The overall initial increase in Rs observed in this study in response to soil warming agrees with common knowledge that warmer soils generally lead to greater Rs in an array of ecosystems, and that soil temperature is the abiotic variable that exerts the greatest control over spatiotemporal variations in Rs [3, 6–8]. The observed increase in Rs in response to soil warming also corroborates with the meta-analysis by Carey et al. [22] which suggests a positive effect of artificial warming on Rs across several ecosystems. Specifically for temperate forests, previous studies highlighted a strong positive curvilinear relationship between Rs and soil temperature, under either natural conditions [9, 11, 15, 51] or artificial warming [10, 52–54].
Although the relationship between Rs and soil temperature were significant for both spatial controls and heated plots of the three forest types, exponential regression models only explained 30 to 39% of the variability in Rs in spatial controls and 35 to 45% of the variability in heated plots. In comparison, exponential models in Bélanger et al. [9], which used the same experimental design prior to artificial warming, were more robust, explaining 55 to 65% of the variability in Rs. The latter models were developed from data collected in all plots in 2019 and 2020 (i.e., temporal controls). On the one hand, the decrease in predictability of Rs based on soil temperature can be attributed to smaller sample sizes in 2021 and 2022 compared to 2019 and 2020 as well as the lower number of observations in spatial controls compared to heated plots in 2021 and 2022 (about half of the measurements). On the other hand, the threshold of 15°C, beyond which a weakening of the influence of soil temperature on Rs was observed in the heated plots, may partly explain the lower coefficients of determination of exponential regression models that include the full range of measured soil temperature (i.e., from 0 to 25°C). Indeed, for all forest types and treatments, exponential relationships between temperature and Rs were remarkably more statistically robust when considering a range from 0 to 15°C compared to relationships within the range of 15 to 25°C.
A relatively consistent decline in the response of Rs to artificial warming was observed at a specific threshold. For most stands, Rs slowed down in response to artificial warming when Rs in spatial controls reached a threshold from 300 to 500 mg CO2 m-2 h-1. Based on exponential regression models, this range in Rs values corresponded to a range of soil temperature from 12 to 18°C, with a median of 15°. For the mixedwoods and hardwood-beech stands, the lower Q10 values of the regression models in heated plots compared to spatial controls also illustrated a decrease in Rs sensitivity to higher temperatures due to artificial warming (Table 2). Soil temperatures produced by artificial warming in this study were not unusually high compared to other temperate hardwood stands in the literature, mainly because the study site is at the very northern limit of the temperate deciduous forest with generally cooler soils. However, the decrease in predictability in Rs at a higher temperature range with a threshold set at 15°C is not common in the literature. In their meta-analysis, Carey et al. [22] determined that Rs rates increase with increasing soil temperature up to a threshold of approximately 25°C in both heated and non-heated plots, above which Rs rates started to slow down. Previous studies that focused on the effect of artificial warming on Rs generally demonstrated a decrease in sensitivity (i.e., a decrease in Q10) over the long term, after at least 5 years of experiment, due to a combination of phenomena such as microbial thermal acclimation, reduction in labile carbon reserves and soil drying [10, 55, 56].
The weakening of the influence of temperature on Rs in response to artificial soil warming in this study could depend on the influence of several other climatic and environmental variables, such as water availability and forest type, and the interactions between them. We hypothesized that the slowing down of Rs at the 300–500 mg CO2 m-2 h-1 threshold (or a decrease in the robustness of exponential models above 15°C) is due to increased evaporative demand and soil drying during artificial warming. Soil drying induced by artificial warming was shown to offset the dependence of Rs on soil temperature [57], and moist soils were shown to exhibit a greater increase in Rs than dry soils [58]. Indeed, Rs depends on the balance between temperature and soil water availability [19]. Microbial activity and Rs are at their optimum within a certain range of soil water content and can be reduced or inhibited when soil water content is too low or too high [1, 59, 60]. For well-drained forest soils such as those in this study, water deficits during very specific periods can modulate Rs and the measured CO2 flux is no longer related to soil temperature [9, 15]. However, in this study, integrating soil water potential along with temperature into a multiple linear regression model did not improve the prediction of Rs in any stand. Even after dividing the database into ‘‘cold’’ and ‘‘warm’’ temperature ranges (i.e., either below or above 15°C), soil temperature remained the main variable modulating Rs, despite a considerable weakening of the relationship above 15°C. Overall, no statistical test performed could clearly demonstrate that soil water potential governed, at least partially, Rs at the study site, not even under higher soil temperatures. This outcome does not imply that water availability does not influence Rs. Rather, it can mean that drought characteristics (i.e., frequency and intensity) were not entirely conducive to statistical testing with Rs, and/or that the effect of water availability may be confounded by other variables and processes that are a function of forest type and species composition.
At first glance, our results suggest that the magnitude of the positive feedback loop between climate warming and soil CO2 emissions from forest ecosystems could be limited [8, 61], at least for Quebec temperate deciduous forests. However, before the effect of warming on Rs becomes limited by induced soil drying, Rs could increase in response to rising soil temperature and contribute to radiative forcing in these systems. Further monitoring of Rs under artificial warming is planned at the study site to verify these responses, and ideally, more sites with other species compositions should be tested for a better portrait of the bioclimatic domain in question.