Soil Warming Shifts Seasonal and Soil Depth‐Related Root Growth Dynamics in Subarctic Grasslands

ABSTRACT

Climate warming in northern ecosystems is accelerating, yet the responses of belowground growth dynamics under warming remain poorly understood. We assessed how soil warming affects seasonal root growth dynamics in subarctic grasslands, aiming to predict belowground responses under future climate scenarios. We investigated root growth across soil depths and seasons in geothermally warmed subarctic grasslands, comparing two warming durations: medium‐term (MTW; 11–14 years) and long‐term (LTW; > 60 years) and three warming magnitudes (ambient: +0°C, < +2°C, and > +2°C). Root length and projection area were quantified from minirhizotron images using the automated segmentation model RS_SAG. The automated segmentation model trained and used in this study performed robustly, with accuracy comparable to manual labels. Warming duration and magnitude effects on root length and projection area were season‐ and soil‐depth‐dependent. Root growth responses were more pronounced under LTW than MTW, particularly in the subsoil under warming > +2°C. Significant differences from ambient conditions occurred during summer, where root length and projection area were larger than in ambient conditions. Warming shifted root growth phenology by accelerating peak‐season growth and intensifying winter decline. Enhanced subsoil root growth was related to a higher topsoil rhizome biomass, which likely provided additional energy to support deeper root growth and a higher abundance of Kobresia myosuroides —a deep‐rooted rhizomatous sedge species with extensive belowground biomass. Higher subsoil root length and projection area under LTW indicate a shift towards deeper rooting, potentially associated with changes in species composition or nutrient availability. Enhanced root growth, proliferation, temporal shifts and seasonal amplitude may increase litter inputs, influencing microbial activity and organic matter dynamics in subsoils. These deep‐root responses may partially offset warming‐induced organic matter decomposition but could also affect the stability of soil carbon pool, emphasising the role of deep‐soil roots in carbon cycling in subarctic soils.

Soil warming effects on root growth in subarctic grasslands varied with soil depth, season, warming magnitude, and warming duration. The most pronounced root growth response was observed in the subsoil (10–30 cm) under long‐term warming (LTW), particularly at warming magnitudes exceeding +2°C above ambient, where the response remained consistently positive across seasons and three study years. Long‐term warming shifted root growth phenology in the subsoil by accelerating peak‐season growth and intensifying winter decline. Enhanced subsoil root growth was related to a higher topsoil rhizome biomass, which likely provided additional energy to support deeper root growth and a higher abundance of Kobresia myosuroides—a deep‐rooted rhizomatous sedge species with extensive belowground biomass.

1. Introduction

The climate in northern high‐latitude ecosystems is warming faster than the global average (IPCC ²⁰²²), making it increasingly important to understand how these changes will impact key ecosystem processes in these regions (Rixen et al. 2022). Grassland ecosystems cover approximately 25% of high‐latitude regions (Chapin et al. 2011) and have a larger proportion of belowground plant biomass than aboveground biomass (Mokany et al. 2006; Yang et al. ²⁰¹⁰). In particular, in high‐latitude grasslands, like in the subarctic and alpine, belowground plant biomass is up to 90% of the total biomass (Bhattarai et al. 2023; Körner ²⁰²¹), which is much higher than the average for grasslands globally (67%; Ma et al. ²⁰²¹).

Climate is the primary factor influencing phenological events, which refers to the timing of recurring biological processes in plants (Cleland et al. 2012; Lieth ²⁰¹³). This includes growth phenology, which is usually studied by monitoring changes in plant organs over time (Hilty et al. 2021). Under climate warming, plants in high‐latitude ecosystems are shown to shift the start and end of their growth earlier (Leblans et al. 2017; Möhl et al. ²⁰²²; Mortier et al. ²⁰²⁴), and such warming‐induced shifts in vegetation phenology can create strong feedback on the global climate system (Peñuelas et al. 2009; Richardson et al. ²⁰¹³). However, most phenological studies have focused on aboveground plant organs, while belowground phenology remains largely overlooked (Abramoff and Finzi 2015; Piao et al. ²⁰¹⁹; Radville et al. ²⁰¹⁶), resulting in uncertainty in predicting biogeochemical fluxes at the ecosystem scale (Abramoff and Finzi 2015). Importantly, belowground and aboveground phenology are not synchronised (Abramoff and Finzi 2015; Blume‐Werry et al. ²⁰¹⁶; Radville et al. ²⁰¹⁶) and meta‐analyses suggest that such mismatches may intensify under future climate, as roots and shoots respond differently to warming (Blume‐Werry ²⁰²²; Liu et al. ²⁰²²). Moreover, roots and shoots exhibit distinct seasonal growth patterns: shoots are typically dormant over winter, whereas roots can remain active (Fernandez and Caldwell 1975; Malyshev et al. ²⁰²³). Studying belowground processes is therefore crucial for understanding plant adaptation mechanisms, biogeochemical cycles, and biotic interactions, particularly in the context of climate warming (Abramoff and Finzi 2015; Radville et al. ²⁰¹⁶).

Roots play a vital role in nutrient and water uptake, as well as in soil organic matter (SOM) formation (Kuzyakov and Domanski 2000; Lucas et al. ²⁰²³; Pausch and Kuzyakov ²⁰¹⁸; Yang et al. ²⁰²³). During plant growth, roots continuously contribute carbon to the soil through rhizodeposition, including root exudates, sloughed root cells, root hairs, and fine root litter, making them a major driver of soil organic carbon inputs (Kuzyakov and Domanski 2000; Kuzyakov and Schneckenberger ²⁰⁰⁴). Because roots represent the dominant pathway of carbon transfer to soils, changes in root growth dynamics, quality, and quantity can strongly influence organic matter accumulation and carbon sequestration in soils (Bardgett et al. 2014; Freschet et al. ²⁰¹³; Jobbágy and Jackson ²⁰⁰⁰; Poirier et al. ²⁰¹⁸).

Temperature is a crucial driver of plant growth and development, influencing root physiology (Atkin et al. 2000), biomass allocation (Wang et al. 2016), and belowground carbon dynamics (Poeplau et al. 2020; Verbrigghe et al. ²⁰²²). Warming affects root growth dynamics, turnover rates (Sullivan et al. 2008), and biomass allocation, ultimately impacting the quantity and timing of soil carbon input and storage in ecosystems (Pregitzer et al. 2000; Robinson ²⁰⁰⁷; Srivastava and Yetgin ²⁰²⁴). Warming‐induced changes in root carbon inputs may either enhance SOM formation through increased root production and rhizodeposition or accelerate SOM decomposition through priming effects, in which greater inputs of labile root‐derived carbon stimulate microbial activity (Dijkstra and Cheng 2007; Kuzyakov ²⁰¹⁰; Pausch and Kuzyakov ²⁰¹⁸).

Despite the critical role of root systems in ecosystem processes, the influence of environmental changes, such as soil warming, on root growth dynamics remains inadequately understood (Iversen et al. 2015). Warming can influence root dynamics both directly by stimulating growth (Malhotra et al. 2020) and indirectly by enhancing carbon allocation to the belowground plant organs (Bai et al. 2010). However, responses are highly variable, ranging from stimulation to inhibition of root development (Gill and Jackson 2000; Wang et al. ²⁰¹⁶) and are likely to depend on factors such as soil depth, season, and the magnitude and duration of warming (Salmon et al. 2025). In particular, phenological responses to warming are also largely influenced by the plant species composition, as there is a prevalence of contrasting temperature sensitivities among species (Diez et al. 2012; Wu et al. ²⁰²⁵).

