,46. These simplifications may conceal important variation across
space and time and overlook, or cancel out, the often-contrasting impacts of human drivers on
different parts of the water cycle. For example, irrigation decreases streamflow but increases
300 soil moisture, and hydropower operations increase evaporation by reservoir impoundment, as
well as shift water flows temporally from the wet to the dry season and blue water scarcity
hotspots from upstream to downstream47. Therefore, the narrow and one-dimensional outlook
on freshwater change by the existing freshwater PB estimates is due for a more comprehensive
replacement.
305 Following Wang-Erlandsson et al.25, we argue that our approach and results are fitting to be
used in setting a new PB for freshwater change (as opposed to the current PB for freshwater
use) because we can capture global water cycle change by analysing several of its components,
using metrics that relate to its Earth system impacts, and adopting meaningful (Holocene-like)
baselines and long time scales. Although streamflow and soil moisture represent only a part of
310 the freshwater cycle, they are connected, either directly or indirectly, to all major anthropogenic
modifications of the freshwater cycle and associated Earth system functions (SI text).
We propose the upper end (95th percentile) of pre-industrial variability in global land area with
local streamflow (blue water) and soil moisture (green water) deviations as the two components
of the PB for freshwater change. This definition represents global conditions occurring
315 expectedly once in 20 years in the pre-industrial reference state and corresponds to 10.2% of
the global land area for blue water (Fig. 2a) and 11.1% for green water (Fig. 2b). This boundary
placement assumes that widespread deviations from pre-industrial conditions (which represent
longer-term Holocene-like conditions) pose elevated risks to freshwater’s Earth system
functions. According to our results, the freshwater change PB has been transgressed for a long
320 time – the current status is 18.2% for the blue and 15.8% for the green water component (Fig.
2). This is in stark contrast with the key PB publications’ estimates of the freshwater use PB’s

11
 This is a non-peer reviewed preprint submitted to EarthArXiv

safe status27,48 but agrees with other, more nuanced studies on the freshwater use PB18,49, which
consider the boundary to be transgressed.
It should be noted that there are major uncertainties associated with setting the freshwater
325 change PB (Methods, SI text). While data-related uncertainties have been mitigated by using a
large ensemble of state-of-the-art global hydrological models that have been validated against
observations50–53, perhaps the most important uncertainty stems from the lack of robust
quantifications of Earth system-wide responses to water cycle modifications that surpass pre-
industrial variability. These responses may be too complex to be quantified with our current
330 knowledge, and it is also possible that the level of freshwater modifications that drives state
changes and increases risks in the Earth system can only be determined in retrospect.
Nevertheless, because the freshwater cycle upkeeps many life-supporting Earth system
functions, risks of environmental degradation can be assumed to elevate along with change
away from a stable, Holocene-like state1 – a change that has occurred in the freshwater cycle
335 within the last century (Fig. 2). While it is possible that buffers or stabilising feedbacks (such
as CO2 fertilisation) in the Earth system allow for increased variability in the water cycle (such
as for the currently observed change), there is also evidence that self-reinforcing feedbacks
(such as forest dieback and moisture recycling feedbacks54) may amplify the impacts of
freshwater change on Earth system functioning. Considering also freshwater’s central role in
340 Earth system process interactions55, the precautionary principle of the PB framework48 thus
motivates a conservative placement of the new PB for freshwater change at the lower level of
scientific uncertainty. Here, we consider this to be the upper end of pre-industrial variability.
The major transgression and remaining uncertainties of this boundary placement, however,
warrant continued research on the role of freshwater in the Earth system.

345 Concluding remarks

We have presented here the trajectory of anthropogenically driven freshwater change since the
stable pre-industrial period and shown that this change has been pervasive across spatial and
temporal scales. Globally, land area in which streamflow and soil moisture deviates from local
pre-industrial reference conditions has increased by 78–94% and 42–61%, respectively, within
350 the 145-year industrial period. This major shift in the global water cycle is in accordance with
the few other studies reporting long-term (>100 years) changes, such as groundwater depletion
of at least ~4,500 km3, corresponding to >6% of total sea-level rise16 (and possibly much
greater15), and at least a 21% loss of wetlands since the year 170013,14. Our results align well
also with studies reporting decadal changes and trends in increasing hydrological extremes5.
355 Locally, evidence of freshwater-triggered ecological and climatic shifts has been mounting as
the main components of the freshwater cycle have moved further away from their pre-industrial
variability. Our findings indicate a transgression of the planetary boundary for freshwater
change already around the mid-20th century, while climate change56, deforestation57 and many
other human pressures on the water cycle continue to pose a major risk of further change.
360 Decreasing these pressures are thus imperative to reducing human disturbance of the water
cycle and to avoiding compromising the life-supporting functions of freshwater.

12
 This is a non-peer reviewed preprint submitted to EarthArXiv

Methods
Data selection
365 Our main data source was the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP)
data repository (available at: https://data.isimip.org, last accessed March 30th, 2023), from
which we used data of the ISIMIP 2b simulation round58 experiments (Table S1). We used
root-zone soil moisture (hereinafter soil moisture; ISIMIP output variable rootmoist) to
represent green water and river discharge (hereinafter streamflow; ISIMIP output variable dis)
370 to represent blue water as the two main components of the freshwater cycle. The models in
ISIMIP 2b have been validated against observations, showing adequate performance especially
when estimates from individual models are combined in an ensemble modelling approach50–53.
By constructing ensembles as large as possible, given the availability of different simulation
scenarios, we can partially tackle the uncertainty to which global modelling is always subject
375 to, to a certain degree59.
Two distinct simulation scenarios were required to determine the pre-industrial reference state
(Fig. 1a–d) and compare the industrial and pre-industrial states of the water cycle (Fig. 1e–f).
For the pre-industrial period (1661–1860), we used model outputs forced with picontrol climate
and 1860soc land use and socio-economic conditions, which described our pre-industrial
380 reference state with a constant 286 ppm CO2 concentration and fixed pre-industrial land use
and socio-economic conditions58. The pre-industrial simulation years are, however, nominal,
as the CMIP5 data used to force ISIMIP 2b models represent 200 control years with climate
variability at pre-industrial levels (with no correspondence to the actual weather of individual
years)58. At the transition from pre-industrial to industrial time (as defined by the ISIMIP
385 protocol) in 1860, we switched to historical climate combined with histsoc land use and socio-
economic conditions, in which carbon-induced climate change and human influences (e.g. land
use change, water use, and dam operation) were dynamically represented in the data58. The
ISIMIP 2b industrial period covers years from 1861 to 2005, the end year owing to estimates
of e.g. irrigation extent not being available33.
390 In ISIMIP 2b, soil moisture and streamflow were simulated by global hydrological models
(GHMs), which were forced with bias-adjusted output of modelled climate from general
circulation models (GCMs). The GCM outputs from CMIP5 models60 were bias-adjusted with
an observational dataset covering the years 1979–201358. We selected all GHM-GCM
combinations, for which output data were available for both picontrol-1860soc and historical-
395 histsoc scenarios (Table S1). For soil moisture, outputs from four GHMs (CLM5061, LPJmL62,
MPI-HM63, and PCR-GLOBWB64) were available, and for streamflow, outputs from six
GHMs (H0865, LPJmL, MATSIRO66, MPI-HM, PCR-GLOBWB, and WaterGAP267) were
available. For all GHMs, outputs of four GCMs were available: GFDL-ESM2M68, HadGEM2-
ES69, IPSL-CM5A-LR70, and MIROC571, except for MPI-HM, which lacked data forced with
400 HadGEM2-ES for both streamflow and soil moisture. Hence, our soil moisture data ensemble
consisted of 15 members (GHM-GCM combinations) and our streamflow data ensemble
consisted of 23 members. For streamflow, we aggregated daily values to monthly values by
taking the mean of daily streamflow, whereas soil moisture data were readily delivered in
monthly time resolution. The spatial resolution of all data was 0.5 degrees (ca. 50 x 50 km at
405 the Equator).

