We are losing soil moisture, why?

New data show a global soil moisture loss equal to several weeks of plant transpiration—something unexplained without accounting for ecosystem decline. It's time to move beyond the CO₂-only narrative.

Before starting this blog, I spent a few years reading what Professor Ugo Bardi has been writing and was profoundly impacted—especially by his humanism, which I am now also trying to adopt (with, at best, only partial success, as I often have to fight my tendency toward sarcasm—and sarcasm is, after all, inherently misanthropic—as well as a sign of weakness). In a recent post, Ugo observes that when a war begins, other disasters spontaneously follow, like an avalanche, before a full-fledged, devastating crisis unfolds. He laments that it is very difficult to stop this avalanche of extermination (especially by standing in its way), but adds that this does not mean we should not try.

A similar situation occurs with the “CO₂-only” narrative. Here, going against the current feels like hitting a stone wall. Should we abandon our efforts, or should we continue trying to change it? Today, I will discuss another study published in Science that reports a global loss of soil moisture— one that is not, and cannot be, explained by global warming based on the data presented. Nevertheless, to align this finding within the CO₂-only narrative, the accompanying Perspective begins—can you guess how? As accustomed as I am to this kind of thing, I was still genuinely surprised. I’ll return to it at the end of the post.

What we learn from the abstract

The abstract of the study in Science published 28 March 2025, “Abrupt sea level rise and Earth’s gradual pole shift reveal permanent hydrological regime changes in the 21st century” by Seo et al., reads as follows (with my emphasis):

Rising atmospheric and ocean temperatures have caused substantial changes in terrestrial water circulation and land surface water fluxes, such as precipitation and evapotranspiration, potentially leading to abrupt shifts in terrestrial water storage. The European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5) soil moisture (SM) product reveals a sharp depletion during the early 21st century. During the period 2000 to 2002, soil moisture declined by approximately 1614 gigatonnes, much larger than Greenland’s ice loss of about 900 gigatonnes (2002–2006). From 2003 to 2016, SM depletion continued, with an additional 1009-gigatonne loss. This depletion is supported by two independent observations of global mean sea level rise (~4.4 millimeters) and Earth’s pole shift (~45 centimeters). Precipitation deficits and stable evapotranspiration likely caused this decline, and SM has not recovered as of 2021, with future recovery unlikely under present climate conditions.

Science is paywalled (modern science is a business), so most readers never get beyond the abstract. What impression do they get from the six sentences above? The opening sets the stage: the paper should be about shifts in terrestrial water storage due to rising temperatures. The middle sentences report a significant soil moisture decline. The final sentence warns that no compensation is to be awaited under continued warming. So the reader, as always (please see “Why it is important to read scientific papers beyond their abstracts”), walks away with the impression that global warming has brought us yet another disaster—this time, a decline in global soil moisture. A scientifically inclined reader might also be struck by the precision of the reported losses: 1,614 and 1,009 gigatonnes. Not, say, 1,615 and 1,003, but exactly 1,614 and 1,009—presented without any mention of uncertainty. Climate science published in Science must be very precise!

As I will now argue, this is a grossly inaccurate—if not entirely distorted—ideologization of the paper’s important results. In short, for two main reasons. First, the observed soil moisture decline is attributed to decreasing terrestrial precipitation, while global climate models actually predict increasing precipitation over land as the planet warms—that is, the opposite of what the paper finds. Second, the atmospheric dataset used (ERA5) does not account for changes in vegetation cover that occurred during the study period.

Sea level rise, Earth’s rotation and the ERA5 dataset

Some readers may have noticed a mismatch between the focus of the paper’s title—which highlights sea level rise and changes in Earth’s rotation as the main data sources—and the abstract, which instead emphasizes a soil moisture product from a global climate reanalysis (ERA5), with the other data merely 'supporting' conclusions drawn from that dataset.

What is a reanalysis like ERA5? A reanalysis is a dataset created by assimilating independent observations of various climate variables into a climate model, producing a mutually consistent set of data. To clarify: suppose we independently measure evapotranspiration (difficult but theoretically possible), precipitation, runoff, and changes in soil moisture. Since each measurement comes with its own uncertainties, combining them formally might result in violations of basic physical principles, such as mass conservation. A reanalysis corrects for this by adjusting the observed values—within their uncertainty ranges—to ensure consistency with conservation laws.

