Quantifying wetness variability in aapa mires with Sentinel‐2: towards improved monitoring of an EU priority habitat

Aapa mires are waterlogged northern peatland ecosystems characterized by a patterned surface structure where water‐filled depressions (‘flarks’) alternate with drier hummock strings. As one of the EU Habitat Directive priority habitats, aapa mires are important for biodiversity and carbon cycling, harbouring several red‐listed species and supporting unique species communities. Due to their sensitivity to hydrological disturbances, reliable, up‐to‐date and systematic information on the hydrological condition and responses of mires is crucial and required for multiple purposes ranging from carbon exchange modelling to EU Habitats Directive reporting and conservation and ecosystem restoration planning. Here, we demonstrate the usability of Sentinel‐2 satellite data in a semi‐automatic cloud‐based approach to retrieve large‐scale information on aapa mire hydrological variability. Two satellite‐derived metrics, soil moisture index and the extent of water‐saturated surfaces based on pixel‐wise classification, are used to quantify monthly and interannual wetness variation between 2017 and 2020 across Natura 2000 aapa mires in Finland, including responses to the extreme drought of 2018. The results revealed high temporal variability in wetness, particularly in the southern parts of the aapa mire zone and generally in the late summer months interannually. Observations from the drought summer showed that one third of usually year‐round wet flark surfaces may be exposed to drying during climatic extremes. Responses varied between sites and regions, implicating the significance of environmental factors for drought resistance: some sites maintained high levels of moisture, whereas others lost wet surfaces completely. Our study provides the first comprehensive national‐level representation of seasonal and interannual wetness variability and drought‐sensitivity of pristine aapa mire sites. The approach and methods used here can be directly upscaled outside protected areas and to other EU countries. Thus, they provide a means for harmonized, systematic large‐scale monitoring of this priority habitat, as well as valuable information for other applications supporting peatland conservation and research.

The Habitats Directive requires EU Member States to achieve a favourable status for the priority habitats and their biota (Delbosc et al., 2021).The conservation status of boreal aapa mires is unfavourable-inadequate.This emphasizes the need for developing systematic, spatially comprehensive monitoring of the ecological status of aapa mires.The ecophysiological functioning of aapa mires depends on good hydrological conditions and requires sufficient water flows from adjacent areas and persistent wet surfaces in summer (Horsáková et al., 2018;Rehell, 2017).Particularly sensitive parts to decreasing water availability and extended droughts are minerotrophic, sparsely vegetated fens and open water surfaces ('flarks') and their species (Kolari et al., 2021;Rehell, 2017).Monitoring the variability in moisture and extent of wet flark surfaces can provide information on the ecological condition of aapa mires, and for the impacts of the warming climate such as increasing drought risks (Gong et al., 2012;Mäkiranta et al., 2018;Swindles et al., 2019).
Traditional field sampling alone is insufficient for operational monitoring across large areas (e.g.Pereira et al., 2010;Wang & Gamon, 2019).Thus, developing satellite remote-sensing (RS) technologies with repeatable and spatially extensive measurements is crucial to support biodiversity assessments and monitoring (Cavender-Bares et al., 2022;Reddy, 2021).Particularly useful RS information can be derived for mapping habitat extents and for indicating changes in ecologically relevant variables (Corbane et al., 2015).Thus, the integration of RSderived information to biodiversity monitoring with other biodiversity indicators is encouraged in concepts such as Convention on Biological Diversity (CBD, the Global Biodiversity Framework) and Essential Biodiversity Variables (EBV) (Pereira et al., 2010).
Mire conservation and monitoring is one of the fields where the potential of RS-derived information is not fully utilized, despite the growing number of promising methods (Kolari et al., 2021).Indeed, remote and vast aapa mires are an ideal target for RS-based monitoring as their treeless central parts provide an undisturbed view for aerial and satellite sensors.Moreover, several RS indices and models are available to estimate surface moisture conditions.Applying RS approaches to assess water tablelevel (WTL) fluctuations in boreal mires has provided generally promising results (e.g.Burdun et al., 2020;Meingast et al., 2014), with both optical and radar sensors (Räsänen et al., 2022).Of optical RS indices, those incorporating short-wave infrared (SWIR) wavelengths have been shown to correlate best with mire WTL (Burdun et al., 2020;Meingast et al., 2014).Variability in seasonally inundated ecosystems can also be described by monitoring the extent of water-covered areas in imagery through automated pixel classification (Lefebvre et al., 2019).
In this study, we apply optical remote sensing and cloud computing to semi-automatically assess the spatiotemporal dynamics of wetness in ≥2000 aapa mires in Finland.We focus on flark-dominated areas in aapa mires, where WTL typically stays close to the mire surface, maintaining very wet conditions throughout the year.Drought periods, however, may decrease WTL below normal levels, which is reflected in moisture content and optical properties of the mire surface.We use data from Sentinel-2 satellites of the European Space Agency (ESA) Copernicus programme, which provide open-access imagery at high spatial resolution (10-20 m), including two bands in SWIR wavelengths (20 m).To assess site-level seasonal and interannual wetness variation in 2017-2020, we use two different RS metrics: soil moisture index SWIR transformed reflectance (STR) and the extent of wet flark surfaces based on pixel-wise classification.An extreme drought occurring in summer 2018 offers valuable insights for aapa mires in terms of their climatic sensitivity and allows tentative assessment of their vulnerability to the projected future decrease in summertime water availability.
Our aims are to: (1) develop an RS-based approach to monitor hydrological conditions in boreal aapa mires across wide areas; (2) quantify seasonal and interannual wetness variability in aapa mire flark habitats using highresolution satellite data; (3) assess the sensitivity of aapa mire flarks to anomalously dry summer; and (4) discuss possibilities for further upscaling and applying this RS approach in operational biodiversity monitoring.

