The four sites were chosen to represent a mixture of land use and geologic history, with Illinois and Nebraska each having paired agricultural and restored prairie sites on similar soils. These sites are part of the Critical Interface Network (CINet), which examines the role of hydrological, biological, ecological, geological, and chemical processes within the critical zone of these intensively managed landscapes (Kumar et al., 2023). The Illinois sites (Illinois, USA) are both located in the Upper Sangamon River Basin (Figure 1), and include ILAG (Illinois Agriculture), which is an active farm with an annual crop rotation of corn and soybean, and ILPR (Illinois prairie), which is a tallgrass prairie that was restored in 2007. The Illinois soils have a similar average bulk density of 1.47 ± 0.20 g cm⁻³ in agricultural (ILAG) and 1.47 ± 0.22 g cm⁻³ in prairie (ILPR). However, the ILAG soils have a higher average bulk density near the surface at 1.31 ± 0.10 gm cm⁻³ in the 0–20 cm layer, compared to the ILPR at 1.14 ± 0.20 gm cm⁻³ in the 0–20 cm layer. ILAG is less dense at depth with a bulk density of 1.44 ± 0.05 gm cm⁻³ in the 60–110 cm layer. In contrast, ILPR soils have an average bulk density of 1.65 ± 0.00 gm cm⁻³ in the 60–110 cm layer. The ILAG soil is Sable silty clay loam whereas the ILPR soil is Elliott silty clay loam (USDA) and parent material is loess over glacial till. The average annual air temperature between 1995 and 2023 was 11.5 °C, and the sites receive an average of 101.9 cm yr.⁻¹ of precipitation and are not irrigated (MRCC, 2024). However, 2022 and 2023 were both drought years, receiving only 88.8 and 84.0 cm yr.⁻¹, respectively (NDMC, 2024; USDA, n.d.; NOAA, n.d.). Both the IL sites have slopes less than 0.03 m m⁻¹, and the landscape is generally uniform in this regard (OpenStreetMap Contributors, 2023). Both ILPR site is 365 m from the Sangamon River, whereas the ILAG site is tile drained at 1 m to the adjacent drainage ditch (QGIS, 2023).

The Nebraska sites (Nebraska, USA) are both within the Glacier Creek Preserve (Figure 1), a 4 km² watershed with a mix of restored prairie and agricultural fields. The research stations are at the top of adjacent knolls, and both drain to Glacier Creek, which has a median discharge of < 0.01 m³ sec⁻¹. The slopes of the NE sites are similar on top of the knolls (< 0.05 m m⁻¹), reaching a max slope of 15% on the hillslope between the sites and the creek (Dere et al., 2019). The soils are Contrary-Monona-Ida Complex (Ryan et al., 2018), and more specifically, the agricultural soils are Contrary silt loam, and the prairie soils are Monona silt loam. The parent material of the soils is loess, and the Ag soils have a bulk density of 1.11 ± 0.09 g cm⁻³ versus 1.14 ± 0.05 g cm⁻³ of the prairie soils with both sites having denser soils closer to the surface. The sites receive 78 cm yr.⁻¹ of precipitation with an average annual temperature of 10 °C (Dere et al., 2019). The NEPR (Nebraska prairie) site was restored in 1970 after a century of agriculture and is maintained with periodic (3-year) burns, (Dere et al., 2019) with the last burn in 2024. NEAG (Nebraska Agriculture) has been in annual corn-soybean rotation for the duration of the study and is not tile drained nor irrigated, and has a relatively deep water Table (20 m). ILAG and NEAG were both planted with corn in 2022 and soy in 2023.

Monitoring was conducted with a sensor array installed within the soil profiles measuring hourly soil gases, volumetric water content (VWC), and temperature. Supporting atmospheric sensor data were collected at 1 min to hourly time intervals. In addition to the sensors, manual sampling was conducted on a two-week basis for both soil and atmospheric data. The installation was completed by May 2022 at all sites, with additional atmospheric data available from previous research initiatives. The data used in this work is from May 2022 through Sep 2023 for ILAG, Aug 2021 to Sep 2023 for ILPR, (soil gases operational May 2022), and Jan 2022 to Nov 2023 for NEPR and NEAG (VWC operational May 2022, soil gases operational Sep 2022).