After this adjustment, the reanalysis values for precipitation, runoff, soil moisture, etc., may no longer closely match their original independent observations at all times and locations, but together they will always satisfy the conservation laws. There can be multiple valid ways to perform these optimizations, so different reanalyses and observational datasets may show different trends and spatial patterns for the same variables.

The issue is especially acute for evapotranspiration, which cannot be easily measured directly. (Just consider for a moment how you would measure evapotranspiration in your nearby forest—given that it represents only a small fraction of the large moisture fluxes being transported by wind.)

While the abstract of Seo et al. 2025 emphasizes ERA5, it does not mention that the authors also considered many other similar datasets—several of which did not align with the sea level rise and Earth’s rotation data. To give an idea of the discrepancies between these datasets, below is Fig. S9 from the Supplementary Material of Seo et al. 2025.

(Given such large differences, reporting soil moisture loss with four-digit precision and no uncertainty is hard to understand. It contrasts with the estimate of associated sea level rise, which does include an uncertainty—though that didn’t make it into the abstract. It feels like the paper reflects input from researchers with differing scientific standards or approaches.)

In (a), we see sea level dynamics observed via mean altimetry compared to estimates made using soil moisture data from two reanalyses (ERA5-Land and MERRA2) and three land surface models (LSMs). In (b), the models are averaged. One model in (a) fits the sea level observations better than ERA5. However, as shown in their next figure, Fig. S10, which you can check yourself (since supplementary information is open access), that model poorly matches the Earth’s rotation data, while ERA5 agrees reasonably well with both. This is why Seo et al. 2025 highlighted and used ERA5 in their study.

At the same time, they were explicit about the fact that

ERA5-Land uses static monthly climatological mean values for the leaf area index and surface albedo with fixed land cover.

In other words, ERA5-Land does not account for the land cover changes that occurred during the studied period. Seo et al. speculated that if it did, increased evapotranspiration from re-greening and irrigation could have resulted in an even higher estimate of soil moisture loss. (They omit to consider, however, that deforestation would reduce evapotranspiration, see Figure 4 of Yuan et al. 2019 in the last but one section of this post.) In the meantime, let's examine the ERA5 data (Fig. 4 of Seo et al. 2025), taking them at face value.

The changing moisture budget of the atmosphere and soil

In panel A, we can see that around the year 2000, precipitation (P) steeply declined over land and, for the next two decades, remained about 30-40 mm/year lower than it was on average in 1979-1999. The sum of evaporation and transpiration (ET) also experienced a decline, but a significantly smaller one. As a result, their difference, P - ET, decreased by approximately the same magnitude as P did (panel B).

If we now recall the terrestrial moisture budget, it is as follows (see further discussion in “Biotic Pump, Ecorestoration, and River Runoff”).

The atmosphere over land receives moisture through winds: F_i is incoming moisture, F_o is outgoing. Their difference is the net import, called atmospheric moisture convergence, C = F_i - F_o. Since the atmosphere stores very little moisture (W), the net moisture import must eventually leave as precipitation, P. If there were no evaporation or transpiration, then C = P. However, when evaporation and transpiration return some moisture to the atmosphere, the balance becomes C = P - ET. ET (shown by green arrows) is the sum of evaporation (E), transpiration from soil (T_S), and transpiration from groundwater (T_G). Infiltration, I, is also shown.

What happens to the net liquid moisture, P - ET, that falls on the ground? Part of it returns to the ocean as streamflow (runoff), R, which includes surface runoff (R₀), soil runoff (R_S), and groundwater runoff (R_G), so that R = R₀ + R_S + R_G. Another part replenishes soil moisture, S, at a rate dS/dt, and groundwater, G, at a rate dG/dt—the latter often assumed to be small compared to R and dS/dt. So,

P - ET ≈ R + dS/dt.

When P declined while ET remained nearly constant, it implied a decrease in atmospheric moisture convergence, C = P - ET. The biotic pump over land became less efficient, drawing less moisture inland from the ocean. Since runoff depends on soil moisture (wetter soil produces more runoff), reduced moisture import caused the soil to begin losing water (as soil moisture was relatively stable before 2000, with dS/dt ≈ 0). This is what we observe in Fig. 4C of Seo et al. 2025.

