The study was conducted in a vineyard located on a gentle slope with a 5° incline south of Fellbach (48.80158° N, 9.28113° E) in southwestern Germany. The site is managed by the wine cooperative “Fellbacher Weingärtner” and the Lemberger grape variety is cultivated with continuous cover crops between vine rows. The soil was identified as a Mollic Anthrosol (Relocatic), reflecting anthropogenic modification through deep ploughing, a practice frequently employed in vineyards (IUSS Working Group WRB, 2022). The soil was classified as a clay loam throughout the profile. In the 0–25 cm layer, the soil contained 23 % sand, 39 % silt, and 38 % clay, with a bulk density of 0.96 g cm⁻³. The 25–90 cm layer also had a clay loam texture, comprising 24 % sand, 42 % silt, and 34 % clay, however, the bulk density was not measured here.

The study was carried out at the experimental vineyard described by Gall et al. (2025), using the same treatment framework and sizes. However, it focuses on a different research objective by investigating soil moisture dynamics instead of runoff and erosion processes.

Experimental treatments were established on 17 February 2022 within four vine rows at the study site. Three treatments were applied, each on a surface of 40 cm × 40 cm: moss restoration (moss), bare soil (bare), and cover crops (grass), each replicated four times. The plot size corresponds to the dimensions of the micro-scale runoff plots used in previous rainfall simulations at the same study site (Gall et al., 2025) and was selected to match the width of a single vine row (approximately 40 cm). Larger plots would have extended beyond the vine row and included the adjacent tractor track, which was intentionally excluded from the measurements. The pre-cultivated moss mats used in the experiment were adapted to this size, resulting in complete coverage of each moss treatment plot. The bare treatment was maintained by periodic weeding to keep vegetation cover below ∼ 20 %. Establishment of both the moss and bare treatments involved removing existing cover crops and, where possible, associated roots by gentle, non-soil disturbing weeding. The grass treatment retained the existing cover crop mixture dominated by the grasses Lolium perenne and Trisetum flavescens, with occasional occurrences of Trifolium repens and Achillea millefolium (identified using Jäger and Werner, 2005). The moss treatment involved installing pre-cultivated moss mats (produced by Reinhold Hummel GmbH+Co.KG, Stuttgart, Germany) with five species: Amblystegium serpens (Hedw.) Schimp., Brachythecium rutabulum (Hedw.) Schimp., Funaria hygrometrica Hedw., Homalothecium lutescens (Hedw.) Robins, Oxyrrhynchium hians (Hedw.) Loeske on weeded soil surfaces which were initially watered to promote establishment.

For this study, continuous topsoil moisture data was recorded with time-domain-transmission sensors by TMS-4 dataloggers (TOMST, Czech Republic) installed in all plots on 14 April 2022. Measurements were conducted at a spatial scale corresponding to the 40 × 40 cm treatment plots, with four replicate measurements per treatment to account for local spatial variability. The TMS-4 dataloggers are designed to mimic a small herbaceous plant and capture its microclimate, so that each device measures volumetric topsoil moisture to ∼ 14 cm depth and air/soil temperature at three heights (15 cm above ground, 2 cm above ground, and 6 cm below ground; Wild et al., 2019). Given this fine-scale observation, the measurements are intended to characterize topsoil moisture dynamics at the microsite scale and are not designed to capture the full spatial integration of infiltration and percolation processes across a vineyard row or hillslope. To synchronize measurements and avoid temporal offsets, all dataloggers were installed simultaneously. All TMS-4 measurements were logged in Coordinated Universal Time (UTC). Data were collected over a 22-month period for analysis at 5 min intervals and cross-checked to ensure data validity and quality. Datalogger calibration was performed using soil parameters specific to the study site, including clay, sand, and loam fractions. Soil characteristics were derived from laboratory analyses reported in Gall et al. (2025), encompassing grain size distribution, soil pH, organic carbon content, and bulk density.

Monitoring the development of the treatments served to document how their conditions changed over time and to identify periods during which they remained functionally distinct. These observations were necessary for subsequent analyses but not part of the hypothesis testing.

The study was structured into distinct periods based on management practices and expected vegetation dynamics.

Maintenance period (April–October 2022): During this period, treatment conditions were actively maintained. Grass was manually removed from moss and bare soil plots using cutting or gentle hand-weeding, while avoiding disturbance of the soil.

First succession period (November 2022–February 2023): Maintenance ceased at the start of November 2022, allowing natural succession to proceed during the winter months, when limited vegetation growth was expected.