13
 This is a non-peer reviewed preprint submitted to EarthArXiv

Data preparation
Because the data we used originated from two different scenarios, and the ISIMIP simulations
can be run independently of each other, the outcomes of the two scenarios may not be
continuous (i.e. the end state of the pre-industrial simulation does not necessarily equal the
410 initial state of the industrial simulation at the grid cell scale). Two kinds of discontinuities may
therefore arise: 1) traces of model spinup, with GHMs reaching a water storage equilibrium
with delay at the beginning of the simulation, and 2) abrupt shifts in value distribution around
the transition point at the end of year 1860 (e.g. in the moving mean). Further, a combination
of the two is also possible. Thus, we ensured that the simulation outcomes were harmonious
415 with a two-step approach applied separately for each grid cell, month, and ensemble member.
Extended Data Fig. 2 shows an example of traces of model spinup, whereas Extended Data
Fig. 3 shows our general approach to ensure the fit between the simulation periods along with
an example of a cell in which the value distribution shifts.
First, we checked for any traces of spinups by detecting the most likely changepoint in the
420 mean and variance of the data with function cpt.meanvar in R package changepoint72,73,
separately for the pre-industrial and industrial periods. As the trace of spinup is related to
reaching an equilibrium in the hydrological state in a cell, we assumed that a true trace of
spinup is located between 10 and 30 years after the initiation of each simulation period. Outside
this period, we considered that detected changepoints were true (natural) changes in mean and
425 variance and did not imply a trace of spinup in the data. If a changepoint was detected between
10 and 30 years after initiation (as in Extended Data Fig. 2, for example), all data before the
changepoint were excluded from distribution shift correction; otherwise (as in Extended Data
Fig. 3, for example), data from the ten first years of the simulation periods were discarded.
Depending on the month and ensemble member, changepoints indicating traces of spinup were
430 detected in 1.4–8.7% (pre-industrial streamflow), 1.6–7.2% (industrial streamflow) 0.4–4.6%
(pre-industrial soil moisture), and 0.6–5.7% (industrial soil moisture) of all cells.
Second, we checked and corrected shifts in value distribution around the simulations’ transition
point in 1860, using an iterative technique combining linear extrapolation and quantile
mapping. Because the simulation periods have no temporal overlap, we extrapolated pre-
435 industrial data onto the industrial simulation period (i.e. extended the pre-industrial time series
to nominally cover years past 1860). This was done by fitting a linear regression model74 to
pre-industrial values excluding spinup (i.e. 1671–1860 in most cases, with flexibility of starting
from 1691 at latest), and extrapolating the detected linear trend to the industrial period (blue
lines in Extended Data Fig. 3a). Further, we computed the standard deviation of the pre-
440 industrial values to which the linear model was fit (defined here as σpreind) and added normally
distributed random noise (μ = 0, σ = σpreind) around the linear extrapolation line to create
extrapolated data points (blue circles in Extended Data Fig. 3a).
After extrapolation, we fitted non-parametric quantile mapping using robust empirical
quantiles implemented in R package qmap75,76 between the industrial simulation values and the
445 values that were extrapolated from the pre-industrial period to cover the first years of the
industrial simulation period. Four quantiles were used in fitting the quantile mapping. Here, we
treated the industrial simulation values as ‘observed’ data and the extrapolated pre-industrial
values as ‘modelled’ data, although both were simulated GHM outputs. We chose to fit the
quantile mapping between 30 years of industrial and extrapolated pre-industrial values, starting

14
 This is a non-peer reviewed preprint submitted to EarthArXiv

450 from the first year not excluded after spinup detection (pink shading in Extended Data Fig. 3a).
Hence, quantile mapping was mostly fitted using years 1871–1900 as the ‘observed’ values but
with flexibility of extending the fitting period to 1891–1920 at maximum. For example, the
fitting period for the cell shown in Extended Data Fig. 2 was 1874–1903, whereas for the cell
shown in Extended Data Fig. 3, the fitting period was 1871–1900. Although 30 years is a
455 relatively short period, and some spinups might end only after 1891, we chose to limit the
quantile mapping with these years to prevent the industrial values being substantially affected
by anthropogenic impacts, which becomes more likely during the 20th century.
Finally, pre-industrial values excluding spinup were corrected with the fitted quantile mapping
function (Extended Data Fig. 3b). However, as our correction procedure included extrapolation
460 and had a relatively small sample size (of 30), distinctive individual values or undetected traces
of spinup might have distorted the quantile mapping function. Hence, we checked whether
applying quantile mapping succeeded in improving the fit between the simulation periods. To
do this, we ran our analysis up until detecting local deviations (Fig. 1a–b) – first without
correction and then with correction – and consequently checked whether the number of local
465 deviations (Fig. 1b) decreased owing to the correction. Checking the local deviations was
performed for a period of 50 years, beginning from the first year not excluded after spinup
detection (hatched fill in Extended Data Fig. 3). This was to include data outside the period
that was used in fitting the quantile mapping. If the number of local deviations decreased, we
considered that our correction procedure had improved the fit between the simulation periods
470 and the corrected cell values were accepted, otherwise we discarded the corrected data and
retained the uncorrected data. Globally, depending on month and ensemble member, the
correction was accepted for approximately 6–26% of streamflow cells and 10–29% of soil
moisture cells.
Setting the local baseline range and identifying local deviations
475 In setting the local baseline range (Fig. 1a), identifying local deviations (Fig. 1b), and
computing the percentage of land area with local deviations (Fig. 1c), we followed the general
approach of Wang-Erlandsson et al.25 with some modifications. Like them, we set the 5th and
95th percentiles of pre-industrial streamflow and soil moisture to bound the local baseline range
(separately for each grid cell, month, and ensemble member), but we drew these bounds from
480 the considered grid cell’s values only, whereas Wang-Erlandsson et al.25 drew their local
baseline range from the values in the considered grid cell and its neighbourhood. We chose a
stricter definition of the local baseline range because including neighbourhood values in
determining the local baseline range can potentially set it unrealistically wide – especially in
the case of streamflow, in which neighbourhood values strongly depend on flow directions.
485 Spinups were excluded from setting the local baseline range, which means that for most cells,
dry and wet bounds (Fig. 1a) were drawn from 190 values covering years 1671–1860, though
allowing flexibility up until 1691 (e.g. for the grid cell shown in Extended Data Fig. 2, the
bounds were drawn from values in 1690–1860). We also evaluated the sensitivity of our results
to the definition of the local baseline range by repeating our main global analysis (Fig. 2) using
490 the 2.5th and 97.5th and the 1st and 99th percentiles as the dry and wet bounds, in addition to the
5th and 95th percentiles (Extended Data Fig. 1).

