Thermally driven CO₂ adsorption/desorption and its effect on soil respiration

Abstract

Soil respiration counterbalances global photosynthesis, yet several aspects of this process remain unexplained despite decades of research: (i) nighttime negative soil carbon flux in drylands and (ii) diurnal temperature-dependent hysteresis patterns. We propose that abiotic CO₂ adsorption-desorption in the soil is the cause or contributor to these phenomena. Laboratory isotherms were conducted on desert soil at various diurnal temperatures and CO₂ partial pressure levels. These results, combined with field data from the Mojave Desert, demonstrate that adsorption causes negative nocturnal CO₂ fluxes, while desorption during the day leads to positive fluxes. In addition, by accounting for thermally driven physical CO₂ exchange, the apparent temperature hysteresis commonly observed is reversed. The annual CO₂ adsorption in the thermally active soil layer ranges from 16.7 to 19.3 grams of CO₂-C per square meter per year, indicating a possible abiotic exchange of up to 2.2 to 2.5 billion tonnes of CO₂-C with the atmosphere globally. Our findings suggest that gaseous adsorption should be included in the global carbon cycle to more accurately represent land-atmosphere CO₂ exchange.

INTRODUCTION

Soil respiration—the net flux (release) of CO₂ from soils to the atmosphere (1) (fig. S1)—is the annual transfer of between 68 and 101 billion tonnes (Gt) CO₂-C year⁻¹ to the atmosphere from the soil environment (2). This large flux consists of the combined contributions of plant root respiration and the respiration of heterotrophic microbes that oxidize soil organic carbon (C). Soil respiration counterbalances net photosynthetic inputs to global soils. Advances in the past two decades that allow near-continuous measurements of soil respiration rates over daily to annual cycles have revealed two puzzling phenomena: (i) a diurnal, and seasonal, hysteresis of respiration rate versus soil temperature (3–5) and (ii) nighttime negative soil flux rates in arid to semiarid regions, especially during dry periods (6–8). Here, we propose that these two seemingly different phenomena are linked and driven by the same process: temperature-dependent physical adsorption of CO₂ onto soil minerals and organic matter.

Physical scientists demonstrated more than a century ago that common soil minerals can adsorb substantial quantities of CO₂ and that this is strongly temperature dependent (9). Langmuir (9) in 1918 found that muscovite, a nonexpanding phyllosilicate mineral, adsorbed 0.2 μmol of CO₂ per gram of mineral under ambient conditions and that the quantity adsorbed varied positively with CO₂ pressure and inversely with temperature. More recently, mineralogists have used CO₂ adsorption isotherms, along with the more standard gas N₂, to probe the nature of mineral surfaces and their molecular characteristics (10). Today, considerable effort is devoted to understanding CO₂ adsorption behavior on various minerals at the highly elevated pressures associated with subsurface injection of CO₂ into underground reservoirs (11). Yet, a linkage between these fields of study and the soil C flux community appears largely absent. Only one paper has attempted a back-of-the-envelope estimate of the possible amount of CO₂ adsorbed on global soils (12).

One of the important gaps in CO₂ adsorption studies has been (with some exceptions) experiments on either natural soil or on common soil mineral types, at ambient conditions, and across temperature ranges observed in field conditions, where negative CO₂ fluxes have been observed. Here, we conducted CO₂ adsorption isotherm experiments for natural Mojave Desert soil and combined these with observations of nocturnal negative CO₂ fluxes at a low-elevation Mojave Desert site using continuous in situ soil monitoring data collected over a 2-year period. Using measurements of temperature (T), CO₂ concentrations, and soil water content (θ) from four depths, we (i) calculated surface CO₂ fluxes, (ii) used laboratory isotherm experiments and modeling, to demonstrate that thermally driven adsorption of CO₂ is a more likely driver of observed negative fluxes than aqueous physiochemical or biotic processes, and (iii) examined whether thermally driven physical adsorption, when combined with soil respiration measurements, can help explain the apparent temperature hysteresis of soil respiration rates. We conclude that the thermal adsorption/desorption of CO₂ onto mineral surfaces is a notable unrecognized process in land-air CO₂ exchange, particularly in areas of low biotic activity and large temperature fluctuations.

RESULTS

Negative soil carbon flux rates

The diffusional flux of CO₂ across the soil-air interface (defined here as positive if the net flux is into the atmosphere or negative if the reverse) is driven by the concentration gradient, meaning that for positive fluxes, CO₂ concentrations in soil are higher than those in the atmosphere (fig. S1). A substantial number of studies have identified negative nocturnal soil CO₂ fluxes in desert soils, driven by soil CO₂ concentrations below local tropospheric levels (Fig. 1 and data S1). The negative flux follows the pattern of the diurnal soil temperature cycle, reaching positive values during the warming phase, and becoming negative during the cooling phase. The negative flux rates typically range from −0.02 to −0.1 μmol m⁻² s⁻¹, reaching −1.5 to −1.64 μmol m⁻² s⁻¹ in some locations (Fig. 1 and data S1). Despite its widespread occurrence, the underlying cause of this anomaly remains uncertain (13–16). The annual integrated values of observed negative CO₂ fluxes range from −4 to −65 g C m⁻² year⁻¹ (13, 14), which approach the rates of biologically driven soil CO₂ effluxes: 10¹ to 10³ g C m⁻² year⁻¹ (17). While in situ soil CO₂ production from autotrophic and heterotrophic soil respiration is almost always occurring, in dryland soils where biological production is limited by available C and/or water, its magnitude is smaller than the puzzling consumption processes, leading to net negative fluxes. There is no apparent biological process that should cause net consumption of CO₂ under these well-aerated conditions. Chemoautotrophic carbon fixation may contribute to carbon cycling in soils (18, 19), but its role in desert soil systems has not yet been evaluated (20), and it is uncertain why it would track inversely the diurnal temperature fluctuations. The unlikelihood of this mechanism is supported by several studies that have observed negative soil flux rates in sterilized soils (21–23). Thus, the prevailing explanations center on abiotic mechanisms: CO₂ dissolution into soil water, calcite dissolution (13, 16, 17, 24–26), and soil water vapor adsorption (15). Here, we address the hypothesis that thermally driven adsorption/desorption of CO₂ onto mineral surfaces is the underlying mechanism for not only negative adsorption in low-respiration drylands but also for temperature-driven impacts on diurnal cycles of soil respiration in more mesic environments.

