The status, type and ecology of the vegetation of the southeastern seasonal regions of Indonesia need to be clarified to identify adequate natural resource management and conservation strategies.Tanimbar archipelago represents a group of islands south of the Banda sea in the Moluccas. The largest is the flat Yamdena (7°36′ S, 131°25′ E).Vegetation was interpreted from LANDSAT satellite data, overlaid with geology and topography for pre‐stratification. Within each strata, forest sites were equally systematically sampled using a network of small 0.2 ha survey plots (60 plots, 7,130 trees) and soil pits sampled for 44 of the vegetation plots. Fisher’s alpha diversity index was used together with ordination techniques to assert differences in forest types.The forest covered c. 70% of the island, comprising seasonal evergreen forest (SEF), dry deciduous forest (DDF) and moist deciduous forest’ (MDF). The SEF canopy (Burseraceae, Meliaceae, Oleaceae and Dilleniaceae) sited around 35 m. The density of trees above 10 cm diameter averaged 632 individuals/ha, the basal area (BA) was 32 m2/ha, and species diversity (SD, Fischer α) 14.84. Rattans are abundant in the understorey. The MDF exhibited a mixture of evergreen and deciduous big trees while the lower storey was evergreen. Rattans were less common. Distinct family species associations emerged from the ordination (Combretaceae, Fabaceae, Malvaceae and Apocynaceae or Meliaceae, Gnetaceae, Clusiaceae). The density of trees averaged 491 individuals/ha, BA 26 m2/ha and SD 18.73. The canopy of the DDF (Ebenaceae, Fabaceae, Apocynaceae and Menispermaceae) was around 30 m. During the dry season all tree species shed their leaves. The density of trees averaged 552 individuals/ha, BA 20 m2/ha and SD 11.58.The last natural seasonal forests of Indonesia are nowadays only found in the Tanimbar Archipelago. The existence of three contrasted seasonal forest types on small flat island was remarkable and should be quickly translated into decision making for land zoning, agriculture or forestry development, avoiding approaches applied in the humid region of the country. To succeed however, the ecology of these seasonal forests should quickly become priority areas for research to feed the design of suitable conservation and management strategies.

AIM: The status, type and ecology of the vegetation of the southeastern seasonal regions of Indonesia need to be clarified to identify adequate natural resource management and conservation strategies. LOCATION: Tanimbar archipelago represents a group of islands south of the Banda sea in the Moluccas. The largest is the flat Yamdena (7°36′ S, 131°25′ E). METHODS: Vegetation was interpreted from LANDSAT satellite data, overlaid with geology and topography for pre‐stratification. Within each strata, forest sites were equally systematically sampled using a network of small 0.2 ha survey plots (60 plots, 7,130 trees) and soil pits sampled for 44 of the vegetation plots. Fisher’s alpha diversity index was used together with ordination techniques to assert differences in forest types. RESULTS: The forest covered c. 70% of the island, comprising seasonal evergreen forest (SEF), dry deciduous forest (DDF) and moist deciduous forest’ (MDF). The SEF canopy (Burseraceae, Meliaceae, Oleaceae and Dilleniaceae) sited around 35 m. The density of trees above 10 cm diameter averaged 632 individuals/ha, the basal area (BA) was 32 m²/ha, and species diversity (SD, Fischer α) 14.84. Rattans are abundant in the understorey. The MDF exhibited a mixture of evergreen and deciduous big trees while the lower storey was evergreen. Rattans were less common. Distinct family species associations emerged from the ordination (Combretaceae, Fabaceae, Malvaceae and Apocynaceae or Meliaceae, Gnetaceae, Clusiaceae). The density of trees averaged 491 individuals/ha, BA 26 m²/ha and SD 18.73. The canopy of the DDF (Ebenaceae, Fabaceae, Apocynaceae and Menispermaceae) was around 30 m. During the dry season all tree species shed their leaves. The density of trees averaged 552 individuals/ha, BA 20 m²/ha and SD 11.58. CONCLUSIONS: The last natural seasonal forests of Indonesia are nowadays only found in the Tanimbar Archipelago. The existence of three contrasted seasonal forest types on small flat island was remarkable and should be quickly translated into decision making for land zoning, agriculture or forestry development, avoiding approaches applied in the humid region of the country. To succeed however, the ecology of these seasonal forests should quickly become priority areas for research to feed the design of suitable conservation and management strategies.

