Plant biodiversity assessment through soil eDNA reflects temporal and local diversity

Several studies have shown the potential of eDNA‐based proxies for plant identification, but little is known about their spatial and temporal resolution. This limits its use for plant biodiversity assessments and monitoring of vegetation responses to environmental changes. Here we calibrate the temporal and spatial plant signals detected with soil eDNA surveys by comparing with a standard visual above‐ground vegetation survey. Our approach compares vegetation in an old‐growth boreal forest in southern Norway, surveyed in 100 permanent 1‐m2 plots seven times over a 30‐year period, with a single soil eDNA metabarcoding‐based survey from soil samples collected at the same 100 plots in the year of the last vegetation survey. On average, 60% and 10% of the vascular plants and bryophytes recorded across all vegetation surveys were detected by soil eDNA. Taxa detected by soil eDNA were more representative for the local taxa pool than for the specific plot, and corresponded to those surveyed over the 30‐year period although most closely matched the current taxa composition. Soil eDNA detected abundant taxa better than rare ones although both rare taxa and taxa unrecorded by the visual survey were detected. Our study highlights the potential of soil eDNA assessments for monitoring of vegetation responses over broad spatial and temporal scales. The method's ability to detect abundant taxa makes it suitable for assessment of vegetation composition in a specific area and for broad‐scale plant diversity assessments.


| INTRODUC TI ON
The current global warming crisis and the fast pace of global biodiversity losses relative to its appraisal require innovative and rapid operational approaches to biodiversity assessment like never before. Plants are central to most biodiversity assessments, as they are predominant and ubiquitous (Kier et al., 2005), as well as valuable indicators of associated diversity (Brunbjerg et al., 2018), surrounding abiotic features (Terwayet Bayouli et al., 2021) and human impact (Uuemaa et al., 2013). In addition, plants are suitable organisms for climate changes monitoring as community reshuffling and spatial redistribution towards summits and higher latitudes are wellestablished biotic responses to increased temperatures (Bertrand et al., 2011;Chen et al., 2011;Steinbauer et al., 2018;Wiens, 2016).
Moreover, plant assessments are often required to map habitats, monitor environmental quality and assess habitat changes in space and time (Halvorsen et al., 2020).
Recording the taxonomic composition is the cornerstone of any biodiversity assessment, and this task requires the use of morphological and/or molecular proxies for detection and identification of taxa (Ruppert et al., 2019). Morphological proxies require inspection of plant characters that are diagnostic for the specimens' identity. However, this is a rather lengthy, and thus expensive, process that usually requires participation from trained botanists. In addition, a plant must be noticeably present and/ or have the characteristics necessary to enable observation and identification, thus limiting the seasons in which a majority of different taxa present may be monitored. Historically, censuses of plant diversity, forest inventories and monitoring programmes have relied on morphological proxies (Corona et al., 2011). These have contributed greatly to the discovery and current knowledge of the known diversity.
More recently, molecular proxies for taxon identification based on environmental DNA (eDNA) have been taken into use for biodiversity assessment purposes (Beng & Corlett, 2020). The use of eDNA extracted from soil, water, faeces or bulk samples  grants the possibility of collecting organismal or extraorganismal DNA from multiple individuals and taxa simultaneously, saving lengthy and costly hours in the field collecting specimens.
In addition, this non-invasive and non-destructive method may be useful for the detection of rare, elusive and/or challenging-tocollect taxa (Alsos et al., 2018;Carrasco-Puga et al., 2021;Hartvig et al., 2021). eDNA-based surveys thus may open for rapid assessment and monitoring of biodiversity within a particular region, which is a critical aspect to understand effects of the current climate change crisis and biodiversity losses. Indeed, soil eDNA samples have gained attention as a potentially valuable tool for the assessment of plant diversity, as it may harbour DNA from both above-ground and below-ground signals (i.e. pollen, debris, roots), from active as well as dormant plant tissues . Accordingly, eDNA may provide a series of past and present plant signals that can assist the documentation of local extinctions and long-term ecosystem changes. Soil eDNA has most often been used to assess plant diversity in Arctic and boreal regions where the low temperature facilitates DNA preservation Wang et al., 2021;Willerslev et al., 2014;Yoccoz et al., 2012), although it may also be successful in tropical Zinger et al., 2019), and extreme environments such as deserts (Carrasco-Puga et al., 2021) and geothermal sites (Fraser et al., 2018). Furthermore, soil eDNA assessments have been successfully applied to identify present and past diversity from natural or cultivated areas , assess woody encroachment in grasslands (Sepp et al., 2021) and predict habitats from crime scenes (Fløjgaard et al., 2019).
While eDNA-based detection for assessing diversity is already in widespread use , knowledge about the sources, fate, persistence and transport of eDNA in the environment is scarce and mainly explored in aquatic environments . Knowledge of these properties of eDNA is indispensable to establish the temporal and spatial resolution expected of an eDNA assessment and to assess the utility of soil eDNA-based methods for ecological monitoring, for example, of biotic responses to climate changes . Soil eDNA has been shown to reflect plant diversity at local (Beng & Corlett, 2020;Edwards et al., 2018;Kumpula, 2020;Yoccoz et al., 2012) as well as regional scales (Carrasco-Puga et al., 2021). Furthermore, crop signals from 10 to 50 years into the past have been detected in cultivated soils (Foucher et al., 2020;Yoccoz, 2012). These insights, obtained from diverse environments with different anthropogenic pressures, may indicate that soil eDNA assessments are adequate for monitoring of vegetation. However, exploration of soil eDNA-based methods is still in its infancy and important knowledge gaps still exist as exemplified by the combined effects of spatial and temporal resolution on plant eDNA signals in soils, which to our knowledge has not yet been studied. This and other knowledge gaps have to be filled to establish a calibration field protocol for monitoring vegetation areas in nearnatural state, for example, for assessment of vegetation responses to climate changes.
Here, we use time-series data from intermittent vegetation surveys originally designed to track vegetation responses to climate changes in an old-growth forest in South Norway, to investigate the temporal and spatial resolution of a soil eDNA assessment made in 2018. Our research questions are as follows: (a) Do soil eDNA assessments capture the current diversity or integrate diversity over a longer time period?; (b) Do soil eDNA assessments reflect the plotspecific diversity or the local or regional pool of taxa?; (c) Can soil eDNA assessments provide taxa abundance estimates, that is, are common and abundant taxa detected with higher probability than rare, subordinate taxa? We used metabarcoding analysis of the trnL (UUA) intron P6 loop and identified plant sequences with customed reference libraries previously built by sequencing herbarium collections (Alsos et al., 2020;Soininen et al., 2015;Sønstebø et al., 2010;Willerslev et al., 2014). While the marker region is chosen based on vascular plants, bryophytes are common by-catch, and we also report results for bryophytes.

