Warming Increases Soil CO2 Flux by 42-59%

Experimental warming resulted in substantial, larger than expected increases in soil respiration in plots that were experimentally warmed relative to the unwarmed control plots (Figs. 1 and 2). Warming increased CO2 flux by 42% and 59% in Mid and Lower slope positions, respectively (paired T-test, df = 24,521 pairs [Lower slope]; df = 15,635 pairs [Mid slope]). In addition to warming-induced increases in soil respiration means, warming increased the variability of the rates, which may represent an increased variability in environmental conditions or a synergy between warming and soil moisture controls. The positive response of CO2 flux to in situ warming was substantially higher in the Upper slope position relative to the lower topographic positions (204%; df = 12,910 pairs). For two of the paired plots (Lower and Mid slope), the warming-induced increases in respiration rates were in the upper range of what has been reported for higher latitude forests; however, the relative responses of the Upper slope were greater than rates observed in any field warming experiment, regardless of ecosystem type or methodology7. Given the already high soil respiration fluxes characteristic of this and other tropical forests7, warming resulted in a large amount of additional CO2 released to the atmosphere, with 6.5, 9.7, and 81.7 Mg CO2-C ha-1 yr-1 more CO2 being released in the warmed plots compared to the control plots in the Lower, Mid, and Upper slope positions, respectively. Specifically, the additional soil CO2 released from the Mid slope was equivalent to the total annual net primary productivity (NPP) of a temperate grassland, and the Lower slope additional CO2 released was equivalent to the total annual NPP of a temperate deciduous forest. For the Upper slope, there is nothing on record that compares to the additional CO2 flux other than the conversion of a tropical peatland forest to an oil palm plantation7,23.

The inclusion of warmed understory plant communities combined with warmer soils in this in situ warming experiment allows for an integrated exploration of above versus belowground controls on soil respiration responses. This inclusion is important because plants grown under warmer temperatures could alter carbon allocation to root biomass and/or root exudates that, in turn, may influence the contribution of root respiration to total soil respiration, as well as the available carbon and fuel for microbial processes24. Acclimation of root-specific respiration under warming conditions could also shift the contribution of roots to total soil respiration25. In addition, changes to the microbial community, microbial activity, changes to abiotic or geochemical conditions, or some combination may drive observed increases in soil respiration rates in response to warming26,27. Understanding how plant and soil responses to warming individually and interactively affect soil respiration is critical for accurately modeling and forecasting feedbacks to carbon cycling and future climate. While spatially variable, live fine root biomass did not differ among warmed and control plots prior to starting the warming treatment; however, after just six months of warming, live fine root biomass was 32% lower in warmed relative to control plots (Supplementary Fig. S1; two-way ANOVA P = 0.04, F = 5.83; followed by Tukey’s test P < 0.05). Further, live fine root biomass of the Upper slope plots did not differ significantly from live fine root biomass in the Lower or Mid slope topographic positions (two-way ANOVA P > 0.05), suggesting cross-slope patterns were not due to topographic differences in root biomass. We observed no significant acclimation of root-specific respiration in response to warming, but, when combined with the reduced root biomass, the contribution of roots to total respiration declined28. Although total and extractable carbon pools weren’t altered by warming, soil microbial biomass carbon did increase significantly, growing by over 50% in the warmed plots relative to the unwarmed control plots29.

Taken together, these results suggest that the large increase in soil respiration rates in response to warming was due to an increase in microbial-derived CO2 efflux, while simultaneously experiencing a concomitant decline in the contribution of root-derived CO2. Similarly, a belowground-only in situ soil warming experiment in Panama found a large increase in soil respiration in response to warming, with increases in CO2 efflux that were primarily derived from microbial sources with no change in root contribution22. It is possible that the reduced allocation to roots by plants observed at our site was driven by warming of the aboveground vegetation, and not by warmer soils. These differential results highlight the potential for above- and belowground interactions to play an important role when considering the key drivers of the response of soil respiration fluxes to warmer temperatures in tropical ecosystems.

