A paired-plot litter decomposition experiment was set up in the Yukon Territory in northwestern Canada, at the southern edge of the northern circumpolar permafrost region. The experiment compared litter decomposition and soil temperature in forest and agricultural land (cropland and/or grassland). Since the Klondike gold rush at the end of the 19th century, the Yukon has had an established agricultural sector which allows for studying the effects of land-use change from forest to grassland and forest to cropland in the subarctic. Farms were considered to be suitable for studying the effects of land-use change from forest to grassland or cropland when they (1) originated from forest; (2) were located on mineral soils; and (3) had a remaining native forest adjacent to the agricultural land, i.e. within a distance of approximately 100 m. Furthermore, both forest and agricultural land needed to be located on flat terrain with comparable soil properties. This was checked in an auger-based pre-assessment in consultation with the farmers, focussing on soil texture and the proportion of rock fragments and on bands of organic material as well as on the visibility of hydromorphic properties. A total of 15 farms, covering pairs of forest and cropland and of forest and grassland and triplets with forest, cropland and grassland, were included in this study (for details see Table S1). These farms provided 21 pairs of forest and cropland (n=12) or forest and grassland (n=9) (Fig. 1). As described in detail in Peplau et al. (2022a), forests in the research area were mixed-wood forests of the boreal cordillera ecoregion and croplands were small-scale fields with grains, potatoes or vegetables, and herbs and greens, whereas grasslands were used as pasture for livestock grazing or for hay production.

In order to investigate the effects of land-use change on soil temperature regime and litter decomposition, tea bags and temperature loggers were buried at the chosen farms in summer 2019. Tea bags with green tea (“Bio Grüner Tee”, Paulsen Tee, Fockbek, Germany, charge no. 187896FC) as a standard litter material were weighed, tagged and buried at depths of 10 and 50 cm from the soil surface (n=3 per depth). Before the start of the experiment, 12 tea bags were opened and weighed to determine the weight of the bag material without the tea. This methodology is based on the work of Keuskamp et al. (2013) but is slightly modified in terms of burial depth (10 and 50 cm instead of 5 cm), choice of tea (only green tea instead of green and rooibos tea) and duration of burial (2 years instead of 3 months), as these modifications were necessary to meet the objectives of our study. Temperature loggers (Tinytag Plus 2 TGP-4017, Gemini Data Loggers Ltd) were buried at the same depths (n=1 per depth) and set up to record the soil temperature every 2 h. The tea bags were buried at spots considered representative of the given plots by placing them approximately 30 cm apart from each other around the temperature loggers. After 2 years, the tea bags and temperature loggers were dug out in September 2021. The tea bags were cleaned of roots and soil, dried at 60 ∘C, opened in order to manually pick out fine roots that grew into the tea bag, and weighed again to determine mass loss as a proxy for decomposition. The tea bags from very clayey sites were additionally washed prior to opening to remove clay particles from the tea bag material.

In total, 209 out of 216 tea bags and all 72 temperature loggers were recovered. After downloading the data from the loggers, measurements were checked for plausibility (no abrupt changes that would exceed normal hourly fluctuations) and completeness (no missing data) before further processing.

In addition to the burial of tea bags and temperature loggers, soil samples were taken from every plot to characterize the soils of the sites investigated. The sampling was done in summer 2019, at the same time as tea bag burial. Details about the soil sampling and laboratory analyses can be found in Peplau et al. (2022a). Soil was sampled from depth increments of 0–10 and 40–60 cm, matching the depth of the buried sensors and tea bags. Five field replicates of every depth increment were pooled to a mixed sample and analysed for organic and inorganic carbon (C) and total nitrogen (N) content (Byers et al., 1978), pHH2O (ISO 10390), plant-available phosphorus (Olsen et al., 1954), SOC fractions (Zimmermann et al., 2007), and texture (Köhn, 1929). The soils in the research area were Cambisols and Cryosols (Jones et al., 2009) with pH values between 5.5 and 8.9 (mean: 7.4) and clay contents between 49 and 578 g kg-1 (mean: 178 g kg-1). Soil parameters and values of SOC stocks were obtained from an earlier study at the same sites (Peplau et al., 2022a). In this earlier study, soils were sampled from 0–80 cm, with depth increments of 0–10, 10–20, 20–40, 40–60 and 60–80 cm. The organic C was measured with an elemental analyser (LECO TruMac CN). To distinguish between organic C and inorganic C, samples with pH > 6.2 were heated in a muffle furnace at 440 ∘C overnight before the measurement. The heating removes organic C, and the following measurement thus provides an estimate of inorganic C.