Study system and design
Our study system is boreal aapa mires of Finland, which have an estimated areal cover of 21 000 km 2 and thus make up a significant proportion (60%) of EU aapa mires (Eionet, 2023).Approximately 65% of the reported aapa mire area in Finland has been estimated to be in an unfavourable-inadequate condition.Ditching is a major threat to aapa mire hydrology, driving the unfavourableinadequate conservation status (Environmental Administration, 2019).We focus on mires that are included in the Natura 2000 network (N2K) and are thus relatively unexposed to anthropogenic disturbances.
We studied aapa mires separately for three mire zones (southern, middle and northern aapa mire zones; Fig. 1).Regional differences in aapa mires are reflected in their surface patterning, that is, alternating water-filled flarks and water blocking hummock strings.The southernmost aapa mires are characterized by lawn-level vegetation with no clear or a weak surface patterning.Strings are low and narrow, inhabiting fen vegetation.In the middle aapa mire zone, strings are higher and covered either by sedges and Molinia caerulea in the nutrient-rich sites or by Sphagnum fuscum bog vegetation.Flarks are larger and more abundant.In the north, strings are broader, often covered by Betula nana and affected by seasonal frost.The surface patterning with strings is the key driver of vertical water flows and water blocking in aapa mires, essential for retaining their hydrology (Laitinen et al., 2007;Rehell, 2017).
We focused on the wet, flark-dominated central parts of N2K aapa mires.These were determined by first extracting data on peatlands belonging to the Corine CLC2018 category 'peatbogs' clipped to the aapa mire zone (Fig. 1).Then from these data, we delineated flark fens using the land cover class 'peatland, difficult to traverse: treeless' in the topographical database of the National Land Survey of Finland (NLS).We included only flark-dominated mire areas that were larger than 1 ha in size and that were located on open mires larger than 10 ha (for the number and size distribution of the included mires, see Table S1).Flark-dominated patches located on the same open mire were considered as one entity.