# Aim Application of environmental envelope modelling (EEM) for conservation planning requires careful validation. Opinions of experts who have worked with species of interest in the field can be a valuable and independent information source to validate EEM because of their first-hand experience with species occurrence and absence. However, their use in model validation is limited because of the subjectivity of their feedback. In this study, we present a method on the basis of cultural consensus theory to formalize expert model evaluations. # Methods We developed, for five tree species, distribution models with nine different variable combinations and Maxent EEM software. Species specialists validated the generated distribution maps through an online Google Earth interface with the scores from Invalid to Excellent. Experts were also asked about the commission and omission errors of the distribution models they evaluated. We weighted expert scores according to consensus theory. These values were used to obtain a final average expert score for each of the produced distribution models. The consensus-weighted expert scores were compared with un-weighted scores and correlated to four conventional model performance parameters after cross-validation with test data: Area Under Curve (AUC), maximum Kappa, commission error and omission error. # Results The median consensus-weighted expert score of all species–variable combinations was close to Fair. In general, experts that reached more consensus with peers were more positive about the EEM outcomes, compared to those that had more opposite judgements. Both consensus-weighted and un-weighted scores were significantly correlated to corresponding AUC, maximum Kappa and commission error values, but not to omission errors. More than half of the experts indicated that the distribution model they considered best included areas where the species is known to be absent. One third also indicated areas of species presence that were omitted by the model. # Conclusions Our results indicate that experts are fairly positive about EEM outcomes. This is encouraging, but EEM application for conservation actions remains limited according to them. Methods to formalize expert knowledge allow a wider use of this information in model validation and improvement, and they complement conventional validation methods of presence-only modelling. Online GIS and survey applications facilitate the consultation of experts.

AIM: Application of environmental envelope modelling (EEM) for conservation planning requires careful validation. Opinions of experts who have worked with species of interest in the field can be a valuable and independent information source to validate EEM because of their first‐hand experience with species occurrence and absence. However, their use in model validation is limited because of the subjectivity of their feedback. In this study, we present a method on the basis of cultural consensus theory to formalize expert model evaluations. METHODS: We developed, for five tree species, distribution models with nine different variable combinations and Maxent EEM software. Species specialists validated the generated distribution maps through an online Google Earth interface with the scores from Invalid to Excellent. Experts were also asked about the commission and omission errors of the distribution models they evaluated. We weighted expert scores according to consensus theory. These values were used to obtain a final average expert score for each of the produced distribution models. The consensus‐weighted expert scores were compared with un‐weighted scores and correlated to four conventional model performance parameters after cross‐validation with test data: Area Under Curve (AUC), maximum Kappa, commission error and omission error. RESULTS: The median consensus‐weighted expert score of all species–variable combinations was close to Fair. In general, experts that reached more consensus with peers were more positive about the EEM outcomes, compared to those that had more opposite judgements. Both consensus‐weighted and un‐weighted scores were significantly correlated to corresponding AUC, maximum Kappa and commission error values, but not to omission errors. More than half of the experts indicated that the distribution model they considered best included areas where the species is known to be absent. One third also indicated areas of species presence that were omitted by the model. CONCLUSIONS: Our results indicate that experts are fairly positive about EEM outcomes. This is encouraging, but EEM application for conservation actions remains limited according to them. Methods to formalize expert knowledge allow a wider use of this information in model validation and improvement, and they complement conventional validation methods of presence‐only modelling. Online GIS and survey applications facilitate the consultation of experts.

Aim: Long-term ecological research sites have become essential tools in the study of tropical forest dynamics, but the potential impacts of researcher activity on the permanent plots being studied is often ignored. Here we ask whether seemingly benign repeated surveys themselves significantly affect stand dynamics and plant species diversity when implementing long-term monitoring of permanent plots.