Changes in rooting depth have been observed under warming (Wang et al. 2021) as well as rising atmospheric CO₂ (Iversen 2010), water limitation (Collins and Bras 2007; Schenk and Jackson ²⁰⁰²), and permafrost thawing (Blume‐Werry et al. ²⁰¹⁹). Increased soil temperatures often stimulate root growth in the topsoil, where root density and microbial activity are highest (Gill and Jackson 2000). However, responses may vary in the subsoil due to the differences in thermal and nutrient conditions (Gill et al. 2002; Jobbágy and Jackson ²⁰⁰⁰). Moreover, it remains unclear whether warming effects are consistent across soil depths, given significant differences in nutrient content, moisture, organic matter, and microbial communities from topsoil to subsoil (Fontaine et al. 2007). Deep soil roots are particularly important in the long‐term carbon storage and nutrient cycling (Kwatcho Kengdo et al. 2025), yet their responses to environmental change remain poorly understood due to the methodological difficulties of accessing deeper soil layers, as deep soil excavation is even more laborious and destructive (Iversen et al. 2015; Maeght et al. ²⁰¹³). Medium‐term warming experiments have revealed early responses in root growth, allocation, and turnover, providing insight into potential transient shifts in root phenology (Brunner et al. 2013; Melillo et al. ²⁰¹¹). In contrast, long‐term warming enables the study of sustained warming impacts and potential adaptation or acclimation processes in root systems. For example, in a previous study on geothermally warmed subarctic grasslands that analysed 128 biotic and abiotic variables, long‐term soil warming produced a less extreme response, including root responses, when compared to short‐term warming (Walker et al. 2020).

The inherent challenges in studying roots are further amplified when examining seasonal root dynamics, as frequent monitoring across multiple time points significantly increases the time, effort, and costs required. Minirhizotron imaging provides a powerful solution by enabling non‐destructive, repeated measurements of root growth at different soil depths (Johnson et al. 2001) and is particularly valuable in warming studies, as it allows the detection of depth‐specific warming responses (Brunner et al. 2013). In this study, we employed a minirhizotron imaging system to track root growth dynamics to a depth of 30 cm in the soil. Images acquired with minirhizotron tubes are usually analysed with semi‐manual approaches, such as using the WinRhizo Tron software (Regent Instruments Inc.), which are highly time‐consuming and labour‐intensive. Recent advancements in machine learning algorithms have led to the development of tools that integrate trained models to automatically distinguish plant roots in images, enabling the analysis of large datasets in a short time (Atkinson et al. 2019; Smith et al. ²⁰²²). However, these models perform best on artificial substrates and often struggle with natural soil profiles, particularly in diverse ecosystems where root colour varies among species (Wang et al. 2019), highlighting the need for an ecosystem‐specific root‐detection model. To address this, we developed and implemented a model that was specifically trained on subarctic grassland root images, enabling accurate root detection.

Root growth patterns have often been studied in separation rather than as part of an integrated response across temporal gradients, with combined effects of soil depth, season, and the magnitude and duration of warming being largely unexplored (Salmon et al. 2025). In this study, we address this gap by examining how the warming magnitude and duration interact with soil depth and seasonality to influence root growth, providing a more comprehensive understanding of belowground carbon cycling in future climate scenarios. We hypothesised that (1) soil warming induces depth‐specific changes in root‐growth dynamics, enhancing deeper root growth; (2) soil warming advances seasonal root growth, shifting it earlier; and (3) changes in root growth dynamics depend upon warming magnitude and duration.

2. Materials and Methods

2.1. Study Site

The study was conducted at the ForHot research site in Iceland (64.008° N, 21.178° W; 83–168 m altitude; Sigurdsson et al. ²⁰¹⁶), which comprises two geothermally warmed subarctic grasslands located approximately 2.5 km apart. At one site, soil warming began in 2008, while the other has been exposed to warming for over 60 years, representing two distinct soil warming durations: medium‐term warming (MTW) and long‐term warming (LTW). Both grassland sites are unmanaged and treeless and are dominated by perennial vascular plants such as Anthoxanthum odoratum, Ranunculus acris, Equisetum pratense, Agrostis capillaris, and Galium boreale (Marañón‐Jiménez et al. ²⁰¹⁸; Sigurdsson et al. ²⁰¹⁶). A detailed overview of species abundance at the study sites is summarised in Table S2. Soils at both sites are classified as Silandic Andosols with a silt‐loamy texture (Arnalds 2004). Climate data from the nearest synoptic station (Icelandic Meteorological Office; https://www.vedur.is) report a long‐term (2003–2015) mean annual air temperature (MAT) of 5.2°C, mean annual precipitation (MAP) of 1457 mm, and mean wind speed of 6.6 m s⁻¹. Mean monthly precipitation averages 75 mm from May to July and increases to 135 mm during the remaining months. Mean monthly air temperatures range from −0.1°C in December to 12.2°C in July. The growing season in unwarmed plots typically begins in late May and ends in late September (Leblans et al. 2017). A detailed site description is described in Sigurdsson et al. (2016).

2.2. Experimental Design

The experimental site was set up using a full factorial design. In 2013, ten replicate transects (~50 m long) were established, with five in the MTW grassland and five in the LTW grassland. Along each transect, five permanent sampling plots (2 × 2 m) were located (Sigurdsson et al. 2016). For this study, we selected three of five replicate transects in each grassland, each containing five permanent sampling plots. Each transect had its own ambient and warm plots.

The permanent sampling plots were established perpendicular to soil temperature (T _s) gradients, where T _s ranged from ambient to approximately +10°C. Soil temperature in each plot was recorded hourly at a depth of 10 cm using HOBO TidbiT v2 Water Temperature Data Logger (Onset Computer Corporation, USA). Across both grasslands, the 9‐year (2013–2022) mean ambient T _s was 6.1°C. Despite seasonal and daily fluctuations, the increase in T _s above ambient temperature in warmed plots remained stable throughout the year (Sigurdsson et al. 2016). However, the 10‐year average T _s differed among the corresponding plots along the replicate transects that were initially targeted for the same T _s. Therefore, plot‐level mean increases in T _s relative to the ambient temperature were used to analyse the root growth dynamics. To capture the full range of T _s variation and associated plant community composition across the study area and to enhance the generability of our results, plots in both MTW and LTW grasslands were grouped into three categories with different warming magnitudes: ambient (n = 3), < +2°C (n = 6) and > +2°C (n = 6). This categorisation aligns with IPCC (2022) projections for subarctic and Arctic regions, where medium to high climate risks are associated with mean surface temperature increases of < 2°C and > 2°C. Details about the range of soil warming magnitude in the studied plots are provided in Table S1. In all warmed plots of both MTW and LTW, T _s measured at a 10 cm depth remained relatively stable down to approximately 25 cm, which represents the most active root layer (Sigurdsson et al. 2016). Ambient plots in MTW and LTW were similar in terms of plant community functional composition (abundance of grasses, forbs, ferns, and non‐rhizomatous species), species richness, evenness and diversity, aboveground biomass, and belowground plant biomass, including fine root biomass and root‐shoot ratio (Bhattarai et al. 2023).