| Vegetation plot surveys
In 1988, one hundred 1-m 2 plots were placed across eight transects with a total length of 1,320 m, subjectively selected to cover the broad-scale variation in forest vegetation in response to natural edaphic gradients in the Solhomfjell Forest Reserve, southern Norway, 58°58′N, 8°58′E, at 350-480 m a.s.l. (Figure 1; Økland & Eilertsen, 1993). Fieldwork permission was granted by the Environmental protection authorities at the County Governor's office in Aust-Agder before fieldwork started in 1988. The studied area is situated within the southern boreal zone and harbours protected old-growth forests with overstories dominated by either Norway spruce (Picea abies [L.] H.Karst.) or Scots pine (Pinus sylvestris L.), alternating with mires and rock outcrops. The one hundred 1-m 2 plots were distributed semi-systematically along the eight transects. Every 10th m along each transect was a candidate plot position. From the 132 candidate positions, 100 were selected randomly.
The turnover of the vegetation plot composition was calculated for vascular plants and bryophytes as the sum of proportions of taxa gained and lost from survey time t−1 to survey time t as fraction of the all taxa recorded at the two time points, using the R library 'codyn' (Hallett et al., 2016). Turnover values with 95% confidence intervals for each period were obtained separately for spruce and pine subsets, by averaging across plots.
For all taxa recorded in the vegetation survey, we created a local Solhomfjell sequence reference library for the chloroplast trnL intron (UUA) P6 loop with retrieved sequences from the regional arctic and boreal reference library (Arcborbryo; Soininen et al., 2015;Sønstebø et al., 2010;Willerslev et al., 2014) when available there, otherwise GenBank and for a few taxa unpublished P6 loop sequences were retrieved from PhyloNorway (Alsos et al., 2020, Table S1). Different taxa with identical sequences for this marker were merged at the F I G U R E 1 Location of the study area, Solhomfjell Forest Reserve, within Norway (left), typical interior of forests dominated by Norway spruce (above) and scots pine (below), and the relative number of taxa of different growth forms registered in the vegetation surveys. A map of the 100 survey plots with specific locations is given in Økland and Eilertsen (1993) lowest possible taxonomic level and named accordingly. Presence/ absence values (1 or 0) for the merged taxon were obtained from the original data by summation of presence values for all parent taxa. A sum ≥1 was scored as presence (1), otherwise absence (0). If each parent taxa had subplot frequency ≥ 16, biomass of the merged taxon was obtained by averaging subplot frequency values for the parent taxa, otherwise the sum of subplot frequency values for parent taxa was used (Table S1). Each soil eDNA sample was separately homogenised with ceramic beads and 1 g was used for eDNA extraction. The latter was done in five rounds of two steps: (a) CTAB/chloroform pre-treatment to increase the separation of the organic phase and (b) aqueous phase and using the E.Z.N.A. soil DNA kit following the manufacturer's protocol (Omega Bio-tek; see Data S1 for a detailed protocol). The chloroplast marker trnL (UAA) intron P6 loop was chosen as its short sequence can yield amplification of old DNA material degraded in eDNA samples. This marker was amplified for each sample with the g and h primers by PCR, using three technical replicates (Taberlet et al., 2007; 5′-GGGCAATCCTGAGCCAA-3′, 5′-CCATTGAGTCTCTGCACCTATC-3′). Forward and reverse primers were tagged with a unique 12 bp oligonucleotide on the 5′ end (Fadrosh et al., 2014). Unique combinations of tagged primers were set up in panels for each PCR reaction for a total of 309 samples (100 samples with 3 PCR replicates each, 5 extractions blanks and 4 PCR negatives). The PCR negatives had no DNA template and were placed on the 96th well position in each panel. Composition of PCRs, final volumes and number of cycles can be found in Data S1.