The process will continue until the soil dries to the point where its reduced runoff matches the reduced atmospheric moisture convergence. By 2021, global soil had lost about 25 mm of water. Is that much or little? Given that global terrestrial evapotranspiration is about 500 mm per year—roughly 70% of which is transpiration—the lost moisture could sustain more than three weeks of plant transpiration.

Soil moisture loss and global climate models

Figure S4 from the supplementary data of Seo et al. 2025 shows where the decline took place.

Let us compare this with the predictions of global climate models that focus on CO₂ accumulation and the associated rise in temperature. Below is Fig. 7a from the well-known paper by Held and Soden 2006, “Robust Responses of the Hydrological Cycle to Global Warming”. It shows how climate models predict changes in moisture convergence, C = P - ET, as the planet warms. The units are mm/day. The graph describes changes projected to occur between the first and last twenty years of the 21st century in a modeled warming scenario.

We can see that almost everywhere on land—excluding the western U.S. and Europe—there is an increase in atmospheric moisture convergence, especially pronounced in the tropics. This stands in sharp contrast to the soil moisture trends shown in Fig. S4 of Seo et al. 2025. In that figure, large regions of South America, Africa, and Eurasia—where climate models consistently predict increased moisture convergence—are instead experiencing soil moisture loss. Globally, land shows a decline in imported atmospheric moisture, while the climate models in the graph above suggest that a warming land should, on the contrary, receive more atmospheric moisture (and thus gain moisture overall).

Furthermore, Fig. 1 from Gu and Adler 2023, “Observed variability and trends in global precipitation during 1979–2020” (open access, you are welcome to check it yourself) shows that, according to the ensemble of climate models (CMIP6-histGHG, green curve), precipitation over land should have been higher in 2000–2020 than in the two preceding decades. However, according to ERA5 data (see Fig. 3A in Seo et al. 2025 above), the opposite is true.

It is interesting to see how Seo et al. 2025 address this apparent departure from the CO₂-only narrative:

Although a warming climate favors increased vapor in the atmosphere, leading to an increase in precipitation, the actual trend of precipitation varies regionally. Conversely, the increasing evaporative demand driven by a warming climate is more uniformly distributed across the globe, suggesting a more consistent and widespread trend toward drying as temperatures rise.

The two sentences above describe a result that would be at least partially consistent with the CO₂-only framework—and would likely be presented as a successful prediction: precipitation increases slightly, while evapotranspiration increases even more (following the red potential evapotranspiration curve, or 'the increasing evaporative demand,' in Fig. 3A of Seo et al. above). In this case, P - ET would decline in a way attributable to global warming (though still conflicting with Fig. 7 of Held and Soden 2006).

In contrast, what we are actually observing is precisely the opposite: ET declines (rather than rises), while P declines (rather than rises) even more. The authors acknowledge that—

the complex interplay between precipitation and evapotranspiration, influenced by a broad range of factors, renders the impact of a warming climate on SM highly unpredictable.

Soil moisture loss predictable from land cover change

Indeed, there has not been any viable prediction from global warming science concerning changes in soil moisture. On the other hand, researchers focused on land use and land cover change have long warned that land mismanagement should lead to a loss of soil moisture.

A prominent figure among such researchers and landscape regeneration practitioners, and founder of the New Water Paradigm, Michal Kravčík, in his 2000 book Water for the Third Millennium (published in Slovakia), extrapolated data from Slovakia to a global scale. He predicted that the drainage of continents due to land mismanagement should lead to a global sea level rise of 2.1 mm/year. This estimate closely matches the 1.95±0.29 mm/year reported by Seo et al. 2025 for the period between 2000 and 2002.

What do we mean by land mismanagement, aside from the direct destruction of vegetation cover? It refers to any practices that result in soil becoming less permeable to rain. In such cases, soil moisture can decline even if atmospheric moisture import remains unchanged. This could occur if surface runoff increases while infiltration, and thus soil moisture recharge, decreases. When ecosystems are degraded, the soil surface can harden, reducing its ability to soak up moisture.