Second succession period (March 2023–February 2024): This period encompassed the full growing season and subsequent months, beginning in spring 2023. It allowed evaluation of whether moss treatments could maintain dominance over grasses throughout an entire growing season without intervention.

Target period (September–November 2022): During this three-month transition from maintenance to the first succession phase, moss cover was vigorous and green, mat material fully decomposed, and soil largely bare with minimal grass.

Ground vegetation cover was assessed using the quadrat method, following the approach of Belnap et al. (2001) and as applied by Gall et al. (2022). For each plot, a digital grid of 100 squares was overlaid on a photograph, and each square containing more than 50 % vegetation was counted. Importantly, the assessment considered only vegetation emerging directly from the surface within the grid square, rather than overhanging parts such as blades of grass extending from adjacent areas. For moss treatments, total vegetation cover was further distinguished between moss and grass (Table 2). Each vegetated grid square was carefully evaluated using field experience to determine whether moss or grass dominated the area. All values reported as absolute percentages of the total plot area (not relative to each other).

Meteorological data were obtained from the agricultural weather station in Fellbach (Agrarmeteorologie Baden-Württemberg), located within the vineyard approximately 10 m from the experimental plots (48.80158° N, 9.28113° E). The station records precipitation (mm, 10 min intervals), global solar radiation (Wh m⁻²), air temperature (2 m above ground), relative humidity (%), and other meteorological variables. Due to its close proximity to the experimental plots, the Fellbach station provides a representative characterization of the local microclimate. Therefore, the period 2006–2021 was used as the local reference period for evaluating meteorological conditions during the study period (April 2022–February 2024).

Relative to the local reference period, the study period was warmer (+1.4–1.6 °C) and drier (-10 %–13 %), with slightly higher global radiation in 2022 (+2 %) and lower radiation in 2023 (-9 %) (Table 1; Agrarmeteorologie Baden-Württemberg, 2025a, b). Seasonal differences within the study period were also pronounced, with a hot and dry summer in 2022 (June–August: 21.9 °C, 157 mm) compared with a similarly warm but wetter summer in 2023 (21.3 °C, 211 mm). Winter conditions also differed, with winter 2023/24 being warmer and wetter than winter 2022/23 (Fig. 1).

At the national scale, 2022 was among the warmest and sunniest years since 1881, characterized by multiple summer heatwaves and drought conditions, whereas 2023 was the warmest year on record and had above-average precipitation (DWD, 2022, 2023a, b, 2024). Compared with the German annual average, the Fellbach site showed consistently warmer and drier conditions (Table 1). These contrasting meteorological conditions provide the climatic context for interpreting soil moisture and vegetation dynamics during the experiment.

To characterize precipitation dynamics during the experiment, precipitation data from April 2022 to February 2024 were analysed and classified according to German Meteorological Service (DWD, 2026) intensity thresholds using both 10 and 60 min precipitation records: violent: ≥ 50 mm h⁻¹ and/or ≥ 8.3 mm (10 min⁻¹); heavy: ≥ 10 mm h⁻¹ and/or ≥ 1.7 mm (10 min⁻¹); moderate: ≥ 2.5 mm h⁻¹ and/or ≥ 0.5 mm (10 min⁻¹); light: below the above thresholds.

A total of 422 precipitation events were identified, comprising 285 light, 101 moderate, 34 heavy, and 2 violent events, their distribution over the study periods is shown in Table 2. Of all events, 313 met both the 10 min and 60 min criteria, while 109 were classified based on the 10 min threshold only. For each event, duration, peak timing, and mean precipitation rate were calculated. Event durations ranged from < 1 to > 10 h (median = 1.3 h, mean = 2.7 h), with peak intensities typically occurring within the first hour. Maximum hourly precipitation reached 17.6 mm, and mean event intensities ranged from 0.07 to 10.8 mm h⁻¹ (median = 0.43 mm h⁻¹), indicating that short, low-intensity events dominated the study period.

All analyses were performed in R version 4.3.1 (R Core Team, 2025). Shapiro–Wilk and Levene's tests were used to assess normality and homoscedasticity. As the data did not meet these assumptions, non-parametric methods were applied where appropriate. Statistical significance was set at p<0.05, and mean values are reported ± standard error. Data visualization was conducted with the R packages ggplot2 (Wickham, 2016) and based on wesanderson (Karthik et al., 2018).

The following observations document the development of treatments and provide context for interpreting subsequent hypothesis-driven analyses.