The descriptive variables of annual mean temperature, minimum temperature, maximum temperature, temperature amplitude, number of frost days (i.e. days with a mean temperature lower than 0 ∘C) and cumulative degree days (temperature sum of days with a mean temperature above 0 ∘C) were calculated from the original 2-year temperature dataset. The number of frost days and cumulative degree days were divided by 2 to obtain the average of both years.

To test for significant differences in litter decomposition and soil temperature parameters between forest, cropland and grassland, linear mixed-effects models were used with land-use type and depth as fixed effects, allowing for an interaction, and site as a random effect, allowing for a random intercept. Homoscedasticity, normality of the residuals and linearity of the dataset were given, and no transformation of the data was necessary. Since cropland and grassland soils did not have the identical reference forests, separate models were used for cropland–forest and grassland–forest pairs. After performing the linear mixed-effects models, estimated marginal means were used to obtain pairwise comparisons of all groups of the linear mixed-effects model (confidence level = 0.95).

In order to identify variables that are driving the decomposition of the buried tea bags, Pearson's correlation coefficient was calculated separately for the complete dataset and for every land-use type and depth.

All statistical analyses were conducted using R version 4.0.4 (R Core Team, 2021) with the packages readxl (Wickham and Bryan, 2019), tidyverse (Wickham et al., 2019), dplyr (Wickham et al., 2020), purrr (Henry and Wickham, 2020), ggplot2 (Wickham, 2016), ggpubr (Kassambra, 2020), ggthemes (Arnold, 2021), ggpmisc (Aphalo, 2021), corrplot (Wei and Simko, 2021), lme4 (Bates et al., 2015), lmerTest (Kuznetsova et al., 2017), multcomp (Horthon et al., 2008), multcompView (Graves et al., 2019) and emmeans (Lenth, 2021). Values are given as mean ± standard deviation if not indicated otherwise. The level of significance for all statistical analyses was selected as α=0.05. All data used for this study are openly available at 10.5281/zenodo.7219753 (Peplau et al., 2022b).

Forest soils were cooler and had smaller intra-day variations in temperature than cropland and grassland soils (Fig. 2). During winter, the soil temperature of all land uses did not exceed 0 ∘C and showed very little short-term variations within a couple of days. With the beginning of spring in April, soil temperature at 10 cm depth increased sharply to above 0 ∘C and short-term variations in temperature became larger. At 50 cm, the spring soil temperature increase was visible but less pronounced than at 10 cm. The sharp increase in soil temperature to above 0 ∘C was visible at all sites at the same time, independently of how low the soil temperature was beforehand. On average, grasslands soils were 2.2 ± 0.9 and 1.8 ± 0.5 ∘C warmer than forest soils at 10 and 50 cm and cropland soils were 2.1 ± 1.1 and 2.0 ± 0.9 ∘C warmer than forest soils at 10 and 50 cm (Fig. 3), which was significant with p<0.001 in all four cases. The linear mixed-effects model showed no interaction between land use and depth (p=0.91 and p=0.26 in cropland and grassland), indicating that soil temperature is affected similarly by land-use change at both depths. Moreover, significantly higher cumulative degree days indicated warmer soils under agricultural use than under forest. Cumulative degree days at 10 and 50 cm were elevated by 606 ± 222 and 733 ± 338 in grassland soils and by 768 ± 262 and 779 ± 237 in cropland soils, with p<0.001 in all four cases. This roughly corresponds to a doubling of cumulative degree days upon land-use change. Detailed information about the soil temperature for every site, land-use type and depth can be found in Table S1.