15
 This is a non-peer reviewed preprint submitted to EarthArXiv

Then, again for each grid cell, month, and ensemble member, we checked for all pre-industrial
months whether streamflow or soil moisture would fall within or outside the local baseline
range, to identify local deviations (Fig. 1b). In case a monthly value was lower than the dry
495 bound (5th percentile of pre-industrial values), this month was marked as a dry local deviation.
Contrarily, in cases of the monthly value surpassing the wet bound (95th percentile of pre-
industrial values), the month was marked as a wet local deviation. Local deviations were
evaluated for all months including spinups, and all deviations were determined in a binary
fashion, as ‘deviated’ or ’not deviated’, with no regards to deviation magnitude.
500 Computing the percentage of land area with local deviations and defining pre-
industrial variability
After identifying local deviations (Fig. 1b), we summed up the land areas of grid cells with dry
and wet local deviations, separately for each month and ensemble member. We excluded
Antarctica and other permanent land ice areas using HYDE 3.2.1 anthromes77 and divided the
505 land area with local deviations by total ice-free land area to yield the percentage of land area
with local deviations (Fig. 1c). We transformed monthly percentages into annual percentages
by taking annual means of the monthly percentages. This differs from Wang-Erlandsson et al.25
who marked an individual grid cell as deviating during a given year if a local deviation occurred
in any month of that year, and subsequently computed the global percentage of land area with
510 local deviations only annually. We chose to compute the percentage of land area with local
deviations at a monthly time step to distinguish between cases in which local deviations occur
in only one month or many months within a year – i.e. when freshwater change is temporally
less or more extensive. With our approach, local deviations that spread out over multiple
months of the year have a higher impact on the annual percentage of land area with local
515 deviations.
As approximately 5% of all pre-industrial values in each grid cell were marked as dry and wet
local deviations by the definition of the local baseline range (5th/95th percentiles), the expected
global percentage of land area with local deviations was also approximately 5%, for dry and
wet deviations separately (under conditions with little interannual variance, i.e. not during
520 spinups). Hence, summing the percentages of land area with dry and wet local deviations
together, the expected value for the global percentage of land area with local deviations was
approximately 10%. As spinups were prevalent especially in some GHMs (Extended Data Fig.
4–5), global percentages of land area with local deviations prior to 1691 were excluded from
further analysis. Therefore, we ended up with a time series of n = 170 for the annual pre-
525 industrial percentage of land area with local deviations.
The pre-industrial percentage of land area with local deviations (Fig. 1c) was further used to
define pre-industrial variability (Fig. 1d). In the main results, we took the ensemble median (n
= 15 or n = 23) of the percentage of land area with local deviations to define two main metrics:
the median of pre-industrial variability and the upper end of pre-industrial variability (Fig. 1d).
530 The median of pre-industrial variability was defined as the 50th percentile of the percentage of
land area with local deviations, while the upper end of pre-industrial variability was defined as
the 95th percentile of the percentage of land area with local deviations. Hence, the median
describes the typical percentage of land area with local deviations in the pre-industrial

16
 This is a non-peer reviewed preprint submitted to EarthArXiv

(Holocene-like; presumably natural) reference state, whereas the upper end describes
535 conditions that occur expectedly once in every 20 years.
To assess the industrial (1861–2005) state of the freshwater cycle, we repeated the
identification of local deviations (Fig. 1b), computed the percentage of land area with local
deviations (Fig. 1c), and compared that to pre-industrial variability (Fig. 1d). Again, we
included all years in identifying local deviations, but chose to draw years 1861–1890 as shaded
540 and dashed in the presented results, owing to spinups being prevalent in some GHMs (Extended
Data Fig. 4–5). We took the ensemble median of the global percentage of land area with local
deviations also for the industrial period and applied a moving (trailing) 10-year mean over this
ensemble median time series, with the latest 10-year mean (for year 2005, computed as a mean
of values from 1996–2005) being defined as the current status.
545 Limitations and uncertainties
Global hydrological models (GHMs) are known to be relatively sensitive to their internal
implementation of anthropogenic drivers and impacts51,59. ISIMIP 2b simulations attempt to
minimise uncertainties stemming from this by using state-of-the-art input data and consistent
scenario definitions58. Nevertheless, the different process descriptions of GHMs lead to a
550 considerable spread in our results (Extended Data Fig. 4–5), which means that our global – and
especially local – results are subject to noteworthy uncertainty. We also found relatively
prevalent discontinuities in the data, which we corrected for the purposes of this analysis
(Extended Data Fig. 3). However, we chose not to perform an explicit validation for our
hydrological data because ISIMIP 2b data has shown adequate50–53 (though variable78)
555 performance against observations, observed data to validate pre-industrial values are
unavailable, and the ISIMIP 2b hydrological data used here is forced by GCM outputs instead
of observed climate. We thus consider that using a data ensemble as large as possible captures
the uncertainty of global hydrological modelling adequately. Moreover, our estimate of the
‘current’ status ends in 2005, due to the ISIMIP 2b simulation protocol58. Since the end of our
560 study period, global trends in many of the key drivers of freshwater change, such as irrigated
area79, water use80, dam construction81 and forest loss82, have increased, and therefore, the
results presented here are likely conservative.
As our scenario setup of using dynamic histsoc socio-economic conditions against static
1860soc (Table S1) already implies a change in anthropogenic drivers of water cycle change,
565 it is not an unexpected result that aggregate water cycle changes were manifested in the early
20th century (Fig. 2, Fig. 4, Extended Data Fig. 8). Repeating the analysis using data that are
absent of anthropogenic forcing (ISIMIP nosoc scenarios58) would potentially aid in estimating
how large proportions of freshwater change are due to direct (e.g. water use) or indirect (e.g.
climate change) anthropogenic factors. However, model outputs for both picontrol-1860soc (or
570 picontrol-nosoc) and historical-nosoc scenarios were not available in ISIMIP 2b. Future
modelling efforts (e.g. ISIMIP simulation round 383) could provide an excellent opening to
further develop our analysis.
Despite the related uncertainties, our approach is as coherent as current knowledge and
modelling capacities allow for a global study. Moreover, the magnitude and rate of global
575 freshwater change in the industrial period and the presumably conservative estimation of its

17
 This is a non-peer reviewed preprint submitted to EarthArXiv

current status suggest that our conclusion of the freshwater cycle substantially transgressing its
pre-industrial variability is valid.

Data availability
All data used in this study are gathered from openly available sources, which are appropriately
580 cited in Methods. Aggregated output data produced by our analysis will be deposited in a public
database and released upon publication.

Code availability
The code used in producing the results shown in this study will be deposited in a public
database and released upon publication.

