Do all global climate models neglect land use impacts, because they rely on the same flawed assumption?

Anastassia Makarieva points out a fundamental flaw in all global climate models. She rings the alarm bell about land management decisions that are based on these flawed aspects of our models. And she poses a challenge to the scientific community on how to respond: according to the scientific method, overcoming our emotional bias that wants us to react defensive, hurt, neurotypically human. My blog is directed to climate activists who want to understand how climate modelling could go so wrong.

*Disclaimer – This blog reflects the current state of my knowledge. It does not reflect the opinion of any institution, and my understanding will change as I learn more. The purpose of this blog is to summarize information, stimulate thought, raise conversation, and serve as a basis for further studies.

The climate movement is currently struggling with a small but growing sub movement, the “small water cycle” advocates – often referred to as “eco-warriors” or “ecologists”, as compared to “real” climatologists from the physics departments. These advocates propose the small water cycle as a mechanism that interferes with the greenhouse effect in a complex manner. If very active, the small water cycle reduces or almost turns off the warming from greenhouse gases, whereas a de-activated small water cycle hyper-charges the greenhouse effect. Yet, global climate models are seemingly not simulating the full power of the small water cycle – this blog sheds light on a bombshell that identified why that is so: a flaw that all global climate models have in common, which can be traced back into earliest climate modelling history.  Readers are invited to review the original critique by Makarieva ([1]), which is difficult to understand without a solid understanding of the thermodynamics of moist gases. Blogger Alpha Lo offers a simplified, common-language explanation of the physics behind Makarieva’s critique ([2]). I would refer to these authors, and instead offer a non-mathematical and intuitive explanation in this blog.

I myself learned about the small water cycle, and had a spiritual experience of awe, when I travelled from the tropical island of Romblon to the neighbouring island, Sibuyan – the centre of the Philippine Visayas islands. The two islands are only 12 km apart, each covering a similar area of approx 200km². Romblon’s forests are degraded to bush and shrubs; the island suffered from a multi-year drought. At Sibuyan’s centre, a volcano mountain is covered by rainforest. In Sibuyan, every evening between 5 and 7pm we experienced a short thunderstorm, and then a cool night with dry air. From the boat, I observed how the volcanic forests birthed clouds above the forested volcano: during late afternoons, fog rose and formed thunderstorm clouds. Whereas deforested Romblon island only created hazy mist without clouds or rain. I observed this in 1997 as a young environmental engineering student, and it left me in awe: I saw evening for evening how life begets rainfall, and how life clears up night skies such that heat can radiate out into space and cool the land. This feeling of awe in me inspired me to shift my studies to Earth System Sciences in 1998. And only now, 25 years later, the mystery of how the biosphere regulates our climate is being recognized more widely.

The rational how the vegetation impacts the climate through modifying the small water cycle is relatively simple, has been known since the 1970s, and features in our meteorological weather models:

• With a vibrant small water cycle, vegetation pulls nutrients into its root system from the soil, by transpiring water into the air. Transpiration converts water into vapour, a process that absorbs lots of heat energy from the Earth’s surface – comparable to an evaporation unit of an air-conditioner, located where cooling is desired. As humid air rises into the sky and reaches the height of clouds, vapour condenses back into water and releases its energy as heat – like a condensing unit of nature’s air-conditioning system, typically installed on the outside. Condensation releases heat mostly inside clouds, toward their top as moist air is exposed to cooler temperatures. So much of this heat can radiate toward space.
• If the small water cycle collapsed, vegetation is removed from the Earth surface. Transpiration stops, and nature’s air conditioning unit no longer does its job. Soils are no longer vibrant and spongy, so rain gets lost as surface runoff. Solar energy is mostly absorbed by the ground, heats up the ground, and is then released as heat radiation from the ground up. This surface-level heat radiation gets amplified by low-atmosphere greenhouse gases.