During the maintenance period (April–October 2022), treatments remained functionally stable, with grass plots maintaining full (100 %) vegetation cover, bare soil plots exhibiting a mean cover of 7.16 %, and moss plots averaging 81.58 % cover (Table 3). Moss plots experienced drying during April and June, with parts of the underlying jute fleece still visible and a temporary reduction in moss cover (Fig. 2). By the end of the period, only one moss plot contained any grass (4 % cover), indicating that, despite seasonal stress, mosses effectively prevented grass establishment.

Despite low temperatures and low global solar radiation during the first succession period (November 2022–February 2023; Fig. 1), clear signs of colonization by grasses were visible in both moss and bare plots. Between October and December, vegetation cover in bare soil plots increased sharply from a mean of 11.25 % to 57.5 %. Moss plots remained more resistant, with grass cover reaching only 12.25 % while mosses still occupied a mean of 61 % of the surface (Table 3). Functional differences between treatments were still apparent at this stage.

Over the 11 months of the second succession period (March 2023–February 2024), vegetation cover in bare soil plots increased from 67.5 % to 91.5 %, dominated by grass but also encompassing mosses (Fig. 2). In moss plots, grass cover expanded from 19.25 % to 45.5 %, while moss cover declined from 61.5 % to 29.75 % (Table 3). By this stage, functional differences between treatments had largely disappeared, and vegetation structure in moss and bare plots had converged towards that of the grass treatment. Monitoring was therefore concluded.

The target period (September–November 2022) captured the phase in which treatments were functionally most distinct. During this period, moss plots exhibited high vitality, as indicated by their green colouration (Fig. 2), while vegetation cover in bare soil plots remained minimal (Table 3), providing optimal conditions for assessing treatment effects.

Topsoil-moisture responsiveness varied strongly with precipitation intensity and surface treatment. Non-response rates declined markedly with increasing precipitation intensity across all treatments. During light precipitation, non-response was most frequent under grass (36.7 %), compared with bare soil (11.1 %) and moss (11.1 %). Under moderate precipitation, non-response occurred in 19.5 % of grass sensors, 5.9 % of bare soil sensors, and 6.8 % of moss sensors. Heavy precipitation further reduced non-response to 14.1 % in grass, 0.9 % in bare soil, and 0 % in moss. During violent precipitation events, non-response was observed only in grass (12.5 %), while both bare soil and moss exhibited complete responses (0 %). Pairwise contrasts of estimated response probabilities from the binomial GLMM confirmed that grass differed significantly from both moss (p<0.0001) and bare soil (p<0.0001), whereas moss and bare soil did not differ significantly (p=0.10).

Initial response time (IRT) differed significantly among treatments across the full dataset (Fig. 3), following the pattern grass > moss > bare soil, with grass showing the longest IRT, moss intermediate, and bare soil the shortest (grass vs. moss p<0.0001; grass vs. bare soil p<0.0001; moss vs. bare soil p=0.253). When analysed by period, significant treatment effects were detected during the maintenance period (April–October 2022) between bare soil and grass (p=0.022) and between grass and moss (p=0.048). In the first succession period (November 2022–February 2023), grass again exhibited longer IRT than both moss and bare soil (p<0.0001 for both), with moss also exceeding bare soil (p=0.044). During the second succession period (March 2023–February 2024), grass maintained significantly higher IRT than moss and bare soil (p<0.0001 for both), while moss and bare soil did not differ (p=1). In the target period (September–November 2022), grass again exceeded both moss and bare soil (p<0.0001 for both), with no significant difference between moss and bare soil (p=1).

IRT generally declined with increasing precipitation intensity, with light events producing the slowest responses, followed by moderate and heavy events. This pattern was most pronounced during the maintenance period, when all event categories were present. IRT during light events was significantly longer than during heavy events (p=0.0002), and moderate events also responded more slowly than heavy ones (p=0.01). In the first succession period, only light and moderate events occurred, and their IRTs did not differ significantly (p=0.162). During the second succession period, when all event categories were recorded, IRTs were significantly longer during moderate events than during heavy events (p=0.005). In the target period, when light, moderate, and heavy events occurred, light precipitation events again showed significantly slower responses than moderate ones (p=0.009). Differences involving violent events were not significant in any period, likely due to their low frequency (Table 2).

The difference between grass and moss or bare soil was most pronounced under light and moderate events (all p<0.001), and remains significant, but less strong, under heavy events (grass > bare p=0.0022; grass > moss p=0.0004). The IRT of moss and bare soil converged with increasing intensity but never differed significantly (light: p=0363, moderate p=0.592, heavy p=1).