AIM: Long‐term ecological research sites have become essential tools in the study of tropical forest dynamics, but the potential impacts of researcher activity on the permanent plots being studied is often ignored. Here we ask whether seemingly benign repeated surveys themselves significantly affect stand dynamics and plant species diversity when implementing long‐term monitoring of permanent plots. LOCATION: Old‐growth semi‐deciduous tropical forest of M'Baïki, Central African Republic. METHODS: We compared demographic parameters of trees (DBH≥9.55 cm) and plant community diversity and composition across three zones of three permanent plots: along trails crossing a plot, in areas adjacent to trails and in the forest interior. We assessed differences in radial growth, mortality, recruitment and species diversity using ANOVA. We examined the dynamics of tree death over time using Cox proportional hazards regression models. We analysed plant species composition using permutational multivariate ANOVA and indicator species analyses. RESULTS: Tree mortality, recruitment and radial growth did not differ among the three zones. Species richness and evenness did not differ among the three zones investigated, but species composition did, with significantly different indicator species between trails and forest interiors. In trails, communities were characterized by light‐demanding small trees and lianas and shade‐tolerant herbs of trampled soils, whilst forest interiors exhibited more shade‐tolerant tree species. CONCLUSIONS: Even in observational studies conducted in the wild, visitors may unintentionally but artificially influence the natural patterns and processes under investigation, an influence whose intensity may depend upon study design, habitat type and natural disturbance regimes. Permanent plots may not be a benign influence on the study system, as is generally assumed, and responses to repeated visitation may place fundamental limits on the questions that can be addressed, especially when species composition is assessed under low levels of natural disturbances or high density of trails.

Aims To describe the structural and floristic diversity of New Caledonian mixed tropical rain forest and investigate its environmental determinants. Location New Caledonia (SW Pacific), a biodiversity hotspot. Methods Structural (stem density, basal area) and floristic characteristics (composition, species richness and dissimilarity) were investigated along environmental gradients (elevation, rainfall and slope) on different substrates (ultramafic and non-ultramafic) through the New Caledonian Plant Inventory and Permanent Plots Network (NC-PIPPN, 201 plots each measuring 20 m x 20 m). Results A total of 28,640 trees (DBH ?5 cm) belonging to 749 species, 240 genera and 92 families were inventoried in the NC-PIPPN. The New Caledonian mixed rain forest studied was characterized as having high stem density, basal area and species richness, and many small stems (60% of the trees <10 cm DBH and almost a quarter of species did not exceed this threshold). More than one-third of the species were rare (i.e. inventoried in less than three plots or represented by fewer than three individuals) in the plot network and floristic dissimilarity was high (Bray-Curtis index >0.70). The presence of ultramafic (UM) and non- ultramafic substrates (non-UM) combined with altitudinal and rainfall gradients were the main drivers of floristic dissimilarity, whereas the effect of geographic distance between the plots was surprisingly low. Floristic dissimilarity was very high between UM and non-UM substrates from species up to family level. About 75% of the species occurred on a single substrate type. The mixed rain forest on UM and non-UM substrates differed in floristic composition but not in structure. Conclusions NC-PIPPN proved to be an effective tool for investigating the woody species richness of New Caledonia as containing ca. 46% of its non-herbaceous species. However, the network's design, and more specifically its small plots, restricts its capacity to capture beta diversity