Given the especially high rates of soil CO2 efflux, we used several approaches to confirm that our observed emissions increases were indeed primarily driven by warming and not by stochastic spatial variability or measurement error (Supplementary Materials and “Methods” section). Whereas the Upper slope is clearly a “hotspot” for higher soil respiration rates, throughout the study, all plots exhibited extreme “hot moments” of soil respiration, where observed flux values were several times higher than the mean value (Fig. 3), which is characteristic of trace gas fluxes in this system30. While there was high temporal variation throughout the year, we found no significant diurnal variation in soil respiration. We further conducted a spatial survey of 30 locations randomly selected outside of the plot locations across the 40 m × 60 m TRACE area to quantify the spatial variability of soil CO2 fluxes across the TRACE landscape (Supplementary Fig. S2). We unsurprisingly found variability across space but no evidence that the warmed plots were systematically located on landscape-level hot spots. Additionally, pre-treatment data showed no statistical differences in soil respiration rates between paired warmed and unwarmed plots prior to the initiation of warming (Supplementary Fig. S3), and running a generalized least squares model with and then without the Upper slope (i.e., the control-warming paired plots with particularly large differences in CO2 flux rates) did not change the finding that warmed plots had significantly higher soil respiration rates (Supplementary Table S1, Supplementary Materials and “Methods” section). Field sensitivity analyses failed to detect evidence that invertebrate or animal presence in the chamber could have produced such high respiration rates (Supplementary Materials and “Methods” section).

Despite sustained high rates of soil CO2 flux, we found no significant difference in total or available (i.e., extractable) carbon in the surface soils (0–10 cm) of the plots either before or after 6 months of warming29. It remains unclear whether or for how long soil carbon stocks can be sustained at current levels despite such high CO2 flux rates, and it will be critical to determine the source of additional carbon that is fueling elevated soil respiration rates, particularly in the Upper Slope. For example, there could be greater carbon loss from deeper soils31, which could be especially relevant for the Upper slope, where the soils are considerably deeper32. Inputs from the declining fine root pool, altered litter layer decomposition, or a concentration of preferential flow paths in the more aerated upper slopes could also explain the high fluxes33. There could also be abiotic factors influencing soil carbon availability via an increase in desorption reactions or a shift in the physicochemical environment26,34. Ascertaining which of these potential sources of carbon are driving our responses requires further study. That said, regardless of the carbon source, our observed increases in soil respiration rates in response to warming are conspicuously greater than the 95% range of that observed for warming experiments in northern hemisphere forests (12–31% compared to our observed 42–204%)35, suggesting large temperature sensitivity of the tropical forest carbon cycle. Further, there were notable increases in the variability of soil respiration at higher temperatures (Fig. 1). Increased variability in soil CO2 flux under chronically warmer conditions has substantial implications for the equilibrium soil C stock size if highly variable emissions are not counteracted by similarly variable and large soil C inputs. These responses in both the mean and variation provide critical insight into quantifying tropical forest feedbacks to climate change in a range of conceptual and numerical models, and also contradict the supposition that tropical forests may be relatively insensitive to warming. In addition, our observed increase in microbial biomass carbon (~50%) was greater than the increase observed in the soil warming experiment in Panama or any northern hemisphere ecosystem22, and is contrary to observations from tropical elevation studies19, highlighting the exceptional value of additional data that elucidate in situ tropical forest responses to warming in global assessments19,22,35.

Because temperature and moisture often co-vary, we also explored how the application of the warming treatment affected both soil temperature and moisture at the site. Soil temperatures increased similarly across the warmed plots at all topographic positions in response to experimental warming and were on average 3.99 °C warmer than control plots (Fig. 3a–c). In contrast, while soil volumetric water content (VWC) did not differ among the plots prior to starting the treatment20, soil moisture responded very differently to warming depending on topographic position (Fig. 3d–f). Soils in the Lower and Mid slope plots were drier under the warming treatment (−0.03 g H2O g−1 soil [6.8% lower] for Lower slope and −0.06 g H2O g−1 soil [16% lower] for Mid slope between warmed and control plots, p < 0.001), while the soils of the Upper slope were surprisingly wetter in the warmed plots compared to control (+0.06 g H2O g−1 soil [18% higher], p < 0.001, Fig. 3). There is a range of explanations for the occurrence of greater soil moisture in the warmed Upper slope plot when compared with the control, including lower litter quality combined with extensive drying of the litter layer, which could slow decomposition and create a buffer against soil moisture loss36,37. More work is needed to isolate the cause. Regardless, these data suggest that warming will, in part, regulate soil respiration through interactions with soil moisture, and that in situ experiments could further elucidate interactive controls with longer-term assessments of the connections among soil temperature, moisture, and biological activity.