Peak response time (PRT) differed among treatments across the full dataset regardless of precipitation intensity (Fig. 4). Grass reached peak topsoil moisture significantly faster than bare soil (p=0.0005) and moss (p<0.0001), whereas moss and bare do not differ significantly. When analysed by defined periods, treatment differences were only evident during the second succession period. In this period, grass exhibited significantly faster PRT than bare soil (p=0.0001) and moss (p<0.0001), while moss and bare soil again did not differ significantly.

PRT generally varied with precipitation intensity, though temporal patterns were less consistent than for IRT. Over the entire study period, light and moderate events showed the longest PRTs, with no significant difference between them, and both were significantly longer than heavy events (light > heavy, p=0.0002; moderate > heavy, p<0.0001). Violent events had the shortest PRTs, but observed differences to light (p=0.0036) and moderate (p=0.0014) events should be interpreted cautiously due to low number of events (Table 2). During the maintenance period, moderate events had the longest PRT of all intensity categories. It was significantly longer than under light (p<0.0001) and heavy events (p<0.0001). Other pairwise comparisons were not significant. During the first succession period, light events had significantly longer PRT than moderate events (p<0.0001). During the second succession period, there was no significant difference between the PRT of light and moderate events. Light events had longer PRT than heavy (p=0.008) and violent events (p=0.002), while moderate events had longer PRT than heavy (p=0.027) and violent events (p=0.0064). During the target period, the PRT of moderate events was again significantly longer than that of light and heavy events (both p<0.0001), the latter two were no different from each other.

Differences between treatments within precipitation event categories were only found under light and heavy events. During light events, grass showed significantly longer PRT than bare soil (p=0.0049) and moss (p=0.007), whereas bare and moss did not differ significantly (p=1). During moderate events, no significant differences between treatments were observed (all p>0.1). For heavy events, grass again had the longest PRT, differing significantly from moss (p<0.0001) and bare soil (p=0.0032), while moss and bare soil remained similar (p=1). No significant differences between treatments were detected during violent events, likely reflecting the limited number of violent events recorded (n=2).

To assess whether moss restoration maintains higher topsoil moisture through slower drying compared to bare soil and grass, we analysed topsoil moisture dynamics across the three treatments, focusing on both mean moisture and drying rates. Topsoil moisture exhibited strong seasonal patterns, with lower values in summer and higher values in autumn and winter, and pronounced non-linear temporal variation across all treatments (Fig. 5). GAMM results indicated no significant differences between treatments over the full study period from April 2022 to February 2024 (all p>0.1). Using bare soil as a reference, significant differences were observed only during the maintenance period (April–October 2022), when grass plots had slightly lower topsoil moisture than bare soil (p=0.03). Although not statistically significant, the trend continues in the second succession period (March 2023–February 2024), with bare soil maintaining the highest mean soil moisture (Fig. 6).

Across all periods, between-sensor variability was consistently significant (random effect, p<2×10-16), indicating systematic differences among measurement locations, while the GAMM models explained 84 %–92 % of the deviance. Overall, temporal variation and seasonal trends had a much stronger influence on topsoil moisture than treatment effects, and differences between treatments were relatively minor.

To test whether moss slows topsoil drying compared to bare soil and grass, drying rates were calculated over 24 h following the topsoil moisture peak. Pairwise comparisons of median 24 h drying rates (Fig. 7) revealed significant differences among treatments, following the pattern moss ≈ bare > grass. Drying rates were higher in moss than in grass (p<0.0001) and higher in bare soil than in grass (p=0.0009), while drying rates did not differ significantly between bare soil and moss (p=1).

When extended to a 72 h drying window, the overall pattern (moss ≈ bare > grass) persisted, but differences were no longer statistically significant. The reduced number of events with a full 72 h drying period (65 events) likely limited statistical power compared to the 24 h analysis (135 events).

The generalized additive model (GAM) revealed strong nonlinear seasonal patterns in initial response time (IRT), which differed among treatments and aligned with the LOESS visualization (Fig. 8). Across the full dataset, grass showed the longest overall IRT, while bare soil and moss responded faster (parametric terms: grass > moss > bare, all p<0.0001). Temporal variability of IRT was most pronounced in grass p<0.0001), followed by moss (p<0.009), while bare soil exhibited weaker but still significant seasonal fluctuations (p=0.017). Overall, the model explained about 11 % of the variation in IRT, suggesting that treatment and seasonality together contributed modest but significant effects to the timing of topsoil moisture responses.