To minimize soil disturbance, the sensor arrays were installed into intact soil at depths of 20, 60, 110, and 180 cm. Soil pits approximately two meters deep, three meters long, and 1 m wide, were excavated using a backhoe or by hand (NEPR), allowing for horizontal installation of sensors into the pit wall at the desired depths. Using a hand auger, holes were cut into the side of the profile at an 45^o angle allowing PVC pipes to be inserted so the sensors were stacked vertically as seen in the diagram presented in Figure 2. Additionally, paired temperature and moisture sensors were directly inserted into the soil pit wall at each depth and the cables were sleeved with a protective flexible conduit. Finally, gas wells were installed horizontally into the soil pit wall to allow for manual sampling of gases for laboratory measurements of concentrations and stable isotope values (Frantal, 2024) using methods modified from Brecheisen et al. (2019). However, data from these manual samplings are not included in this study. After installation of all sensors was completed, the soil pit was backfilled replacing the removed soil from each depth in the original positions. This method of installing soil sensor arrays, while intensive in its initial construction, offers substantial benefits, including the ability to easily remove sensors installed in the PVC pipes for calibration, maintenance, and replacement which allows for shorter data gaps and reduced maintenance costs. A detailed installation guide, parts list, and data logger code is available in Supplementary materials.

Each depth hosts a CO₂, O_2, and volumetric water content (VWC) and temperature sensor, and each sensor has a built-in temperature probe for parameter corrections. The continuous CO₂ measurements were taken with an Eosence eosGP sensor with a measurement range of 0–20% and an accuracy of 3.5%. CO₂ data with known sensor errors were removed manually if a problem in the field was noted such as damage to equipment. Some data display characteristics of a sensor construction design flaw which resulted in rapid drops of the measured CO₂ values primarily during the first winter (2022), and these data were removed from the 180 cm data at NEAG and NEPR using an automated detection method. Specifically, we removed points that showed an average hourly drop of greater than 67 ppm hr.⁻¹ in NEAG and 450 ppm hr.⁻¹ in NEPR over a 12 h period. The oxygen sensors are Apogee SO-110 with a range of 0–100% and accuracy of better than 1%. Both the O₂ and CO₂ sensors were inserted into an airtight 63 mm OD PVC pipe perforated at the depth of measurement with the CO₂ sensor on the bottom and the O₂ sensor on top. Campbell Scientific CS655 VWC/temperature sensors were installed directly into the side of the soil pit to record both VWC and soil temperature. The VWC was corrected for temperature based on the manufacturers default equation. Atmospheric sensors were installed next to the soil pit and included a Campbell Scientific EE181-L air temperature sensor, Campbell Scientific 03002-L wind speed and direction sensors, and Campbell Scientific TE525WS precipitation sensor. The observed atmospheric temperatures at the weather stations had values outside of the regional recorded range and were scaled to the maximum and minimum temperatures recorded over the measurement years for the local county using National Weather Service data (National Weather Service, n.d.). Additionally, the Nebraska sites had Campbell Scientific SP230 solar radiation sensors, which measure 360 to 1,120 nm wavelengths, whereas the solar radiation monitoring for the Illinois sites used a Kipp and Zonen CNR4 Pyranometer shortwave radiation sensor (300–2,800 nm) at the Goose Creek Eddy Covariance Flux Tower, located 16.3 and 23.5 km from ILAG and ILPR, respectively, (Hernandez Rodriguez et al., 2023).

The Normalized Difference Vegetation Index (NDVI) was calculated using Planet Lab data (Planet Team, 2022). The planet lab images have a 3 day return period with occasional gaps due to cloud cover. To quantify NDVI an area adjacent to the sensor array, but not including the sensor array, was chosen to represent the field conditions. The image date, location, and NDVI interpolated to a 1 day resolution were then exported and used to interpolate NDVI at a 1 h timescale.