Fig. 1. Reported negative nighttime soil surface CO₂ fluxes across global sites.

The soil surface CO₂ flux rates (micromoles per square meter per second) are in large font, and the references for the observations are in parentheses (6, 7, 8, 13–16, 21–26, 33, 64–80). Values represent mean fluxes where summary statistics were available in the original publications (data S1); otherwise, individual reported negative nighttime flux values are shown. The map was created in ArcGIS Pro.

Our gaseous adsorption investigation was motivated by field observations. Between 2017 and 2019, we continuously monitored soil CO₂ (fig. S2), temperature (T) (fig. S3), and volumetric water content (VWC) (θ) (fig. S4) in a characteristic low-elevation biotic zone (Larrea tridentata community) in the Mojave Desert National Preserve (27) (Materials and Methods). From the measured soil parameters (data S2) (and diffusion modeling described in Materials and Methods), net soil CO₂ fluxes (fig. S5) were calculated over 10-minute intervals and aggregated to 1-hour intervals. During the hot, dry summer months, when net soil CO₂ flux rates were low (−0.07 to 0.35 μmol m⁻² s⁻¹), negative surface fluxes were commonly observed at night (Fig. 2 and table S1). To test the hypothesis that this phenomenon could be explained, at least in part, by CO₂ adsorption onto soil minerals, we subsequently measured CO₂ adsorption on the same soil across the range of observed soil temperatures (2° to 70°C) and partial pressures of CO₂ (Pco₂) with depth [0 to 0.5 kPa, corresponding to 0– to 5000–parts per million (ppm) CO₂ levels] (Fig. 3A and figs. S6 to S8) (28). We also measured adsorption-desorption cycles and found minimal hysteresis (less than 4%), indicating that CO₂ desorption is nearly equal to adsorption under diel soil CO₂ conditions. Both processes occur rapidly, reaching equilibrium within seconds to minutes under typical soil conditions. Notably, after a rain event, negative nighttime soil CO₂ fluxes often disappeared, replaced by pronounced CO₂ pulses from soil to atmosphere (fig. S4), which are the result of moisture-driven increases in soil heterotrophic respiration and related phenomena (a.k.a. the Birch Effect) (29). In this study, we do not examine adsorption dynamics during or after rain events. Under high soil moisture, adsorption is likely negligible, as water occupies most available sorption sites and limits the soil’s capacity to adsorb CO₂.

Fig. 2. Diurnal variation in soil surface CO₂ flux and temperature on 6 June 2018.

Soil surface CO₂ flux (left axis) is shown alongside air temperature, soil surface temperature, and soil temperature at 5-cm depth (right axis). Gray shading denotes periods of negative CO₂ flux, and yellow shading highlights daytime hours. Data were recorded at 10-min intervals by meteorological and in situ soil sensors and aggregated to hourly means. Some components of this figure were created in BioRender. A, A. (2026) https://BioRender.com/fl1e6fc.

Fig. 3. CO₂ adsorption isotherms for several soils and clay minerals.

(A) Temperature dependence of CO₂ adsorption on Mojave Desert soil (Arizo soil series, Mojave National Preserve, CA) at temperatures ranging from −2° to 70°C. (B) Comparison of CO₂ adsorption isotherms at ambient temperature (T = 20°C) for three soils—Snelling soil series (Central Valley, CA), carbonatic Typic Torriorthents (Kyle Canyon, NV), and Mojave Desert soil (Mojave National Preserve, CA)—and two clay minerals (kaolinite, KGa-1b, and montmorillonite, SWy-3). In both panels, CO₂ adsorption (micromoles per gram) is plotted as a function of CO₂ partial pressure (0 to 1 kPa, corresponding to 0- to 10,000-ppm CO₂ levels). Some components of this figure were created in BioRender. A, A. (2026) https://BioRender.com/jzght33.

The Mojave soils (Mojave National Preserve, CA) formed in Late-Pleistocene granitic alluvium, classified as sandy-skeletal with minor pedogenic clay development, Typic Torriorthents (Arizo series), dominated by quartz (30%), Na/Ca feldspar (39.4%), K feldspar (23.3%), chlorite (5.1%), and borate (2%), with largely noncarbonate surface horizons (data S2). Despite the relatively low surface area per mass [Brunauer-Emmett-Teller (BET) surface area 3.3 to 6.5 m² g⁻¹], the soil adsorbs 1.2 μmol CO₂ g⁻¹ at partial pressures of 2000 ppm (equivalent to ∼0.2 kPa) and can adsorb-desorb 1 μmol CO₂ g⁻¹ across a 70°C temperature range (Fig. 3A and figs. S6 to S8) (28). In laboratory experiments, we observed that samples not subjected to heating and degassing adsorbed less CO₂ than treated samples, which we expected, with the difference varying by CO₂ partial pressure (fig. S9). At low pressures, adsorption on nondegassed samples was about ¹/₁₀ of the degassed values, while at higher pressures, it approached ⁷/₁₀ of the degassed values. For our estimates of thermally driven adsorption-desorption in the field, we derived a nonlinear empirical relationship between the fully degassed (corresponding to the maximum CO₂ adsorption capacity) and nondegassed (corresponding to natural occupancy of adsorption sites in soils) conditions (fig. S10). With these data, we applied a monotonic cubic Hermite interpolation (PCHIP) method to construct a three-dimensional (3D) matrix of CO₂ adsorption as a function of both T and CO₂ concentration (figs. S11 and S12 and data S3). This was combined with observed T and CO₂ variations with depth (below), to calculate the total adsorption-desorption of the soil profile for a given time interval (Fig. 4).