|
The PCR products were run on a 2% agarose gel, and the amplicon concentrations were measured via band intensity using ImageLab software (Bio-Rad). The lowest concentration (μM) available for all PCR products and its relative volume was identified and the relative concentrations of the PCR products were adjusted to this same concentration. Amplicons were pooled in one library using a Biomek 4000 automated liquid handler (Beckman Coulter Life Sciences).
The library was cleaned using AMPure XP reagent beads (Beckman Coulter Life Sciences). The length for all amplicons in the library was determined using a Fragment Analyser (Agilent Technologies). The library was sequenced on an Illumina MiSeq platform with 150 bp paired-end reads (Illumina Inc.).

| Sequence analysis and taxonomic identification
Sequence data were analysed and curated using OBITools 2 (Boyer et al., 2016; sequences and a detailed script is available at Ariza et al., 2022) following the wolf tutorial with adaptations for demultiplexing dual indexes from QIIME2 (Caporaso et al., 2010). Sequences were retained with both indexes for dereplication for further analysis.
Similar sequences were clustered with obiclean (Boyer et al., 2016) only when the read count of the less abundant sequence was below 5% of the most abundant sequence. To reduce multiple identifications of the same sequence, taxonomic assignment of dereplicated and denoised sequences was done by matching to three reference sequences databases containing: (a) only taxa registered in the local Solholmfjell reference library (see above); (b) the complete arctic boreal database for vascular plants and bryophytes (Soininen et al., 2015;Sønstebø et al., 2010;Willerslev et al., 2014); and (c) taxa available in the EMBL database (downloaded on 7/02/2020) filtered to sequences with trnL (UUA) intron g-h primers using ecoPCR tool from OBITools (Boyer et al., 2016). Resulting identifications from the three databases were merged by sequence and duplicates were eliminated giving priority to reference databases (a), (b) and (c) in that order. To minimise erroneous taxonomic assignments, only taxa with a 100% match to a reference sequence were retained. We observed that below this threshold, sequences remained without a taxonomic rank assigned. Furthermore, assigned taxa names were changed to the lowest taxonomic rank possible with trnL (UUA) intron and thus are identical to those registered in vegetation surveys. When different sequences were identified with identical taxa names, a unique entry was retained and the read counts within plots and replicates were summed. Read counts were averaged across all samples and negative controls (extraction + PCR).

| Comparison between vegetation surveys and eDNA survey
The vegetation survey composition served as a baseline to assess the overlap with the composition of the soil eDNA survey, and from this, the spatial and temporal resolution was derived. Thus, our comparison framework comprises (a) a 7-point time-series of vegetation surveys in one hundred 1-m 2 plots (spruce and pine data subsets) from the Solhomfjell Forest Reserve, carried out in the years 1988, 1993, 1998, 2003, 2008, 2013 and 2018; and (b)

| Temporal resolution of eDNA survey data
The temporal resolution of soil eDNA assessments, that is, at which temporal scale (survey year) the two inventories were most compa- To determine whether the similarity of the plot number of matching taxa in detectable vegetation survey t across years was due to soil eDNA survey recording the same dominant taxa over vegetation survey years, we investigated the relation of this variable to the unchanged composition between survey t bdvs and t 1 , …,t 6 with a Spearman correlation test. First, we calculated the number of identical taxa between vegetation survey t bdvs and t 1 , …,t 6 by unchanged composition variable. Second, for each plot, we summed the number of taxa between vegetation survey t bdvs and t 1 , …,t 6 by taxon count variable. Thus, six iterations were calculated for both of these variables in each plot. Finally, we calculated the proportion of unchanged composition out of the taxon count t bdvs + t for each iteration.
To investigate plant DNA permanence and the past plant signals stored in a soil eDNA sample, we annotated each taxon from the plot number of matching taxa in total and detectable vegetation survey variable with the vegetation survey year t in which it was recorded.
Since taxa permanence across vegetation survey years will lead to many annotations, we focused on those taxa that were recorded only in 1 year of vegetation survey. We call this subset a temporal number of matching taxa and describe how many years ago (t 1 = 30, t 2 = 25….t 7 = 0) each taxon was recorded.