18
 This is a non-peer reviewed preprint submitted to EarthArXiv

585 References
1. Gleeson, T. et al. Illuminating water cycle modifications and Earth system resilience in
the Anthropocene. Water Resour. Res. 56, e2019WR024957 (2020).
2. Virkki, V. et al. Globally widespread and increasing violations of environmental flow
envelopes. Hydrol. Earth Syst. Sci. 26, 3315–3336 (2022).
590 3. Huntington, T. G. Evidence for intensification of the global water cycle: Review and
synthesis. J. Hydrol. 319, 83–95 (2006).
4. Levia, D. F. et al. Homogenization of the terrestrial water cycle. Nat. Geosci. 13, 656–
658 (2020).
5. Douville, H., et al. In Climate Change 2021: The Physical Science Basis. Contribution of
595 Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on
Climate Change. Chapter 8: Water Cycle Changes (Cambridge University Press,
Cambridge, United Kingdom, 2021).
6. Gudmundsson, L. et al. Globally observed trends in mean and extreme river flow
attributed to climate change. Science 371, 1159–1162 (2021).
600 7. Robertson, F. R. et al. Consistency of Estimated Global Water Cycle Variations over the
Satellite Era. J. Clim. 27, 6135–6154 (2014).
8. Zhang, Y. et al. Multi-decadal trends in global terrestrial evapotranspiration and its
components. Sci. Rep. 6, 19124 (2016).
9. Haddeland, I. et al. Global water resources affected by human interventions and climate
605 change. Proc. Natl. Acad. Sci. 111, 3251–3256 (2014).
10. Jägermeyr, J., Pastor, A., Biemans, H. & Gerten, D. Reconciling irrigated food
production with environmental flows for Sustainable Development Goals
implementation. Nat. Commun. 8, 15900 (2017).
11. Pastor, A. V. et al. Understanding the transgression of global and regional freshwater
610 planetary boundaries. Philos. Trans. R. Soc. Math. Phys. Eng. Sci. 380, 20210294
(2022).
12. Cooley, S. W., Ryan, J. C. & Smith, L. C. Human alteration of global surface water
storage variability. Nature 591, 78–81 (2021).
13. Davidson, N. C. How much wetland has the world lost? Long-term and recent trends in
615 global wetland area. Mar. Freshw. Res. 65, 934 (2014).
14. Fluet-Chouinard, E. et al. Extensive global wetland loss over the past three centuries.
Nature 614, 281–286 (2023).
15. de Graaf, I. E. M. et al. A global-scale two-layer transient groundwater model:
Development and application to groundwater depletion. Adv. Water Resour. 102, 53–67
620 (2017).
16. Konikow, L. F. Contribution of global groundwater depletion since 1900 to sea-level
rise. Geophys. Res. Lett. 38, (2011).

19
 This is a non-peer reviewed preprint submitted to EarthArXiv

17. Grill, G. et al. Mapping the world’s free-flowing rivers. Nature 569, 215–221 (2019).
18. Jaramillo, F. & Destouni, G. Local flow regulation and irrigation raise global human
625 water consumption and footprint. Science 350, 1248–1251 (2015).
19. Adler, R. F., Gu, G., Sapiano, M., Wang, J.-J. & Huffman, G. J. Global Precipitation:
Means, Variations and Trends During the Satellite Era (1979–2014). Surv. Geophys. 38,
679–699 (2017).
20. Ukkola, A. M. & Prentice, I. C. A worldwide analysis of trends in water-balance
630 evapotranspiration. Hydrol. Earth Syst. Sci. 17, 4177–4187 (2013).
21. Berdugo, M. et al. Global ecosystem thresholds driven by aridity. Science 367, 787–790
(2020).
22. Meir, P. et al. Threshold Responses to Soil Moisture Deficit by Trees and Soil in
Tropical Rain Forests: Insights from Field Experiments. BioScience 65, 882–892 (2015).
635 23. Richter, B., Baumgartner, J., Wigington, R. & Braun, D. How much water does a river
need? Freshw. Biol. 37, 231–249 (1997).
24. Wang-Erlandsson, L. et al. Remote land use impacts on river flows through atmospheric
teleconnections. Hydrol. Earth Syst. Sci. 22, 4311–4328 (2018).
25. Wang-Erlandsson, L. et al. A planetary boundary for green water. Nat. Rev. Earth
640 Environ. 3, 380–392 (2022).
26. Rockström, J. et al. A safe operating space for humanity. Nature 461, 472–475 (2009).
27. Steffen, W. et al. Planetary boundaries: Guiding human development on a changing
planet. Science 347, 1259855 (2015).
28. Gleeson, T. et al. The Water Planetary Boundary: Interrogation and Revision. One Earth
645 2, 223–234 (2020).
29. R Core Team. R: A language and environment for statistical computing. (R Foundation
for Statistical Computing, 2021).
30. Lehner, B. & Grill, G. Global river hydrography and network routing: baseline data and
new approaches to study the world’s large river systems. Hydrol. Process. 27, 2171–
650 2186 (2013).
31. Biasutti, M. Rainfall trends in the African Sahel: Characteristics, processes, and causes.
WIREs Clim. Change 10, e591 (2019).
32. Porkka, M. et al. Is wetter better? Exploring agriculturally-relevant rainfall
characteristics over four decades in the Sahel. Environ. Res. Lett. 16, 035002 (2021).
655 33. Siebert, S. et al. A global data set of the extent of irrigated land from 1900 to 2005.
Hydrol. Earth Syst. Sci. 19, 1521–1545 (2015).
34. Lehner, B. et al. High-resolution mapping of the world’s reservoirs and dams for
sustainable river-flow management. Front. Ecol. Environ. 9, 494–502 (2011).
35. Micklin, P. The Aral Sea disaster. Annu Rev Earth Planet Sci 35, 47–7272 (2007).

20
 This is a non-peer reviewed preprint submitted to EarthArXiv

660 36. Hammond, W. M. et al. Global field observations of tree die-off reveal hotter-drought
fingerprint for Earth’s forests. Nat. Commun. 13, 1761 (2022).
37. Cottrell, R. S. et al. Food production shocks across land and sea. Nat. Sustain. 2, 130–
137 (2019).
38. Schöngart, J. et al. The shadow of the Balbina dam: A synthesis of over 35 years of
665 downstream impacts on floodplain forests in Central Amazonia. Aquat. Conserv. Mar.
Freshw. Ecosyst. 31, 1117–1135 (2021).
39. Deemer, B. R. et al. Greenhouse Gas Emissions from Reservoir Water Surfaces: A New
Global Synthesis. BioScience 66, 949–964 (2016).
40. Lawrence, D. & Vandecar, K. Effects of tropical deforestation on climate and
670 agriculture. Nat. Clim. Change 5, 27–36 (2015).
41. Liu, J., Wang, B., Cane, M. A., Yim, S.-Y. & Lee, J.-Y. Divergent global precipitation
changes induced by natural versus anthropogenic forcing. Nature 493, 656–659 (2013).
42. Puma, M. J. & Cook, B. I. Effects of irrigation on global climate during the 20th century.
J. Geophys. Res. Atmospheres 115, (2010).
675 43. Sillmann, J. et al. Extreme wet and dry conditions affected differently by greenhouse
gases and aerosols. Npj Clim. Atmospheric Sci. 2, 1–7 (2019).
44. Sterling, S. M., Ducharne, A. & Polcher, J. The impact of global land-cover change on
the terrestrial water cycle. Nat. Clim. Change 3, 385–390 (2013).
45. Heistermann, M. HESS Opinions: A planetary boundary on freshwater use is misleading.
680 Hydrol. Earth Syst. Sci. 21, 3455–3461 (2017).
46. Bunsen, J., Berger, M. & Finkbeiner, M. Planetary boundaries for water – A review.
Ecol. Indic. 121, 107022 (2021).
47. Veldkamp, T. I. E. et al. Water scarcity hotspots travel downstream due to human
interventions in the 20th and 21st century. Nat. Commun. 8, 15697 (2017).
685 48. Rockström, J. et al. Planetary Boundaries: Exploring the Safe Operating Space for
Humanity. Ecol. Soc. 14, (2009).
49. Gerten, D. et al. Towards a revised planetary boundary for consumptive freshwater use:
role of environmental flow requirements. Curr. Opin. Environ. Sustain. 5, 551–558
(2013).
690 50. Huang, S. et al. Evaluation of an ensemble of regional hydrological models in 12 large-
scale river basins worldwide. Clim. Change 141, 381–397 (2017).
51. Veldkamp, T. I. E. et al. Human impact parameterizations in global hydrological models
improve estimates of monthly discharges and hydrological extremes: a multi-model
validation study. Environ. Res. Lett. 13, 055008 (2018).
695 52. Zaherpour, J. et al. Worldwide evaluation of mean and extreme runoff from six global-
scale hydrological models that account for human impacts. Environ. Res. Lett. 13,
065015 (2018).