Fig. 4. Depth profiles of soil temperature and CO₂ adsorption change on 6 June 2018.

(A) Soil temperature profiles were recorded at depths of 5 and 25 cm and then interpolated for each centimeter. (B) Hourly change in CO₂ adsorption (ΔQ) at corresponding depths. Some components of this figure were created in BioRender. A, A. (2026) https://BioRender.com/bpzndt8.

To assess first-order variability in CO₂ adsorption in different soils, we measured CO₂ adsorption in several additional natural soils and reference clays at 20°C (Fig. 3B and Materials and Methods). The phyllosilicate clay montmorillonite has relatively high adsorption capacities, reaching 6 μmol CO₂ g⁻¹ at 0.2 kPa (corresponding to ∼2000-ppmv CO₂), whereas kaolinite has low CO₂ adsorption values. Soils vary in their adsorption capacities on the basis of their properties and environmental conditions. Soils with high carbonate content from Kyle Canyon near Las Vegas had adsorption values of 2 μmol CO₂ g⁻¹ at 0.2 kPa. The adsorption capacity of Mojave Desert soil lies at the lower end of the observed range among our tested soils and clays (1 μmol CO₂ g⁻¹ at 0.2 kPa). Thus, the quantities of thermally adsorbed CO₂, which we calculate for this site, are conservative, and for other locations, the magnitude of diurnal adsorption/desorption may be much higher.

In arid regions, sparsely vegetated landscapes in the summer and/or dry periods experience a very large diurnal temperature range. At our Mojave site, diurnal amplitudes of ∼40°C are common (Fig. 2 and fig. S3). Soil temperature (due to heat flow) undergoes a diurnal pattern with depth and time of day. We used measured soil temperatures (5, 25, 50, and 125 cm) (fig. S3) and a cubic polynomial fitting to determine the depth at which diurnal temperature changes are effectively damped (Fig. 4A). We consider the layer above this depth to be the thermally active layer in the soil, and we replicated the observed quasi-periodical pattern of soil temperature on a daily cycle (Materials and Methods). Similarly, soil CO₂ concentrations at any depth experience diurnal changes (fig. S2), and we fit depth-dependent models to the observed soil CO₂ concentrations (fig. S13). From the temporal and depth patterns in both temperature and CO₂ concentrations, we used a 3D surface matrix (fig. S11) to calculate depth-resolved T and CO₂ concentration–driven changes in CO₂ adsorption (Fig. 5A).

Fig. 5. Diurnal dynamics of soil CO₂ adsorption and fluxes.

(A) Depth-resolved CO₂ adsorption at 1-cm intervals across the 0- to 25-cm soil profile, expressed per unit soil area (1 m²) and per 1-cm soil depth. Labels indicate selected soil depths (centimeters); data are shown for 5 to 10 June 2018. (B) CO₂ fluxes: (i) net CO₂ adsorption flux integrated over the 0- to 25-cm depth [shown as × (−1) for visual comparison], (ii) soil surface CO₂ flux, and (iii) soil CO₂ flux at the 25-cm depth. All fluxes are in micromoles per square meter per second; data are shown for 1 to 10 June 2018. Some components of this figure were created in BioRender. A, A. (2026) https://BioRender.com/q68c760.

The calculated temporal changes in the quantity of CO₂ adsorption/desorption varied greatly with depth because of the reduction in daily temperature ranges with depth (Fig. 4B). The calculated adsorbed CO₂ at any given time of day was converted to rates of adsorption or desorption by evaluating depth-dependent changes over 1-hour time intervals (Fig. 5B).

There are important insights embedded in Fig. 5. First, there is a strong positive relationship between the observed soil surface CO₂ flux and the calculated temperature-driven rates of physical adsorption/desorption between 0 and 25 cm. In this depth interval, rapid temperature fluctuations strongly influence CO₂ adsorption and desorption onto and from soil particles. With increasing depth, the adsorption-desorption dynamics are damped, while the total amount of CO₂ adsorbed increases because of higher Pco₂ concentrations. Second, the calculated physical adsorption-desorption rates, while inversely mirroring soil surface CO₂ flux rates, at night (when negative fluxes occur) sometimes exceed the negative surface flux. This is because the net physical adsorption includes not only the inward flux of CO₂ from the atmosphere into the soil (which has a lower CO₂ concentration) but also the upward flux of CO₂ from deeper depths into the upper 25 cm (fig. S14). When net adsorption minus the upward flux from deeper in the soil across the 25-cm interface is calculated, the difference is numerically very close to that of the observed negative soil surface CO₂ flux.

We find strong evidence that physical adsorption drives the negative fluxes, and that, additionally, the daily desorption is a strong contributor to the positive soil CO₂ fluxes that occur during the day, but with some interesting complexities. The diurnal cycle shows a clear phase shift between air and soil temperatures at various depths, as well as between CO₂ concentrations at different depths. As a result, CO₂ flux and net adsorption do not respond to temperature alone but reflect the combined, time-integrated effects of temperature and CO₂ pressure. Consequently, adsorption represents a cumulative thermodynamic response to asynchronous variations in thermal and concentration gradients within the soil profile (fig. S15). Despite the complexity of integrating field observations with laboratory-obtained parameters, the agreement between adsorption and flux divergence is strong [coefficient of determination (R²) = 0.78] in dry hot months, for example, in August 2017, June 2018, and August 2019 (fig. S16).