International audience

AIMS: To describe the structural and floristic diversity of New Caledonian mixed tropical rain forest and investigate its environmental determinants. LOCATION: New Caledonia (SW Pacific), a biodiversity hotspot. METHODS: Structural (stem density, basal area) and floristic characteristics (composition, species richness and dissimilarity) were investigated along environmental gradients (elevation, rainfall and slope) on different substrates (ultramafic and non‐ultramafic) through the New Caledonian Plant Inventory and Permanent Plots Network (NC‐PIPPN, 201 plots each measuring 20 m x 20 m). RESULTS: A total of 28,640 trees (DBH ≥5 cm) belonging to 749 species, 240 genera and 92 families were inventoried in the NC‐PIPPN. The New Caledonian mixed rain forest studied was characterized as having high stem density, basal area and species richness, and many small stems (60% of the trees <10 cm DBH and almost a quarter of species did not exceed this threshold). More than one‐third of the species were rare (i.e. inventoried in less than three plots or represented by fewer than three individuals) in the plot network and floristic dissimilarity was high (Bray–Curtis index >0.70). The presence of ultramafic (UM) and non‐ ultramafic substrates (non‐UM) combined with altitudinal and rainfall gradients were the main drivers of floristic dissimilarity, whereas the effect of geographic distance between the plots was surprisingly low. Floristic dissimilarity was very high between UM and non‐UM substrates from species up to family level. About 75% of the species occurred on a single substrate type. The mixed rain forest on UM and non‐UM substrates differed in floristic composition but not in structure. CONCLUSIONS: NC‐PIPPN proved to be an effective tool for investigating the woody species richness of New Caledonia as containing ca. 46% of its non‐herbaceous species. However, the network's design, and more specifically its small plots, restricts its capacity to capture beta diversity and forest structure. High species richness and floristic dissimilarity confirm that New Caledonian mixed rain forest is exceptionally rich.

Aims: Fine roots are essential components of the below-ground layer and play an important role in the carbon cycle. Methods for root extraction and biomass estimation have been proposed, including the temporal prediction method. However, there are doubts if the best model to estimate total root mass varies between study sites. Additionally, there are no records regarding the prediction method's efficiency for shorter collection times than 40 min. Here, we aim to clarify these doubts. Location: Brazilian Atlantic Forest. Methods: We extracted 1080 fine-root samples from two contrasting ecosystems at 60 time intervals of 2 min each. We then performed a model selection to identify the best-fit model and used it to find the shortest time suitable for collecting fine-root samples (40, 32, 24, 16, or 8 min). A further 448 root samples were collected from seven ecosystems by employing the shortest time tested (8 min). We calculated the percentage of estimated mass at 120 min and tested for differences between ecosystems. Results: We found that Weibull was the best-fit model, and it performed well for modeling root extraction at shorter collection times. All collection times tested had excellent goodness of fit, and there was strong evidence that the estimated mass did not differ between them. Moreover, collections at 8 min were enough to make reliable estimates of fine-root mass at 120 min in all ecosystems. Conclusions: Weibull is a flexible model and can accurately estimate fine-root mass at 120 min in different ecosystems. The extraction of fine roots can be reduced to four time intervals of 2 min each when using the temporal prediction method. By reducing the time spent removing roots from each soil core, researchers can increase the number of soil cores extracted per study site and characterize the environment properly.

Source Agritrop Cirad (https://agritrop.cirad.fr/604920/) | International audience | Aims: Fine roots are essential components of the below-ground layer and play an important role in the carbon cycle. Methods for root extraction and biomass estimation have been proposed, including the temporal prediction method. However, there are doubts if the best model to estimate total root mass varies between study sites. Additionally, there are no records regarding the prediction method's efficiency for shorter collection times than 40 min. Here, we aim to clarify these doubts. Location: Brazilian Atlantic Forest. Methods: We extracted 1080 fine-root samples from two contrasting ecosystems at 60 time intervals of 2 min each. We then performed a model selection to identify the best-fit model and used it to find the shortest time suitable for collecting fine-root samples (40, 32, 24, 16, or 8 min). A further 448 root samples were collected from seven ecosystems by employing the shortest time tested (8 min). We calculated the percentage of estimated mass at 120 min and tested for differences between ecosystems. Results: We found that Weibull was the best-fit model, and it performed well for modeling root extraction at shorter collection times. All collection times tested had excellent goodness of fit, and there was strong evidence that the estimated mass did not differ between them. Moreover, collections at 8 min were enough to make reliable estimates of fine-root mass at 120 min in all ecosystems. Conclusions: Weibull is a flexible model and can accurately estimate fine-root mass at 120 min in different ecosystems. The extraction of fine roots can be reduced to four time intervals of 2 min each when using the temporal prediction method. By reducing the time spent removing roots from each soil core, researchers can increase the number of soil cores extracted per study site and characterize the environment properly.