We used a previously published model (Cerling, 1984; Winnick et al., 2020) to simulate the 1-D production of CO₂ within each soil profile (Equation 1). This model is based on Fick’s second law of diffusion and has been used in many past studies (Davidson and Trumbore, 1995; Tang et al., 2005; Davidson et al., 2006; Winnick et al., 2020). The model equation is

where C is the observed concentration of CO₂ (mol cm⁻³), t is time (s), D* is effective diffusivity (cm² s⁻¹), z is depth (cm), and P is net CO₂ production (mol cm⁻³ s⁻¹) (Table 1). This temporally-transient equation was solved using a first-order centered difference approximation at a 1 h timestep for the 20 cm, 60 cm, and 110 cm depths (Supplemental information). Effective diffusivity was estimated as a linear function (Equation 2) of volumetric water content (VWC, cm³ water cm⁻³ dry soil) and dry soil diffusivity or D (cm² s⁻¹) (Campbell, 1985; Steefel et al., 2015; Winnick et al., 2020) as follows:

where VWC_o (cm³ cm⁻³) is the residual soil water content, which was assumed to be the minimum value observed across all soil depths at a site. The VWC (cm³ cm⁻³) at saturation or porosity (VWC_sat, cm³ cm⁻³) was assumed to be the maximum VWC at each site across all soil depths. T is the soil temperature in degrees Kelvin (K) and 283 is a reference temperature (Winnick et al., 2020). Dry soil diffusivity D (cm² s⁻¹) was calculated from soil physical properties using the following Equation 3 (Davidson et al., 2006).

where D_CO2 is the diffusion coefficient of CO₂ in air at STP (0.162 cm² s⁻¹), and ∅ g is the total soil porosity (cm³ cm⁻³). Total soil porosity was calculated from (Equation 4) bulk density (BD, g cm⁻³) and weighted average particle density (PD, 2.61 g cm⁻³) using the following equation (Davidson et al., 2006).

The bulk density was measured at all sampling depths (20, 60, 110, and 180 cm) using the core method. The core volume was 98.17 cm⁻³ across all sites; however, replicates at ILAG were made with an additional core volume of 242.63 cm⁻³. A bulk density measured at each depth was used in the model.

Each site was characterized using summary statistics such as the mean ± 2 standard deviations of CO₂ and CO₂ production. To determine when and if factors such as temperature, VWC, solar radiation, and plant cover, covary with CO₂ concentrations or production, we calculated Pearson’s correlations between these variables for each soil depth over a range of timescales and data resolutions (Pearson, 1920). Specifically, CO₂ production and concentration were tested for correlations with temperature, solar radiation, and VWC at hourly, mean daily, and mean monthly scales for each soil depth across the entire data set and by season. The concentration and production of CO₂ were also tested for correlations with daily NDVI, both seasonally and across the data record, as hourly data was unavailable. For correlations with CO₂ concentration, the logarithm of CO₂ (log₁₀ CO₂) was used as CO₂ data is often right skewed. Additionally, as the environmental variables are interrelated, we tested for correlations between temperature, VWC, solar radiation, and NDVI. Furthermore, multiple linear regressions between CO₂ and NDVI, temperature, and VWC were applied to determine how interactions between parameters affected CO₂ responses. We used t-tests to determine statistical significance (95% confidence) (Student, 1908). Beyond correlations, we also studied instances of hysteresis between soil CO₂ and soil VWC observed over wetting events to determine the effects of storms on soil CO₂. These individual events were analyzed in conjunction with summary statistics and antecedent conditions to infer how the system reacted to wetting in different seasonal conditions. The data analysis was conducted in R statistical language version 4.4.3 (R Core Team, 2025).