Within the context of a warming climate, it is worth noting that lowland tropical forests experience a very narrow temperature range compared to all other terrestrial environments on Earth. In this system, a mean experimental increase of 4 °C nearly doubled the total ecosystem temperature range. Specifically, the mean diurnal temperatures of the control plots (~20–26 °C) overlap less than half of the range of the warmed plots (~22–32 °C, Figs. 2 and 4). Therefore, while the magnitude of warming that these soils experienced was in line with other ecosystem warming studies38, the temperature increase relative to the temperature range was well beyond that imposed in other aboveground plus belowground field warming studies and pushed the warmed plots into a new climate space. The future climate warming expected in tropical forests1,4,8,9,10, will also have these large proportional increases in temperature.

Using diurnal averaged soil respiration rates, temperature, water content, and topographic position, we evaluated soil temperature and soil moisture effects on soil respiration using a generalized least squares modeling approach39 (Supplementary Materials and “Methods” section). Model results revealed that, regardless of warming treatment, the relationship between normalized respiration response and soil moisture on Lower and Mid slope positions was significantly negative (p < 0.001, Supplementary Table S1, Supplementary Fig. S4), while Upper slope positions, while also having a significantly negative relationship between soil respiration rate and moisture, exhibited a significantly positive interaction term between normalized respiration response and soil moisture (p < 0.05, Supplementary Table S1, Supplementary Fig. S4). In other words, soil respiration in the lower topographic positions declined during times when soils were wetter (i.e., after rainfall), consistent with decreased soil aeration40, but this relationship was much weaker in Upper slope positions. The change in this statistical relationship was driven by occasional extreme respiration rates, which were observed in Upper slope plots during periods of high soil moisture values, regardless of warming status (Figs. 3 and 4). This points to more complex moisture and temperature interactions than expected for this system.

Over time, there is potential for organisms to adapt or acclimate to their new environment41. Thus, we derived the Q10 of soil respiration for each of the plots (defined as the multiplicative change in soil respiration for every 10 °C increase in temperature, Fig. 4b, Supplementary Materials, and “Methods” section) to explore if organisms in the warmed plots had an altered relationship between warming and respiration rates. Q10 was significantly reduced with warming from a Q10 of 2.51 ± 1.23 (controls) to a Q10 of 0.71 ± 1.30 (warmed; Fig. 4b, p < 0.001). The Q10 values of the control plots fell within the range of the global mean of 1.6–37,42,43,44, while the mean Q10 of the warmed plots falls well below that range. These Q10 values indicate that soil respiration rate decreased per unit of increased temperature in the warmed plots (i.e., Q10 < 1), despite the fact that respiration rates themselves were higher in the warmed plots. In summary, while the sensitivity to temperature was reduced in response to warming, the higher respiration rates suggest a shift toward overall higher basal metabolic rates. Crucially, this finding suggests that the microbial community is exhibiting a highly plastic response to chronic +4 °C warming, with acclimatization or adaptation responses that both compensate for (reduced Q10) and enhance (higher respiration rates) the response to temperature14. This has important implications not only for the relationship between temperature and CO2 efflux for such carbon-rich forests, but also for the use of a single Q10 in Earth System Model forecasts of future climate.

Overall, the higher respiration rates in the warmed relative to control plots, independent of temperature sensitivity, coincided with a significantly higher soil microbial biomass in the warmed plots, which has been shown to correlate with higher soil respiration rates14. While soil respiration in the warmed plots was higher across all temperatures (Figs. 2 and 3), the sensitivity (i.e., slope) of soil respiration in the warmed plots was flat to negative (Fig. 4b), suggesting that warming induced a systemic shift in function. Given the decline in root biomass observed in the warmed plots, we attribute this shift in function to heterotrophic rather than autotrophic responses. These functional shifts could include a change in microbial carbon use efficiency, a shift in the microbial community composition17,45, and/or a change in the distribution of biotic activity vertically through the soil profile31 (e.g., driven by warming-induced changes to soil hydrology).