| Spatial resolution of eDNA survey data
The spatial resolution of taxa registered in the soil eDNA survey in a given plot was assessed by categorising each taxon name according to the spatial scale on which was recorded: matching taxa in the best detected vegetation survey if recorded within the plot (<1 m 2 ), vegetation survey match if also registered in the best detected vegetation survey but in another plots (>1 m 2 ), or regional flora match if registered in artsdatabanken.no for the 41.77 km 2 Solhomfjell Forest Reserve area (Figure 1; >1 m 2 ). In addition, taxa with match to nonnative plants or with higher mean number of reads in the negative controls than across samples were categorised as false positives.
Sequence reads assigned to taxa within all categories were counted, and taxon count and proportions were calculated. Moreover, the spatial scales of both soil eDNA and vegetation surveys were compared plot-wise by correlating the number of taxa from above categories to the number of taxa registered in the best detected vegetation survey. Correlations were fitted to a linear model and predicted with a 95% confidence level interval.

| Representation of vegetation in soil DNA survey
Each taxon name in the best detected vegetation survey was categorised as a matching taxa if also present in the soil eDNA survey, no trnL reference if the reference sequence for the trnL (UUA) intron with g-h primers was not available or undetected by eDNA if not present in the soil eDNA survey but a reference sequence was available.
Taxon counts and proportions within all categories were calculated.
Finally, we investigated if soil eDNA detections of taxa occurring within 1 m 2 plots (matching taxa) reflected their abundance. For this, we summarised the subplot frequency (used as proxy for biomass; Porté et al., 2009;Wilson, 2011) separately for matching taxa and taxa undetected by eDNA. The distribution of biomass data for each taxon in each plot from these two categories was compared with a Spearman's rank correlation test. We also investigated the correlation between the number of reads assigned to each matching taxa within major growth forms and the biomass registered in the best detected survey year.

| Vegetation surveys
Counts of taxa registered in plots, both in each survey and across all seven vegetation survey years can be found in Data S3. Across all vegetation surveys, the spruce subset registered more unique vascular plants and bryophytes than the pine subset. The number of vascular plant taxa recorded per plot in each vegetation survey was on average 9.23 ± 5.12 SD and 5.23 ± 1.40 SD in the spruce and pine subsets, respectively, while the corresponding numbers for bryophytes were 9.29 ± 3.33 SD and 5.48 ± 2.04 SD.
Similarly, higher turnover was found for pine than for spruce subsets for most 5-year periods ( Figure 2). Across consecutive vegetation survey years, 7%-15% and 4%-14% of the total bryophytes and vascular plants were gained, and 10%-20% and 10%-15% lost, respectively ( Figure S1). Accordingly, the number of taxa of both groups declined over the 30-year survey period. Differences between bryophytes and vascular plants were more pronounced for gains than those for losses ( Figure S1).
Reference sequences for the trnL (UUA) intron were available for 84% (133/157) plant taxa from the total vegetation survey. Of the remaining 13% (24/157), 20 were bryophytes and four were vascular plants (Table S1), and these taxa were filtered out from most analyses. Several taxa, 21% (28/133), had identical sequences (mainly bryophytes), and these were lumped into 12 unique taxa. Our final vegetation dataset thus used for comparison with the soil eDNA survey consisted of 65 vascular plants and 68 bryophytes (117 in total; Table S1).

| Soil eDNA survey
The Illumina MiSeq PE150 run yielded approximately 10 million reads assigned either to plots or blanks (mean reads/plot for all replicates: >100,000; mean reads/blank: 384). Of these, more than 4 million reads matched 100% to 130 taxa in the reference libraries. A summary of the soil eDNA survey is shown in 29.66 ± 9.08 SD; bryophytes: 4.41 ± 2.51 SD; Table S3).

| Matching taxa and temporal resolution of eDNA survey data
A total of 53 vascular plants and 11 bryophytes were registered both in the vegetation and the eDNA survey (matching taxa), accounting for 81% (53/65) and 13% (8/68) of the total vegetation survey, respectively (a summary is found in Table S3 Table S3). The total number of taxa in this group was 51. These are mainly perennial shrubs, herbs and graminoids typical of boreal forests (Table S3 for a taxonomic overview and descriptive statistics). False positives, all vascular plants, were registered in 61 plots accounting for 8% of the soil eDNA survey (10/129) and less than 3% of the total reads ( Figure S4). A summary figure is found in Figure S5.