By moving energy upward in the form of latent vapour energy, vegetation can turn down the greenhouse effect. Without vegetation, greenhouse gases start trapping heat bellow cloud level near the surface. With vegetation, solar energy is mainly released above clouds, and radiated back into space. These impacts of the small water cycle are undisputed in the climate field of research. However, disagreement remains how relevant these processes are in shifting an entire regional climate system, in combination with other biological mechanisms that are truly complex. For example, vegetation also releases cloud condensation nuclii that accelerate condensation, increase the overall cloud cover, increases the relative share of low clouds. This cloud cover directly reflects approx 50% of all solar energy as visible light back into space, without transforming the sun’s energy into sensible or latent heat. Another effect is that large forests lower air pressure because condensation removes gas in the atmosphere. This lowering of pressure, in very large forests, sucks in oceanic air and pulls in new vapour from the ocean, which can keep entire forest systems hydrated – Makarieva and Gorshkov call this the “biotic pump” ([3], [4], [5]). While climate modellers noticed her work (e.g. [6], [7]), they don’t believe in the relevance of these processes, because their models don’t reproduce them (e.g. [8], also personal communication, Potsdam Institute for Climate Adaptation Research). Modellers proclaimed that the Biotic pump is an interesting but irrelevant detail… all climate models cannot fail together, was their logic for refuting Makarieva’s theory.

After a few earlier rebuttal attempts, Makarieva responded with a bombshell in January 2023: all climate models carry the same inaccurate approximation that skews simulated interactions between the biosphere, the small water cycle, and the climate. Makarieva pointed at an approximation that shows how moist air, as vapour starts to condense and release its latent energy as heat, continues to rise and carry energy upward. Whereas global climate models utilize the “radiative-convective approximation”, going back to one of the first weather and climate models by Manabe and Wetherald ([9]). This radiative-convective approximation suppresses convective patterns that everyone can observe as cumulus clouds or as towering cumulonimbus thunderstorm systems. Instead, models combine two simplifications: the atmosphere is treated as homogeneous within each model grid cell – artificial “boxes” that are 100km long and cover 10,000km² of area. Secondly, models approximate the vertical temperature-pressure gradient in the atmosphere as being “in equilibrium”, such that condensing moist air would no longer lead to convective rise of air.

With these approximations, a climate model will initially miss-represent energy fluxes above land – satellite data show how the land warms faster than models predict. So modellers may impose a calibration factor that adjusts greenhouse gas reflections within the models and other uncertain energy transfer processes, and may have attributed the “excess warmth” to greenhouse gas emissions. And they impose the same ‘calibration factor’ to the entire world, suppressing model sensitivity for localized land degradation. Whereas, in reality, much of this “excess warmth” should be attributed to the localized disruption of the small water cycle from land degradation – it impacts many processes listed above that remain “highly uncertain” in our climate models, from albedo, energy transfer, cloud formation. Calibration then increases the amount of heat that is re-radiated from the surface – the model now reproduces global measurements, but is internally biased to suppress impacts from water cycle degradation, while slightly escalating the role of GHG warming over land.

As land degradation is happening globally and effect most of our Earth’s land surface, and in the same time-frame as GHG emissions, potential model biases are very hard to quantify. If localized “collapse of natural cooling” all across the Earth’s landmass accumulates into a significant driver of global warming, models cannot reflect that accurately. In a way, climate modellers unintentionally “appropriated” reduced vegetative cooling for their own field of concern: greenhouse gas reflections of heat. And, within their models, they inadvertently deny life’s agency for regulating the climate.

Everyone has observed how moist air creates turbulence when it starts condensing, thus successively releasing its latent vapour energy. We experience this when watching cumulus clouds. Condensing vapour heats up the internal air within the cloud, lifting it upwards – leading to more condensation as air rises, in particular toward the top of the cloud. We can observe how turbulence give clouds their roundish shapes. As the atmosphere cools during evening hours, condensation creates so much upward convection that towering cumulonimbus storm systems build up, carrying latent energy even higher, releasing their heat even further up into the sky toward space. Because cumulus clouds and cumulonimbus storm systems both release most heat at their top, this mechanism by-passes lower greenhouse gases and “turns the greenhouse effect down to a simmer”, as Walter Jehne explains.