The CO₂ concentration was generally highest at deeper depths, 110 cm or 180 cm, across all sites, with CO₂ at 180 cm often falling just below the peak at 110 cm during mid to late summer. These peaks reached 59,551 ppm, 44,825 ppm, 25,201 ppm, and 33,492 ppm in ILAG, ILPR, NEAG, and NEPR, respectively, (Figure 3). The minimum recorded values at each site were 255 ppm, 452 ppm, 374 ppm, and 113 ppm in ILAG, ILPR, NEAG, and NEPR, respectively, all of which were at 20 cm. Conversely the largest seasonal variability of CO₂ ppm was at the 110 cm depth. The mean CO₂ concentration at 20 cm was 5,485 ± 12,270 ppm at ILAG, 3946 ± 5,074 ppm at ILPR, 2936 ± 5,734 ppm at NEAG, and 2,453 ± 6,324 ppm at NEPR. Despite having higher concentrations at the 20 cm depth in agricultural soils, CO₂ production has the opposite pattern. The mean CO₂ production at 20 cm at each site was 0.93 ± 5.06 g CO₂ m⁻³ h⁻¹ in ILAG, 1.99 ± 4.01 g CO₂ m⁻³ h⁻¹ in ILPR, 0.44 ± 2.67 g CO₂ m⁻³ h⁻¹ in NEAG, and 0.58 ± 3.18 g CO₂ m⁻³ h⁻¹ in NEPR (mean ± 2x standard deviation). Therefore, although ILPR soils are producing more CO₂ at 20 cm then ILAG, the rate of evasion must be higher at ILPR to account for the lower concentrations observed at the ILPR sites.

NDVI peaked in June through August in NE and IL with both site-pairs showing higher peaks in the agricultural sites in 2023 and prairie sites in 2022. NDVI peaks reached 0.86 in NEAG, 0.81 in NEPR, 0.93 in ILAG and 0.85 in ILPR. The Planet Lab NDVI data correlated significantly to both air temperature (ILAG R = 0.39, ILPR R = 0.52, NEAG R = 0.59, and NEPR R = 0.68) and solar radiation (ILAG R = 0.13, ILPR R = not significant, NEAG R = 0.15, and NEPR R = 0.17). Overall, NDVI shows a consistent seasonal pattern varying between 0.20 in winter and 0.93 in summer with sharp seasonal changes in spring and fall.

The summer months had the largest daily fluctuations in CO₂ concentrations. The amplitude of the diurnal cycles seen in the CO₂ concentrations were up to 25,000 ppm at 20 cm and persisted to depths of 180 cm across all sites. When comparing the daily cycle of CO₂ to that of solar radiation we used minimum pCO₂ per day. This was done as the maximum daily CO₂ values are generally at night causing a bias to midnight if there was an increasing or decreasing trend in the data. Most solar radiation peaks were within 2 h of the minimum daily CO₂ concentration at 20 cm across the sites in summer months, with a minimum CO₂ concentration between 11:00 and 13:00, whereas solar radiation maximums were later at 12:00 to 13:00 (Figure 4).

Soil wetting events show a distinct response in CO₂ concentrations, with sharp declines often occurring at the onset of wetting, then rapidly returning to pre-storm or elevated conditions (Figure 5). These rapid dips in CO₂ concentration were present at all depths when VWC increased and were more pronounced when antecedent CO₂ concentrations were higher and wetting events were larger. Similar patterns were seen in O₂ with soil wetting events resulting in rapid increases in soil O₂, when soil O₂ was depleted before the rain event. In essence, these wetting events caused O₂ to increase towards atmospheric values and CO₂ to decrease towards atmospheric values because rain water entering the soil was previously equilibrated with atmospheric gas concentrations. O₂ approached but never exceeded typical atmospheric values, and CO₂ always remained elevated above atmospheric values (Figure 5). Hysteresis plots between soil CO₂ and soil VWC showed that some wetting events led to prolonged increases in CO₂ during the following days, particularly those in mid-summer when NDVI was high (Figure 6).