2.3. Minirhizotron Installation and Imaging

Thirty minirhizotron tubes (50 cm long, 5 cm inner diameter) were installed at one corner of each permanent plot in October 2019. The tubes were inserted into the soil at an angle of approximately 35°–45°, with ~40 cm of each tube inserted into the soil, corresponding to a vertical soil depth of ~30 cm. This soil profile was subsequently divided into topsoil (0–10 cm) and subsoil (10–30 cm). We avoided rocks and bedrock and used tape and a cap on the exposed part of each tube to prevent light from entering the tubes.

Regular minirhizotron observations started 7 months after installation, allowing sufficient time for root regrowth in the soil adjacent to the tubes. Images were taken using the BTC‐100X system (Bartz Inc., USA). Each tube was imaged at 16 time points from May 2020 to August 2022, with images taken from both left and right sides of the tube. For each side of the tube, 26 images were acquired at different soil depths. Throughout the study, a total of ca. 24,960 images, each with a resolution of 480 × 640 pixels, were collected. We used the same minirhizotron camera throughout this study, so all images collected had the same image window dimensions (13.5 × 18 mm). The image quality varied widely, ranging from greyscale to overexposed to very dark images.

2.4. Image Processing and Automatic Root Segmentation

From the collected images, 239 random images representing a range of root and soil colours, soil temperatures, soil depths and seasons were manually annotated for roots using the RS_SAG tool (Baykalov 2025). These manual labels were used to train an automated root image segmentation model based on an artificial neural network approach (Baykalov et al. 2023).

Images were preprocessed to denoise, normalise, and scale the images before being fed to the models. Denoising was performed to reduce lattice noise by converting each image to the LAB (L*a*b*) format, where L represents lightness and a and b represent the colour information. Then, a fast non‐local means denoiser was applied to each element of the LAB format, after which the elements were reassembled and converted back to the RGB (Red, Green, Blue) format. The images were scaled to a range of 0–1 to ensure compatibility with the model. All images and their corresponding masks generated during training were resized to 320 × 320 pixels to standardise the dimensions.

Data augmentation was applied during training using two zoom‐based strategies: (i) cropping full images to 480 × 480 pixels followed by resizing to 320 × 320 pixels, and (ii) dividing full 480 × 640 pixel images into four 320 × 320 pixel patches with some overlap. To predict the full dataset, an additional preprocessing step was performed because the randomly selected training data did not capture all variations in image quality, leading to differences in predictions related to light intensity and colour conditions (data not shown). To address this, histogram matching (Sawayama and Nishida 2018) was performed on each tube separately for each depth location (Kwiek and Jakubowska 2024). After histogram matching, an overexposure capping was applied by clipping RGB channels to prevent overexposed artifacts. Additional augmentations included rotations (90° and 180°), flipping/mirroring, and vertical blurring for the colour model. Two models, a coloured model and a grey‐scale model, were trained using a pre‐trained U‐Net EfficientNet‐b6 model (Yakubovskiy 2019) trained on different root images from another study (Baykalov et al. 2023).

It was observed that the pre‐trained model already produced reliable predictions on the dataset and was particularly effective at detecting fine, small roots that were difficult and time‐consuming to annotate manually. To preserve this advantage of the pre‐trained model while incorporating expert manual labelling, a hybrid masking strategy was adopted, in which new masks were generated by combining manual labels with predictions from the pre‐trained model. This approach retained the pre‐trained model's ability to detect finer, smaller roots while also incorporating the accuracy of manual labels for obvious and medium‐to‐large root structures, thereby ensuring biologically validated truth. These new masks were used to retrain and validate both models. Model performance was then evaluated on an independent, unseen dataset (n = 89) consisting of images not used during training, and the model predictions were compared with manual root labels. The test dataset consisted of an equal number of images from topsoil and subsoil (44 and 45, respectively) collected from both grassland sites and included variation in quality and differences in lighting, colour, soil texture, and soil structure.

2.5. Root Length and Projection Area Calculation

Belowground plant organs in the studied subarctic grasslands consist of a substantial proportion of fine roots and rhizomes (Bhattarai et al. 2023). Rhizomes differ from fine roots based on their anatomy and morphology, particularly by the presence of nodes, internodes, and axillary buds. However, the minirhizotron camera window (13.5 × 18 mm) was too small to observe these features, preventing us from distinguishing rhizomes from roots. As rhizomes are generally larger in diameter than roots (see Figure 1), our visual inspection of the images indicated that rhizomes were rare (detected in < 5% of the images), whereas most observed structures were roots. Therefore, from this point onward, all belowground plant organs analysed in the image will be collectively referred to as ‘roots’.

FIGURE 1.

(a) Original image; (b) manually labelled roots; (c) segmentation of roots from RS_SAG.

For root length and projection area estimation, images from all tubes were preprocessed and used as input to both the coloured and grayscale models. The output masks from both models were combined into a single merged mask. Total root projection area was obtained by summing all pixels labelled as root within the merged mask. Total root length was determined by skeletonising the merged mask and summing the pixels within it. Pixel values were then converted to physical units (mm for length and mm² for projection area) using the known dimensions of the minirhizotron image window (13.5 × 18 mm). Changes in root length and projection area were calculated as the increase or decrease between two consecutive seasons and were standardised as mm/day for length and mm²/day for projection area. It should be noted that the root length and projection area analysed in our study are expressed in millimetres (mm) and square millimetres (mm²) per window of the minirhizotron images. The intervals between consecutive sampling campaigns reflect logistical constraints and equipment maintenance requirements. Thus, to account for this inconsistency, we grouped the resulting measurements into seasonal cohorts, thereby tying our analyses to ecologically meaningful periods (spring, summer, autumn, and winter) rather than arbitrary calendar dates. For this, we followed regional climatology and defined spring as April–May, summer as June–August, autumn as September–November, and winter as December–March. Figure 2 shows seasonal soil temperatures for these four seasons over three consecutive study years (2020–2022; Figure 2).

FIGURE 2.

Soil temperatures across four seasons (spr, spring; sum, summer; aut, autumn; win, winter) over three consecutive years (1st: 2020; 2nd: 2021; 3rd: 2022) measured at the topsoil (0–10 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient: Black; < +2°C: Orange; > +2°C: Pink) and durations ((a) medium‐term warmed [MTW; 11–14 years] and (b) long‐term warmed [LTW;> 60 years]). Mean soil temperatures, averaged across study plots, are presented as vertical error bars representing mean ± standard error (SE).

2.6. Species Abundances and Fine Root and Rhizome Biomass

To relate the belowground growth dynamics to the biomass of belowground plant organs, we used previously published data on fine root and rhizome biomass (Bhattarai et al. 2023) where belowground plant biomass had been measured using a soil corer, washed in the laboratory, separated into fine roots and rhizomes, and then dried and weighed. As root growth traits vary among species (Freschet et al. 2017; Weemstra et al. ²⁰²¹), collective growth responses at the plot level would be influenced by the composition of plant communities. Thus, we incorporated plant species abundance data from each plot (Meynzer 2017) into our analysis. We included only those species that occurred in at least 20% of the studied plots (Table S2) to enable more robust statistical analyses.

2.7. Model Implementation and Data Analysis

The models were implemented in Python using the PyTorch library. R version 4.1.0 (R Core Team ²⁰²²) was used for statistical analysis. Data visualisation was performed using the ‘ggplot’ package (Wickham 2007). Statistical significance was determined at p < 0.05. Shapiro–Wilk and Kolmogorov–Smirnov tests were used to test the normality of the data. Data are presented as mean ± standard error (SE) throughout the manuscript.