The model error related to Manabe & Wetherald’s radiative-convective approximation has been known for decades. In 1982, Lindzen et al. proposed an improved alternative approximation method that would catch cumulus-type convection and water vapour energy transfer, at least at significantly higher spatial resolution with smaller grid cells. Authors call the height profile of temperature and pressure the ‘lapse rate’, and the imposed equilibrium a ‘fixed’ lapse rate (^[10]):  “The freedom of a variable lapse rate allows radiative perturbations to be accommodated locally near the tropopause, without being carried through a fixed lapse rate to the surface.”  Earlier in their paper, Lindzen had already stated: “As a result, the perturbation response of a cumulus model tends to concentrate at the cloud-top levels rather than the uniform response of a fixed-lapse rate model; this would then lead to a smaller greenhouse feedback.” ([11]). Lindzen and colleagues believe that their equations are “not much more complicated” than the Manabe Weatherald approximation. So Makarieva simply re-discovered what meteorologists have known for long – a flaw that cannot be corrected easily due to runtime considerations.

Lindzen and colleagues fail to acknowledge that models would require far higher resolution to do this well – instead of using 100kmx100km grid cells that we use today (Figure 1), modellers would need to move to much smaller gridding (maybe 5kmx5km). This would not be practical for climate models, as it would increase spatial model resolution 400-fold, and increase model runtime ~8,000-fold due to smaller time steps. If a climate model runs 1 week with their current resolution on a super computer, this would increase the runtime with finer grid to 150 years! (for current grid resolution in IPCC’s 6th Assessment Report, see Figure 1-19).

Gridded models and calibration with effective variables

In global climate models, cumulus or cumulonimbus clouds don’t exist. The atmosphere is cut into grid cells (curved rectangles) that models consider to be homogeneous, usually with a side length of 100km and covering an area of 10,000 square kilometres. All atmospheric variables are treated as identical within this area: temperature, humidity/moisture, rainfall, wind speed, or air pressure. If turbulence circulate air up and down within a grid cell, the model only measures the “effective” or net movement of air. For example, in reality, the air in half of the grid area may rise upward and in the other half air downward at the same flow. A gridded model would estimate the “effective air and vapour movement” as zero – hence no transport of vapour or energy.

Reality does not follow this grid pattern, and much happens below grid scale. For example, let’s imagine that moisture is lifted in a chess board pattern within the grid cell – lifted by 1000m in all black parcels, 50% of the grid area, and no lift on white parcels. The “average” moisture lift in a gridded model should be ½ * 1000m = 500m, and satellite data can measure that precisely. Yet, a model that cannot resolve subgrid circular fluxes may estimate only 400m airlift… leaving a discrepancy of 100m airlift between model outputs and observation data.

Modellers will note this discrepancy between model outputs and observations, and turn to calibration. Calibration is an external intervention by the modeller who “trains” the computer simulation such that it reproduces observation data. Modellers do that by introducing one or several “fudge factors” that “correct” variables in ways that computer simulations reproduce real-world observations as good as possible. All models use such calibration process.

Modellers try their best to choose calibration factor that leaves the model interpretable. However, every calibration factor imposes more assumptions to a model: new sources of errors such as undue smoothing, each inadvertently imposes certain causality, and fuged or effective variables at grid scale, which are hard to interpret and impossible to observe:

• Undue smoothing: Because a universal land use-related calibration factor is applied to the entire Earth, models are less sensitive to localized variations of land use change. Instead, the same averaged land use impacts are attributed equally to every grid cell – whether it is forest, desert, or ocean. Actual land use changes are still parameterized and impact some earth surface parameters, but no longer impact energy transfer in the cloud layer.
• Imposed causality: With calibration factors, modellers also add implicit assumptions about the causes of an observed phenomenon. In the case of the radiative-convective approximation, the calibration factor attributes observed warming to either vegetative cooling or the greenhouse gas reflectivity. If the modeller chooses the wrong attribution, then the model may no longer have physical meaning. It merely reflects a calibration choice by the modeller, which is likely biased by personal factors – interest, research expectations, peer opinion, and belief.
• Effective variables are difficult to interpret: Grid-based models operate with effective variables internally – observable variables that are modified by a fudge factor (or “corrective transformation”). It is impossible to interpret or parameterize effective variables with observation data. Yet, many modellers communicate model results without considering the role of effective variables ([12]).