Seo et al. 2025 discussed possible changes in infiltration in qualitative terms but did not address the role of land use. Their argument was that global warming intensifies individual rain events, which could lead to more surface runoff and less soil moisture storage. At the same time, they acknowledged an opposite effect—more intense rain over very dry soils may enhance infiltration (a weaker rain would simply evaporate). But what man does to soil was neglected in their discussion.

This was discussed, instead, in a study by Auerswald et al. 2024 “HESS Opinion: Floods and droughts – Land use, soil management, and landscape hydrology are more significant drivers than increasing temperatures”, which I highly recommend. Auerswald et al. (2024) published their work in a journal of the European Geophysical Union, where the reviewers' comments are openly available, offering a glimpse into the process behind peer-reviewed science. In this particular case, we see how any significant departure from the CO₂-only narrative is often met with resistance—one reviewer recommended 'toning down' the statements about the importance of land use.

In contrast, the significance of CO₂ is consistently overemphasized. Despite the fact that the findings of Seo et al. 2025 cannot be meaningfully linked to global warming science, the Perspective article “Permanent shifts in the global water cycle,” aimed at explaining the relevance of Seo et al. 2025 to Science readers, begins as far from the water cycle topic as one could possibly imagine:

Understanding the relationship between atmospheric carbon levels and global temperature dates back to 1895, when Swedish scientist Svante Arrhenius argued that variations in carbon dioxide concentrations could affect Earth’s heat budget.

Turning to the water cycle, it admits in passing that “anthropogenic influences such as farming, large dams, and irrigation systems” are yet to be considered in the next-generation models and ends with a call to improve our understanding of “the impacts of climate change on the global water cycle” (neglecting those other anthropogenic influences yet another time).

So, what is happening?

As you read this text, our species continues tearing up the green and black fibers of our life, destroying natural ecosystems and killing soil.

Fig. 4 from Potapov et al. 2017 “The last frontiers of wilderness: Tracking loss of intact forest landscapes from 2000 to 2013”

In 2000-2013 alone, the world lost about one million square kilometers of intact forest landscapes, i.e., the most hydrologically competent, functionable ecosystems. As the degradation continues worldwide at similar rates everywhere, one can expect a tipping point when the remaining natural ecosystems are unable to compensate for the accumulating disturbances.

Figure 4 from Yuan et al. 2019 “Increased atmospheric vapor pressure deficit reduces global vegetation growth” shows how the greening trend underwent a reversal around the year 1999 in many regions of the world, resulting in a slight globally declining trend.

Comparison of NDVI (Normalized Difference Vegetation Index) trends over the globally vegetated areas between two periods of 1982–1998 (A) and 1999–2015 (B), units 1/year.

As vegetation collapses, land faces an increasing risk of shifting to a dry regime, where the continental biotic pump breaks down and moisture transport from the ocean stalls—similar to what occurred in Australia, but this time it could happen globally.

Conclusions

The study by Seo et al. 2025 is a fascinating attempt to use independent constraints—such as sea level rise and changes in Earth’s rotation—to extract insights into the changing terrestrial water cycle. While they should have reported their (presumably substantial) uncertainties, the very idea of applying constraints from entirely different fields of knowledge is great.

Their results indicate a considerable decline in soil moisture on land in the 21st century—about 25 mm. This could sustain nearly three weeks of terrestrial evapotranspiration (which is around 500 mm/year on a global average). Given the declining precipitation and nearly stable evapotranspiration, these findings imply a reduction in ocean-to-land moisture transport, i.e., a global weakening of the biotic pump. This decline is not predicted by CO₂-focused global climate models but parallels the progressive degradation of natural ecosystems on land that Seo et al. did not consider.

I want to emphasize again: this is the first time in human history that we are amassing such a wealth of information about our planet. To exclude life from consideration and force every new finding into the Procrustean bed of the CO₂-only narrative is not only unproductive, but also unworthy of a thinking species—and dangerous as a strategy.

We must open our eyes to the importance of biospheric processes and their profound impact on every aspect of the global environment and climate change. The harm we have been doing to life is a major contributor to most of the disasters that now come our way, including the global decline in soil moisture.