21
 This is a non-peer reviewed preprint submitted to EarthArXiv

53. Zaherpour, J. et al. Exploring the value of machine learning for weighted multi-model
combination of an ensemble of global hydrological models. Environ. Model. Softw. 114,
700 112–128 (2019).
54. Zemp, D. C. et al. Self-amplified Amazon forest loss due to vegetation-atmosphere
feedbacks. Nat. Commun. 8, 14681 (2017).
55. Chrysafi, A. et al. Quantifying Earth system interactions for sustainable food production
via expert elicitation. Nat. Sustain. 5, 830–842 (2022).
705 56. Stevenson, S. et al. Twenty-first century hydroclimate: A continually changing baseline,
with more frequent extremes. Proc. Natl. Acad. Sci. 119, e2108124119 (2022).
57. Boulton, C. A., Lenton, T. M. & Boers, N. Pronounced loss of Amazon rainforest
resilience since the early 2000s. Nat. Clim. Change 12, 271–278 (2022).
58. Frieler, K. et al. Assessing the impacts of 1.5 °C global warming – simulation protocol of
710 the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP2b). Geosci. Model
Dev. 10, 4321–4345 (2017).
59. Telteu, C.-E. et al. Understanding each other’s models: an introduction and a standard
representation of 16 global water models to support intercomparison, improvement, and
communication. Geosci. Model Dev. 14, 3843–3878 (2021).
715 60. Taylor, K. E., Stouffer, R. J. & Meehl, G. A. An Overview of CMIP5 and the
Experiment Design. Bull. Am. Meteorol. Soc. 93, 485–498 (2012).
61. Lawrence, D. M. et al. The Community Land Model Version 5: Description of New
Features, Benchmarking, and Impact of Forcing Uncertainty. J. Adv. Model. Earth Syst.
11, 4245–4287 (2019).
720 62. Schaphoff, S. et al. LPJmL4 – a dynamic global vegetation model with managed land –
Part 1: Model description. Geosci. Model Dev. 11, 1343–1375 (2018).
63. Stacke, T. & Hagemann, S. Development and evaluation of a global dynamical wetlands
extent scheme. Hydrol. Earth Syst. Sci. 16, 2915–2933 (2012).
64. Sutanudjaja, E. H. et al. PCR-GLOBWB 2: a 5 arcmin global hydrological and water
725 resources model. Geosci. Model Dev. 11, 2429–2453 (2018).
65. Hanasaki, N. et al. An integrated model for the assessment of global water resources –
Part 1: Model description and input meteorological forcing. Hydrol. Earth Syst. Sci. 12,
1007–1025 (2008).
66. Takata, K., Emori, S. & Watanabe, T. Development of the minimal advanced treatments
730 of surface interaction and runoff. Glob. Planet. Change 38, 209–222 (2003).
67. Müller Schmied, H. et al. Variations of global and continental water balance components
as impacted by climate forcing uncertainty and human water use. Hydrol. Earth Syst. Sci.
20, 2877–2898 (2016).
68. Dunne, J. P. et al. GFDL’s ESM2 Global Coupled Climate–Carbon Earth System
735 Models. Part I: Physical Formulation and Baseline Simulation Characteristics. J. Clim.
25, 6646–6665 (2012).

22
 This is a non-peer reviewed preprint submitted to EarthArXiv

69. Collins, W. J. et al. Development and evaluation of an Earth-System model – HadGEM2.

Geosci. Model Dev. 4, 1051–1075 (2011).
70. Dufresne, J.-L. et al. Climate change projections using the IPSL-CM5 Earth System
740 Model: from CMIP3 to CMIP5. Clim. Dyn. 40, 2123–2165 (2013).
71. Watanabe, M. et al. Improved Climate Simulation by MIROC5: Mean States,
Variability, and Climate Sensitivity. J. Clim. 23, 6312–6335 (2010).
72. Chen, J. & Gupta, A. K. Parametric statistical change point analysis: With applications
to genetics, medicine, and finance. (Birkhauser, Boston, USA, 2014).
745 73. Killick, R. & Eckley, I. A. changepoint: An R Package for Changepoint Analysis. J. Stat.
Softw. 58, 1–19 (2014).
74. Chambers, J., Hastie, T. & Pregibon, D. Statistical Models in S. in Compstat (eds.
Momirović, K. & Mildner, V.) 317–321 (Physica-Verlag, Heidelberg, Germany, 1990).
75. Boé, J., Terray, L., Habets, F. & Martin, E. Statistical and dynamical downscaling of the
750 Seine basin climate for hydro-meteorological studies. Int. J. Climatol. 27, 1643–1655
(2007).
76. Gudmundsson, L. et al. Comparing Large-Scale Hydrological Model Simulations to
Observed Runoff Percentiles in Europe. J. Hydrometeorol. 13, 604–620 (2012).
77. Klein Goldewijk, K., Beusen, A., Doelman, J. & Stehfest, E. Anthropogenic land use
755 estimates for the Holocene – HYDE 3.2. Earth Syst. Sci. Data 9, 927–953 (2017).
78. Gädeke, A. et al. Performance evaluation of global hydrological models in six large Pan-
Arctic watersheds. Clim. Change 163, 1329–1351 (2020).
79. Food and Agriculture Organization of the United Nations (FAO). Food and agriculture
data statistics (FAOSTAT) (FAO, 2022).
760 80. Wada, Y. & Bierkens, M. F. P. Sustainability of global water use: past reconstruction and
future projections. Environ. Res. Lett. 9, 104003 (2014).
81. Zarfl, C., Lumsdon, A. E., Berlekamp, J., Tydecks, L. & Tockner, K. A global boom in
hydropower dam construction. Aquat. Sci. 77, 161–170 (2015).
82. Keenan, R. J. et al. Dynamics of global forest area: Results from the FAO Global Forest
765 Resources Assessment 2015. For. Ecol. Manag. 352, 9–20 (2015).
83. Frieler, K. et al. Scenario set-up and forcing data for impact model evaluation and impact
attribution within the third round of the Inter-Sectoral Model Intercomparison Project
(ISIMIP3a). Preprint at doi:10.5194/egusphere-2023-281.

23
 This is a non-peer reviewed preprint submitted to EarthArXiv

Acknowledgements
770 We would like to acknowledge Will Steffen – who sadly passed away during the writing of
this paper – and Katherine Richardson for the discussions related to planetary boundaries, and
the ISIMIP team and all participating modelling teams for making the outputs available.

Funding
European Research Council under the European Union’s Horizon 2020 research and
775 innovation programme (grant no. 819202) (MP, MK)
Erling-Persson Family Foundation (MP)
Aalto University School of Engineering Doctoral Programme (VV)
European Research Council through the “Earth Resilience in the Anthropocene” project
(grant no. ERC-2016-ADG 743080) (LWE, IF, AT, JR)
780 IKEA Foundation (LWE)
Formas - A Swedish Research Council for Sustainable Development (grant no. 2022-02089)
(LWE)
Academy of Finland-funded project WATVUL (grant no. 317320) (MK)
The Netherlands Organization for Scientific Research (NWO), project number
785 016.Veni.181.015 (RvdE)

Competing interests
Authors declare that they have no competing interests.

Additional information
Supplementary material includes Supplementary text and Table S1. Extended Data Figures
790 1–8 are attached to this document.