Significant differences between forest and grassland at both depth increments were found in mean temperature, minimum temperature, maximum temperature, total amplitude and cumulative degree days but not in the number of frost days (Table 1). The comparison between forest and cropland resulted in significant differences at 10 and 50 cm for mean temperature, maximum temperature, total amplitude, cumulative degree days and frost days. In contrast to grasslands, croplands did not have different minimum temperature but had fewer frost days than forests. Temperature differences between forest and grassland were smaller in soils with high clay content, while there was no such correlation observed in cropland soils (Fig. 4).

We observed significant differences in litter mass loss between cropland and forest soils but not between grassland and forest soils (Fig. 5). The linear mixed-effects model showed no significant interaction between land use and depth (p=0.11 and p=0.43 for cropland and grassland), indicating that litter mass loss is affected similarly by land-use change at both depths. In forest–cropland pairs, the mean proportional loss of added litter was lower in forest soils (70 ± 7 % and 52 ± 4 % at 10 and 50 cm) than in cropland soils (73 ± 8 % and 61 ± 8 % at 10 and 50 cm), with p=0.321 and p=0.003. In forest–grassland pairs, mean decomposition of added litter in forest soils was 70 ± 4 % and 60 ± 12 % at 10 and 50 cm, while it was 67 ± 7 % and 53 ± 18 % in grassland soils at 10 and 50 cm, with p=0.517 and p=0.085. Detailed information about mass loss for every site, land-use type and depth can be found in Table S1.

The correlation between litter mass loss and soil temperature, site characteristics and soil properties strongly differed between agricultural land and forest. In forest topsoils, only minimum temperature was significantly negatively correlated with mass loss of tea, while mass loss in subsoils was significantly correlated with minimum temperature, number of frost days and total temperature amplitude (Fig. 6). In cropland soils, significant correlations were only observed in topsoils. In contrast to forest soils, mass loss in cropland soil was positively correlated with minimum temperature. Furthermore, there was a significant negative correlation between mass loss and silt content. In grassland soils, tea mass loss was only correlated with temperature parameters (mean temperature, maximum temperature, amplitude, number of frost days and cumulative degree days). In contrast to croplands, there was no significant correlation between decomposition and SOC fractions in grassland soils. Across all land-use types and depths, mass loss correlated significantly with soil temperature parameters, except for mean temperature and number of frost days. Weaker, yet significant, correlations were observed between mass loss and soil organic matter (C and N content as well as SOC fractions). This was not observed when separating the sample set into the different land-use types and depths, except for forest subsoils (soil organic matter parameters) and grassland subsoils (temperature parameters). Besides elevated mean and maximum temperature, which may be biased by single extreme values, cumulative degree days also increased in cropland and grassland soils compared to forest soils. This increase had a highly significant effect on litter decomposition (Fig. 7) (p<0.001). Furthermore, there was a good correlation between SOC stocks and mean soil temperature in forest soils with higher SOC stocks in colder soils, something that was not observed in agricultural land (Fig. 8).

The higher temperatures, higher cumulative degree days and greater amplitude that we measured in grassland and cropland soils supported our hypothesis that deforestation is shifting the soil temperature regimes from a moderate temperature amplitude, with relatively low summer temperatures in forest soils, to more extreme amplitudes in agricultural soils, with particularly warm summer temperatures. The observed soil warming upon deforestation is in line with results from earlier studies, reporting temperature increases of 2.0 ∘C in the tropics (Jiménez et al., 2007), between 2.5 and 3.3 ∘C in the temperate zone (Morecroft et al., 1998), and up to 5.0 ∘C during summer in boreal Alaska (Grünzweig et al., 2003). Similarly to our results, forest soils in all of the three studies mentioned were on average cooler in summer compared to agricultural land, which is due to shading by the forest canopy. During summer, we observed that forest soils were on average 4.0 ∘C cooler in topsoil and subsoil than croplands and 3.8 and 4.2 ∘C cooler than grasslands, which is slightly less than observed by Grünzweig et al. (2003). Also, cumulative degree days increased between 600 and 800, which is almost a doubling of cumulative degree days upon land-use change. This is a similar increase to that reported by Grünzweig et al. (2003), which was between 500 and 650 annually. As shown in a modelling study for all of Canada, warmer winter soil temperatures can be related to thicker snow cover in deforested land compared to forest (Zhang et al., 2005). Snow cover, along with vegetation, is the most important factor determining soil temperature patterns (Qian et al., 2011; Zhang et al., 2005). Warmer winter soil temperatures in agricultural land compared to forest as a consequence of thicker snow cover on open land than on forests were also reported by Grünzweig et al. (2003). Agricultural soils of the Yukon were as cold as (cropland) and slightly warmer than (grassland) the forest soils during winter. This emphasizes the importance of vegetation for the soil temperature. Snow cover on bare soil might insulate the soil in a similar way to natural forest vegetation, but the combination of dense grasses and overlaying snow cover adds an additional insulation effect.