F I G U R E 4
Matching taxa in each detectable vegetation survey and temporal resolution of soil eDNA survey. The number of matching taxa (the number of identical taxon names registered in both detectable vegetation and soil eDNA surveys) are calculated per plot for each plant group, and for compositions from each vegetation survey (t 1 , …,t 6 ) and for the total vegetation survey ('all'). The proportion of matching taxa in a vegetation survey t is the fraction of number of matching taxa out of the detectable taxon count in a vegetation survey t or in the total vegetation survey. Points and bars indicate means with 95% confidence intervals for the means. Identical small letters indicate non-significant differences (p > 0.05) in pairwise Friedman multiple comparison tests

| Representation of vegetation in soil DNA
On average, 6% ± 21 SD and 16% ± 19.00 SD of the vascular plants and bryophytes proportion registered in the best detected vegetation survey (t 7 = 2018), respectively, were also present in the soil eDNA survey (matching taxa; Figure 5b Figure S7).

F I G U R E 5
Spatial resolution of the soil eDNA survey and best detected vegetation survey. (a) Plant taxa detected with a single soil eDNA sample from each plot is assessed according to the spatial scale on which was recorded by categorising into: Matching taxa (taxa registered in the plot in the best detected vegetation survey t 7 = 2018); vegetation survey match (taxa recorded in any other plot in the best detected vegetation survey t 7 = 2018); regional flora match (taxa recorded in the Solhomfjell area outside plots); and false positives (taxa not recorded in the Solhomfjell area or with higher mean read count in PCR negatives than across samples and replicates). (b) Plant taxa recorded in each plot at the last vegetation survey t 7 = 2018 are categorised into: Matching taxa (taxa also registered in the soil eDNA survey); undetected by eDNA (trnL reference sequence available, but taxa were not registered in the soil eDNA survey); and no trnL reference available (no reference sequence was available for identification with eDNA data). For both panels, results are shown for taxon count proportions from each category and subsets with boxplots: horizontal bold lines depict median, the boxes show interquartile range and vertical lines indicate non-outlier range; the dots represent the outliers 4 | DISCUSS ION

| Vegetation survey inventories and its change across time
More bryophytes than vascular plants were registered in plots from the spruce subset than pine subset (Figure 1), and its average taxon counts in the total vegetation survey follow the same plot trends across all survey years (Data S3). Overstory dominance by pine or spruce influences the understory vegetation by different effects on ground-level light conditions (Felton et al., 2020), with lower light levels in the spruce forest (Esseen et al., 1997). Reduced richness in pine forest from the Solhomfjell area has been attributed to severe effects of drought from its shallow soils (Økland & Eilertsen, 1996).
Furthermore, nutrient-demanding taxa are restricted to the richer and deeper soils that in the Solhomfjell area are found in spruce forests only (Økland & Eilertsen, 1993). Dwarf shrubs such as bilberry and lingonberry, and mosses such as feather moss and shaded wood-moss, registered in the majority of plots, are common and dominant species in boreal understory forest vegetation (Nilsson & Wardle, 2005;Økland et al., 2004).
An average of one to two vascular plants and bryophytes were either lost or gained between consecutive vegetation surveys, respectively, and fluctuations observed in plots located below overstories dominated by pine were generally lower than spruce (Figure 2). Both trends were also observed in Solhomfjell plots analysed every year from 1988 to 1993 (Økland & Eilertsen, 1996), and in other boreal forests in the south of Norway between 1988 and 2003 (Økland et al., 2004). This suggests these turnover rates are representative for Norwegian boreal forests over the last 30 years. The latter studies also detected a decrease in vascular plant richness, notably in spruce forests, and increase in large bryophytes in both forest types, attributed to a combination of past soil acidification due to higher deposition of air pollutants and longer and warmer growth seasons.
Though we did not detect steady trends of gain or loss of vascular plants and bryophytes between consecutive years of vegetation survey, we detected peaks that may correspond to exacerbation of these climatic conditions (Figures S1).