The most pronounced difference occurred between grass and bare soil, emerging in September 2022 (p=0.004) and remaining highly significant from October 2022 to December 2023 (p<0.0001), before gradually declining in early 2024 (January p=0.0008; February p=0.0453). The grass–moss contrast followed a similar pattern, becoming significant in September 2022 and persisting until January 2024 (September 2022–December 2023 p=0.0009, January (p=0.0018). Differences between bare and moss were limited, occurring only in December 2022 (p=0.035) and January 2023 (p=0.0317). It should be noted that data from April 2022 and February 2024 represent only partial months, resulting in a reduced number of observations.

Variability in IRT remained low and similar among treatments during the early study period but diverged in later periods (Fig. 9). Across the maintenance period (April–October 2022), variability did not differ among treatments (Levene's p = 0.424). In the first succession period (November 2022–February 2023), grass exhibited greater dispersion than moss and bare soil, with variance differences reaching significance (p = 0.020). Variability further diverged during the second succession period (March 2023–February 2024), with significant differences among treatments (p < 0.001) and grass showing the highest variance. In the target period (September–November 2022), differences in variability were also significant (p = 0.017), primarily driven by the higher dispersion under grass. These temporal patterns suggest that the heterogeneity of IRT increased over time, particularly under grass cover.

Discussing the natural succession of treatments provides essential context for interpreting topsoil moisture responses.

Grass establishment in bare-soil plots resumed rapidly after maintenance ceased, likely reflecting favourable late-autumn conditions, including mild temperatures and high humidity, as well as reduced disturbance following the harvest. Similar growth responses of Lolium perenne, the dominant species in our grass treatment, under moist and mild conditions have been reported by Celette et al. (2008), although primarily for spring growth.

Moss plots initially suppressed grass colonization, consistent with findings summarized by Qu et al. (2023), who reported that mosses can inhibit seed germination and early seedling establishment. Grass gradually expanded over subsequent months, while moss cover declined; however, mosses persisted for over a year, highlighting their ability to maintain dominance under limited disturbance. These dynamics indicate that mosses can delay grass colonization during early succession but are ultimately outcompeted over time.

Seasonal and environmental conditions further shaped these patterns. Mosses remained green and vigorous throughout winter, likely supported by persistently high relative humidity, moderate temperatures and limited solar radiation exposure, which reduce desiccation stress (Figs. 1 and 2). This underscores the importance of moisture availability and microclimatic buffering for moss persistence under field conditions.

By the end of the experiment, functional differences between all treatments had largely converged, with bare-soil plots recolonized by grasses and moss plots containing a mixed cover. This convergence demonstrates that moss restoration can persist for extended periods but may require occasional interventions, such as selective grass removal or seasonal trimming, to maintain long-term dominance and achieve desired under-vine management outcomes.

The observed succession patterns highlight that moss restoration outcomes are strongly influenced by environmental conditions and management practices. While the present study provides insights into moss persistence and associated soil moisture responses under the investigated vineyard conditions, further studies across contrasting climates, soil types, and management systems are needed to evaluate the transferability of these findings. In addition, the feasibility of implementing pre-cultivated moss mats at larger spatial scales, including full under-vine or between-row coverage, are subject to further investigation.

Our results support the hypothesis that moss restoration does not significantly delay topsoil moisture response times after precipitation events. While grass cover slowed both initial and peak response times, moss treatments showed only minor delays compared to bare soil, indicating that moss did not impede infiltration under the conditions of this study.

To analyse response times accurately, it was necessary to exclude non-response events. Such thresholds are common: da Silva-Dias et al. (2024) reported topsoil moisture responses only above 1.2 mm h⁻¹, whereas in our study the threshold was much lower (< 0.5 mm h⁻¹). This highlights the importance of determining site-specific thresholds in future studies to ensure meaningful interpretation of topsoil moisture dynamics.

Initial response times (IRT) differed consistently among treatments, indicating that vegetation structure modulates the onset of infiltration. This supports our hypothesis that moss restoration does not significantly delay topsoil moisture response, even though vegetation cover per se can slow infiltration. Similar observations under high-intensity precipitation have been reported in experimental studies, where mosses did not act as a barrier to infiltration (Li et al., 2016; Thielen et al., 2021; Zi et al., 2026). The delayed response under grass likely reflects an interception effect, where dense vegetation prevents precipitation from reaching the soil surface, thereby delaying infiltration (Bodner et al., 2008; Huang et al., 2023; da Silva-Dias et al., 2024).