Any experienced modeller should be fully aware of these basic problems of calibration and effective variables. The statistician George Box proclaimed in the 1980s that “all models are wrong, but some are useful”. When interpreting the results of any model, modellers must consider whether their interpretation is based on solid aspects of a model, or on uncertain aspects. And coming back to Makarieva’s point, climate modellers seem to have miss-calibrated land use in ways that limits their model’s usefulness for evaluating landuse, and especially for land management decisions.

Why did modellers not catch the inadequacies of their model when doing uncertainty analysis?

Modellers have developed rigorous methods to assess the accuracy of environmental models by quantifying uncertainty (e.g. [13]), and climate modelling agencies are at the forefront of model uncertainty management (compare Arnold et al., 2020). Model uncertainty analysis is extensively applied to climate models, but the flaw from the Manabe-Wetherald approximation was apparently missed completely. How could this happen?

A full (or “global”) model uncertainty analysis tests how every single assumption impacts every output variable, and estimates “error bars” for each simulated output variable– just how students learned to draw error bars when making physics experiments at school. Global uncertainty analysis for bulky climate models is particularly cumbersome: with thousands of variables, and weeks of model runtime even on a super computer, uncertainty analysis can occupy the most advanced cloud computing for decades. To avoid this runtime costs, climate modellers have found an alternative approach that is practical and also treats researcher groups democratically: researchers compare the results of many differently structured climate models by several institutions – scientists call that multi-model ensemble experiments ([14]). And here is exactly where Makarieva’s argument sets in: if models in an ensemble experiment are built on the same (false) approximation, then even the most advanced ensemble experiments cannot identify discrepancies. And because the radiative-convective approximation is apparently embedded deeply within all global climate models (neither Makarieva nor other scientists could confirm or refute this claim!), the ensemble modelling experiments will reproduce the same systematic error across all models, with errors in the same direction. Our current way to assessing uncertainty in climate models then fails us.

Assessing the relevance of Makarieva’s point will require further academic corroboration and research. It is well-known and well-acknowledged by the IPCC how climate models fair poorly in predicting cloud cover and other processes related to water’s shape-shifting between the states of ice, snow, liquid droplets, and vapor. Global climate models also fail to predict heat domes and associated storms, or wildfires. They cannot handle the ways how forests (or other large vegetated areas) modify wind systems. How many of these inconsistencies are related to the Manabe-Wetherald approximation? – Makarieva recommends constructing new models with entirely different structure.

The political implications of her critique cannot be understated: Whereas humanity focuses “climate mitigation action” almost entirely on greenhouse gas emission reductions, or carbon sequestration, far more importance should be given to the energetic regime shift that follows land degradation (e.g. when turning a forest into a parking lot). Furthermore, climate modellers should no longer offer recommendations on land management, as these would be based on the flawed conceptualization of vegetation’s role in regulating the climate. As Makarieva lays out, these recommendations are dangerous because they turn a flawed model assumption into a real-world policy — flawed recommendation with potentially deadly consequences for humanity.

How will the climate modelling community respond?

Makarieva challenges the academic consensus by pointing out one flaw in climate models that could likely lead to wrong modelling results and potentially disastrous model-based policy recommendation. But her point does not make climate models useless in their entirety – she just points at a simplification that no longer is appropriate as humanity needs to consider the interactions of landuse, climate, and greenhouse gases. In fact, global warming and the greenhouse effect remains equally relevant. Especially above the oceans, surface vegetation plays no role in climate regulation (maybe algal abundance does?). And humans remain the main drivers of this global warming.