To assess the strength of association between model‐predicted and manually labelled root length and projection area, a Pearson correlation test was performed on a testing subset of images that were both manually annotated and model‐predicted. In addition, simple linear regression was performed with manual measurements as the independent variable and model predictions as the dependent variable. The estimated regression slope and intercept were then tested against the 1:1 line (slope = 1, intercept = 0). Model's predictive accuracy was assessed using the root mean square error (RMSE) and the mean absolute error (MAE).

Repeated‐measures ANOVA was used to analyse the effects of duration, magnitude, soil depth, and season (categorical factors) on root length and projection area. Measurements were taken repeatedly on the same experimental units (plots) at nine time points, with time included as the within‐subject (repeated‐measures) factor. Relative changes in root length and projection area were calculated between consecutive time points, resulting in eight repeated measurements.

Comparisons of root length, projection area, and change in length and projection area between seasons, between warming magnitudes, between warming durations, and between soil depths were made using ANOVA followed by a Tukey's Honest Significant Difference test. The responses of root length, root projection area, change in length, and change in projection area to warming magnitude were calculated using a base‐10 logarithmic scale by subtracting ambient values from those in warming magnitudes of < +2°C and > +2°C across soil depths, seasons and warming duration. To assess the relationship between root length and root projection area, Spearman's rank correlation coefficients were calculated separately for each combination of season, soil depth, warming duration, and warming magnitude. Seasonal slopes were estimated between seasons and compared across soil‐warming magnitudes using pairwise comparisons of estimated slopes.

Redundancy analysis (RDA) (Braak and Šmilauer 2002) was used to explore and test the contribution of the abundance of plant species (n = 14) and belowground plant biomass (n = 4) to explain the variation in root length (Table S2) separately for MTW and LTW. All explanatory factors in each dataset were tested using forward selection, and significant factors (p < 0.05) from each dataset were selected and used as a matrix for RDA, which was then presented as an ordination biplot. Data were log‐transformed. The significance of the RDA results was tested using a permutation test (Monte Carlo test [999]).

3. Results

3.1. Model Performance

Our results showed that, when evaluated on completely new images that were not used during model training, the measurements from the trained models were positively correlated with those from manual labels for both root length (r = 0.93; p < 0.0001) and root projection area (r = 0.93; p < 0.0001). Furthermore, a significant linear relationship was observed between the manually labelled and model‐predicted root length and projection area (Figure 3a,b). The model showed good predictive performance, with root mean square errors (RMSEs) of 6.1 and 1.1 and mean absolute errors (MAEs) of 4.7 and 0.9 for root length and projection area, respectively, indicating small differences between model‐predicted and manually‐labelled values and minimal influence of larger errors. Similar model performance metrics, significant positive correlations, and significant linear relationships were observed between manually labelled and model‐predicted root length and projection area when analyses were conducted separately for topsoil and subsoil (Figure S1).

FIGURE 3.

Comparison of root length (a) and projection area (b) obtained from manual labelling and from the trained model. The red line indicates the 1:1 line. The model showed no systematic over‐ or under‐estimation (mean residuals for both length and projection area models were approximately zero [length: 2.9 × 10⁻¹⁷; projection area: −1.4 × 10⁻¹⁷]).

3.2. Proportion of Root Length and Root Projection Area in the Topsoil and Subsoil of Subarctic Grasslands in Ambient Conditions

Our results showed that in subarctic grasslands under ambient conditions, root length and root projection area were concentrated in the topsoil throughout the sampling seasons. On average, 63% ± 1% of root length and 65% ± 1% of root projection area were located in the topsoil. A seasonal variation was observed, with the percentage of topsoil root length ranging from 59% ± 3% in the third summer to 71% ± 5% in the first summer, and the percentage of topsoil root projection area ranging from 60% ± 3% in the third summer to 73% ± 4% in the first summer. When analysed separately for medium‐term warming (MTW) and long‐term warming (LTW), a similar seasonal pattern was observed, with the highest proportion of topsoil root length and projection area in the first summer and the lowest in the third summer (Figures S2 and S3). However, the two grassland sites differed in the vertical distribution of root length and projection area: root length and projection area were statistically similar between topsoil and subsoil in MTW, whereas in LTW, they were consistently higher in topsoil across all sampling seasons (Figures S2 and S3).

3.3. Responses of Root Length and Projection Area to Soil Warming Magnitude and Duration Are Season‐ and Depth‐Dependent

The repeated‐measures analysis of variance showed that soil depth affected both root length and projection area, whereas warming duration only affected root length (Table 1). Interestingly, soil depth alone did not affect either the rates of change in root length or root projection area from the previous season (Table 1), whereas the magnitude of warming affected only the rates of change in root length (Table 1).

TABLE 1.

The effect of soil warming duration (D; medium‐term and long‐term), magnitude (M; ambient, < +2°C, > +2°C), soil depth (Depth_soil; topsoil [0–10 cm] and subsoil [10–30 cm]) and seasons (repeated measured factor) on the root length, root projection area, and change in root length and change in the root projection area from the previous season (repeated measures ANOVA).

	Root length (mm)		Root projection area (mm2)		Change in root length (mm/day)		Change in root projection area (mm2/day)
	F	p	F	p	F	p	F	p
D	5.2	*	2.2	ns	1.4	ns	0.6	ns
M	0.3	ns	0.4	ns	3.4	*	2.2	ns
Depthsoil	16.4	****	24.0	****	0.1	ns	0.9	ns
D × M	1.8	ns	2.3	ns	1.7	ns	1.9	ns
D × Depthsoil	0.7	ns	0.5	ns	0.03	ns	0.0	ns
M × Depthsoil	0.5	ns	1.0	ns	1.8	ns	1.4	ns
D × M × Depthsoil	1.0	ns	0.5	ns	1.3	ns	0.7	ns
Season	37.2	****	32.5	****	17.8	****	24.9	****
Season × D	3.3	***	2.4	*	8.0	****	5.3	****
Season × M	3.6	****	3.1	****	5.1	****	2.7	***
Season × Depthsoil	4.2	****	4.6	****	4.2	***	4.7	***
Season × D × M	2.6	***	3.2	****	2.1	**	1.2	ns
Season × D × Depthsoil	0.3	ns	0.6	ns	0.2	ns	0.5	ns
Season × M × Depthsoil	1.9	*	2.1	**	0.9	ns	1.0	ns
Season × D × M × Depthsoil	0.8	ns	0.7	ns	0.7	ns	0.6	ns

Note: Significance was determined at the 0.05 level and presented with ns (p > 0.05), *(p < 0.05), **(p < 0.01), ***(p < 0.001), ****(p < 0.001).

Seasons had a highly significant effect on all measured root traits, including root length, projection area, and their rates of change. The seasonal effects on root traits were further indicated by significant interactions with soil depth, warming duration, and magnitude (season × depth, season × duration, and season × magnitude; Table 1). Furthermore, the three‐way interaction between season, duration, and magnitude (Season × D × M) was highly significant for root length, projection area, and change in root length, suggesting that the influence of duration and magnitude of soil warming on these traits was strongly season‐dependent (Table 1).

Moreover, root length and projection area showed a strong positive correlation across seasons, soil depths, warming durations, and warming magnitudes (Figure S4), indicating tightly coupled responses of these two root traits across all experimental factors in this study. In addition, root length and projection area exhibited similar responses to soil warming (Figures 4, 5, 6; Figures S5–S7). Given this consistency, only root length results are presented in the main text from this point onward, while projection area results are provided in the Supporting Information.