At the same time, the influence of moss on infiltration is context dependent. Previous work has demonstrated that infiltration responses reflect not only surface cover but also underlying soil structure, with well-developed, stable topsoil horizons promoting rapid infiltration irrespective of vegetation type (Leonard and Andrieux, 1998). Reviews of biocrusts further emphasize that moss-dominated surfaces can either enhance or inhibit infiltration depending on coverage, thickness, and developmental stage, with early or less dense crusts generally facilitating water entry (Ya et al., 2025). In the present study, the moss was applied as cushion-like mats rather than compact crusts, which may partly explain the absence of infiltration delays observed here. Experimental and modelling studies confirm this complexity, identifying crust thickness, organic matter content, and biocrust maturity as key controls, with infiltration rates often declining as moss-dominated crusts become more developed (Yang et al., 2025; Zhao et al., 2025).

Precipitation intensity further modulated IRT across treatments, with faster responses under higher intensities, consistent with threshold-like infiltration behaviour. Similar intensity-dependent effects have been reported for biocrusts, where moss-dominated surfaces can enhance infiltration during prolonged, low-intensity precipitation but reduce it during short, high-intensity events (Ju et al., 2024). More generally, precipitation characteristics often outweigh vegetation effects in controlling soil moisture response, particularly under variable antecedent conditions (Zhang and Liu, 2022). Together, these findings indicate that while moss restoration does not delay initial topsoil moisture response under the conditions studied here, its hydrological role is contingent on both precipitation and plant development.

Peak response times (PRT) were generally longest under grass, while moss and bare soil exhibited similarly rapid peaks, further supporting the hypothesis that moss restoration does not substantially delay topsoil moisture responses following precipitation. These differences indicate that vegetation structure influences not only the onset but also the temporal evolution of topsoil moisture dynamics.

The delayed peak observed under grass likely arises from mechanisms similar to those influencing initial response times. Although grass has been shown to reduce surface runoff compared to bare soil (Rodrigo Comino et al., 2016; Duan et al., 2022; Gall et al., 2025), dense aboveground biomass and surface residues may temporarily reduce water entry into the soil. In contrast, grass root systems can enhance soil structure by increasing porosity and infiltration capacity over time (Duan et al., 2022).

The delayed peak observed under grass likely reflects altered infiltration and redistribution processes induced by the grass root system. Although grass cover generally reduces surface runoff compared to bare soil (Rodrigo Comino et al., 2016; Duan et al., 2022; Gall et al., 2025), root architecture can influence the dominant water flow pathways by modifying soil porosity, infiltration capacity, and preferential flow development (Duan et al., 2022, 2026). Moss restoration has likewise been shown to reduce surface runoff (Xiao et al., 2015; Tu et al., 2022; Gall et al., 2025), but due to mosses' poikilohydric characteristics and lack of deep rooting systems (Proctor et al., 1998), infiltration processes differ fundamentally from those under grass cover.

PRT generally varied with precipitation intensity, but temporal patterns were less consistent than for IRT, partly due to unexpectedly high estimates for moderate events during the maintenance period. This variability may also reflect the inherent complexity of soil moisture response dynamics, as emphasized by da Silva-Dias et al. (2024), who showed that the magnitude and timing of soil moisture responses can represent partly independent processes. To our knowledge, studies explicitly focusing on peak response times in moss-dominated systems are lacking, which limits broader generalization of the observed patterns.

Across both IRT and PRT, our ability to assess extreme precipitation is limited because only two violent events occurred during the study. This is an inherent constraint of field experiments relying on natural precipitation. Targeted precipitation simulations could help determine whether extreme storms override vegetation effects or amplify differences between treatments.

Results show that moss restoration does not increase topsoil moisture through slower drying and our second hypothesis was thus rejected. Mean topsoil moisture did not differ significantly among treatments, except during the maintenance period, when grass plots exhibited slightly lower values than bare soil. Although not significant, this pattern re-emerged during the second succession period (March 2023–February 2024).

Although bare soil lacks protective cover and would therefore be expected to dry more rapidly, the absence of vascular vegetation and associated plant water uptake may counterbalance enhanced evaporative losses, resulting in higher mean topsoil moisture. This interpretation is consistent with findings by Capri et al. (2023), who showed that grasses generally exhibit the highest daily water uptake due to evapotranspiration, but that mowing can substantially reduce water use, in some cases to levels below those of bare soil. Similarly, Monteiro and Lopes (2007) observed higher topsoil moisture in tilled vineyard soils compared to plots with resident vegetation or sown cover crops, which they attributed to reduced transpiration and interception losses. Together, these results highlight that vegetation effects on topsoil moisture reflect a balance between surface protection and plant water uptake rather than vegetation cover alone.