With a more complex understanding of the climate’s regulation by the biosphere, humanity also has a larger range of actions to address global warming. Actions now include land use change. Yet, according to Makarieva’s hypothesis, climate modellers have unintentionally “appropriated” the power of vegetation to regulate the global climate, by miss-assigning the causality for observed global warming to the greenhouse gas narrative, while invalidating (or dampening down) the narrative of a self-regulating biosphere. In our broadened understanding, carbon dioxide emissions only contribute a share of the global warming that we are experiencing! Whereas the exact attribution of global warming to two drivers, landuse and greenhouse gas emissions, needs to be re-assessed, with new or modified tools. Makarieva follows the scientific method that humanity learns with – science as an iterative process of observation and hypothesis formation, hypothesis testing, finding and correcting errors in our hypothesis, formulating counter-hypothesis, and re-evaluating our knowledge and our assumptions.  Her hypothesis around the errors of the Manabe-Wetherald lapse rate approximation is testable, but Makarieva herself is in no position to find out how this assumption is buried within the codes of our huge climate models.

Now, the climate community should to respond scientifically – by perceiving and respecting her critique, assessing its relevance, revisiting their model assumptions, digging into their model codes. And then, by honestly communicating what they learn. Until then, climate scientists should speak out clearly that their land use parameterization may contain a systemic conceptual error that makes all predictions questionable and unsuitable for global policy recommendations.

I hope that two decades of “climate consensus” against the attacks by the fossil fuel industry has not scarred climate scientists – scarred them such that scientists no longer act scientifically but respond emotionally.  An emotional response starts with ascertaining hierarchy by stressing that they are experts – in their bout against GHG denial, climate researchers learned to exude confidence and trust worthiness similar to medical doctors. The emotional response would continue with ignoring a well-founded critique, then defensive dismissiveness, then bullying and aggressive attacks. Such behavior is human but not scientific. Denial of Makarieva’s critique would take legitimacy from the climate movement that rests solidly on its “scientific basis”. It would fundamentally question the institution of academia during our era, a time when humanity depends on the scientific method for distinguishing truth from fiction.

In the end, our climate movement needs to reconvene around a shared climate narrative. The ‘old’ or ‘dominant’ climate models set the climate movement up against the fossil fuel industry, and against humanity’s hunger for fossil fuels. Whereas the new (and ancient) interpretation sets humanity up against additional corporate behemoths: the corporations dominating our land use, in particular the chemical input suppliers that shape our global industrial food systems, the timber and logging industry, increasingly biofuels that are legislated as “climate action” on our road to “net zero”, and the financial industry that profits from all of these. And let’s not to forget humanity’s craving for cheap food that is consistently available at all seasons, and our craving for enormous houses and other forest-based products.

PS: During my own academic career, I also experienced how life defies being modeled – I can deeply empathize with Anastassia Makarieva, who is facing an academic consensus around how to model the climate. When I was an academic modeller myself, I investigated how agricultural irrigation water flows can modify watershed hydrology. And I identified problems that are quite similar to those that Makarieva is pointing at: When water flows circulate within a single grid cell, or if processes vary dramatically within a grid cell, then models go wrong. So wrong, that even calibration cannot make them right. These effects happen when farmers irrigate ineffectively, letting much of the water flow back from their fields into the canal system. These effects happen when soils are healthy and alive: then water infiltration is mostly localized in macropores, whereas soils in most other areas quickly saturate ([16]). In both cases, the success of modelling would mainly require models to capture this exceptional behaviour of life, which is lost in spatial averaging. Averaging ignores how life opens up pores in the soil, how farmers allow water to circulate back into the canal system. Modellers are uneasy (or even unaware) how their calibration process adds uncertainty, even though it determines the predictive power of any integrated environmental model. I initially observed this problem in 2007 during my PhD thesis ([15]), but only managed to publish it years later ([17]).

References

[1] Makarieva AM, Nefiodov AV, Rammig A, Nobre AD. Re-appraisal of the global climatic role of natural forests for improved climate projections and policies. Frontiers in Forests and Global Change. 2023 Jul 20;6:1150191.