| Soil eDNA survey and detection of taxa
The soil eDNA survey consisted of 127 taxa assigned mostly to species level (63.77%), from which 53 were vascular plants not registered in the total vegetation survey (Table S2). In all, 57 taxon detections accounted for 81% and 13% of vascular and bryophyte taxa registered in the total vegetation survey, respectively (Table S2). Soil eDNA surveys in temperate and tropical forests, tundra and deserts have routinely found 'hidden taxa' that were not observed when surveying above-ground diversity using vegetation surveys (Carrasco-Puga et al., 2021;Edwards et al., 2018;Osathanunkul et al., 2021;Palacios et al., 2021;Yoccoz et al., 2012). Small herbaceous plants F I G U R E 6 Correspondence between taxon detectability and abundance in the year of the best detected vegetation survey (t 7 = 2018). The composition of each plot from the best detected vegetation survey is categorised as matching taxa and undetected by eDNA. The distribution of summed subplot frequency (proxy for biomass) from each taxon in these two categories is shown by violin plots: Shape width is proportional with the frequency of observations. Subplot frequency corresponds to taxon presence in 625 cm 2 grids, and values >16 correspond to taxa with identical trnL (UUA) intron sequences that were merged into one single taxon with their subplot frequencies summed. Significance of differences between biomass from taxa within both categories was tested using a Wilcoxon-Mann-Whitney rank-sum test and are shown with ** and *** for p values <0.05 and <0.01, respectively and seedlings can be easily overlooked while surveying vegetation, and eDNA-based assessments detect DNA rather than organisms (Deiner et al., 2017). The pool of plant DNA in top soils is composed of locally deposited debris, roots, rhizomes and seeds. In addition, it may contain local and/or exotic pollen, but this does not seem to contribute to the local eDNA signal .
Metabarcoding analysis of root diversity has highlighted the DNA contribution of many perennial plants that persist below-ground even in the temporary absence of above-ground parts, which, in turn, increases below-ground richness estimates compared to aboveground Rucińska et al., 2022;Träger et al., 2019).
In our study, soil eDNA detections of taxa not registered in the vegetation surveys but present in the Solhomfjell area are indeed mostly perennial plants (Figure 5a; Table S2), suggesting that these are most likely local signals from plants growing in between vegetation survey plots or seedlings not recruited in the plot.
Reference sequences for the trnL (UUA) intron P6 loop were available for about 95% and 75% of the vascular and bryophyte taxa registered across all vegetation surveys, respectively, and the taxonomic resolution was 75% and 65% identified to species level. While longer markers such as ITS, matK and rbcL in general may provide higher taxonomic resolution, the actual taxonomic resolution obtained depends on the marker region used, the representation in the reference library and the size and nature of the local flora (Hollingsworth et al., 2016). The P6 loop of the chloroplast trnL (UUA) intron (Taberlet et al., 2007) is the most commonly used marker for soil eDNA studies targeting vascular plants (Capo et al., 2021;Parducci et al., 2017), as its short sequence may be found in the degraded DNA that is typically present in underground decomposed material and sediments (Taberlet et al., 2007).
This primer is designed to target vascular plants, and our results on vascular plants show high detection and high taxonomic resolution similar to other studies that are based on this primer (e.g. Alsos et al., 2018;Edwards et al., 2018). Other primer pairs for the P6 loop such as c-d (Taberlet et al., 2007) and Bryo_P6 (Epp et al., 2012) are conserved from Bryophytes to Angiosperms, but these markers have not been widely used and very few reference sequences exist in public repositories (Boukhdoud et al., 2021;Soininen et al., 2017). Though other nuclear ribosomal (ITS) and chloroplast (rbcL) markers may yield higher specificity in bryophytes (Lang et al., 2014;Liu et al., 2010), detection of bryophytes with eDNA-based assessments is limited by low intraspecific variation in marker regions (Hassel et al., 2013) and particularly for boreal forests when relying on non-exhaustive reference libraries.
However, the fact that we did detect bryophytes in almost every soil eDNA sample suggests that there is a great potential for eDNA also for this group, but we recommend further development of primer design and build up of reference library The detection of taxa was related to abundance, which, in turn, is an expression of biomass. The soil eDNA survey failed to detect some taxa registered in vegetation surveys even when reference sequences were available (Figure 6), but most of these had significantly lower plot abundance than those that were detected (Figure 6). A positive relation between detectability and plant biomass has been also observed in tundra and temperate sites (Alsos et al., 2018;Edwards et al., 2018;Yoccoz et al., 2012) and in aquatic environments (Alsos et al., 2018;Anglès d'Auriac et al., 2019;Matsuhashi et al., 2016). The relation is often attributed to the greater chance of deposited or suspended plant DNA that can be detected with higher organismal biomass. However, our study also reports detections of taxa present in only one out of the 16,625 cm 2 subplots of a vegetation plot (Figure 6), demonstrating that soil eDNA metabarcoding also detects some rare taxa. In addition, the apparent stochasticity of rare taxa raises questions on how biomass differences between root and shoot at different life-history stages (Qi et al., 2019) can potentially underlie the detection of less abundant taxa. Furthermore, in metabarcoding studies, sequence read counts are often interpreted as a proxy for abundance since DNA template availability for PCR amplification covariates with biomass (Amend et al., 2010;Beng & Corlett, 2020;Deagle et al., 2019). However, the signal of biomass can be diluted by technical and biological biases in marker recovery rates across different taxa (Deiner et al., 2017). In our study, sequence read counts assigned to taxa from all considered growth forms registered in the last year of vegetation survey correlated positively to their plot abundance, and these correlations were significant for forb, graminoid, moss and shrub taxa as well as for all life-forms combined (See Figure S7 for figures and p values). The correlations for ferns and trees were not significant, and this might be due to the abundance being skewed by the larger aerial vegetation cover in relation to the smaller underground cover of these taxa, as may decrease the DNA contribution of roots and rhizomes to the soil eDNA pool in relation to other growth forms (Qi et al., 2019).