AIMS: Fine roots are essential components of the below‐ground layer and play an important role in the carbon cycle. Methods for root extraction and biomass estimation have been proposed, including the temporal prediction method. However, there are doubts if the best model to estimate total root mass varies between study sites. Additionally, there are no records regarding the prediction method's efficiency for shorter collection times than 40 min. Here, we aim to clarify these doubts. LOCATION: Brazilian Atlantic Forest. METHODS: We extracted 1080 fine‐root samples from two contrasting ecosystems at 60 time intervals of 2 min each. We then performed a model selection to identify the best‐fit model and used it to find the shortest time suitable for collecting fine‐root samples (40, 32, 24, 16, or 8 min). A further 448 root samples were collected from seven ecosystems by employing the shortest time tested (8 min). We calculated the percentage of estimated mass at 120 min and tested for differences between ecosystems. RESULTS: We found that Weibull was the best‐fit model, and it performed well for modeling root extraction at shorter collection times. All collection times tested had excellent goodness of fit, and there was strong evidence that the estimated mass did not differ between them. Moreover, collections at 8 min were enough to make reliable estimates of fine‐root mass at 120 min in all ecosystems. CONCLUSIONS: Weibull is a flexible model and can accurately estimate fine‐root mass at 120 min in different ecosystems. The extraction of fine roots can be reduced to four time intervals of 2 min each when using the temporal prediction method. By reducing the time spent removing roots from each soil core, researchers can increase the number of soil cores extracted per study site and characterize the environment properly.

Question: Do remote sensing signals represent β‐diversity? Does β‐diversity agree with community types? Location: UNESCO Man and the Biosphere Reserve, La Palma, Canary Islands. Methods: We recorded perennial, vascular plant species abundances in 69 plots (10 m × 10 m) in three pre‐defined community types along an elevational gradient of 2,400 m: succulent scrubland, Pinus canariensis forest and subalpine scrubland. The remote sensing data consists of structural variables from airborne Light Detection and Ranging (LiDAR) and multispectral variables from a time series of Sentinel‐2 (S2) images. Non‐metric Multidimensional Scaling was used to assess β‐diversity between plots. K‐means unsupervised clustering was applied to remote sensing variables to distinguish three community types. We subsequently quantified the explanatory power of S2 and LiDAR variables representing β‐diversity via the Mantel test, variation partitioning and multivariate analysis of variance. We also investigated the sensitivity of results to grain size of remote sensing data (20, 40, 60 m). Results: The β‐diversity between the succulent and pine community is high, whereas the β‐diversity between the pine and subalpine community is low. In the wet season, up to 85% of β‐diversity is reflected by remote sensing variables. The S2 variables account for more explanatory power than the LiDAR variables. The explanatory power of LiDAR variables increases with grain size, whereas the explanatory power of S2 variables decreases. Conclusion: At the lower ecotone, β‐diversity agrees with the pre‐defined community distinction, while at the upper ecotone the community types cannot be clearly separated by compositional dissimilarity alone. The high β‐diversity between the succulent scrub and pine forest results from positive feedback switches of P. canariensis, being a fire‐adapted, key tree species. In accordance with the spectral variation hypothesis, remote sensing signals can adequately represent β‐diversity for a large extent, in a short time and at low cost. However, in‐situ sampling is necessary to fully understand community composition. Nature conservation requires such interdisciplinary approaches. | acceptedVersion