Contrary to the long-held paradigm that tropical forest responses to increased temperature will be relatively muted in their role in climate change feedbacks11,13, we found large increases in soil respiration rates in response to in situ experimental warming that interacted in complex ways with soil moisture. Due to the amount of carbon released, these increased rates have substantial implications for forecasts of future climate at the global scale. While soil respiration showed signs of acclimation with respect to Q10, the respiration rates were substantially higher in warmed relative to control plots (+42–205%) across all three topographic positions. This study contributes the second field-based observation from an in situ experiment that a warming climate may lead to large increases in CO2 fluxes from carbon-rich tropical soils, adding observations from a a-seasonal wet tropical forest on highly weathered, deep soils to the previous results from a seasonally dry tropical forest with relatively shallow soils22. Taken together, the work demonstrates a potential for large carbon losses from tropical forest ecosystems in a warmer world, and highlights the immense value in evaluating the response of soil respiration to warmer temperatures across a range of tropical forested ecosystems, which include 30 Holdridge Life Zones spanning lowland dry deciduous to montane wet evergreen46,47. Each of these forest types reflects an incredible level of diversity in forest structure, community composition, and soil conditions that make it unlikely these systems will exhibit a single response trajectory to a changing climate7. While experimental warming resulted in a substantial increase in soil respiration at both this forest site and the site in Panama, the underlying controls, as well as the magnitude of the response, appear to be different. For example, our data suggested some acclimation/adaptation of respiration rates combined with a decrease in root respiration contributions to CO2 flux, whereas the seasonally dry forest in Panama found no significant effect of warming on root-derived CO2 flux and no signs of adaptation/acclimation. At both sites, large amounts of additional carbon continued to be released to the atmosphere from warmed plots with no signs of declining in the first 1–2 years after initiating warming, though how this will moderate with time remains to be seen. Nevertheless, understanding the underlying mechanisms driving the response of soil respiration to warmer temperatures is critical for accurate representation of tropical ecosystems in global models and assessing the magnitude and duration of feedback to future climate over the long term.

The TRACE experiment is located in the Luquillo Experimental Forest in northeastern Puerto Rico (18.32465° N, 65.73058° W, 100 m a.s.l., 24 °C mean annual temperature, 3500 mm mean annual precipitation)20. Three open-air replicate 12 m2 hexagonal plots were warmed with an array of six infrared heaters in each plot, increasing soil temperatures by +4 °C above average ambient. Three control plots were created with the same dimensions and infrastructure, but with no warming9. Plots were paired by topographic position within the site (Lower slope, Mid slope, Upper slope). The warming treatment began September 28, 2016, using a feedback control system that acts concurrently and independently at the plot scale to maintain a fixed temperature above the mean ambient temperature of all control plots20. Each plot was instrumented with Campbell CS655 soil moisture and temperature probes at three depths (10 cm, 30 cm, 50 cm; Campbell Scientific, Logan, Utah, USA). On average, we achieved 24-h per day soil warming of 3.69–4.02 °C above ambient at 10 cm depth with significant warming to 50 cm. The warming treatment was stopped on September 5, 2017, when Hurricane Irma and then Maria passed over the island of Puerto Rico, and power to the experiment was lost.

Soil respiration was measured with a soil CO2 flux analyzer control unit (LI-8100A) connected to a multiplexer (LI-8150) and six Long-Term Chambers (8100-104; LiCor Biological Sciences; Nebraska, USA). A permanent PVC soil collar (21.34 cm o.d.) was inserted 5 cm near the center of each plot one month prior to starting measurements. Prior to the start of the warming treatment, several long-term chambers were under repair, and thus, we collected soil respiration measurements from each plot in field campaigns such that three continuous measurements were collected per day per plot for a total of nine days. This survey was conducted with the same long-term chamber and hose/cable extension assembly as used in the automated measurements, as well as the same programming configuration used during the long-term automated measurements, changing the chamber offsets before each measurement to account for the repositioning of the chamber. The field campaigns provided us with a reliable baseline comparison of all the plots prior to the warming treatment, with no pre-treatment differences in soil respiration rates between warmed and control plots during this period (Supplementary Fig. S3). On the day that the warming treatment began (September 23, 2016), one long-term chamber was installed in all six plots, and half-hourly respiration measurements were begun. Equipment ran almost continuously except for a period in November when all six long-term chambers were removed for maintenance. Flux validation efforts included both computational and field checks (Supplementary Materials and “Methods” section).