| Matching taxa and temporal calibration
Each plot has a measure of the total number of unique taxa registered during the seven survey years as well as the number of taxa detected in the single eDNA survey. On average, 60% and 10% of the vascular and bryophyte taxa from the total vegetation survey matched with a single soil eDNA survey, respectively ( Figure 4; Table S3).
Conversely, 18% and 30% of the vascular and bryophyte taxa registered by a single soil eDNA survey matched with the total vegetation survey, respectively (Table S3). Similar rates of undersampled vegetation, that is, taxa that were detected in only one of the two surveys , and matching taxa values, have been found when comparing surveys at similar and even larger plot scales (1-m 2 plots in alpine subarctic vegetation Kumpula, 2020; 1-4 m radius from circular plots in Svalbard tundra Edwards et al., 2018;and 15-m 2 plots in Varanger boreal forest Yoccoz et al., 2012), highlighting the ability of both proxies to assess the total vegetation of a site.
In our study, we show that the number of matching taxa increased with richness registered in the total vegetation survey in both pine and spruce subsets (Figure 3), and this suggests that differences in soil and vegetation properties in these two environments probably have no effect on detectability. Furthermore, surveys built by both proxies detected the same dominant taxa. This highlights the ability of both proxies to recover the main components of the vegetation.  (Table S4). Altogether, our study indicates that a single soil eDNA survey can detect taxa from multiple vegetation surveys and its turnover fraction across time with the same power, and pinpoint how soil eDNA samples can encapsulate the vascular composition going back at least 30 years. Detections of past signals from 30 up to 50 years ago have also been found in crop soils (Foucher et al., 2020;Yoccoz et al., 2012), though these are more likely to be detected since plant biomass exponentially increases in monocultures. However, unique past signals detected in our study correspond to taxa with median to low abundance, registered in 50% or fewer subplots. This suggests that biomass may not play a role in the detectability of past signals in natural environments (Data S4). Nevertheless, detections of past signals can be also attributed to the resurfacing of deep soil particles through bioturbation by biotic underground DNA transporters such as insects, moles, worms, etc. (Prosser & Hedgpeth, 2018). The

| Spatial patterns of detection
On average, about 55% of the soil eDNA survey composition in each plot matched the composition registered in any other (near) plots or in the Solhomfjell area, whereas about 22% matched the vegetation plot composition (Figure 5a), suggesting that soil eDNA samples reflect mainly local vegetation rather than plot specific signals. Our results contrast those from Edwards et al. (2018) in Svalbard tundra where soil eDNA signals were highly specific to those recorded within a circular plot of 1 m radius and no taxa existing beyond a 4 m radius were found. Instead, our results are more in concordance with <1 km signals speculated by Yoccoz et al. (2012), as these authors did not find signals in uncultivated meadows from crops located a kilometre away but found signals that are likely part of the regional species pool. Differences in taxa richness and vegetation distribution between tundra (low, homogeneous) and temperate forests (high, heterogeneous) such as that of Solhomfjell area led us to consider how distribution patterns of vegetation may hinder the spatial recovery of soil eDNA signals. Furthermore, the hilly Solhomfjell landscape in which the plots are located may contribute to DNA transport from one plot to another via snow-melt, rainfall run-off and through-flow, thus enabling detections from other plots.
Although we attempted to calibrate each eDNA detection with a match to an area, that is, a match to the vegetation survey is a match to an area >1 m 2 , our categories may disguise a temporal match within plots (a match to taxa detected in the same plot back in time).
Our study is the first to assess simultaneously the temporal and spatial resolution of soil eDNA samples in natural environments, yet our approach cannot disentangle the contribution of each signal separately. This limitation highlights the need for studies in controlled microcosms where plant signals can be followed with biomarkers, thus enabling the possibility of tracing both the spatial and temporal signals. Nevertheless, our taxa comparison in space allowed us to identify plant richness detected by both methods with similar sampling efforts (surveying 1-m 2 plot vs taking a soil sample at the centre of 1-m 2 plot; Figure S5) and so provide a baseline for decision-making when designing sampling for soil eDNA assessments.