FIGURE 4.

Root length across seasons (spr, spring; sum, summer; aut, autumn; win, winter) over 3 years (1st: 2020; 2nd: 2021; 3rd: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient: Black; < +2°C: Orange; > +2°C: Pink) and durations ((a, c) medium‐term warmed [MTW; 11–14 years] and (b, d) long‐term warmed [LTW; > 60 years]). Error bars represent standard errors, and points indicate the mean root length. Seasonal differences within each warming duration, magnitude, and soil depth are indicated by lowercase letters (a, b), while differences between warming magnitudes within each season, duration, and soil depth are indicated with uppercase letters (X, Y) above the error bars. Only significant differences are shown.

FIGURE 5.

Change in root length across seasons (spr, spring; sum, summer; aut, autumn; win, winter) over 3 years (1st: 2020; 2nd: 2021; 3rd: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient: Black; < +2°C: Orange; > +2°C: Pink) and durations ((a, c) medium‐term warmed [MTW; 11–14 years] and (b, d) long‐term warmed [LTW; > 60 years]). Error bars represent standard errors, and points indicate the mean change in root length. Seasonal differences within each warming duration, magnitude, and soil depth are indicated by lowercase letters (a, b), while differences between warming magnitudes within each season, duration, and soil depth are indicated with uppercase letters (X, Y) above the error bars. Only significant differences are shown.

FIGURE 6.

Response ratio of mean root length (log₁₀(warmed/ambient)) across seasons (spr, spring; sum, summer; aut, autumn; win, winter) over three consecutive years (1st year: 2020; 2nd year: 2021; 3rd year: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (< +2°C orange; > +2°C: pink) and durations ((a, c) medium–term warmed [MTW; 11–14 years] and (b, d) long–term warmed [LTW;> 60 years]). Positive values indicate an increase in root length, and negative values indicate a decrease in root length compared to ambient conditions. The horizontal line at zero indicates ambient.

Seasonal dynamics in root length and projection area, and their daily rates of change, varied among sampling years, warming magnitudes, and warming durations (Figures 4 and 5; Figures S5 and S6). In the first year, root length and projection area in both soil depths increased in response to long‐term soil warming during the first half of the growing season (Figure 4; Figure S5b,d). However, these positive responses were transient; the daily rate of change in root length and projection area declined more rapidly during the second half of the growing season, particularly in the subsoil (Figures S8d and S9d).

Seasonal variability in daily rates of change in root length and projection area was strongest in the first year, particularly in the subsoil of > +2°C long‐term warming; however, these seasonal fluctuations became less evident by the third year (Figure 5d and Figure S6d). In the subsoil of > 2°C long‐warmed plots, root length and projection area were similar to ambient conditions at the start of the growing season (spring) but increased significantly by the peak of the growing season (summer) across all three study years (Figure 4d; Figure S5d). Enhanced summer root length and projection area under LTW were associated with higher positive rates of change, except in the third summer, when rates of change were similar to ambient conditions (Figure 5d; Figure S6d), although the root length and projection area remained higher (Figure 4d and Figure S5d).

Across all seasons, subsoil root length and projection area showed a consistent positive response to soil warming, with the most pronounced increase observed under long‐term warming of > 2°C (Figure 6d; Figure S7d). In contrast, root length and projection area in topsoil remained relatively unchanged (Figure 6b; Figure S7b), resulting in similar root length and projection area between the two soil depths in most sampling sessions, except during the first spring and summer when topsoil values were significantly higher (Figures S2 and S3).

3.4. Relationship Between Changes in Root Length, Projection Area, and Plant Species Abundance

RDA forward selection analysis with two sets of environmental predictors (Table S2) revealed that the observed variation in root length and projection area in LTW was explained by both plant species abundance and belowground plant biomass. In MTW, however, only belowground plant biomass was significant, whereas plant species abundance had no significant effect (p > 0.05).

In LTW, the abundance of plant species such as Kobresia myosuroides and Pilosella islandica explained 39.4% and 36.9% of the variation in root length and area, respectively, while topsoil rhizome biomass explained 26% and 22% of the variation in root length and projection area (Figure 7; Figure S10). RDA with all significant explanatory variables showed that a total of 57.8% (axis 1: 35.6%; axis 2: 13.8%) of the variation in root length and a total of 52.2% (axis 1: 32.5%; axis 2: 12.7%) of the variation in root projection area in LTW was explained by plant species abundance and belowground plant biomass (Figure 7; Figure S10).

FIGURE 7.

Ordination biplot based on redundancy analysis (RDA) of the root length in the topsoil (black solid arrows) and subsoil (grey dashed arrows) in relation to soil warming magnitude under long‐term warming (LTW; ambient [black], < +2°C [orange] and > +2°C [pink] triangles), rhizome biomass in the topsoil and abundance of plant species (blue arrows). Root length is shown across seasons (spr = spring, sum = summer, aut = autumn, win = winter) over 3 years (1 = 2020, 2 = 2021, 3 = 2022); for example, ‘spr_1’ = first spring (2020). In total, the model described 57.8% of the variation in root length (999 Monte Carlo permutation test, p < 0.05).

In LTW, root length and projection area in subsoil increased towards soil warming of > +2°C, which was related to higher abundance of Kobresia myosuroides and higher topsoil rhizome biomass (Figure 7; Figure S10). In contrast, root length and area in the topsoil did not show a clear response to soil warming and were negatively related to the abundance of Pilosella islandica (Figure 7; Figure S10).

4. Discussion

Our study provides the first evidence from a subarctic grassland that long‐term soil warming, particularly at higher intensities (> +2°C above ambient), drives a shift towards deeper rooting and affects seasonal root growth patterns. By comparing medium‐term (11–14 years) and long‐term (> 60 years) soil warming across multiple magnitudes, we revealed that root growth responses are not uniform but emerge from complex interactions between warming duration, warming magnitude, soil depth, and seasonality. Our findings underscore the importance of incorporating phenological dynamics and soil depth when assessing belowground responses to climate warming in northern ecosystems. Additionally, the automated root segmentation model used in this study performed robustly, with accuracy comparable to manual analyses, providing a tested and scalable solution for quantifying depth‐specific root growth dynamics across many time points, thereby overcoming a key methodological bottleneck in belowground climate change research.

4.1. Long‐Term Soil Warming Reshapes Root Distribution Across Soil Depths

Our study showed that soil depth and season were the most significant factors in describing variation in root length and projection area (Table 1). Under ambient conditions, across all seasons, the proportion of root length and projection area was consistently higher in the topsoil than in the subsoil (Figures S2 and S3). This shallow rooting pattern is a typical characteristic of high‐latitude grassland and tundra ecosystems (Jackson et al. 1996) and is commonly attributed to higher nutrient availability, higher microbial activity, and higher root decomposition rates in the topsoil (Gill et al. 2002), which supports higher root biomass (Bhattarai et al. 2023) and root densities (Bonin et al. 2013) in the upper soil layer.