SWIR transformed reflectance
Two RS metrics (Fig. 2) were used to indicate mire wetness: (1) STR soil moisture index, and (2) areal proportion of wet flark surfaces (%) based on supervised classification of image pixels.STR is a physical-based transformation on SWIR reflectance (R SWIR ), which correlates with soil and vegetation moisture (Sadeghi et al., 2015(Sadeghi et al., , 2017)).It is calculated as follows: STR has been used in mire WTL modelling as the moisture-sensitive part of the optical trapezoid model, OPTRAM (Burdun et al., 2020) and in a multi-metric study by Räsänen et al. (2022), with promising results especially in open mires.STR and SWIR are usually combined with some other band or index to derive moisture estimations, like with NDVI vegetation index in OPTRAM.Here we opted for using STR alone.OPTRAM would be ideal, but it requires visual local parameterization which hinders automated global applications (Burdun et al., 2020).Other common vegetation moisture indices like NDMI (Gao, 1996), however, would be problematic in heterogenous aapa mire environment, where the moisture index should be applicable to mixtures of various vegetated surfaces, sparsely vegetated mud-bottom fens and open water surfaces.NDMI does not recognize moisture in bare soils and is sensitive to phenology (Gao, 1996).STR is sensitive to moisture variation in both bare and vegetated soils, and it further increases on inundated surfaces (Sadeghi et al., 2017), thus providing a consistent gradient.Here, STR was calculated using SWIR band 11 of Sentinel-2 imagery ( $ 1600 nm), that was recognized as a highly important variable by Räsänen et al. (2022), outperforming SWIR-band 12. Instead of absolute prediction of moisture content or WTL, we aim to assess relative interannual variation in surface wetness.STR shows sensitivity to vegetation cover (Sadeghi et al., 2017) and its seasonal changes, but this should not hamper between-year comparisons.

Supervised classification
The extent of wet flark areas was estimated from the imagery using supervised classification of pixels.We trained a simple decision tree model following Lefebvre et al. (2019), who mapped inundated surfaces in wetlands based on satellite imagery.We used aerial orthoimagery derived from NLS to compose a set of 20 × 20 m training data pixels, including wet flarks and various drier mire surfaces.Two N2K mires from each aapa mire vegetation subzone (n = 8) were selected by an experienced mire researcher to represent regional variation, and in each site 53-138 training pixels were manually classified.Pixels were selected to represent a range of surfaces, from open water and sparsely vegetated wet flarks to drier lawn-and hummock-level surfaces and mire margins with sparse tree cover.Additional dry pixels from nearby peat extraction sites were included to represent very dry, unvegetated peat surfaces, which were absent or uncertain in aerial images of N2K sites.Pixels were selected from preferably homogenous surfaces to avoid mixed pixels.However, some mixed pixels were deliberately included in the wet training pixels with a condition that they are dominated by wet flark surfaces ( [ 50% pixel area), to ensure detection of wet flarks in mires with dense string patterning.Our final training dataset consisted of 1460 pixels.
Sentinel-2 satellite images were selected to maximally correspond with the collection dates of aerial imagery.Majority had a difference of 1 day or less, and maximum difference was 11 days.Aerial images were spatially aligned with S2 images using Q-gis Georeferencer tool.Reflectance values of Sentinel-2 bands and a set of common vegetation and moisture indices were extracted and calculated for training points.The R software package 'rpart' (Recursive Partitioning and Regression Trees) (Therneau & Atkinson, 2022) was used to train the decision tree.The final classifier was validated with leave-one-mire-out repeated cross-validation, using metrics of overall accuracy (OA), Fscore (F 1 ) and Matthew's correlation coefficient (MCC) in evaluation.The model's ability to reflect temporal variability in wetness was further confirmed by producing multitemporal classifications for the 16 training mires and by expert evaluation.In addition, in situ WTL measurements available at one southern zone mire (Olvassuo, Figure S4) in 2017 and 2018 were used to test the model sensitivity for seasonal and interannual WTL variability.
Detailed documentation of model development is provided in Supporting Information (Text S1) and the training data in project repository (https://doi.org/10.5281/ZENODO.7599117).