Comparisons with other studies show that the hydrological role of mosses is highly site-dependent, influenced by soil structure and climate. Hu et al. (2023) found that in alpine forest soils, bare surfaces retained more moisture than moss cover in the upper 0–10 cm, whereas moss increased retention at 10–30 cm depth, raising the question of whether the higher surface moisture in our bare plots might invert at greater depths. In contrast, Lulu and Dongdong (2025) reported that in karst soils, moss establishment reduced surface evaporation by 48 %–72 % and increased water storage by 11 %–22 %, partly through surface sealing within the top 5 cm. In addition, Li et al. (2026) recently demonstrated that moss carpets grown in a greenhouse on burlap increased soil moisture by 16.11 % at a depth of 10 cm when planted in the field on an aeolian sandy soil in comparison with a bare soil control.

Drying rates followed the pattern moss ≈ bare > grass. Contrary to our expectations, grass did not exhibit the fastest drying despite its active root system; instead, moss and bare soil dried more rapidly. This pattern suggests that grass primarily accesses water from deeper soil layers beyond the 14 cm measurement depth, while its dense vegetation cover reduces direct soil evaporation. This interpretation is consistent with findings by Liu et al. (2020), who identified root length density at 5–30 cm depth and root surface area at 10–20 cm depth as key factors influencing infiltration rates, indicating that root effects on soil hydraulic processes are strongly depth-dependent. Similar depth-dependent root activity has been observed in vineyards by Celette et al. (2008), who reported that a perennial cover including Lolium perenne, the dominant species in our cover crop (grass), concentrates most of its roots within the upper 0.5 m but can extend down to 1 m. This rooting strategy enables efficient water uptake from deeper layers while simultaneously limiting near-surface drying through shading and reduced evaporation.

In contrast, the higher drying rates observed under moss and bare soil indicate a stronger coupling between the soil surface and the atmosphere. Rice et al. (2018) showed that moss layers are thermally structured, with cooler shoot tips creating vertical temperature gradients. Moreover, coupling with the atmosphere is strong when mosses are wet but weak when dry, and the structural characteristics have a greater influence on these dynamics than species identity. The size and surface roughness of moss cushions may create a threshold beyond which further increases in roughness either enhance or restrict water loss, with the leaf area-to-volume ratio reflecting the balance between evaporative surface and water storage capacity (Rice and Schneider, 2004). In our study, this suggests a high fluctuation in water storage within the moss treatments over time, as mosses were visibly dry in early 2022 (April and June; see Sect. 3.2), yet topsoil moisture beneath moss remained similar to that under grass and lower than bare soil. Controlled laboratory experiments could help disentangle the specific mechanisms by which moss traits, such as structure, roughness, and vitality, affect topsoil moisture, independent of other environmental factors.

In parallel, interactions between ground vegetation and grapevine root systems further complicate soil moisture dynamics. Although grapevines generally develop deeper root systems, grasses can still compete for water in superficial soil layers (Smart et al., 2006; Lehnart et al., 2008; Lanari et al., 2025). Grass cover has been shown to reduce vine root density near the surface while promoting deeper rooting, resulting in vertical root stratification that may partially alleviate competition in upper soil horizons (Lanari et al., 2025). However, because grapevines access water across all soil compartments, competition for shallow soil moisture can still be substantial (Celette et al., 2008). These interactions are likely to become most critical during drought periods or at sites with limited deep soil water reserves, underscoring the need to integrate both above- and below-ground processes when evaluating vegetation-mediated soil moisture regulation in vineyards.

The hypothesis that differences in topsoil moisture response between moss restoration, grass, and bare soil diminish over time after maintenance ended is only partially supported. While vegetation cover clearly converged, as discussed in Sect. 4.1, topsoil moisture dynamics tell a more nuanced story, with treatment-specific responses persisting in certain periods.

Monthly comparisons of initial response time (IRT) showed that significant treatment differences only emerged after September 2022 (Fig. 9). At first, this may seem counterintuitive, since all treatments were actively maintained from April to October 2022, and the strongest contrasts might have been expected during this period. However, as detailed in Gall et al. (2025), moss establishment progressed more slowly than anticipated due to unusually hot and dry conditions during summer 2022, which delayed jute mat disintegration and hindered moss development.