The temperature difference between forest and agricultural land appeared to be influenced not only by insulation of the soil by vegetation or snow but also by inherent soil thermal properties, which are regulated by soil texture, and therefore indirectly soil moisture. Measurements of soil moisture have not been conducted in this study, but since the study region has only little precipitation and farmers irrigate their croplands regularly, it is likely that the irrigated land is on average wetter than the forest (Peplau et al., 2022a). Clayey grassland sites had smaller temperature differences upon deforestation than sandy sites. This can be related to the differences in the thermal properties of air, water and different minerals, with clayey sites having the lowest and sandy sites having the highest thermal conductivity (Dong et al., 2015). Overall, clayey soils have a larger pore volume and are, therefore, more buffered thermally than sandy soils if the pore space is not water filled but contains a lot of air. Under wet conditions, heat exchange between the soil and the atmosphere is increased and soils cool down more strongly than under dry conditions. In cropland soils, no relationship between soil texture and temperature was supported statistically. Since croplands in the Yukon are irrigated regularly, the thermal insulation of the soils might be reduced at all cropland sites, independently of soil texture, which was not the case in unirrigated grassland sites. Moreover, the effects of soil properties on the soil temperature regime might be masked by differences in vegetation.

As we have shown, land-use change has a soil warming effect of around 2.1 ∘C. Due to climate change, Canadian soils warmed by 0.6 ∘C during the 20th century, with regional differences of between -2 and +5 ∘C, underlining the great importance of spatially distributed soil temperature measurements (Zhang et al., 2005). Climate-change-related alterations in the temperature of Canadian soils have been observed to be greatest in spring (0.26–0.30 ∘C per decade since the 1950s), while winter soil temperatures have not changed significantly (Qian et al., 2011). In contrast, air temperatures in Canada have increased most strongly in winter (2.3 ∘C between 1950 and 2010) and less so in spring (1.7 ∘C between 1950 and 2010). The annual mean air temperature increased by 1.5 ∘C between 1950 and 2010 or 0.25 ∘C per decade (Vincent et al., 2012), which is slightly less than the increase in soil temperature within the same time. Our results imply that land-use change from pristine forest to agriculture exceeds the effect of climate change on soil temperature and is particularly strong during summer, when biological activity is highest in subarctic ecosystems. Common models of SOC turnover are fed by air temperature instead of soil temperature (Balesdent et al., 2018; Kaczynski et al., 2017; Crowther et al., 2016). Thus, one of the most important drivers of microbial activity and SOC mineralization, that is temperature, is assumed to be independent of the vegetation cover. The present study highlights that this might be a severe shortcoming in such models, which often fail to capture land-use change effects (Boysen et al., 2021; Gottschalk et al., 2010).