QUESTION: Do remote sensing signals represent β‐diversity? Does β‐diversity agree with community types? LOCATION: UNESCO Man and the Biosphere Reserve, La Palma, Canary Islands. METHODS: We recorded perennial, vascular plant species abundances in 69 plots (10 m × 10 m) in three pre‐defined community types along an elevational gradient of 2,400 m: succulent scrubland, Pinus canariensis forest and subalpine scrubland. The remote sensing data consists of structural variables from airborne Light Detection and Ranging (LiDAR) and multispectral variables from a time series of Sentinel‐2 (S2) images. Non‐metric Multidimensional Scaling was used to assess β‐diversity between plots. K‐means unsupervised clustering was applied to remote sensing variables to distinguish three community types. We subsequently quantified the explanatory power of S2 and LiDAR variables representing β‐diversity via the Mantel test, variation partitioning and multivariate analysis of variance. We also investigated the sensitivity of results to grain size of remote sensing data (20, 40, 60 m). RESULTS: The β‐diversity between the succulent and pine community is high, whereas the β‐diversity between the pine and subalpine community is low. In the wet season, up to 85% of β‐diversity is reflected by remote sensing variables. The S2 variables account for more explanatory power than the LiDAR variables. The explanatory power of LiDAR variables increases with grain size, whereas the explanatory power of S2 variables decreases. CONCLUSION: At the lower ecotone, β‐diversity agrees with the pre‐defined community distinction, while at the upper ecotone the community types cannot be clearly separated by compositional dissimilarity alone. The high β‐diversity between the succulent scrub and pine forest results from positive feedback switches of P. canariensis, being a fire‐adapted, key tree species. In accordance with the spectral variation hypothesis, remote sensing signals can adequately represent β‐diversity for a large extent, in a short time and at low cost. However, in‐situ sampling is necessary to fully understand community composition. Nature conservation requires such interdisciplinary approaches.

Aims: The accurate classification of habitats is essential for effective biodiversity conservation. The goal of this study was to harness the potential of deep learning to advance habitat identification in Europe. We aimed to develop and evaluate models capable of assigning vegetation-plot records to the habitats of the European Nature Information System (EUNIS), a widely used reference framework for European habitat types. Location: The framework was designed for use in Europe and adjacent areas (e.g., Anatolia, Caucasus). Methods: We leveraged deep-learning techniques, such as transformers (i.e., models with attention components able to learn contextual relations between categorical and numerical features) that we trained using spatial k-fold cross-validation (CV) on vegetation plots sourced from the European Vegetation Archive (EVA), to show that they have great potential for classifying vegetation-plot records. We tested different network architectures, feature encodings, hyperparameter tuning and noise addition strategies to identify the optimal model. We used an independent test set from the National Plant Monitoring Scheme (NPMS) to evaluate its performance and compare its results against the traditional expert systems. Results: Exploration of the use of deep learning applied to species composition and plot-location criteria for habitat classification led to the development of a framework containing a wide range of models. Our selected algorithm, applied to European habitat types, significantly improved habitat classification accuracy, achieving a more than twofold improvement compared to the previous state-of-the-art (SOTA) method on an external data set, clearly outperforming expert systems. The framework is shared and maintained through a GitHub repository. Conclusions: Our results demonstrate the potential benefits of the adoption of deep learning for improving the accuracy of vegetation classification. They highlight the importance of incorporating advanced technologies into habitat monitoring. These algorithms have shown to be better suited for habitat type prediction than expert systems. They push the accuracy score on a database containing hundreds of thousands of standardized presence/absence European surveys to 88.74%, as assessed by expert judgment. Finally, our results showcase that species dominance is a strong marker of ecosystems and that the exact cover abundance of the flora is not required to train neural networks with predictive performances. The framework we developed can be used by researchers and practitioners to accurately classify habitats.

International audience | Aims The accurate classification of habitats is essential for effective biodiversity conservation. The goal of this study was to harness the potential of deep learning to advance habitat identification in Europe. We aimed to develop and evaluate models capable of assigning vegetation-plot records to the habitats of the European Nature Information System (EUNIS), a widely used reference framework for European habitat types.Location The framework was designed for use in Europe and adjacent areas (e.g., Anatolia, Caucasus).Methods We leveraged deep-learning techniques, such as transformers (i.e., models with attention components able to learn contextual relations between categorical and numerical features) that we trained using spatial k-fold cross-validation (CV) on vegetation plots sourced from the European Vegetation Archive (EVA), to show that they have great potential for classifying vegetation-plot records. We tested different network architectures, feature encodings, hyperparameter tuning and noise addition strategies to identify the optimal model. We used an independent test set from the National Plant Monitoring Scheme (NPMS) to evaluate its performance and compare its results against the traditional expert systems.ResultsExploration of the use of deep learning applied to species composition and plot-location criteria for habitat classification led to the development of a framework containing a wide range of models. Our selected algorithm, applied to European habitat types, significantly improved habitat classification accuracy, achieving a more than twofold improvement compared to the previous state-of-the-art (SOTA) method on an external data set, clearly outperforming expert systems. The framework is shared and maintained through a GitHub repository.Conclusions Our results demonstrate the potential benefits of the adoption of deep learning for improving the accuracy of vegetation classification. They highlight the importance of incorporating advanced technologies into habitat monitoring. These algorithms have shown to be better suited for habitat type prediction than expert systems. They push the accuracy score on a database containing hundreds of thousands of standardized presence/absence European surveys to 88.74%, as assessed by expert judgment. Finally, our results showcase that species dominance is a strong marker of ecosystems and that the exact cover abundance of the flora is not required to train neural networks with predictive performances. The framework we developed can be used by researchers and practitioners to accurately classify habitats.