NDVI had a significant positive correlation to Log₁₀ CO₂ concentrations in NE (NEAG R = 0.71, NEPR R = 0.72) and IL (ILAG R = 0.78), and log₁₀ CO₂ production NE (NEAG R = 0.27, NEPR R = 0.55) and IL (ILAG R = 0.60) with the NEPR having a stronger coefficient of determination than NEAG, unfortunately not enough data was available in ILPR (Table 2). Furthermore, log₁₀ CO₂ concentrations at 20 cm show statistically significant (p < 0.05) negative correlations with solar radiation when preforming a correlation for each day in summer, yielding mean correlation coefficient of the daily correlations at each site of −0.60, −0.35, −0.46, and −0.52 in ILAG, ILPR, NEAG, and NEPR, respectively. Similarly, CO₂ production had significant (p < 0.05) positive daily correlations with solar radiation in summer, yielding mean R of the daily correlations at each site of 0.43, 0.54, 0.42, and 0.43 in ILAG, ILPR, NEAG, and NEPR, respectively. We also detected correlations between soil temperature and log₁₀ CO₂ concentration at 20 cm in NE (NEAG R = 0.68, NEPR R = 0.73) and IL (ILAG R = 0.75, ILPR R = 0.37) and weaker but still statistically significant correlations between soil temperature and CO₂ production at 20 cm with R of 0.36, 0.53, 0.32, and 0.20 in NEAG, NEPR, ILAG, and ILPR, respectively. Furthermore, we find that correlations between soil temperature and CO₂ concentration or production are both inconsistent with depth and data coarseness (Table 3). In general, CO₂ concentration correlations with soil temperature increase with depth at ILPR and NEAG but fluctuate at ILAG and NEPR. Soil temperature correlations with production generally decrease with depth, however the 60 cm correlations are generally negative (Figure 7, Supplementary Table 1). Additionally, correlations become weaker when only considering summer months at all sites except for CO₂ production at ILAG at 110 cm depth. We also note an increase in the correlation coefficient and decrease in Akaike Information Criterion between soil temperature and CO₂ concentration or production when coarser data is used, such as daily or monthly data (Table 3).

VWC shows an annual cycle with declining soil moisture through the summer growing season (yellow shading in Figure 8), with pulses of rapid wetting due to rain events followed by drying. Between both IL sites, 99 individual events over the course of the study increased VWC at 20 cm, of which only 42% reached 60 cm, 27% reached 110 cm, and 12% reached 180 cm. Across both NE sites, 64 events increased VWC at 20 cm with 34, 16, and 8% of the 64 events reaching 60, 110, and 180 cm, respectively. In both regions, the AG sites had at least 1.4 times more wetting events at depths of 60 cm or greater than PR sites. In NE, the deepest 180 cm soils were wettest in all seasons except spring when 20 cm was wettest. ILPR generally had wetter conditions at the 20 cm and 60 cm depths, however 20 cm was the driest in summer. At ILAG, soil was wettest at the 60 depth for all seasons. Seasonal differences in VWC were highest in the shallower soils, with the largest variability at 20 cm. Specifically, VWC ranged from 0.123 to 0.499 in NE and from 0.2 to 0.435 in IL at the same depths. Additionally, NEAG had the largest swings in VWC, and ILAG had the smallest (Figure 8, Supplementary Table 2). Correlation between CO₂ concentrations and VWC were generally weak especially in the shallow soils (20 and 60 cm) and were stronger in summer than across the entire year (Figure 9). However, only the summer (Jun-Sep) correlations for ILPR at 20 cm and NEAG at 110 cm had correlations above 0.50 (Figure 9, Supplementary Table 3). Furthermore, the correlations between CO₂ production and VWC were generally weaker than those with CO₂ concentration. Of the sites, only ILPR showed correlations above R = 0.50 (Figure 9, Supplementary Table 3).

Temperature also showed a seasonal and daily cycle driven by solar radiation, and correlations between radiation and air temperature were noted in IL (R = 0.49 AG, R = 0.42 PR) and NE (R = 0.41 AG, R = 0.42 PR). In Nebraska, air temperatures ranged from −27 to 40 °C, whereas Illinois was on average warmer and ranged from −24 to 37 °C. Daily swings in air temperature were generally lowest in December and January and largest in April and October, with daily temperature swings as large as 32 °C and as small as 1 °C. Peak air temperatures were offset from solar radiation by an average of 2.5 h after maximum solar radiation in IL and 1.9 h in NE. This was consistent throughout the year in NE, whereas in IL some winter months had small temporal offsets. Soil temperatures at the agricultural sites showed greater variability, especially at the shallowest 20 cm depths where soil temperature ranged from −8.8 to 30 °C in NE and −1.8 to 32 °C in IL. Additionally, 20 cm was the only depth to show a clear diurnal oscillation in soil temperature. Deeper soils showed seasonal variability, which was muted and lagged with depth (Figure 10).