Using tea bags instead of naturally occurring litter had the advantage that the material was standardized, eliminating effects on decomposition caused by differences in litter material between sites or within the soil profile. However, the decomposition values obtained can only be interpreted as potential litter decomposition. It was hypothesized that warmer summer temperatures foster the decomposition of fresh soil organic matter in agricultural soils compared to forest soils. Indeed, there was a greater mass loss of tea in cropland soils than in forest soils. Particularly in subsoils, mass loss was around 8.7 % higher in croplands than in forests, while it was 2.9 % in topsoil, which might be related to the fact that the temperature effect was more masked by agricultural management in the topsoil, where tilling, harvesting and other practices led to regular soil disturbance. However, the hypothesis must be rejected for grassland soils, since there was no significant difference in litter decomposition between forest and grassland soils, despite warmer soil temperatures in grasslands than in forests. This might suggest that not only did the deforestation induce soil warming but also agricultural management controlled litter decomposition. In a global study, Djukic et al. (2018) reported that precipitation is the most important climatic factor for litter decomposition and temperature appeared to be less important. However, their study was conducted only during summer, where water availability, not temperature, was the limiting factor. Our results suggest that, in temperature-limited regions, the temperature may play an important role in litter decomposition, as shown by the relationship between cumulative degree days and litter mass loss. However, missing evidence for this relationship in grassland–forest pairs might underline the importance of water availability as a prerequisite for decomposition. Soil moisture was not quantified in this study, but we can assume that croplands had higher soil moisture than grasslands, as croplands are irrigated regularly due to the dry climate in the research area (on average, 262 mm annual precipitation; Environment Climate Change Canada, 2020) and grasslands remain rainfed, according to a farmers' questionnaire conducted in Peplau et al. (2022a). However, litter decomposition might be underestimated in grassland plots, since there were more fine roots potentially growing into the tea bags, which might not have been removed entirely before weighing. Furthermore, despite the fact that no soil particles were visible within the tea bags after opening the bags after burial, there might be a slight underestimation of the decomposition caused by small clay particles sticking between the remaining tea leaves.

Various studies have reported losses of SOC after land-use change (Grünzweig et al., 2003, 2004; Guo and Gifford 2002; Wei et al., 2014; Poeplau, 2011). A certain proportion of these losses can be assigned to deforestation-induced removal of the uppermost soil layers, including litter and parts of the topsoil (Grünzweig et al., 2003). C input quantity (Luo et al., 2017) and quality (Cotrufo et al., 2019) as well as frequent soil disturbances and changes in aggregate stability (Six et al., 2000) can add to the alterations in SOC stocks driven by land-use change. Here, we were able to show that microclimatic changes and their effect on litter decomposition are another relevant driver of SOC stock change after deforestation. Under natural conditions, as represented by the forest sites, SOC stocks were related to mean soil temperature to some extent. The coldest forest soils stored significantly more SOC than the warmest forest soils, with a linear decrease in SOC of 38.6 ± 17.2 Mg C ha-1 ∘C-1 (p<0.05) warming (Fig. 8). This slope is 1 order of magnitude higher than values observed in warming experiments (Peplau et al., 2021: 1.9 Mg C ha-1 ∘C-1; Verbrigghe et al., 2022: 2.8 Mg C ha-1 ∘C-1). This might indicate that the observed range in forest SOC stocks cannot solely be explained by a direct temperature effect. Instead, it has been observed that the coldest sites, which were also characterized by shallow permafrost (detectable ground ice in summer within the upper 50 cm), were rather wet sites with thick, C-rich A horizons. Farmers reported that the waterlogging ceased after deforestation, i.e. with the deepening of the permafrost layer (Peplau et al., 2022a). Despite a high general variability in SOC stocks across forest sites, our data suggest that, in the presence of permafrost, warming might have a much more severe effect than in non-permafrost soils. This is because water infiltration is hampered by underlying ice layers and soils remain wetter during summer than in sites without permafrost. The linear relationship between soil temperature and SOC stocks was not apparent in agricultural soils, although SOC stocks were reduced significantly in cropland soils (Peplau et al., 2022a). The decoupling of SOC stocks from soil temperature in agricultural soils shows that the natural, climate-driven balance between C input, mineralization and C storage is heavily disturbed by agricultural activity. Depending on agricultural practices (i.e. cultivated crops, soil management, irrigation), the amount and quality of C input vary greatly from natural habitats and also between sites. This makes the quantification of alterations of C mineralization induced by land-use change more complicated. Given that land-use change will increase soil temperatures in subarctic soils, the future spread of agriculture on permafrost soils may therefore additionally accelerate the climate–carbon feedback, causing more SOC loss than already caused by climate change and the warming of pristine forests. The observed different effects of land-use change on soil temperature and decomposition in croplands and grasslands indicate that changes in the soil water regime might be essential for prospective land-use–carbon feedbacks. Since warmed soils have a deeper (or entirely thawed) active layer, soils naturally dry upon warming due to drainage effects (Andresen et al., 2020).