Soil warming altered this vertical root distribution, with effects that depended on both the duration and magnitude of warming. Root growth response to warming differed between medium‐term and long‐term exposure, likely due to changes in nutrient dynamics, microbial communities, and plant physiological adaptations over time (Marañón‐Jiménez et al. ²⁰¹⁹; Metze et al. ²⁰²⁴; Walker et al. ²⁰¹⁸). Previous studies suggested that phenological responses, particularly in aboveground, are pronounced at the early stage of warming and decline over time due to plant acclimation or depletion of soil nutrients over time (Barrett and Hollister 2016; Lu et al. ²⁰²⁵). In contrast, our results show that medium‐term warming exhibited relatively moderate responses, whereas long‐term warming led to pronounced changes in subsoil root growth, especially at warming magnitudes exceeding 2°C above ambient (Figures 4 and 6; Figures S5 and S7). This suggests that belowground growth responses become more pronounced after long‐term exposure to warming, whereas medium‐term responses may reflect transient growth dynamics rather than stabilised root growth responses.

Under long‐term warming above +2°C from the ambient, subsoil root length and projection area increased substantially, whereas topsoil roots remained relatively stable (Figure 4; Figure S5), resulting in similar lengths and projection areas between the topsoil and subsoil (Figures S2 and S3). This finding suggests that prolonged soil warming shifted root distribution towards deeper soil layers, potentially due to changes in soil nutrient availability (Marañón‐Jiménez et al. ²⁰¹⁹), which also shifted biomass allocation to deeper layers. The proliferation of new roots in deep soil layers has the potential to stimulate the mineralisation of older organic matter (Dijkstra and Cheng 2007) and influence the microbial community composition and activity in the long‐warmed soils (Metze et al. 2024; Walker et al. ²⁰¹⁸). Warming is generally shown to increase root growth (Gill and Jackson 2000), and deeper soil layers become suitable for root growth as warming extends downward (Kaspar and Bland 1992). The subarctic research site where we conducted our study does not have permafrost, but it does have temporarily frozen soils during the winter months (Sigurdsson et al. 2016). Hence, the warming may have improved conditions for enhanced root growth towards deeper soil layers.

Soil physical properties may further modulate root responses to warming, as factors such as soil bulk density (BD) influence rooting depth and fine root traits, with less dense soils allowing penetration by thin fine roots (Clarke et al. 2013; Materechera et al. ¹⁹⁹²; van Eekeren et al. ²⁰¹⁰). However, in our study site, subsoil BD was significantly higher than that of the topsoil (Verbrigghe et al. 2022) and BD in warmed soil was slightly higher than in the ambient (Verbrigghe et al. 2022) suggesting potentially thinner roots and a finer root system under long‐warmed subsoil, allowing them to proliferate in smaller soil pores of dense soils. The previously observed increase in specific root area (SRA) in warmed soils (Fang et al. 2023) suggests a change in root morphology for better proliferation and nutrient acquisition (Freschet and Roumet 2017; Ostonen et al. ²⁰¹⁷). Such changes in fine root morphology and the resulting stratification of roots between topsoil and subsoil may affect microbial activity and nutrient cycling, as fine‐root morphological traits are closely linked to root respiration and nutrient mineralisation rates (Han and Zhu 2021; Kwatcho Kengdo et al. ²⁰²⁵). Consequently, these responses likely influence carbon and nitrogen dynamics in the belowground compartments of subarctic grasslands.

Despite pronounced changes in the subsoil, root length and projection area in the topsoil were comparable to those in ambient conditions under long‐term warming exceeding +2°C (Figure 4; Figure S5). This suggests a depth‐dependent warming response in roots, with surface soil roots becoming less responsive to prolonged warming while deeper soil roots remain highly sensitive. This potentially indicates that root growth in the topsoil may have reached a threshold or acclimated to sustained warming and that over long timescales, other constraints, such as nutrient availability, may have become a more important contributing factor than soil temperature alone (Barrett and Hollister 2016; Ernakovich et al. ²⁰¹⁴).

4.2. Soil Warming Affects the Seasonal Root Growth Dynamics

Our results showed that root growth response to warming was season‐dependent, supporting the general idea that root growth is dynamic and time‐dependent. Seasonal fluctuations in the vertical distribution of root length and projection area were most pronounced in the first summer after the minirhizotrons were installed, with topsoil root length and projection area reaching their highest values of 71% and 73%, respectively, before declining in subsequent summers. This pattern is consistent with observations from other minirhizotron studies, where initial root growth stimulation is often followed by stabilisation or decline due to changes in carbon allocation priorities in response to warming (Feng et al. 2022). Moreover, the highly significant three‐way interaction (Season × D × M) across all measured root traits (Table 1) suggested that warming‐induced changes in root dynamics vary across seasons. Previous research has shown that warming effects on root dynamics vary with seasonal shifts in soil moisture, temperature, and microbial activity (Abramoff and Finzi 2015; Piao et al. ²⁰¹⁹). Warming extended the aboveground growing season at the same site (Leblans et al. 2017; Mortier et al. ²⁰²⁴). Our findings suggest earlier growth in warmed plots, as indicated by higher root length and projection area in the first spring under long‐term warming (Figure 4; Figure S5).

Seasonal dynamics in the subsoil differed markedly from those observed in the topsoil. In the subsoil of long‐warmed plots (> +2°C), root length and projection area were similar to ambient conditions at the start of the growing season, but by summer, they were significantly higher than those in ambient conditions across all three study years (Figure 4d; Figure S5d). This pattern suggests that the influence of high long‐term soil warming may not be immediate, but rather manifest later in the growing season. In addition, subsoil roots under long‐term high warming also had faster growth rates in the summer (Figure 5d; Figure S6d), and the slope of increase in growth rate from spring to summer was steeper (Figures S8d and S9d). Following the increase in summer, subsoil roots declined sharply in winter (Figures S8d and S9d). Overall, these results suggest that long‐term warming altered the seasonal dynamics, as roots start similarly to those in ambient conditions in spring, grow faster to reach a larger size in summer, and then decline sharply afterwards. Such sharper declines likely reflect accelerated root turnover under warming, consistent with previous studies showing root growth is promoted but also leads to earlier root senescence due to increased respiration (Eissenstat et al. 2000; McCormack and Guo ²⁰¹⁴). Consistent with this interpretation, previous studies have shown that soil warming reduces fine‐root longevity and increases root litter inputs to the soil (Leppälammi‐Kujansuu et al. ²⁰¹³). However, direct measurements of root turnover rates were not made in this study, highlighting an important knowledge gap and the need for future work to explicitly quantify how warming influences root turnover rates.

By the third year, variability in the rate of seasonal root growth was reduced in long‐warmed plots (Figure 5d; Figure S6d), but warmed plots still had higher root length and projection area (Figure 4d; Figure S5d), suggesting possible acclimation or system stabilisation over time. Overall, warming amplified the seasonal dynamics of root growth: roots in warmed plots exhibited faster spring recovery, larger summer peaks and stronger winter declines, especially in deeper soil layers. These results underscore the importance of considering phenological timing when assessing root responses to soil warming.