Our results underscore the strong influence of plants, likely from root respiration and microbial respiration stimulated by root exudates, on soil CO₂ dynamics. This influence is evident across seasonal and daily timescales, especially during the growing season when plant phenology and activity plays a dominant role (Table 3). Previous studies have highlighted the importance of autotrophic drivers in soil CO₂ production (Lei et al., 2023; Liu et al., 2024), as root growth processes (root respiration and microbial respired exudates) can account for 29–90% (Dugas et al., 1999; Li et al., 2021) and 22% of soil CO₂ fluxes in prairie and corn systems, respectively, (Nichols et al., 2016) and indices like Enhanced Vegetation Index strongly predict shallow soil CO₂ in ecosystems such as rocky mountain meadows (Winnick et al., 2020). Similarly, soils with active root systems under tree cover exhibit significantly higher respiration compared to adjacent areas without roots (Tang et al., 2005). Within our data, soil CO₂ production and concentration show distinct and sharp seasonal shifts that indicate plant interactions are a significant driver (Figure 3). Primarily, soil CO₂ concentration in the agricultural sites correlate more strongly to the sharp seasonal trends of NDVI than the gradual change in temperature (Table 2). This indicates that the onset of plant growth is a stronger driver than temperature driven microbial respiration alone. This stronger correlation to NDVI is seen at both agricultural sites and is consistent across measurement years, indicating this is a consistent pattern across the agricultural landscapes.

Further support for the importance of plant mediated respiration is seen in the daily oscillation of CO₂ concentration and production within the data and the negative correlation between solar radiation and CO₂ concentrations at 20 cm in summer months. These daily oscillations are roughly 12 h offset from solar radiation availability, resulting in the lowest CO₂ around peak solar radiation and the highest CO₂ at night (Figure 4). Hirano et al. (2003) and Tang et al. (2005) observed similar patterns with CO₂ minimums occurring during the day. Tang et al. (2005) also noted soil CO₂ production peaks and photosynthesis peaks were offset by 7–12 h in forested environments which was attributed to the slow transport of carbohydrates from leaves to the roots. This matches with other research that shows grasses have a 12.5 ± 7.5 h (mean ± standard deviation) time lag in carbohydrate transport from leaves to the roots through the phloem (Kuzyakov and Gavrichkova, 2010). The presence of these oscillations in CO₂ concentrations, which potentially match the pulses of carbohydrates from roots, means that during the growing season the availability of easily respired C limits respiration instead of temperature. As this represents a potentially large shift in how many models handle soil CO₂ respiration, further research is needed to corroborate these findings.

Soil temperature is likely not the primary driver of measured total soil respiration, as soil temperature does not show diurnal oscillations at depths greater than 20 cm (Figure 10), while CO₂ concentrations show these oscillations at much deeper depths. If temperature influences on microbial respiration were the primary driver of soil CO₂ concentration variability, then we would expect soil temperature and CO₂ concentration to jointly oscillate across the same depths, as temperature enhances microbial activity and drives respiration (Han et al., 2007). Similar mismatches between temperatures and CO₂ concentration variability have been documented in the literature (Makita et al., 2018; Liu et al., 2006), and recent findings have shown that diurnal cycles play an important role in soil respiration response to rising temperature with larger daily temperature swings resulting in muted responses of respiration (Adekanmbi and Sizmur, 2022; Adekanmbi et al., 2022). Furthermore, these CO₂ oscillations are present in the late spring, summer, and early fall at all sites but not in early spring, late fall, or winter when plant cover is low, indicating that these oscillations are caused by plant cover. Additionally, our data show that the daily oscillations of CO₂ concentration persist to depths of 60, 110, and 180 cm, far deeper than daily soil temperature swings, which are only present to 20 cm (Figures 4, 10). These findings are strong indicators that although temperature plays a role in CO₂ production in soils, plant induced respiration represents a primary driver during the growing season. This research is not the first study to show plant mediated soil CO₂ production (Liu et al., 2006; Han et al., 2007; Makita et al., 2018; Li et al., 2021; Lei et al., 2023), it does so at a depth (1.8 m) and magnitude (25,000 ppm) of CO₂ oscillations not often seen. Our data show that accurate predictions of soil respiration at sub daily scales rely on the influence of plant soil interactions such as exudates to accurately understand the respiration rates of microbial communities and should be considered when determining soil C budgets even at these rapid time scales or within deep soils.