Undoubtedly, thermally driven adsorption-desorption on soils and other solid surfaces is a global phenomenon, but its presence is obscured in soils with high biological CO₂ production. In addition, soil relative humidity and water content likely affect CO₂ adsorption capacity. However, studies of CO₂ adsorption under environmentally relevant conditions, where both CO₂ and water vapor are present, are experimentally difficult and, therefore, remain limited. Experiments on pure minerals have shown that increasing relative humidity reduces CO₂ adsorption by competing for adsorption sites (30). In contrast, studies of polymer-based sorbents have reported enhanced CO₂ adsorption under humid conditions due to water-induced pore opening and sorbent swelling (31).

We also considered the role of CO₂ partitioning into soil water driven by temperature-dependent carbonic acid speciation and carbonate mineral dissolution/precipitation (fig. S17 and data S4). We found that because of very low soil water contents at the time negative fluxes are observed, the role of CO₂ consumption by aqueous phase carbonate system reactions is small relative to the amount and rates calculated for physical adsorption. It was at least an order of magnitude lower than physical gaseous adsorption during the dry season (fig. S17). The relative impact of aqueous CO₂ processes may be greater during different seasons of the year in the Mojave or in more mesic environments.

Hysteresis of soil respiration

As in many studies (3, 7, 13, 32), we observed a temperature-dependent diurnal hysteresis on soil respiration rates (Fig. 6A). The pattern is that soil respiration increases with increasing temperature, but the observed respiration rate during the morning warming phase is higher than at the same temperature during the evening cooling phase. A number of mechanisms have been proposed for this phenomenon, including lag times between measured temperature and the observed flux at the soil surface, temperature effects on diffusion-related parameters, and/or temperature (and time) lags in plant substrates available for respiration (3, 4).

Fig. 6. Diurnal hysteresis between soil surface CO₂ flux and soil temperature at 5-cm depth during June 2018.

(A) Soil surface CO₂ flux as a function of soil temperature, showing clockwise hysteresis. (B) Net CO₂ flux (surface flux minus adsorption flux) as a function of soil temperature, illustrating the contribution of CO₂ adsorption-desorption to the observed hysteresis. Some components of this figure were created in BioRender. A, A. (2026) https://BioRender.com/uxrb375.

Here, we test the hypothesis that thermally driven adsorption-desorption is a contributor to this phenomenon. For the month of June 2018, we plot the soil respiration rates for all days for hourly intervals. The clockwise temporal trajectory of the respiration reflects the commonly observed pattern described above (3). We then subtracted from the fluxes the calculated net thermally driven adsorption-desorption for the time interval (Fig. 6B). Once we have subtracted physical adsorption/desorption from the June flux data, the CO₂ flux data still maintain an apparent temperature dependence, but at higher temperatures the hysteresis becomes counterclockwise.

This comparison has several important implications. First, the observation of higher CO₂ fluxes during the daily warming phase is frequently attributed to thermal acceleration of microbial respiration. However, this work indicates that a substantial contributor to this behavior is the thermal desorption of CO₂ sorbed to soil particle surfaces during the previous night. Second, similarly, the apparent lower temperature sensitivity of respiration during the daily cooling phase could be explained by the physical adsorption of CO₂ during this interval, which counters the biological flux. A third feature is that once the physical movement of CO₂ is removed from the net fluxes, the hysteresis takes on a new, counterclockwise pattern. This suggests that respiration is not as sensitive to temperature during the morning warmup as it is during the evening cooldown. The adsorption-corrected pattern is consistent with recent research suggesting that plant respiration is substrate dependent (2) and that after a nighttime dark period, plants will be more deficient in respirable substrates than after the daytime photosynthetic period. Thus, from theory, plant respiration should follow the pattern of the adsorption-corrected flux patterns in Fig. 6B. Also, thermal convection may play a role in the observed hysteretic behavior (33).

DISCUSSION

Thermally driven gaseous adsorption onto synthetic materials deployed in desert environments has been used for atmospheric water harvesting (34), but the adsorption of other gases onto natural soils has received little attention. Davidson et al. (12) estimated that adsorption contributes only marginally to CO₂ exchange, leading Schlesinger (35) to conclude that it accounts for a minor fraction of the apparent CO₂ sink. However, Davidson et al. (12) did not address the temperature-driven dynamics of adsorption, focusing instead on total amounts potentially adsorbed. Here, we focus on the magnitude of gross fluxes driven by diurnal temperature ranges. We used our calculated annual average values of the total amount of CO₂ thermally adsorbed/desorbed per day from the upper 25 cm of the Mojave soil (16.7 to 19.3 g CO₂-C m⁻² year⁻¹) and multiplied this by the global desert area (17.64 × 10⁶ km²) (36) and the ice-free land area of the planet (130 × 10⁶ km²) (37). This extrapolated estimate suggests that from 0.3–0.34 to 2.2–2.5 Gt CO₂-C may be exchanged annually through purely physical processes across the land-air interface. Although our global estimate is very uncertain, it indicates that the daily amounts of CO₂ adsorbed and released by surface soils are substantial, contributing substantially to CO₂ cycling, particularly in arid ecosystems. This estimate is based on measurements from Mojave Desert soils, which represent the lower bound of the gaseous adsorption potential. Soils richer in organic matter, clay, and carbonates likely exhibit even greater adsorption capacity, highlighting the need for studies across diverse geographic areas. Furthermore, the addition of biochar and other adsorptive amendments to arid soils has been proposed as a nature-based strategy for enhancing carbon capture (38–40).

Desert ecosystems with strong daily temperature variability are likely to exhibit the most pronounced CO₂ adsorption-desorption rates. Consequently, along a climatic gradient from dry to humid ecosystems, the magnitude of physical CO₂ adsorption likely decreases while biological soil respiration generally increases. Under a warming climate, when the entire diurnal temperature cycle shifts toward warmer conditions (assuming the same day-night temperature amplitudes), CO₂ adsorption capacity will decrease (fig. S12). However, the net effect of this on the annual atmospheric CO₂ budget is near zero (assuming that the adsorbed CO₂ is not involved in some subsequent chemical reaction).