Questions: Restoration of disturbed alpine ecosystems is difficult due to harsh environmental conditions. Transplanting of vegetation turfs into disturbed areas has been used as a restoration method in disturbed alpine sites. The aim of this study is to investigate which environmental factors influence the vegetation recovery in turf surroundings and how turf attributes contribute to vegetation recovery. Location: Restored roads in a former military training area, Dovrefjell mountain range, central Norway. Methods: We recorded species richness, vegetation cover and soil characteristics of transplanted turfs and turf surroundings in roads restored between 3 and 14 years ago.Linearandgeneralizedlinearmixedmodelswereusedtoinvestigatetherelative importance of turf attributes and soil factors for recovery of turf surroundings. Results: Time was the most important factor for vegetation recovery, but soil conditions in turf surroundings were also highly important. Species richness and vegetation cover in turf surroundings were almost twice as high on silt dominated soil and with presence of soil organic matter compared to on coarser soils and without organic matter. Species richness in turfs and turf surroundings was almost equal after 14years,andthesimilarityofthespeciescompositionwashigh.Neitherturfsize, distance to the second closest turf or species richness and vegetation cover of the turfs were important factors for vegetation recovery in the turf surroundings. Conclusion: This study demonstrates the importance of preparing the restoration sites before using turf transplants in road and infrastructure restoration. Of particularimportanceisensuringsoilorganiccontentandafinesoilgrainsizetoincreaserates of vegetation recovery in short time scales. Time is the most important factor for recovery in this ecosystem, and this should be communicated to project owners andtothepublictoensurerealisticexpectationsonrecoverytime. ecosystem management, low-alpine ecosystems, turf transplants, vegetation recovery, vegetation restoration | acceptedVersion

QUESTIONS: Restoration of disturbed alpine ecosystems is difficult due to harsh environmental conditions. Transplanting of vegetation turfs into disturbed areas has been used as a restoration method in disturbed alpine sites. The aim of this study is to investigate which environmental factors influence the vegetation recovery in turf surroundings and how turf attributes contribute to vegetation recovery. LOCATION: Restored roads in a former military training area, Dovrefjell mountain range, central Norway. METHODS: We recorded species richness, vegetation cover and soil characteristics of transplanted turfs and turf surroundings in roads restored between 3 and 14 years ago. Linear and generalized linear mixed models were used to investigate the relative importance of turf attributes and soil factors for recovery of turf surroundings. RESULTS: Time was the most important factor for vegetation recovery, but soil conditions in turf surroundings were also highly important. Species richness and vegetation cover in turf surroundings were almost twice as high on silt‐dominated soil and with presence of soil organic matter compared to on coarser soils and without organic matter. Species richness in turfs and turf surroundings was almost equal after 14 years, and the similarity of the species composition was high. Neither turf size, distance to the second closest turf or species richness and vegetation cover of the turfs were important factors for vegetation recovery in the turf surroundings. CONCLUSION: This study demonstrates the importance of preparing the restoration sites before using turf transplants in road and infrastructure restoration. Of particular importance is ensuring soil organic content and a fine soil grain size to increase rates of vegetation recovery in short time scales. Time is the most important factor for recovery in this ecosystem, and this should be communicated to project owners and to the public to ensure realistic expectations on recovery time.