24
 This is a non-peer reviewed preprint submitted to EarthArXiv

Extended Data Fig. 1. Percentage of global ice-free land area with local streamflow (a, c, e) and soil moisture
(b, d, f) deviations, using variable bounds of the local baseline range. Panels show the percentage of land area
with local deviations (Fig. 1c) when the local baseline range (Fig. 1a) is defined as the 5th–95th (a–b), 2.5th –97.5th
(c–d) or 1st–99th (e–f) percentile range. Panels a–b correspond to Fig. 2 and Extended Data Fig. 6a, d. Shown is
795 the annual percentage, which is computed as an average of monthly percentages. The ensemble median (solid
black line) and interquartile range (grey shading) are computed from n = 23 (streamflow) and n = 15 (soil moisture)
ensemble members. The horizontal dashed lines drawn in each panel denote the median and the upper end of pre-
industrial variability (Fig. 1d), and the current (mean of 1996–2005) percentage of land area with local deviations
is annotated at the end of the red 10-year moving (trailing) mean line. Values prior to 1691 are excluded and the
800 ensemble median line for 1861–1890 is shaded and dashed due to traces of model spinups being common during
these years (Extended Data Fig. 2–3).

25
 This is a non-peer reviewed preprint submitted to EarthArXiv

Extended Data Fig. 2. An example case of detecting the most likely changepoints in mean and variance.
Values shown in the plot describe soil moisture in one grid cell, for one month, and for one ensemble member.
As the changepoints are detected between 10 and 30 years from the initiation of each simulation period, they are
805 considered as traces of model spinup (Methods). Hence, values shaded in grey are excluded from distribution shift
correction (Extended Data Fig. 3) and from setting the local baseline range (Fig. 1a).

26
 This is a non-peer reviewed preprint submitted to EarthArXiv

Extended Data Fig. 3. Correcting shifts in cell-wise value distribution around the year 1860 transition point
between pre-industrial and industrial ISIMIP 2b simulations. Values shown in the plot describe soil moisture
in one grid cell, for one month, and for one ensemble member. Black circles show data prior to correction (a) and
810 after the correction (b). Linear trend (blue line) is estimated and extrapolated excluding the ten first pre-industrial
values (i.e. years 1671–1860). For the period highlighted with pink shading, quantile mapping is fitted between
industrial simulation values and values following the linear extrapolation line with added random normal noise
(blue circles). Once quantile mapping is applied for pre-industrial data points, resulting in panel b, the number of
local deviations (purple filled circles) outside the local baseline range (red lines) is compared pre- and post-
815 correction for a period of 50 years (striped shading). Should the number of local deviations decrease, as is the case
here, the corrected values (panel b) are accepted, otherwise the uncorrected values (panel a) would be retained.
See Methods for details.

27
 This is a non-peer reviewed preprint submitted to EarthArXiv

Extended Data Fig. 4. Percentage of global ice-free land area with local streamflow deviations, separately
for global hydrological models (GHMs) included in this study. Panels describe the percentage of land area
820 with local deviations (Fig. 1c) for H08 (a), LPJmL (b), MATSIRO (c), MPI-HM (d), PCR-GLOBWB (e),
and WaterGAP2 (f). Shown is the annual percentage, which is computed as an average of monthly percentages.
The horizontal dashed lines drawn in each panel denote the median and the upper end of pre-industrial variability
(Fig. 1d), and the current (mean of 1996–2005) percentage of land area with local deviations is annotated at the
end of the red 10-year moving (trailing) mean line. The ensemble median and interquartile range (grey shading)
825 are computed from n = 3–4 members (number of GCMs), as data forced with HadGEM2 were not available MPI-
HM. The ensemble median line for 1861–1890 is shaded and dashed due to traces of model spinups being common
during these years (especially a, e).

28
 This is a non-peer reviewed preprint submitted to EarthArXiv

Extended Data Fig. 5. Percentage of global ice-free land area with local soil moisture deviations, separately
for global hydrological models (GHMs) included in this study. Panels describe the percentage of land area
830 with local deviations (Fig. 1c) for CLM50 (a), LPJmL (b), MPI-HM (c), and PCR-GLOBWB (d). Shown is
the annual percentage, which is computed as an average of monthly percentages. The horizontal dashed lines
drawn in each panel denote the median and the upper end of pre-industrial variability (Fig. 1d), and the current
(mean of 1996–2005) percentage of land area with local deviations is annotated at the end of the red 10-year
moving (trailing) mean line. The ensemble median and interquartile range (grey shading) are computed from n =
835 3–4 members (number of GCMs), as data forced with HadGEM2 were not available for MPI-HM. The ensemble
median line for 1861–1890 is shaded and dashed due to traces of model spinups being common during these years
(especially a, d).

29
 This is a non-peer reviewed preprint submitted to EarthArXiv

Extended Data Fig. 6. Percentage of global ice-free land area with local streamflow (a–c) and soil moisture
(d–e) deviations, dry and wet local deviations summed (a, d), and for dry (b, e) and wet (c, f) local deviations
840 separately. Shown is the annual percentage, which is computed as an average of monthly percentages. The
annotated years mark the 10-year moving (trailing) mean transgressing the upper end of pre-industrial variability
(95th percentile of the pre-industrial percentage of land area with local deviations; Fig. 1d). The ensemble median
and interquartile range are computed from n = 23 (streamflow) and n = 15 (soil moisture) ensemble members.
Values prior to 1691 are excluded and the ensemble median line for 1861–1890 is shaded and dashed due to traces
845 of model spinups being common during these years (Extended Data Fig. 2–3).

30
 This is a non-peer reviewed preprint submitted to EarthArXiv

Extended Data Fig. 7. Statistically significant changes in dry and wet local deviation frequency for
streamflow (a) and for soil moisture (b). Changes in the frequency of local deviations are computed by
comparing the ensemble median frequency of local deviations during 1976–2005 against 1691–1860, and
significance of change is tested at a 95% confidence level (p = 0.05) with R package stats function prop.test29.
850 The changes are classified according to the direction of change (decreasing or increasing frequency of deviations)
but with no regards to magnitude. Percentage shares of ice-free land area covered by each class are represented in
the bivariate legend.

31
 This is a non-peer reviewed preprint submitted to EarthArXiv

Extended Data Fig. 8. Statistically significant increases in dry and wet local deviation frequency for
streamflow (a), for soil moisture (b), and combined for streamflow and soil moisture (c). Changes in the
855 frequency of local deviations are computed by comparing the ensemble median frequency of local deviations
during 1931–1960 against 1691–1860, and significance of change is tested at a 95% confidence level (p = 0.05)
with R package stats function prop.test29. Colours denoted with + indicate statistically significant increases with
magnitude less than 5 pp (minor), whereas colours denoted with ++ indicate statistically significant increases with
magnitude greater than 5 pp (major). Colours denoted with +(+) pool together any statistically significant increase
860 (minor or major).