These findings suggest that partitioning soil respiration into direct root respiration or root-exudate stimulated microbial respiration is important to fill current knowledge gaps within the soil carbon cycle. This could be accomplished through isotope-based age classifications or other partitioning methods as plant mediated respiration may come from a multitude of sources that would have different responses to climatic or land use changes. Additionally, the contributions of plant species, genus, or functional type should be considered in future work as previous studies have shown that individual species (Johnson et al., 2008) or root sizes (Bahn et al., 2006) may contribute to respiration at higher rates than others. This difference will be important to consider when planning restoration projects, determining how species composition changes post agricultural abandonment may alter soil carbon (Ladouceur et al., 2023) and how climate shifts in plant ranges could alter soil carbon feedback. Furthermore, the links between root exudates, plant diversity, and microbial respiration are an important component to consider as many studies have shown connections between species richness, microbial activity, and soil carbon storage all of which are susceptible to rapid changes with both climate and land use change (Zak et al., 2003; Lange et al., 2015; Steinauer et al., 2016; Chen et al., 2019; Prommer et al., 2020). These plant and microbial interactions in the soil represent a diverse array of possible outcomes that depend on human as well as climatic drivers and therefore need to be better characterized so that we can accurately predict carbon stocks and cycling.

Temperature effects on soil CO₂ production are difficult to determine, as many factors co-vary with both temperature and soil CO₂ production (Han et al., 2007). In the past, temperature was viewed as a primary driver of soil respiration (Ouyang et al., 2015) as it speeds up microbial respiration through its effects on enzyme-mediated reactions (Conant et al., 2011). However, there are many confounding influences on soil respiration that may make temperature an indirect and potentially inaccurate determining feature of soil respiration (Auffret et al., 2016; Lei et al., 2023). The common covariance between temperature and other seasonal changes led to the significant correlation between soil temperature and CO₂ production seen in our data (Figure 7, Table 2). This is easily seen when we compare hourly to daily and monthly correlations between soil CO₂ production and 20 cm soil temperature as R values often double for monthly compared to hourly data (Table 3). Furthermore, despite a correlation existing, the magnitude and direction of this correlation change with depth is not consistent at individual sites through time (Figure 7, Supplementary Table 1). Additionally, during summer months (Jun-Sep) the temperature to CO₂ production correlations become weaker or not statistically significant (Figure 7). The inconsistency of the temperature and CO₂ production correlation plus the bias in coarser data leads us to similar conclusions as past studies (Conant et al., 2011; Winnick et al., 2020), which find that temperature is likely not a strong direct driver of soil CO₂ production during summer months. This finding is further supported by the lack of diurnal temperature patterns in the deeper soils despite diurnal CO₂ patterns. Future studies are necessary to explore specific mechanisms related to temperature as the covariations among temperature and other factors pose a large unknown in soil respiration modeling and future climate feedbacks.

Like temperature, VWC does not show a strong correlation to CO₂ production; furthermore, the correlations we do find are inconsistent in both their direction and magnitude (Figure 7, 9, Supplementary Table 3). The nonsignificant effects of VWC are similar to the findings of other research (Bouma et al., 1997). This inconsistency suggests that VWC may affect a range of factors that alter CO₂ production, causing differences in the correlation depending on which factors are currently limiting. Furthermore, these inconsistencies in correlations suggest that antecedent conditions play an important role in determining how the system reacts to rain. For example, VWC is not a strong driver of CO₂ production during the summer drought (Figure 9, Supplementary Table 3). This finding is different from past research, which has found that VWC becomes a more important driver of soil CO₂ production with drought, particularly at the expense of plant-related metrics (Huang et al., 2014). This divergence from past research might be due to ecosystem differences as these studies were conducted in deciduous forests and mountain meadow environments, or methodological differences such as the timescales of the data. Alternatively, the moderate severity of the drought at our sites could be such that plants were not sufficiently stressed for VWC to become the limiting resource in surface soils.