[2] Video: https://youtu.be/ZnrYy5T3-uk Article: https://climatewaterproject.substack.com/p/carbon-warming-water-cooling

[3] https://www.science.org/content/article/controversial-russian-theory-claims-forests-don-t-just-make-rain-they-make-wind

[4] Makarieva AM, Gorshkov VG. Biotic pump of atmospheric moisture as driver of the hydrological cycle on land. Hydrology and earth system sciences. 2007 Mar 27;11(2):1013-33.

[5] Makarieva AM, Gorshkov VG. The Biotic Pump: Condensation, atmospheric dynamics and climate. International Journal of Water. 2010 Jan 1;5(4):365-85.

[6] Douglas Sheil , Daniel Murdiyarso, How Forests Attract Rain: An Examination of a New Hypothesis, BioScience, Volume 59, Issue 4, April 2009, Pages 341–347, https://doi.org/10.1525/bio.2009.59.4.12

[7] Hatton I, Galbraith E. Commentary on “Key ecological parameters of immotile versus locomotive life” by VG Gorshkov and AM Makarieva. Rus. J. Ecosyst. Ecol. 2020;5.

[8] Meesters, A. G. C. A., Dolman, A. J., and Bruijnzeel, L. A.: Comment on “Biotic pump of atmospheric moisture as driver of the hydrological cycle on land” by A. M. Makarieva and V. G. Gorshkov, Hydrol. Earth Syst. Sci., 11, 1013–1033, 2007, Hydrol. Earth Syst. Sci., 13, 1299–1305, https://doi.org/10.5194/hess-13-1299-2009, 2009.

[9] Manabe, S. and Wetherald, R. T. (1967). Thermal Equilibrium of the Atmosphere with a Given Distribution of Relative Humidity. Journal of the Atmospheric Sciences, 24(3):241{259

[10] Lindzen RS, Hou AY, Farrell BF. The role of convective model choice in calculating the climate impact of doubling CO 2. Journal of Atmospheric Sciences. 1982 Jun;39(6):1189-205.

[11] Thanks to Alpha Lo for finding this paper and suggesting this quote. See https://climatewaterproject.substack.com/p/carbon-warming-water-cooling

[12] This example explains the difficulty of interpreting effective variables: it may rain over a 10,000km2 grid cell – some areas get no rain at all, others areas more than 100mm. The average rainfall is 10mm. How much of this rain will be taken up by the soil, how much will run off into the rivers?  Assuming that soils can readily absorb 10mm of rain, an “averaged” rainfall value will predict zero runoff into rivers, all rainfall is absorbed by soils and available to plants. With the real spatial distribution of rainfall, some soils are not wetted at all, while soils in other areas rapidly saturate completely such that the remaining rainfall runs off into the river system and may cause flooding.  So how can modellers predict flooding if it predicts “effective” 10mm of rainfall in an area with 10mm soil absorption capacity? Should modellers issue a flood warning even though their models predict no runoff at all? Modelers can only add more assumptions how 10mm averaged rainfall may create effective runoff.

[13] Refsgaard JC, van der Sluijs JP, Højberg AL, Vanrolleghem PA. Uncertainty in the environmental modelling process–a framework and guidance. Environmental modelling & software. 2007 Nov 1;22(11):1543-56.

[14] Tebaldi C, Smith RL, Nychka D, Mearns LO. Quantifying uncertainty in projections of regional climate change: A Bayesian approach to the analysis of multimodel ensembles. Journal of Climate. 2005 May 15;18(10):1524-40.

^[15] Arnold T. (2008). The WASIM-ETH watershed irrigation extension. https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=6e98c4cb2b729077982fc1771eeb8c18bd984f7e

[16] Norman JM. Fifty years of study of SPA systems: past limitations and a future direction. Procedia Environmental Sciences. 2013 Jan 1;19:15-25.

[17] Fu B, Horsburgh JS, Jakeman AJ, Gualtieri C, Arnold T, Marshall L, Green TR, Quinn NW, Volk M, Hunt RJ, Vezzaro L. Modeling water quality in watersheds: From here to the next generation. Water resources research. 2020 Nov;56(11):e2020WR027721.