Processing satellite imagery
We used level-2A data (atmospherically corrected surface reflectance imagery) from Sentinel-2 optical satellites, available from November 2016 onwards, to retrieve wetness metrics for the period 2017-2020.The processing of imagery was executed in the Sentinel Hub Batch Statistical API cloud service (Sentinel Hub, https://www.sentinelhub.com,Sinergise Ltd, Ljubljana, Slovenia), that enables users to request aggregated statistics for multiple geometries for long image time series.Using the API service, we could efficiently calculate and extract the values for our wetness metrics for over two thousand mires and across the 4-year study period.Mires ≤1 km 2 were processed using 10-m spatial resolution imagery and larger mires using a coarser 20 m resolution for cost-effectiveness.
Processing was performed for snow-free dates with \80% cloud coverage in Sentinel image tiles (100 × 100 km 2 ) to maximize observations, for the period May-September.In mire bounding box areas we used a combination of L2A-scene classification and 160 m resolution s2cloudless cloud flag for cloud and snow-cover masking.Masking cloud shadows introduced a risk of misclassifying flark pools as shadows.Thus, instead we estimated coverage of clouds that potentially might cast shadows over the mire by applying a second cloud mask in a buffer area reaching 5000 m south and 500 m in other directions in the mire area.
The classification tree trained in R was recorded as a service compliant script ('evalscript') in the API service, and together with cloud masking the algorithm was implemented for each mire and each observation date.Finally, for each date, the following properties were retrieved per mire: (1) number of pixels in dry and waterlogged classes; (2) SWIR band B11 reflectance mean for STR calculation; and (3) cloud cover percentages estimated for mire area and the extended cloud shadow risk area.Scripts for classification and cloud masking, and an example of the data retrieval process are available in the project repository: https://doi.org/10.5281/ZENODO.7599117.
Next, cloud percentage information was used to filter the retrieved data by image quality in R. Observations were excluded in case the cloud coverage exceeded 1% in the mire bounding box area or 10% in the extended area.After the filtering process, there were on average 2.3 observations per mire in each month.Some of the sites lacked highquality observations in certain months but the vast majority had cloudless observations in at least 1 month in both early and late summer every year, indicating the magnitude of seasonal variation.The final data consisted of 2169 mires.
Observations of wetness metrics were aggregated to monthly mean values and visualized as graphs and maps, and relevant statistics of wetness temporal variability were extracted for each mire zone.Mire-specific monthly values were visualized to maps in a 10-km resolution grid.Additionally, STR and classification rasters were produced for one representative example mire in each zone to demonstrate seasonal and interannual variability visually (Figures S4-S6).Difference between drought-year wetness metrics means and normal year monthly averages in each zone was tested using paired t-tests.Normality was tested with Shapiro-Wilk test.
To support the interpretation of wetness data, monthly climatic water balance (WAB; precipitation sum À potential evapotranspiration) values were retrieved for the study mires based on daily gridded 1 × 1 km spatial resolution climatology for 1961-2020, that is, ClimGrid dataset (Aalto et al., 2016;Pirinen et al., 2022).Values were presented in a cumulative manner as a net balance for months following snowmelt, starting from May.Monthly net WAB and its variability in respect to normal variation (3rd, 10th and 30th percentiles) were calculated additionally for the reference period 1961-2020 to assess normality of climatic conditions during 2017-2020.

Variability in climatic WAB
In the reference period , summertime evaporative water loss in the aapa mire region is usually low and WAB reaches its minimum, typically À75 to 100 mm, in July (Fig. 3A).In 2018, multiple dry months caused exceptionally high negative WAB culminating in July.The net WAB minimum in 2018 was the third lowest recorded in the 60-year reference period, the net climatic water loss being more than double ( [ 200 mm) the normal amount.In the northern zone, summer 2018 was less exceptional than in the two other zones.Monthly WAB maps (Figure S7) show considerable spatial and temporal variation within the zones.

STR moisture index
Monthly variation in the STR moisture index reflected WAB variation, not only especially the 2018 drought but also in the sporadic wetter events in the late summer months in 2020 and 2017 (Figs.3B and 4).In the southern zone, STR values showed a clear seasonal unimodal pattern, reaching the lowest values in July and August.Annual STR variation was greatest in late summer in all zones, showing variability of AE22% during climatically normal years (2017, 2019, 2020) (Fig. 3b; Table 1).In the southern zone, variability was also high in June and September.
The dry year of 2018 stands out with relatively lower STR values.In the middle and southern zones, the months from May to August were drier than normal, with statistically significant (P \ 0.05) differences to the averages of respective months for 2017-2020.Driest conditions occurred in July, when STR was 36% lower than late summer average for the three other years.Mean STR values across zones decreased to very low levels (\2) that are more characteristic of hummock and lawn level than flarks according to STR distribution in training pixels.Regional differences were substantial: in the southern zone mires were 52% drier, in the middle zone 42% and in the northern zone Wetness Variability in Aapa Mires 25% drier than average.Variation in mire drought responses was high also within zones (Figs. 4 and 6).