In addition, CO₂ adsorption may have impacts on the stable isotope composition of atmospheric CO₂. Physical adsorption of CO₂ onto mineral surfaces may place it in more direct contact with mineral-associated soil water and contribute to the well-studied effect of soil water on the stable O isotope composition of atmospheric CO₂ (41, 42). In addition, there are a few experiments that suggest that adsorbed CO₂ on minerals may induce a unique C isotope effect on the remaining CO₂, with the present data suggesting that ¹²CO₂ may be preferentially adsorbed onto basalt (43) and onto clay (44) minerals. This process (whatever the sign of the isotope effect) may have impacts on the measured δ¹³C value of soil respiration over diurnal time courses. These potential isotope effects require much further study, but recognizing that physical processes occur in concert with biologically driven CO₂ fluxes is important as land-air C exchange studies expand in complexity and spatial scope. Thermally driven atmospheric CO₂ adsorption has long been considered an important process on Mars (45–47), and our study identifies it as a notable process in arid environments and likely globally.

MATERIALS AND METHODS

Laboratory CO₂ (and N₂) adsorption isotherm measurements

We conducted main CO₂ adsorption isotherm experiments on soils from the Mojave Desert (CA), and additionally on soils from Kyle Canyon (NV) and the Central Valley (CA), and on the clay minerals Na-montmorillonite (SWy-3) and kaolinite (KGa-1b). The gas adsorption isotherms were measured using a three-port Micromeritics 3Flex gas adsorption analyzer via a manometric method. This instrument allowed us to measure at very low gas pressures and featured a high-sensitivity pressure transducer, enabling a measurement range from 1.3 × 10⁻⁹ to 1.0 P/P₀. This capability allowed us to accurately replicate ambient atmospheric and soil CO₂ partial pressures, spanning from 0.04 to 0.5 kPa (∼400- to 5000-ppmv CO₂ partial pressure in soil air). For Mojave Desert soils, we conducted experiments at eight temperatures to obtain CO₂ adsorption isotherms that reflect seasonal and diurnal soil surface changes: −2°, 2°, 5°, 10°, 20°, 40°, 60°, and 70°C. For other soils and clays, we obtained the isotherms at T = 20°C. The temperature was controlled and stabilized using a Micromeritics Iso-Controller unit. The full dataset is available in a Zenodo repository (28).

The average soil sample for the adsorption isotherm experiments was ∼3 to 4 g. Before measurement, each sample was purged with high-purity He or N₂ gas (at 1-atm pressure) at 90°C for 1 hour, followed by at least 4 hours at 350°C. This process, performed with a Micromeritics Flow Prep 060 unit, ensured consistent background measurements.

To ensure that the sample preparation process (degassing and heating) did not change soil properties related to adsorption, we also performed thermogravimetric analysis (TGA) on the Mojave Desert soil using the METTLER TOLEDO Thermal Analysis System TGA/DSC ³⁺ under an inert Ar atmosphere. This analysis allowed us to measure the weight loss caused by heating. The initial sample weighed 78.1020 mg, and it was placed on a Platinum 70-μl holder with a mass of 278.4460 mg. Results showed a very low weight loss (∼2%), primarily due to water evaporation (fig. S18).

After degassing, the sample tubes were moved to the instrument analysis ports, and the analysis runs were conducted. The initial and final absolute pressures, equilibration intervals, and gas dosage amounts were set manually, on the basis of the soil CO₂ partial pressure range identified during field observations. The typical starting pressure was about 0.001 kPa, and the equilibration interval at this pressure was 10 seconds. The amounts of gas absorbed are reported in millimoles per gram of soil; measurement details are included in the measurement reports (28).

We used the same set of soil samples to measure adsorption isotherms at various temperatures. Each experiment was performed in triplicate; however, only Port 2 had an advanced transducer capable of detecting CO₂ adsorption at very low pressures (below 1.3 kPa). For Mojave Desert soils, we repeated the experiments at different temperatures to generate a second set of triplicates and assess measurement uncertainties. These replicates were conducted at four temperatures (T = −2°, 20°, 40°, and 60°C). The pressure points during adsorption measurements varied slightly across replicates, leading to nonaligned data. To compute the SD between replicates, we interpolated the adsorption values onto a common pressure grid using spline interpolation. We then calculated the mean, SD, and 95% confidence intervals (fig. S6, A to D, and data S5). The adsorption isotherms from replicates generally follow consistent patterns and are well aligned, though some variability in the absolute CO₂ amounts was observed—likely because of soil heterogeneity. The soil surface areas ranged from 3.3 to 6.5 m²s.a. g⁻¹, and this variation in sorption sites contributed to the spread in CO₂ adsorption values among replicates (28).

For Mojave Desert soils, adsorption isotherms at the other four temperatures studied (T = 2°, 5°, 10°, and 70°C) were measured in triplicate, with no repeats (fig. S7, A to D). Consequently, only a single set of data was available at very low pressures, obtained with the advanced transducer (Port 2). Calculating average values above 1.3 kPa and extrapolating to lower pressures introduced minor distortions. Therefore, we relied on the high-resolution data from the advanced pressure transducer (Port 2) for these temperatures, as it offers the most precise measurements in the low-pressure range. The isotherms at T = 5° and 10°C exhibited unexpectedly higher adsorption than at −2°C, especially at low pressures (fig. S8). We repeated the adsorption analysis at T = −2°C, which confirmed strong reproducibility (fig. S6A), supporting the reliability of the −2°C isotherm. Because of physical inconsistencies, the isotherms at T = 5° and 10°C were excluded from further analysis. The final dataset included mean isotherm values for −2°, 20°, 40°, and 60°C, along with high-resolution data from the advanced transducer at 2° and 70°C.