Rain events over the monitoring period were relatively infrequent as the region experienced drought conditions during the summer and fall seasons in 2022 and 2023. However, the data still capture some rain events and show different patterns depending on antecedent conditions. Often, responses to rain in soils are interpreted as a stimulation of microbial respiration (Griffiths and Birch, 1961; Unger et al., 2010; Winnick et al., 2020) however, physical processes play a prominent role in determining the immediate responses to rain events, while biological processes affect respiration hours to days after the event. At the onset of storms in the growing season where soil CO₂ was greater than 5,000 ppm and VWC increased, our data showed a rapid drop in CO₂ lowering concentrations by up to 7,000 ppm. This drop is interpreted as the equilibration of rainwater with soil gases, resulting in dissolution of soil CO₂ into the aqueous phase (carbonic acid). After a wetting event, soil CO₂ concentration often returned to pre-storm or higher levels within a few hours, reaching a new (often elevated) state which could last hours to days. These responses are due to biological responses often referred to as the Birch effect which describes an increase in microbial activity after a wetting event (Birch, 1958; Griffiths and Birch, 1961).

Our data show evidence of CO₂ production responses to events during the growing season when the CO₂ concentration responses were much larger. Post-wetting respiration CO₂ concentration responses often appeared to be delayed by hours due to the equilibration of atmospheric rainwater with soil gases. Without the high resolution of our data, these could be misinterpreted as a delay in the microbial response if captured during infrequent sampling, neglecting the buildup of CO₂ in the aqueous phase. As soil CO₂ is dynamic, representing the balance between production within the soils and losses due to diffusion to the atmosphere or from the dissolution of weatherable minerals, these physical controls on soil CO₂ need to be considered when interpreting changes in soil CO₂ concentrations to accurately predict how soil respiration responds to environmental conditions. To mathematically model soil CO₂ production, soil VWC and CO₂ concentration were used to predict the rate at which CO₂ is being produced or consumed (Davidson and Trumbore, 1995; Davidson et al., 2006; Steefel et al., 2015; Winnick et al., 2020). This production can then be interpreted as the net result of environmental factors such as microbial respiration rate and the dissolution of weatherable minerals (Gallagher and Breecker, 2020). However, this interpretation is incomplete, as CO₂ dissolution into carbonic acid is neglected, this additional sink could be directly accounted for if soil pore water pH was measured. If these models are used without accounting for this sink, the rate of soil respiration and the magnitude of the response to wetting events would be underestimated by the magnitude of this sink.

Previous studies have shown that plant growth stage plays an important role in how soil respiration reacts to changes in temperature and VWC (DeForest et al., 2006; Winnick et al., 2020). Here we find that this role is amplified in prairie environments and that plant mediation of soil respiration shows distinct differences in the magnitude of CO₂ production between agriculture and prairie environments. This is seen in the stronger correlations between NDVI and CO₂ production at prairie in the NE sites. Additionally, larger variability of CO₂ concentration and significantly higher mean CO₂ concentrations in spring and fall months were noted in the NE prairie site. These differences were likely due to the ability of the PR sites to respond more readily to favorable conditions during colder months as the plant composition is more diverse and made up of primarily perennial species (Prairie et al., 1993; Chimner and Welker, 2005; Polley et al., 2005; Johnson et al., 2008). In contrast, agricultural soils are planted with an annual monoculture that is harvested in the fall for the winter and planted again in late spring, reducing the ability of soil microbes to respond to favorable conditions as plant roots and associated exudates are not present. Beyond the differences in the plant communities, the lack of ground cover in agricultural sites leads to often colder soils in the fall and winter, likely minimizing the microbial response during colder months. In the summer months, we note significantly higher CO₂ production in prairie sites.