4.3. Root Growth Dynamics Are Related to Changes in Plant Species Abundance Under Long‐Term Warming

Rooting depth largely depends on plant species and ecosystem type (Jackson et al. 1996). Species‐specific traits explain a substantial proportion of variation in root length across communities, often more than within‐species plasticity (Wang et al. 2024; Weemstra et al. ²⁰²¹). Thus, shifts in species composition would alter not only root length but also the spatial distribution of roots in soil, as species differ in their rooting depth and branching patterns (Bardgett et al. 2014). Warming has been shown to induce shifts in species composition towards deep‐rooted species (Liu et al. 2018), likely because species with deeper root systems can better exploit scarce nutrients under warming (Odone et al. 2024). Other studies have similarly reported shifts in belowground biomass to deeper soil layers associated with warming‐induced changes in nutrient availability and plant community composition (Li et al. 2011). At the ecosystem level, plant community composition influences belowground growth dynamics (Malhotra et al. 2020), and warming may favour stress‐tolerant or pioneer plant species (Cowles et al. 2016). Sedges are generally considered to be stress‐tolerant (Li et al. 2024). Consistent with these findings, the increase in subsoil root length and area under LTW (> +2°C) in our study was associated with higher topsoil rhizome biomass and a greater abundance of Kobresia myosuroides (Figure 7; Figure S10). Kobresia myosuroides is a sedge species characterised by dense turf formation, extensive rhizomes, and belowground biomass (Hinzke et al. 2021), as well as ectomycorrhizal associations (Lipson et al. 1999; Mühlmann and Peintner ²⁰⁰⁸). Sedges are also capable of deep rooting and exploiting subsoil nutrient pools down to depths of 70 cm (Peng et al. 2025). Increased rhizome biomass may provide additional carbon resources to support deeper root growth, while colonisation with mycorrhizal fungi can further facilitate nutrient uptake in soils (Read and Perez‐Moreno ²⁰⁰³). Thus, it is likely that as species composition and abundances changed under long‐term warming, root length and projection area increased more substantially in the subsoil.

4.4. Implications for Carbon Cycling in Subarctic Belowground Under Long‐Term Warming

Rooting patterns strongly influence how organic matter and nutrients are vertically distributed in the soil (Jobbágy and Jackson 2000; Souza et al. ²⁰²³). Deeper root systems enhance plant access to water and nutrients in subsoil layers, potentially increasing plant resilience to environmental variability (Fan et al. 2017). Changes in root biomass distribution also affect soil carbon dynamics, as root‐derived carbon inputs are a major contributor to soil organic matter formation (Verbruggen et al. 2016). Thus, the warming‐induced root proliferation into deeper soil layers may activate organic matter pools that were previously resistant to decomposition (Friggens et al. 2025; Henneron et al. ²⁰²²). Root inputs in deep soils can increase the C and N stocks, but the addition of labile C and N from root exudates (De Graaff et al. 2007) or from root turnover (Iversen et al. 2008) may also enhance the decomposition of older carbon stored in deeper soil layers (Fontaine et al. 2007). This phenomenon, known as the rhizosphere priming (Kuzyakov 2010; Kuzyakov et al. ²⁰⁰⁰), occurs when root‐derived substrate stimulates microbial activity and accelerates organic matter decomposition (Dijkstra and Cheng 2007) and C and N mineralisation (De Graaff et al. 2007; Shahzad et al. ²⁰¹⁸).

Changes in root growth and distribution may influence soil microbial communities and nutrient cycling, further impacting ecosystem‐level carbon and nutrient dynamics (Bardgett and van der Putten ²⁰¹⁴). Interestingly, soil depth did not significantly affect changes in root length and projection area from the previous season, suggesting that root growth and mortality rates may be similar across depths, despite geothermal warming in the subsoil potentially being slightly more intense (Sigurdsson et al. 2016). However, long‐term warming of more than 2°C had a more pronounced effect on root length and projection area in the subsoil, implying that deeper roots may be more responsive to long‐term exposure to extreme warming. It has been shown that warming‐induced topsoil C‐loss has stabilised at a new “steady state” in both MTW and LTW between 2013 and 2018, and no warming‐induced losses have been detected in the subsoil of LTW (Verbrigghe et al. 2022). These findings indicate that, over the long term, increases in root litter inputs may outweigh any initial warming‐induced priming of soil organic matter decomposition. Although deep root responses to warming imply potential changes in subsoil C cycling (Hicks Pries et al. 2017), it remains unclear whether turnover and decomposition rates of subsoil roots are affected by warming. Given the implications for global carbon storage, future research should focus on the turnover and decomposition rates of roots in deeper soils, a perspective also emphasised by Iversen (2010) to better predict C cycling dynamics under a warmer climate.

4.5. Methodological Strengths, Constraints and Future Directions in Root Studies

A well‐recognised challenge in root ecology is that belowground growth dynamics are still far less understood than aboveground dynamics due to the time‐consuming and labour‐intensive nature of root sampling and analysis (Jackson et al. 1996; Iversen ²⁰¹⁴; Freschet et al. ²⁰²¹). As a result, temporal root dynamics remain comparatively underexplored across ecosystems. The automated image analysis approach used in this study addressed this limitation by enabling rapid, large‐scale quantification of root traits from image datasets. Rather than a separate technical development, automated root image analysis serves as an integral tool in ecological root studies, addressing a key methodological bottleneck that has long constrained root research.

In the studied subarctic grassland, the high species diversity, with approximately 50 species, resulted in roots exhibiting a wide range of colours and shapes. While colour and morphological changes are commonly used as indicators of root senescence, this approach was not applicable in our study due to the substantial variation in root traits. Consequently, we could not distinguish between newly grown, pre‐existing, and dead roots; thus, our measurements reflected both root growth and mortality that occurred between sampling sessions. Thus, a key limitation of our study is that, in analysing root images, we could only measure root length and projection area at each sampling session without tracking individual roots throughout the seasons and determining root turnover rates. Future research could overcome this limitation by integrating advanced imaging techniques with computational models that enable the tracking of individual roots over time across diverse ecosystems, such as subarctic grasslands. In this study, only plant species present in at least 20% of plots were included to ensure robust statistical analyses. However, the effects of rare species on plot‐level growth dynamics cannot be overlooked, as they may disproportionately contribute to community responses. Future research should aim to sample a larger number of plots or employ targeted sampling approaches to capture rare plant species, thereby enabling a more comprehensive assessment of species‐specific and community‐level root growth patterns.

5. Conclusion

Root growth responses to the duration and magnitude of soil warming were strongly depth‐ and season‐dependent. Root growth in the topsoil acclimated to long‐term soil warming, while increased root length and projection area in the subsoil indicated a redistribution of biomass, potentially due to a shift in plant community composition towards deep‐rooted species or changes in nutrient availability and resource acquisition strategies. Under long‐term warming (> +2°C), subsoil roots exhibited clearer seasonal growth patterns, with the largest difference observed at the peak of the growing season. Warming shifted root growth phenology by accelerating peak‐season growth and intensifying late‐summer decline, highlighting the importance of considering phenological timing when assessing root responses to warming. Increased root growth in subsoil could partially offset warming‐induced increases in SOM decomposition, but if deep roots decompose more rapidly under high temperatures, subsoil carbon loss would accelerate. These findings underscore the need for future research to quantify root turnover and decomposition in deeper soil layers. Overall, our findings highlight the potential role of deep‐soil root responses to long‐term warming in shaping belowground carbon dynamics in subarctic grassland ecosystems.