To measure hysteresis behavior between adsorption and desorption at soil-relevant CO₂ concentrations, we conducted an experiment with several adsorption-desorption subcycles across pressure ranges of 500 to 1500 ppm (0.05 to 0.15 kPa) and 5000 to 6500 ppm (0.5 to 0.6 kPa), then reached 10 kPa and repeated the first two pressure loops. The measured diurnal adsorption-desorption hysteresis was very small, ranging from ∼0.3 to 4%. After equilibration at 10-kPa CO_2, the diurnal hysteresis remained similarly low, at ∼0.3 to 3% of the adsorbed amount. In contrast, the comparison of the adsorption and desorption branches at 10-kPa CO₂ showed a relatively larger hysteresis, ranging from 8 to 14%. However, such pressures substantially exceed those typically observed in the soils investigated in this study (fig. S19 and data S6). Therefore, we assume that at soil-relevant CO₂ diurnal cycles, adsorption and desorption are symmetrical.

Adsorption time depends on the material’s surface complexity, specific surface area, and temperature and pressure conditions. The soil texture in the Mojave Desert is sandy loam, which has a relatively low surface area (BET surface area 3.3 to 6.5 m² g⁻¹). Consequently, adsorption is assumed to occur very quickly. On the basis of the adsorption isotherm measurements, we estimated that CO₂ adsorption takes from seconds to minutes (fig. S20). Specifically, achieving adsorption equilibrium at up to 10 kPa at T = 2°C took around 100 min, indicating that the system reaches equilibrium within roughly 1 min for a pressure increase of 0.1 kPa (±1000-ppm CO₂). This includes the adsorption process itself and equilibration-sensing measurements, which likely take up most of this time. The equilibration interval (the time over which pressure must remain stable within a very small range) was set to 10 s, and the pressure tolerance (how closely the actual pressure must match the target pressure to be acceptable) was below 0.01% of the weighted average. After each dose, the instrument took 11 pressure readings and calculated the weighted-average pressure drop to track the rate of change in pressure, which must be below 0.01% of the weighted average to make a measurement. CO₂ adsorption rates decrease at lower temperatures, although the total amount adsorbed is greater. The instrument finished the adsorption isotherm measurement at T = 60°C more quickly than at T = 2°C; however, the time intervals were quite similar (fig. S20). The comparison of the adsorption and desorption time increments showed that they were approximately equal.

Adsorption under laboratory conditions (Q_lab) was measured for samples degassed to remove water vapor and other gases from the absorbent surfaces. To investigate the natural occupancy of adsorption sites in soils (Q_nat), we performed a series of isotherm measurements on nondegassed samples equilibrated at various relative humidity levels, controlled by saturated salt solutions of lithium chloride (LiCl for ∼15% RH) and potassium chloride (KCl for ∼86% RH), following the method in (48), and on uncontrolled samples stored under ambient laboratory conditions (28). In addition, adsorption isotherms were measured for soil samples heated and nitrogen purged at 150°C to partially degas and 350°C to fully degas, thereby removing most of the adsorbed water vapor. The adsorption capacities of the nondegassed soils were compared with measurements on degassed samples (fig. S9). This relationship was numerically represented by a transformation equation (Q_nat = 452.06 × Q_lab^1.9644) (fig. S10).

Site description and soil sampling

The field site (geographic coordinates: 35.10613, −115.55074; elevation, 935 m) is situated in the Mojave National Preserve in California, within a creosote shrubland ecosystem typical of low-elevation Mojave Desert environments. The soil sampling and site were obtained under the National Park Service Research Permit MOJA-2024-SCI-0034. The climate is arid: The mean annual temperature is ∼18.3°C, and the mean annual precipitation is ∼156 mm (49). The vegetation composition at the site is dominated by creosote (L. tridentata) and burroweed (Ambrosia dumosa) shrubs (50). The soils have formed in Late-Pleistocene granitic alluvium (51) and are mapped as the Arizo series, a sandy-skeletal, mixed, thermic Typic Torriorthents with modest carbonate contents and minor pedogenic clay accumulation. Two soil profiles were excavated to a depth of 125 cm. The first soil profile was located on a patch of intercanopy bare ground; the second was under the canopy of a creosote (Larrea tridentata) bush (50). Soils were sampled at the surface and at four depths (5, 25, 50, and 125 cm) for laboratory analyses, and instrumentation was installed at the same depths to continuously monitor soil conditions (fig. S21).

In addition, for gas adsorption isotherm analysis, we used archived soil samples from Kyle Canyon (NV) and the Central Valley (CA), as well as common clay minerals, including well-crystallized kaolin (GA) (KGa-1b) (52) and Na-montmorillonite (WY) (SWy-3), purchased from the Clay Minerals Society. Kyle Canyon carbonate soils (NV) developed on a limestone-dominated alluvial fan, where 90 to 95% of clasts are derived from Paleozoic-Mesozoic limestones and dolomites. These sandy-skeletal, carbonatic Typic Torriorthents contain 10 to 25% CaCO₃ in the A horizon (53). The Snelling soil (Central Valley, CA) is a noncalcareous soil from granitic Sierra Nevada glacial outwash (∼200 thousand years ago). The A horizon contains moderate clay (8 to 10%), near-neutral pH (6.7), and low organic matter (∼0.75%) (54). Na-montmorillonite (SWy-3) is a pure smectite clay mineral, characterized by a 2:1 phyllosilicate structure, high cation exchange capacity (76.4 mequiv 100 g⁻¹), and elevated surface area (31.8 m² g⁻¹). Kaolinite (KGa-1b) is a 1:1 phyllosilicate, with a low cation exchange capacity (2.0 mequiv 100 g⁻¹) and a modest surface area (10.05 m² g⁻¹).