| Limitations and considerations for soil eDNA for plant diversity assessments
Soil and sediments are suitable substrates for eDNA-based plant assessments in terrestrial environments, as most extra-organismal and organismal plant DNA from both active and dormant tissues are gathered or ultimately deposited in these substrates (Rodriguez-Ezpeleta et al., 2021). Accordingly, soil eDNA assessment is a valuable tool for identification of plant diversity at any season, especially when non-destructive and easy sampling is needed. Our study has shown how a single eDNA sample can signal local dominant flora and thus might be useful for general plant diversity assessments. However, when detection of less abundant and/or rare taxa is desired, collecting multiple eDNA soil samples is recommended. Moreover, our study shows how a single soil eDNA survey can provide a series of local, regional, past and present plant signals that can help track long-term responses to climate and ecosystem changes. However, as with any method, there are some limitations to consider before embarking on a plant soil eDNA study (see Figures S8 for a summary of methodological steps). Since eDNA-based organismal detections are dependent on both DNA presence (intracellular or extracellular) and environmental conditions that may enhance or diminish DNA permanence, degradation and/or decay (Nagler et al., 2018;Pietramellara et al., 2009;Rodriguez-Ezpeleta et al., 2021), an evaluation of the potential state of DNA given the study environment is essential before choosing an appropriate approach. Plant eDNA from soil substrates is particularly subject to degradation or decay from decomposition processes of organic matter by both underground and above-ground biota (Pietramellara et al., 2009). Thus, long DNA fragments are expected to account for the lowest fraction of target soil eDNA that can be isolated. In tropical areas, warmer environments and richer decomposing communities may exacerbate DNA degradation rates in soils (Pietramellara et al., 2009). Furthermore, DNA decay increases with time and past plant signals may be only in the form of short DNA fragments (Kistler et al., 2017). For these reasons, it is generally recommended to employ markers targeting short DNA sequences. Indeed, most plant eDNA-based studies employ the chloroplast trnL (UAA) intron which amplifies on average a 50 bp region that has been robustly catalogued for the flora of Fennoscandia (Alsos et al., 2020). Noteworthy, in temperate areas, studies have successfully amplified matK, rbcL, ITS2 markers with target regions of more than 490 bp (Fahner et al., 2016). Furthermore, the use of short markers may be hindered by their low variability across species resolutions, that is, that identification of related taxa is supported by a few base pairs only (Taberlet et al., 2007). Employing short markers may thus require stricter thresholds of OTU (head sequence) matching to a reference sequence as the probability of identification mismatching resulting from polymerase errors is amplified. Although the prospecting of new plant DNA markers with targeted capture of multiple informative genes is promising, eDNA-based assessments can only identify taxa present in a reference sequence library.
Thus, an eDNA assessment is only as good as its reference library.
Although correlations between plant biomass and DNA concentration in the environmental samples are poorly understood, several studies-including this one-show that read counts may be used as a proxy for biomass of some plant life-forms (Deagle et al., 2013;Deiner et al., 2021). Finally, as DNA may remain in the environment after the organism is no longer present (Harrison et al., 2019), taxon detections provided by eDNA-based assessments should be interpreted merely as detections of organismal DNA. If one is interested in detections of live organisms, RNA approaches should be considered. These limitations highlight the importance of considering the current literature carefully to ensure that the study design is suited to address feasible and measurable questions.

| CON CLUS IONS
In this study, we investigated the temporal and spatial resolution of soil eDNA surveys of plant biodiversity to interpret the utility of this approach to effectively assess biodiversity and monitor vegetation changes through time and space. Our results show that a combination of aboveground vegetation surveys and soil eDNA surveys yields the most comprehensive inventory of plant diversity for a site.
In particular, a single soil eDNA sample mainly detects local plant diversity rather than site specific diversity. In addition, a soil eDNA sample captures plant diversity going back at least 30 years in time while matching most closely with current diversity. Similarly, we find that soil eDNA samples can be useful to detect both rare and unrecorded taxa, but are best at detecting abundant taxa. Our results highlight the potential of soil eDNA surveys to monitor vegetation responses over broader spatial and temporal scales, and encourage a rethinking of the optimal strategies for assessment of vegetation if soil eDNA is used as a method.

ACK N OWLED G EM ENTS
Morphological surveys were conducted as a part of a project from the Natural History Museum, University of Oslo. We recognise the great effort put into these surveys by their participants over 30 years. We like to thank Andreas Wollan for his support during soil sampling and plot recognition in the Solhomfjell Forest Reserve, Jarl Andreas Anmarkrud, Lisbeth Torbek and Audun Schrøder-Nielsen for their support in the laboratory for DNA extraction, library building and sequencing, Youri Lammers for his support in metabarcoding bioinformatics, Marcella Orwick Rydmark for her support and valuable feedback in manuscript writing, Rebecca Angelica Blakeney for proof-reading the manuscript. This research is part of the H2020 MSCA-ITN-ETN Plant.ID network, and has received funding from the European Union's Horizon 2020 research and innovation programme under the grant agreement No 765000.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/2041-210X.13865.

DATA AVA I L A B I L I T Y S TAT E M E N T
Raw sequences from both MiSeq runs (F and R) and bash and R codes are available in Ariza et al. (2022).