32
 This is a non-peer reviewed preprint submitted to EarthArXiv

Supplementary material for

Global water cycle shifts far beyond pre-industrial

865 conditions – planetary boundary for freshwater change
transgressed

Miina Porkka†, Vili Virkki†, Lan Wang-Erlandsson, Dieter Gerten, Tom Gleeson, Chinchu
Mohan, Ingo Fetzer, Fernando Jaramillo, Arie Staal, Sofie te Wierik, Arne Tobian, Ruud van
870 der Ent, Petra Döll, Martina Flörke, Simon N. Gosling, Naota Hanasaki, Yusuke Satoh,
Hannes Müller Schmied, Niko Wanders, Johan Rockström, Matti Kummu

†These authors contributed equally to this work

Correspondence to: [email protected], [email protected]
875
This Supplementary material includes:
Supplementary text
Table S1
References

33
 This is a non-peer reviewed preprint submitted to EarthArXiv

880 Supplementary text

Freshwater supports terrestrial and aquatic ecosystems, mediates flows of material and energy,
and regulates the Earth’s climate. It is therefore tightly coupled with the core planetary
boundaries (PBs) of climate change and biosphere integrity, as well as the land-system change
and biogeochemical flows boundaries1,2. As the novel entities and biogeochemical flows PBs
885 can be considered to (implicitly) cover quality aspects of freshwater, the freshwater PB focuses
on the quantity of water. While justifiably included in the planetary boundaries framework, the
current freshwater planetary boundary definition has been criticised for not reflecting all
critical Earth system functions of water, not capturing the impact of all major anthropogenic
drivers and failing to recognise the interplay between local and global drivers and impacts3–6.
890 Initially, the freshwater use PB was defined by annual global blue water (surface and
groundwater) consumption as a proxy control variable that reflects precipitation partitioning
into blue and green water (terrestrial evaporation and soil moisture) flows and stocks, and
integrates the anthropogenic drivers impacting them7. The original assessment7 estimated
human modifications of freshwater to be within safe boundaries, but later estimates with more
895 nuanced quantification of water availabilities8 and water consumption – by explicitly including
also consumptive use of green water5 – suggested that the PB is either being rapidly approached
or has already been transgressed. Nevertheless, an update of the PB framework9 still concluded
that the globally aggregated status of the freshwater use PB is safe. However, transgressions
were recognised at the basin scale, where safe boundaries were set based on minimum
900 acceptable levels of river flows to maintain an adequate state of aquatic ecosystems9. Recently
– to respond to the ambiguity and criticisms around the freshwater PB – studies have proposed
sub-dividing the freshwater PB into blue and green water components and adopting a bottom-
up, area-based approach3,10,11. The provisional analysis of Wang-Erlandsson et al.11 proposed
that the freshwater PB should describe change rather than use, and found a likely transgression
905 of the green water component.
We follow Wang-Erlandsson et al.11 in proposing to replace the original freshwater use PB
with two control variables for freshwater change, measuring the global area with streamflow
(blue water) or root-zone soil moisture (hereinafter referred to as soil moisture; green water)
deviations from the range of local pre-industrial variability. Our approach is based on Wang-
910 Erlandsson et al.11 who suggested complementing the freshwater PB with a provisional green
water component, while we unify the blue and green water components under one assessment.
Widespread deviations from the relatively undisturbed pre-industrial state can be considered to
pose elevated risks to the Earth system functions of freshwater. For example, terrestrial and
freshwater ecosystems have adapted to specific quantities and timing of water flows12–14, and
915 wetness of landscapes regulates climate from micro- to regional and global scales, such that
changes in (timing of) wetness in one location can impact rainfall and consequently streamflow
and soil moisture both locally and remotely15.
By extending the freshwater PB to cover both blue and green water components, our approach
allows 1) representing the extent and impacts of freshwater change not only with respect to
920 aquatic ecosystems but also across terrestrial biomes and ecosystems; and 2) allowing
consistent comparison of the blue and green water components. Neither of these can be
represented by the previous freshwater PB (water use & environmental flows). Another key

34
 This is a non-peer reviewed preprint submitted to EarthArXiv

advantage of our approach is that calculating the blue and green water metrics using coherent
methodology allows us to assess their coevolution. This is very important for a comprehensive
925 analysis of freshwater change, as comparable metrics allow for inferring the nature of change
(e.g. how other elements of the water cycle may have changed or what might be driving the
change). Furthermore, the methodology applied here is transferable to additional variables in
future studies. The previous methods of assessing the freshwater PB lack these factors, which
is why we use a more general approach.
930 Our green water control variable aligns with Wang-Erlandsson et al.11 who suggested, after an
extensive literature review, that root-zone soil moisture is the most suitable control variable for
representing the Earth system functions of green water. For blue water, we select streamflow
as our control variable because streamflow alterations are either directly or indirectly connected
to most major anthropogenic modifications and Earth system functions of blue water. These
935 include e.g. water withdrawals, flow regulation, and groundwater depletion, as well as key
Earth system functions, such as aquatic habitat provision and storage and transport of carbon
and nutrients10. As streamflow is also impacted by anthropocentric climate change16, it is a
suitable control variable to represent the Earth system functions of blue water in aggregate. In
addition to an adequate representation of the human drivers and Earth system impacts of
940 freshwater change, the selected control variables are routinely simulated by global hydrological
models and can therefore be updated on an annual basis. The variables can, in principle, also
be observed directly, e.g. by in situ measurements or remote sensing17,18, though spatial and
temporal coverage of observational data is currently not sufficient for a comprehensive global
analysis such as the one presented here.
945 The planetary boundaries framework7,9 considers the Holocene geological epoch as the desired
reference conditions prior to extensive anthropogenic forcing. However, as major spatial shifts
of the water cycle have occurred since the Holocene19,20, strictly grid cell scale comparisons –
as those done here – are not meaningful using a Holocene baseline. Hence, we selected the pre-
industrial period as our comparison baseline, also because of the wide availability of consistent
950 and comparable ISIMIP 2b data21. By grounding our analysis in the pre-industrial period, we
were able to establish our reference conditions before substantial anthropogenic modifications
of the water cycle, including e.g. large dams22, large scale irrigation systems23, the rapid
increase of intensive land uses24, and anthropocentric climate change16. Therefore, we assumed
that the pre-industrial period represents Holocene-like conditions – and that widespread
955 deviations from that ‘pristine’ state are problematic.
Based on our analysis, the freshwater change PB has been transgressed since the early 20th
century, for both blue and green water components. Currently, 18.2% and 15.8% of global ice-
free land area experience local deviations that occurred only on approximately 9–10% of land
area during the pre-industrial time. Given that variability in this percentage during the pre-
960 industrial period is small (~ 3 pp), these changes are substantial and clearly indicate a shift in
the global water cycle. This means that extreme conditions – both wet and dry – are occurring
more often than before the onset of major anthropogenic drivers of change. The water cycle
has thus shifted to a state in which it has not been in during the development of contemporary
human societies. Therefore, humanity’s risks related to the Earth system functions of
965 freshwater may be elevated.

35
 This is a non-peer reviewed preprint submitted to EarthArXiv

Our new freshwater change PB definition is associated with two main kinds of uncertainties:
1) the ability of our control variables to represent Earth system change and 2) placement of the
boundary value of freshwater change and assessing what its transgression means for the Earth
system. Streamflow and soil moisture do not represent all freshwater changes – such as water
970 quality aspects – and omit explicit consideration of other freshwater flows and stocks, such as
groundwater and land ice. From the perspective of the PB framework, however, these choices
can be justified, as water quality is (implicitly) considered in the PBs of Biogeochemical flows
and Novel entities7,9, and because such an overarching framework requires expressing very
complex Earth system processes in relatively simple terms. Nonetheless, a more
975 comprehensive assessment of humanity’s overall impact on the freshwater cycle would require
careful consideration of the many complex relationships between drivers, hydrological
responses and their Earth system impacts that weren’t explicitly considered in this analysis.
Major uncertainties remain also regarding the planetary boundary value of freshwater change
– that is, which level of freshwater change would trigger state changes in the Earth system or
980 increase gradual risks at an unacceptable rate. Although impacts of freshwater change on the
Earth system functions of freshwater have been studied extensively at local to regional scales10,
regional to planetary scale impact assessments are still lacking. Furthermore, the PBs’ control
variables – including freshwater – interact strongly with each other1,2, which should be
considered when setting a PB for freshwater change. These factors complicate the boundary
985 setting, as ideally, PBs should be set based on control variable–response variable relationships
to minimise the risk of abrupt or irreversible state shifts. In the absence of such assessments,
we chose to set the boundary at the upper end of the pre-industrial, Holocene-like variability
range, as it represents conditions that were present before major anthropogenic alterations of
the freshwater cycle. It is possible, however, that the intrinsic resilience of the Earth system
990 will maintain a Holocene-like state even if the pre-industrial range of variability is exceeded –
as it has already been for a long time. However, as the extent of a ‘safe’ exceedance cannot be
robustly defined without quantifications of Earth system-wide responses of freshwater change,
we applied the precautionary principle of the PB framework7,25, and set the boundary at the
lower end of scientific uncertainty. We consider this to be at the upper end (95th percentile) of
995 pre-industrial variability range, which can be regarded as highly precautionary – although an
even stricter boundary placement could also be justified, due to signs of weakened Earth system
resilience11 and the complex interactions between freshwater and other PBs1,2.