Cover of wet surfaces
The classification model showed good accuracy in classifying pixels to wet and dry classes (OA = 93.6%,F 1 = 94.9%,MCC = 84.5%)and showed also visually plausible temporal variability.Correlation was high (r = 0.88) between measured WTL and the estimated extent of wet surfaces based on classification in the test mire (Figure S1).Even during the maximum spring-time extent, wet surfaces did not cover the entire flark-dominated area but on average 82% in the southern, 70% in the middle and only 45% in the northern zone sites (Fig. 3C).Of this extent, an average of 90% remained wet throughout summer in normal years (Table 1).Seasonal variation was larger in the south, where the wet area decreased on average 22% during summer.The total area of permanently wet flark surfaces in N2K mires in normal years was estimated at c. 71 000 ha. Almost half of these, 31 000 ha, were in the middle aapa mire zone, 22 000 ha in the southern, and 18 000 ha in the northern zone respectively (Fig. 3D; Figure S8).Similar to STR values, the cover of wet surfaces followed monthly and interannual WAB variation and had greatest annual variability in late summer (Fig. 3).During the normal years, wet surface cover in late summer varied annually largely across zones, on average from a À46.7% decrease from spring reference maximum coverage to a +11.2% increase in years with wet late summer conditions (Table 1).
As with STR, the dry summer of 2018 led to notably lower wetness levels compared to the average of the normal years (Fig. 3C).In July, the wet surface area in mires had shrunk on average to 48% of their spring reference maximum extent: to 59% in the northern, 52% in the middle, and as low as 25% in the southern zone mires.The decrease was more than double the normal average seasonal decrease.In the northern zone, 2018 values fall within the variation of the other years.In the middle zone, a similar magnitude of wet surface decrease is observed both in 2018 and 2019, but with an earlier starting point and a longer duration of drought in 2018 (Fig. 3; Figure S7).Of the normally permanently wet flark area, roughly two thirds maintained wet conditions throughout the year in 2018.In the southern zone, less than half remained wet.
There was a large variation in responses within the zones (Figs. 5 and 6).Some of the sites appeared to completely lose their wet surfaces, whereas some maintained large extents of wet surfaces during the driest periods, even in the most sensitive regions in the south.