The delayed divergence of IRT between bare soil and grass prior to September remains unclear, as both were expected to be fully established from the outset. Bare soil plots were prepared through careful weeding and cutting of existing vegetation, suggesting that residual grass roots might have persisted. However, all treatments were installed two months before the start of the experiment to minimize such legacy effects. While a longer establishment period might have further reduced residual influences, this alone does not fully explain the absence of early-season differences. A more likely factor is the hot and dry weather (see Sect. 2.1), which generated high evaporative demand and may have masked treatment-specific topsoil moisture responses. Seasonal variation in vineyard soil moisture is also strongly influenced by vine phenology: Monteiro and Lopes (2007) reported that soil moisture declines sharply from budbreak to harvest due to increasing vine water uptake. This pattern aligns with our volumetric topsoil moisture data (Fig. 7), which declined across all treatments from May to July/August, potentially obscuring early-season differences and explaining why significant treatment effects in response times only appeared after September.

Variability in topsoil moisture response revealed a clear pattern: moss plots consistently exhibited the lowest dispersion in IRT throughout the study, reflected in narrow standard deviations and interquartile ranges. In contrast, grass plots displayed higher variability, reaching significance during the first succession period (November 2022–February 2023) and remaining elevated thereafter. This likely reflects seasonal fluctuations in grass root water uptake driven by growth dynamics, mowing, and environmental conditions (Capri et al., 2023; Celette et al., 2008), whereas mosses, being poikilohydric and lacking true roots, maintain more consistent water absorption and release via rhizoids (Proctor et al., 1998). For vineyard management, this suggests that moss restoration not only imposes minimal infiltration barriers but also promotes more stable and predictable soil moisture dynamics, highlighting its potential as a reliable under-vine ground cover.

To summarize, vegetation cover clearly converged, and by February 2024, bare soil and moss treatments no longer exhibited their distinct characteristics, justifying the termination of the experiment. However, topsoil moisture responses only began to show signs of hydrological convergence toward the end of the study. Extending the experiment by several months could have clarified whether these differences would have continued to diminish, providing a more definitive test of the hypothesis and offering deeper insight into the temporal dynamics of vegetation-mediated topsoil moisture under natural succession.

Our findings should be interpreted with an understanding of the spatial scale and measurement approach underlying them, as these factors constrain the extent to which the findings can be transferred to larger spatial scales. First, the 40 × 40 cm plots and point-scale TMS-4 dataloggers, positioned at consistent locations relative to the vine row, characterize topsoil moisture dynamics at a specific microsite. They do not capture spatial variability in throughfall, stemflow, and infiltration across the row or hillslope. Therefore, our results should be interpreted as evidence of treatment effects at this scale rather than a scaled estimate of vineyard-wide hydrologic response.

Second, undocumented subsurface processes may also shape the patterns we observed. The gentle slope of 5° and textural similarity between the 0–25 and 25–90 cm horizons at the study site make pronounced lateral interflow at the horizon boundary unlikely to dominate the results. However, lateral subsurface flow has been shown to constitute a substantial component of total flow even on comparable or steeper slopes, with its magnitude strongly dependent on land use (Teschemacher et al., 2019). Particularly in vineyard soils on steep slopes, subsurface runoff and the associated transport of particles were identified as an important pathway distinct from surface runoff (Kögler et al., 2025), underscoring that such processes cannot be dismissed even where topography is comparatively gentle. The low topsoil bulk density (0.96 g cm⁻³) and clay-rich texture at our study site are further consistent with the potential for macropore and preferential flow pathways that point measurements with the TMS-4 dataloggers cannot resolve. Since TMS-4 dataloggers provide a single integrated moisture value for a small volume of soil around the probe (Wild et al., 2019), they cannot distinguish whether a particular reading reflects water stored in the soil matrix or water that has flowed through or accumulated in macropores. Depending on the exact position of the sensor relative to biopores or root channels, this can lead to additional, unresolved variations between replicates within a treatment and distort comparisons of moisture response times, independent of the actual treatment effects.

Third, while moss persisted with minimal management once established in our study, the wider literature suggests that this outcome is not yet reliably low-maintenance. Bu et al. (2018) achieved > 90 % moss coverage on a 15° slope, but required sustained soil moisture, weekly nutrient application, and 70 % summer shading. Fan et al. (2025) achieved 55 % coverage after 90 d on a steeper (60°) stony slope, but required daily irrigation and weekly nutrient application. Similarly, Li et al. (2026) demonstrated successful field establishment of prefabricated moss carpets, although intensive initial maintenance was required. These studies indicate that moss establishment is feasible under challenging conditions, but remains maintenance-intensive during establishment. Thus, low-maintenance establishment and persistence under realistic vineyard management have yet to be demonstrated, warranting further research on the scalability of moss-based applications.