Field measurements

At the Mojave National Preserve site, each soil profile was equipped with in situ sensors to measure temperature, VWC, and electrical conductivity (CS615, Campbell Scientific, Logan, UT, USA) at four depths (fig. S21). The sensors were installed directly into the exposed wall of the soil excavation before back-filling. Soil CO₂ sensors were also placed at the same depths. A cylindrical cavity was dug 20 cm into the exposed pit face to hold a 1–inch (2.54-cm)–diameter PVC tube, which was connected to a vertical PVC tube running to the soil surface via a 90° elbow. The pits were then refilled with soil, and solid-state soil CO₂ sensors (GMP 252, Vaisala Inc., Vantaa, Finland) were lowered into the PVC elbow. The vertical PVC tubing was capped and sealed with silicone sealant above the soil surface. The VWC sensors detected changes in soil dielectric permittivity, mainly affected by water content. However, water’s permittivity depends on temperature, making VWC measurements sensitive to temperature. Following (55), we used periods with minimal VWC change to estimate the temperature sensitivity (dVWC/dT) on the basis of diurnal fluctuations in temperature and VWC, then corrected the data by subtracting dVWC/dT(VWC) × (T − 25) from the raw VWC measurements. Meteorological and in situ soil sensor data were recorded at 10-min intervals using a Campbell Scientific data logger (CR1000, Campbell Scientific Inc., Logan, UT, USA). The full dataset is available on a HydroShare repository (27).

Determining the thermally active soil layer

In desert environments, large daily fluctuations in air temperature result in large daily temperature changes in shallow soil depths. The soil temperature record shows that this effect is strongest at soil depths above 20 to 25 cm (Fig. 4A). Below 20 to 25 cm, the temperature remains stable and does not fluctuate with day/night cycles. Therefore, we defined the thermally active layer as the soil layer above 25 cm, which we used as the lower boundary for studying adsorption depth profiles, with the upper boundary at the land surface (fig. S22).

Soil analysis

Soil samples were dried in a 50°C oven and sieved to separate the gravel-sized fraction from the <2-mm-size fraction. Soil physical properties—gravimetric water content, percent gravel by weight, and bulk density—were measured following published methods (56) for both profiles. Particle size distribution and bulk mineralogy were analyzed using powder x-ray diffraction (57). The pH was determined with a saturated paste method using water at the UC Davis Analytical Laboratory. Organic carbon content was measured through continuous-flow dual-isotope analysis with a CHNOS Elemental Analyzer connected to an IsoPrime 100 mass spectrometer at the Center for Stable Isotope Biogeochemistry at UC Berkeley. The full dataset is available in the Supplementary Materials (data S2).

Calculation of CO₂ flux in soil

Soil CO₂ fluxes across different depths, including soil respiration—the flux of CO₂ across the soil-air interface (fig. S1), were calculated using the flux gradient method, making the explicit assumption that CO₂ production (and consumption, where it occurs) and transport via molecular diffusion are the dominant processes controlling CO₂ concentrations in the soil profile (58, 59). Soil CO₂ diffusivity, D_s (square meters per second), was calculated at each depth as a function of soil texture, temperature, pressure, and water content, following (60). The CO₂ flux across various layers in the soil profile, including the surface, was then calculated using flux gradient theory and Fick’s first law of diffusion

$$F_{\left(z\right)}=-D_{\mathrm{s}} \frac{\mathrm{\partial }C}{\mathrm{\partial }z}$$ (1)

where F is the soil CO₂ flux (micromoles per square meter per second), Ds is soil CO₂ diffusivity (square meters per second), z is soil depth (0, 5, 25, 50, and 125 cm), and C is the molar CO₂ concentration (micromoles per cubic meter).

$$D\mathrm{s}=\mathit{Da} \left({\mathrm{\varphi }}^2\right){\left(\frac{\mathrm{\epsilon }}{\mathrm{\varphi }}\right)}^{\mathrm{\beta }\times F_{\mathit{cp}}}$$ (2)

where ϕ is the porosity, ε is the air-filled porosity $(\mathrm{\varphi }-\text{VWC})$, β is an empirically fitted parameter taken to be 2.9, $F_{\mathit{cp}}$ is the sand and silt content of the <2-mm-size fraction $(0\le F_{\mathit{cp}}\le 1)$, and $D_{\mathrm{a}}$ is the diffusivity of the CO₂ in the air, corrected for temperature and pressure

$$D_{\mathrm{a}}=D_{\mathrm{a}0} {\left(\frac{T}{293.15}\right)}^{1.75.} \left(\frac{1.013\times {10}^5}{P}\right)$$ (3)

where $D_{\mathrm{a}0}$ is 1.47 × 10⁻⁵ m² s⁻¹ (60), T is the temperature, measured in kelvins, and P is the pressure in pascal.

Calculation of soil CO₂ flux divergence

Soil CO₂ flux divergence, FD (figs. S1 and S14), in the entire thermally active soil layer (in the 0 to 25 cm) (not normalized to horizon thickness) was calculated from the difference between the CO₂ fluxes across the 25-cm layer and across the soil-air interface (z = 0 cm) (61)

$$\text{FD}=F_{(z=25 \text{cm})}-F_{(z=o \text{cm})}$$ (4)

Calculation of the mass balance model of net CO₂ adsorption/desorption in soil

To understand how CO₂ adsorption corresponds to soil CO₂ fluxes, we formulated and applied a mass balance model. Our experimental results showed that adsorption is a very temperature-sensitive process. As the temperature drops, the amount of CO₂ adsorption onto the soil particles increases. We can define the rate of CO₂ adsorbed/desorbed by a soil volume of a thermally active layer as the difference of CO₂ adsorbed amounts between two observation time points [t_n, t_(n−1)]

$$\mathrm{\Delta }{Q}_{\mathrm{d}}^{t_n,t_{n-1}}=Q_{\mathrm{d}}^{t_n}-Q_{\mathrm{d}}^{t_{n-1}}$$ (5)

The net CO₂ adsorbed or desorbed in the thermally active soil layer per time increment is equal to the CO₂ flux divergence

$$\mathrm{\Delta }{Q}^{t_n,t_{n-1}}\approx (F_{\text{in}}-F_{\text{out}})\times \mathrm{\Delta }t$$ (6)

where $\mathrm{\Delta }Q$ is the amount of adsorbed/desorbed CO₂ in the thermally active soil layer per time increment,$F_{\text{in}}$ is soil-atmosphere CO₂ flux (micromoles per square meter per second), calculated using the gradient method in the 0- to 5-cm soil layer, $F_{\text{out}}$ is soil CO₂ flux soil layer (micromoles per square meter per second), calculated using the gradient method in the 25- to 50-cm soil layer, and $∆t$ is the time increment (for example, 1 hour).