A further 152 flux measurements were taken in fall 2020 (from November 4 to 19, 2020) across the full TRACE field site to assess spatial heterogeneity in soil CO2 flux rates and characterize whether the rates seen in the six experimental TRACE plots were representative of the overall spatial heterogeneity. We collected CO2 measurements from 30 collars installed outside the TRACE plots and from the 6 collars installed previously in each experimental plot. We took 4 measurements from most collars located outside of the plots, 5 measurements from the collars inside of the plots, and 3 observations from 2 of the collars outside of the plots due to inclement weather during field work. Spatial interpolation was used to create a flux spatial variability map (Supplementary Fig. S2).

Root biomass and soil carbon and nutrient analyses

Three soil cores of 10 cm depth and 5 cm diameter were collected from each of the six plots in March 2016 (prior to warming), March 2017 (6 months after warming), and September 2017 (one year after warming). Within 24 h of collection, fine roots (<2 mm diameter) were hand-sorted from cores and subdivided into live and dead. Roots were then washed and dried to obtain total root biomass for each of the cores. Root biomass data were not obtained from the September 2017 cores due to the imminent arrival of Hurricane Maria; however, the complete suite of biogeochemical analyses was performed.

After roots were removed, total soil carbon concentrations were assessed on oven-dried (60 °C for 48 h) soils using an elemental analyzer (Elementar Americas, Mt. Laurel, NJ, USA). Extractable carbon was assessed by shaking fresh soils (on the same day of collection) with 0.5 M K2SO4 for one hour, and filtering using Whatman GF/F glass fiber filter paper (Whatman International, Springfield Mill, UK). Extracts were analyzed using a Shimadzu TOC-Vcpn/TN-1 (Shimadzu Corporation, Kyoto, Japan).

To determine the effect of infrared warming on soil temperature (10 cm depth), moisture (VWC; 10 cm depth), and respiration, measurements of each variable from paired warmed and control plots were paired in time, and then the difference between the respective variables in warmed and control plots at each time step was calculated. We then used a one-sample T-test in R version 3.5.3 (“Great Truth”, R Core Team 2019) to evaluate whether the differences were >0. The application of the T-test to these data was based on the following: the very large sample size permitted some deviation from the assumption of normality of small-sample T-tests, based on the central limit theorem48. It is important to note that while measurements of temperature and moisture within each time series are not independent, the differences between the paired time series can indeed be considered independent. We compared data from before (March 2016) and after warming (March 2017) with a two-way ANOVA (Treatment*Time), using the mean biomass of three replicate cores per plot to assess treatment effects on root biomass. To evaluate whether topographic position affected live fine root biomass, a second two-way ANOVA (Treatment*Slope) was performed, using the March 2017 data of all replicates per plot to obtain enough replication per slope (upper, mid, lower). For both analyses, Tukey’s post-hoc test was used for pair-wise comparisons within groups of significant main effects. All analyses were conducted with the aov and TukeyHSD functions of the stats package in R version 3.6.0 (“Planting of a Tree”).

We modeled respiration as a function of temperature and moisture using a mixed modeling approach as follows. Using diurnal averaged soil respiration, temperature, and water content, we evaluated treatment effects on soil respiration using a mixed effects model implemented in the ‘nlme’ package in R39 in R version 3.6.0 (“Planting of a Tree”). We log-transformed the respiration rate and used the 2nd-order polynomial fit of soil moisture, based on a priori expectations of the temperature and moisture responses of soil respiration. We evaluated both linear mixed effects, with chamber random effects, and a generalized least squares model with different variance structures and approaches to temporal autocorrelation within a chamber. Model selection was conducted using Bayesian Information Criteria (BIC) to determine the variance and autocorrelation structure for repeated measurements within a chamber across time. After the best variance structure was found, backward selection was applied (using BIC) to determine the best set of fixed effects to include in the model. Parameter estimates for fixed effects (including warming treatment level, soil moisture, soil temperature, and hillslope location) were based on the subset of parameters and interactions that were significant in the best-fit model. The best model was fitted via generalized least squares and accounted for autocorrelation of fluxes across days within a chamber as a first-order autoregressive process with an estimated phi = 0.78, where variance was estimated as a function of measurement temperature and was allowed to vary between chambers49. The best model excluded interactions between soil temperature and hillslope location, as well as the interaction between warming treatment and soil moisture, but retained the main effects of each of these, as well as the interaction between hillslope location and vwc, and warming treatment and soil temperature. Q10 values were estimated by applying the transformation Q10 = exp(10 * β) to the temperature response parameters (β) in the GLS analysis. The model is described in detail in the Supplementary Information (Supplementary Materials and “Methods” section).