Author Contributions

Biplabi Bhattarai: methodology, formal analysis, investigation, visualization, writing – original draft, writing – review and editing, conceptualization, validation, data curation. Páll Sigurdsson: methodology, writing – review and editing, investigation. Bjarni D. Sigurdsson: methodology, writing – review and editing, project administration, resources, funding acquisition, investigation. Pavel Baykalov: methodology, software, writing – review and editing, formal analysis, data curation, validation. Ivika Ostonen: conceptualization, methodology, investigation, validation, supervision, resources, project administration, writing – review and editing, writing – original draft, funding acquisition, formal analysis, visualization. Ivan Janssens: funding acquisition, writing – review and editing, project administration, methodology, resources. Boris Rewald: writing – review and editing, investigation, methodology. Coline Le Noir de Carlan: methodology, writing – review and editing.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Comparison of root length (a, b) and projection area (c, d) obtained from manual labelling and from the trained model in topsoil (0–10 cm) and subsoil (10–30 cm). The red line indicates the 1:1 line. The model showed no systematic over‐ or underestimation (mean residuals for both length and area were approximately zero for both topsoil and subsoil [topsoil length: 1.5 × 10^–16; topsoil area: 1.5 × 10⁻¹⁷; subsoil length: 5.0 × 10⁻¹⁷; subsoil area: 6.4 × 10⁻¹⁷]).

Figure S2: Root length across seasons (spring: spr; summer: sum; autumn: aut; winter: win) over 3 years (1st: 2020; 2nd: 2021; 3rd: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient; < +2°C; > +2°C) and durations ((a, c, e) medium–term warmed [MTW; 11–14 years] and (b, d, f) long–term warmed [LTW; > 60 years]). Root lengths are presented as bar plots with mean ± SE. Statistical comparisons of root length between topsoil and subsoil within each season, duration and magnitude of warming are presented as ns (p > 0.05), * (p < 0.05), ** (p < 0.01) and *** (p < 0.001).

Figure S3: Root projection area across seasons (spring: spr; summer: sum; autumn: aut; winter: win) over 3 years (1st: 2020; 2nd: 2021; 3rd: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient; < +2°C; > +2°C) and durations ((a, c, e) medium–term warmed [MTW; 11–14 years] and (b, d, f) long–term warmed [LTW; > 60 years]). Projection areas are presented as bar plots with mean ± SE. Statistical comparisons of root projection area between topsoil and subsoil within each season, duration and magnitude of warming are presented as ns (p > 0.05), * (p < 0.05), ** (p < 0.01) and *** (p < 0.001).

Figure S4: Correlation matrix showing Spearman correlation coefficients between root length and root projection area across seasons (spring: spr; summer: sum; autumn: aut; winter: win) over 3 years (1st: 2020; 2nd: 2021; 3rd: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient; < +2°C and > +2°C) and durations(medium–term warmed [MTW; 11–14 years] and long–term warmed [LTW; > 60 years]). Tile colours represent the strength and direction of correlation (blue = negative, red = positive). Asterisks (*) denote statistically significant correlations (p < 0.05).

Figure S5: Root projection area across seasons (spring: spr; summer: sum; autumn: aut; winter: win) over 3 years (1st: 2020; 2nd: 2021; 3rd: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient: black; < +2°C: orange; > +2°C: pink) and durations ((a, c) medium–term warmed [MTW; 11–14 years] and (b, d) long–term warmed [LTW; > 60 years]). Mean root projection area is indicated by dots, and error bars represent standard errors. Seasonal differences within each warming duration, magnitude and soil depth are indicated with letters (a, b), whereas differences between warming magnitudes within each season, duration and soil depth are indicated with capital letters (X, Y) above the error bars.

Figure S6: Change in root projection area across seasons (spring: spr; summer: sum; autumn: aut; winter: win) over 3 years (1st: 2020; 2nd: 2021; 3rd: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient: black; < +2°C: orange; > +2°C: pink) and durations ((a, c) medium–term warmed [MTW; 11–14 years] and (b, d) long–term warmed [LTW; > 60 years]). Mean change in root projection area is indicated by dots, and error bars represent standard errors. Seasonal differences within each warming duration, magnitude and soil depth are indicated with letters (a, b), whereas differences between warming magnitudes within each season, duration and soil depth are indicated with capital letters (X, Y) above the error bars.

Figure S7: Response ratio of mean root projection area [log₁₀(warmed/ambient)] across seasons (spring: spr; summer: sum; autumn: aut; winter: win) over three consecutive years (1st year: 2020; 2nd year: 2021; 3rd year: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (< +2°C: orange; > +2°C: pink) and durations ((a, c) medium–term warmed [MTW; 11–14 years] and (b, d) long–term warmed [LTW; > 60 years]). Positive values indicate an increase in root projection area, and negative values indicate a decrease in root projection area compared to ambient conditions. The horizontal line at zero indicates ambient.

Figure S8: Slope of daily change in root length across seasons (spring: spr; summer: sum; winter: win) over 3 years (1st: 2020; 2nd: 2021; 3rd: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient: black; < +2°C: orange; > +2°C: pink) and warming durations ((a, c) medium–term warmed [MTW; 11–14 years] and (b, d) long–term warmed [LTW; > 60 years]). Error bars represent standard errors, and points indicate the mean slope. Seasonal differences within each warming duration, magnitude and soil depth are presented as lowercase letters (a, b), whereas differences between warming magnitudes within each season, duration and soil depth are indicated with * (p < 0.1) and * (p < 0.05). Only significant differences are shown.

Figure S9: Slope of daily change in root projection area across seasons (spring: spr; summer: sum; winter: win) over 3 years (1st: 2020; 2nd: 2021; 3rd: 2022) in topsoil (0–10 cm) and subsoil (10–30 cm) of subarctic grasslands exposed to different soil warming magnitudes (ambient: black; < +2°C: orange; > +2°C: pink) and warming durations ((a, c) medium–term warmed [MTW; 11–14 years] and (b, d) long–term warmed [LTW; > 60 years]). Error bars represent standard errors, and points indicate the mean slope. Seasonal differences within each warming duration, magnitude and soil depth are presented as lowercase letters (a, b), whereas differences between warming magnitudes within each season, duration and soil depth are indicated with * (p < 0.1) and * (p < 0.05). Only significant differences are shown.

Figure S10: Ordination biplot based on redundancy analysis (RDA) of the root projection area in the topsoil (black solid arrows) and subsoil (grey dashed arrows) in relation to soil warming magnitude under long‐term warming (LTW; ambient [black], < +2°C [orange] and > +2°C [pink] triangles), rhizome biomass in the topsoil and abundance of plant species (blue arrows). Root length is shown across seasons (spr = spring, sum = summer, aut = autumn, win = winter) over 3 years (1 = 2020, 2 = 2021, 3 = 2022); for example, ‘spr_1’ = first spring (2020). In total, the model described 52.2% of the variation in root projection area (999 Monte Carlo permutation test, p < 0.05).

Table S1: The range of soil warming magnitude from the ambient (2013–2022 average;°C) at grassland sites subjected to two soil warming durations (medium–term warmed [MTW; 11–14 years], and long–term warmed [LTW; > 60 years]). Soil warming was categorised into three classes according to the level of additional impact/risk depending on the magnitude and rate of warming (IPCC ²⁰²²): ambient, < +2°C and > +2°C of T _s increase.

Table S2: Dataset consisting of two sets of environment variables: Fine root and rhizome biomass (FRB and RHB) in topsoil (0–10 cm) and subsoil (10–40 cm) and abundance of plant species averaged over three soil warming magnitudes (ambient [+0°C], < +2°C and > +2°C) and two warming durations: (medium‐term warming [MTW; 11–14 years] and long‐term warming [LTW; > 60 years]). Fine root and rhizome biomass (Bhattarai et al., 2023) and plant species abundance (Meynzer 2017) were measured in previous studies.

Data Availability Statement

The data that support the findings of this study are openly available in Zenodo repository at https://doi.org/10.5281/zenodo.20429148.