Calculation of the dissolution of CO₂ into soil water

The amount of dissolved inorganic carbon (DIC) in the soil water at each 1-hour time step was determined through carbonic acid system speciation calculations (62). For these calculations, temperature, CO₂ partial pressure, and VWC were measured at 5- and 25-cm soil depths and then interpolated to 1-cm depth. Measurements were collected at a 10-min time resolution and averaged to a 1-hour resolution. We measured the soil pH at each soil horizon in the studied soil profile and then interpolated it at 1-cm intervals. The soil pH changed from 7.2 in the topsoil to 8.4 at a depth of 25 cm. We assumed that pH was at a steady state.

The carbonate chemistry is as follows

$${\text{CO}}_2\left(\mathrm{g}\right) k_{\mathrm{H}}\leftrightarrow \mathrm{C}{\mathrm{O}}_2\left(\text{aq}\right)$$ (7)

where $k_{\mathrm{H}}$ is the Henry equilibrium coefficient and ${\text{CO}}_2\left(\mathrm{g}\right)$ is the partial pressure of CO_2.CO₂ (aq) represents the combination of dissolved CO₂ and H₂CO₃

$${\text{CO}}_2\left(\text{aq}\right) k_1\leftrightarrow {\mathrm{H}}^{+}\left(\text{aq}\right)+\text{HC}{{\mathrm{O}}_3}^{-}\left(\text{aq}\right)$$ (8)

$$\text{HC}{{\mathrm{O}}_3}^{-}\left(\text{aq}\right) k_2\leftrightarrow {\mathrm{H}}^{+}\left(\text{aq}\right)+{{\text{CO}}_3}^{2-}\left(\text{aq}\right)$$ (9)

For the calculations, we applied the Henry’s law constant ( K_H) and equilibrium constants (K₁ and K₂), which were calculated according to (63)

$$\begin{array}{c}\text{log }{K}_{\mathrm{H}}=108.3865+0.01985076\times \hfill \\ T-\frac{6919.53}{T}-40.45154\times \text{log} T+\frac{669365}{T^2}\hfill \end{array}$$ (10)

$$\begin{array}{c}\text{log} K_1=-356.3094-0.06091964\times T\hfill \\ +\frac{21834.37}{T}+126.8339\times \text{log} T-\frac{1684915}{T^2}\hfill \end{array}$$ (11)

$$\begin{array}{c}\text{log} K_2=-107.8871-0.03252849\times T\hfill \\ +\frac{5151.79}{T}+38.92561\times \text{log} T-\frac{563713.9}{T^2}\hfill \end{array}$$ (12)

The total DIC is the sum of the concentrations of the carbonate ions (62)

$$\text{DIC}=\text{pC}{\mathrm{O}}_2\ast K_{\mathrm{H}}\ast (1+\frac{K_1}{\left[{\mathrm{H}}^{+}\right]}+\frac{K_1\times K_2}{{\left[{\mathrm{H}}^{+}\right]}^2})$$ (13)

where Pco₂ is the partial pressure of CO₂ in equilibrium with the solution (atmosphere), K_H is the Henry equilibrium coefficient, K₁ and K₂ are the first and second dissociation constants of carbonic acids, and [H⁺] is the hydrogen ion concentration (moles per liter). These calculations are available in the Supplementary Materials (data S4).

Data processing

Laboratory-derived adsorption data were compiled into a standardized format to create a temperature- and pressure-dependent CO₂ adsorption matrix (fig. S11). The calculated adsorption values were interpolated across a regular grid using monotonic cubic Hermite interpolation (PCHIP) to construct a 2D adsorption surface, Q (T, P), for each temperature-pressure pair. The interpolation was carried out in two steps: First, CO₂ adsorption (Q) was interpolated across pressure steps (1-Pa resolution) for each temperature; second, the resulting interpolated adsorption values were used to generate a temperature-smoothed matrix at 1-K intervals from 271 to 343 K. A 3D wireframe plot of the adsorption matrix was created to visualize trends across the full temperature and pressure space (fig. S11). This dataset serves as input for further modeling (data S3).

To characterize vertical soil temperature and CO₂ concentration profiles at high temporal resolution, the field-measured temperature and Pco₂ data were first aggregated to hourly means to reduce noise while preserving natural variability. For each timestamp, temperature values measured at discrete soil depths (5, 25, 50, and 125 cm) were used to fit a cubic polynomial curve. The fitted curve was then used to interpolate continuous temperature estimates with soil depths at 1-cm intervals, generating synthetic soil temperature profiles.

The laboratory-derived adsorption Q (T, P) matrix was used to determine the amount of CO₂ adsorbed (Q, millimoles per gram) within the vertical soil profiles. For each hourly timestamp in the field dataset, CO₂ concentrations and temperatures at the soil surface and at depth intervals were used as inputs to a bilinear interpolation function, which returned the corresponding adsorption value from the laboratory matrix.

The resulting gravimetric Q values (millimoles per gram) were further converted to volumetric units (mmol 10,000 cm⁻³) using depth-resolved soil bulk density and gravel content obtained from core samples. Gravel corrections accounted for the reduced adsorption volume based on the known gravel density (2.6 g cm⁻³). Final volumetric CO₂ adsorption profiles were generated for each site and depth interval.

Supplementary Materials

The PDF file includes:

Figs. S1 to S22

Table S1

Legends for data S1 to S6

References

Other Supplementary Material for this manuscript includes the following:

Data S1 to S6