36
 This is a non-peer reviewed preprint submitted to EarthArXiv

simulation climate socio-economic ISIMIP output

GHM GCMs
period scenario conditions variable

pre-industrial picontrol 1860soc rootmoist CLM50 GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5

pre-industrial picontrol 1860soc rootmoist LPJmL GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
pre-industrial picontrol 1860soc rootmoist MPI-HM GFDL-ESM2M, IPSL-CM5A-LR, MIROC5
pre-industrial picontrol 1860soc rootmoist PCR-GLOBWB GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
pre-industrial picontrol 1860soc dis H08 GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
pre-industrial picontrol 1860soc dis LPJmL GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
pre-industrial picontrol 1860soc dis MATSIRO GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
pre-industrial picontrol 1860soc dis MPI-HM GFDL-ESM2M, IPSL-CM5A-LR, MIROC5
pre-industrial picontrol 1860soc dis PCR-GLOBWB GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
pre-industrial picontrol 1860soc dis WaterGAP2 GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
historical historical histsoc rootmoist CLM50 GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
historical historical histsoc rootmoist LPJmL GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
historical historical histsoc rootmoist MPI-HM GFDL-ESM2M, IPSL-CM5A-LR, MIROC5
historical historical histsoc rootmoist PCR-GLOBWB GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
historical historical histsoc dis H08 GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
historical historical histsoc dis LPJmL GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
historical historical histsoc dis MATSIRO GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
historical historical histsoc dis MPI-HM GFDL-ESM2M, IPSL-CM5A-LR, MIROC5
historical historical histsoc dis PCR-GLOBWB GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5
historical historical histsoc dis WaterGAP2 GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC5

1000 Table S1. Data sets used in this study, adopting ISIMIP 2b nomenclature21. The pre-industrial simulation period
represents 200 years of pre-industrial conditions, nominally covering years 1661–1860, and the historical
simulation period covers years 1861–2005. The climate scenario picontrol uses pre-industrial climate and 286
ppm CO2 concentration, while the historical climate scenario uses historical climate and CO2 concentration.
Socio-economic conditions for the scenario 1860soc are fixed at the pre-industrial level, while the histsoc scenario
1005 uses varying historical land use and socio-economic conditions. The ISIMIP output variables rootmoist and dis
refer to root-zone soil moisture (soil moisture) and discharge (streamflow), respectively.

37
 This is a non-peer reviewed preprint submitted to EarthArXiv

References
1. Chrysafi, A. et al. Quantifying Earth system interactions for sustainable food production
1010 via expert elicitation. Nat. Sustain. 5, 830–842 (2022).
2. Lade, S. J. et al. Human impacts on planetary boundaries amplified by Earth system
interactions. Nat. Sustain. 3, 119–128 (2020).
3. Gleeson, T. et al. The Water Planetary Boundary: Interrogation and Revision. One Earth
2, 223–234 (2020).
1015 4. Heistermann, M. HESS Opinions: A planetary boundary on freshwater use is misleading.
Hydrol. Earth Syst. Sci. 21, 3455–3461 (2017).
5. Jaramillo, F. & Destouni, G. Local flow regulation and irrigation raise global human
water consumption and footprint. Science 350, 1248–1251 (2015).
6. Molden, D. Planetary boundaries: The devil is in the detail. Nat. Clim. Change 1, 116–
1020 117 (2009).
7. Rockström, J. et al. Planetary Boundaries: Exploring the Safe Operating Space for
Humanity. Ecol. Soc. 14, (2009).
8. Gerten, D. et al. Towards a revised planetary boundary for consumptive freshwater use:
role of environmental flow requirements. Curr. Opin. Environ. Sustain. 5, 551–558
1025 (2013).
9. Steffen, W. et al. Planetary boundaries: Guiding human development on a changing
planet. Science 347, 1259855 (2015).
10. Gleeson, T. et al. Illuminating water cycle modifications and Earth system resilience in
the Anthropocene. Water Resour. Res. 56, e2019WR024957 (2020).
1030 11. Wang-Erlandsson, L. et al. A planetary boundary for green water. Nat. Rev. Earth
Environ. 3, 380–392 (2022).
12. Berdugo, M. et al. Global ecosystem thresholds driven by aridity. Science 367, 787–790
(2020).
13. Meir, P. et al. Threshold Responses to Soil Moisture Deficit by Trees and Soil in
1035 Tropical Rain Forests: Insights from Field Experiments. BioScience 65, 882–892 (2015).
14. Richter, B., Baumgartner, J., Wigington, R. & Braun, D. How much water does a river
need? Freshw. Biol. 37, 231–249 (1997).
15. Wang-Erlandsson, L. et al. Remote land use impacts on river flows through atmospheric
teleconnections. Hydrol. Earth Syst. Sci. 22, 4311–4328 (2018).
1040 16. Gudmundsson, L. et al. Globally observed trends in mean and extreme river flow
attributed to climate change. Science 371, 1159–1162 (2021).
17. Birkinshaw, S. J. et al. Daily discharge estimation at ungauged river sites using remote
sensing. Hydrol. Process. 28, 1043–1054 (2014).
18. Peng, J., Loew, A., Merlin, O. & Verhoest, N. E. C. A review of spatial downscaling of
1045 satellite remotely sensed soil moisture. Rev. Geophys. 55, 341–366 (2017).

38
 This is a non-peer reviewed preprint submitted to EarthArXiv

19. Dallmeyer, A., Claussen, M., Lorenz, S. J. & Shanahan, T. The end of the African humid
period as seen by a transient comprehensive Earth system model simulation of the last
8000 years. Clim. Past 16, 117–140 (2020).
20. Dallmeyer, A. et al. Holocene vegetation transitions and their climatic drivers in MPI-
1050 ESM1.2. Clim. Past 17, 2481–2513 (2021).
21. Frieler, K. et al. Assessing the impacts of 1.5 °C global warming – simulation protocol of
the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP2b). Geosci. Model
Dev. 10, 4321–4345 (2017).
22. Lehner, B. et al. High-resolution mapping of the world’s reservoirs and dams for
1055 sustainable river-flow management. Front. Ecol. Environ. 9, 494–502 (2011).
23. Siebert, S. et al. A global data set of the extent of irrigated land from 1900 to 2005.
Hydrol. Earth Syst. Sci. 19, 1521–1545 (2015).
24. Ellis, E. C. et al. People have shaped most of terrestrial nature for at least 12,000 years.
Proc. Natl. Acad. Sci. 118, (2021).
1060 25. Kriebel, D. et al. The precautionary principle in environmental science. Environ. Health
Perspect. 109, 871–876 (2001).

39