Discussion
Adequate hydrological conditions and persistence of wet flark surfaces is fundamental for the good ecological condition in aapa mires (Mäkiranta et al., 2018;Tahvanainen, 2011).WTLs vary in aapa mires dynamically, with  seasonal fluctuation and temporal variability being a natural part of their hydrology (Solantie, 2006).There is also notable between-mire variability in their wetness responses to meteorological conditions, with some mires being more sensitive than others to drought (Rinne et al., 2020).This variability complicates assessments of the hydrological status of aapa mires across large regions based solely on in situ measurements from individual sites, or on temporally scarce data.However, the significant links of aapa mire biodiversity and greenhouse gas fluxes to hydrology, and disturbances from land use and projected changes in climate, underline the importance of developing systematic wetness monitoring.Here, we demonstrated the possibilities of satellite remote sensing in acquiring such information and introduced a costeffective approach to monitoring aapa mire hydrological  conditions across wide areas.The WAB conditions during the four study years include one exceptionally dry and three normal years compared to the 1961-2020 reference data.Thus, our Sentinel-2 imagery analyses provide useful new information on the regional and temporal variability in the wetness of pristine aapa mire flark habitats, and the impacts of drought events on them.
Observed wetness variability and drought responses Regional differences in the wetness seasonality of the N2K aapa mires agreed with the earlier knowledge of regional mire characteristics (e.g.Laitinen et al., 2007).Seasonal changes were greatest in the southern aapa mire zone characterized by a large proportion of hydrologically intermediate lawn-level surfaces.In contrast, the flark fen-dominated middle zone had the largest area of stable, permanently wet surfaces.
The monthly and interannual variation shown by both RS metrics imply a short-term climatic sensitivity of aapa mires.The range of interannual variability was generally smaller in early summer and autumn, while the greatest variations occurred in late summer.This pattern reflects the effect of flooding in spring and early summer, when water storage of mires is saturated and climatic WAB plays a relatively small role.In the southern zone, where flooding occurs earlier, observed interannual variability in wetness was also large in June.
Drought events had profound impacts on flark fen wetness.In summer 2018, a remarkable drop in WTL was observed in boreal mires in Europe (Rinne et al., 2020).Our results suggest that the impact was indeed widespread in the aapa mire zone and substantially affected the flarkdominated parts.This climatic 'stress test' revealed that one third of permanently wet flark area in N2K mires are exposed to extreme drought disturbances.From an ecological perspective, surface drying affects fen-and flarkdependent species, including many red-listed mire plant species (Saarimaa et al., 2019), and the biological processes dependent on water-saturated conditions.Mosses are especially vulnerable to even slight changes in WTL (Gerdol et al., 1995;Laitinen et al., 2008).In addition, the oxidation of peat and disturbance of photosynthesis stemming from lowered WTL can alter the greenhouse gas balance.In 2018, multiple mires turned from net sinks to carbon sources, with a net warming effect on the climate (Rinne et al., 2020).Climatic variability and frequency of extreme events are projected to increase in the future (Ruosteenoja et al., 2018), suggesting that similar effects on aapa mire wetness are likely to increase.
Importantly, mire wetness responses to climatic variability and drought varied largely between and within regions.The southern and middle zones experienced nearly identical net WAB dynamics in 2018 from May to July, but the moisture index and wetness levels decreased more and more rapidly in southern mires.Differences in snowmelt timing and flood regimes are likely to be major drivers of these regional differences (Luomaranta et al., 2019;Sallinen et al., 2023).In addition, variation between mires was remarkable, particularly within the southern aapa mire zone.Our results provide large-scale evidence for the substantial spatial variability in drought sensitivity of aapa mires, in agreement with in situ WTL measurements (Rinne et al., 2020).Across mire zones there are mires that maintain high levels of moisture and wet flark surfaces, even under exceptional summer conditions, and respectively, sites that lose their water surfaces completely.Some within-zone variation is explained by variation in climatic conditions, but local environmental factors probably also have a role here.These include mire structure and vegetational characteristics such as weaker string-flark structure (Laitinen et al., 2007), and characteristics of the surrounding watershed, like its size and inclination as well as the amount of anthropogenic disturbance (Rehell, 2017;Sallinen et al., 2019).Although the aapa mires in the N2K network are mostly undrained, intensive ditching in the watersheds, especially in the southern aapa mire zone, can pose carryover impacts on the water flows (Rehell, 2017), which may make mires more reactive to climatic variability (Gong et al., 2012).

Applications of RS information in biodiversity conservation
Remote sensing enables the monitoring of habitat changes and supports the detection of mires that are sensitive to environmental changes.At the global level, one successful example of satellite-based ecosystem condition mapping is the Global Surface Water Explorer (GSWE, https://globalsurface-water.appspot.com/)which provides statistics on the distribution and temporal changes to the world's surface waters.Our Sentinel-2 study provides similar indepth regional RS information on surface wetness and ecological status in priority peatland ecosystems.
The drought responses revealed here can be applied as indicators of mire ecosystem vulnerability to climatic disturbances.Regions and sites that expressed strong wetness responses to the 2018 drought and short-term variability will arguably be the most sensitive to future climatic changes as well.Sensitivity information can be utilized in climate-wise nature conservation planning to enhance the protected area network, as well as targeting ecosystem restoration.A particularly useful application would be RSbased monitoring in restored aapa mires, as peatland restoration is becoming a focal point for EU-level nature conservation (Farrell et al., 2022).Spatial information on wetness and climatic responses of mires is also needed in upscaling carbon exchange estimations and predicting the effects of climatic changes (Laine et al., 2019;Rinne et al., 2020).Moreover, the high temporal variability in flark wetness observed here highlights the importance of spatially extensive multi-year data for monitoring of ecosystem processes and setting condition reference levels in a reliable manner.
Extensive information is especially important for remote, highly variable and seasonally dynamic systems such as northern mires.From an ecological perspective, the monitoring of changes in wet surface coverage in aapa mires is crucial as the heterogeneity of habitats supports the high local mire species richness, and climatic changes and land use threaten to decrease the cover of wet flark habitats (Kolari et al., 2021;Tahvanainen, 2011).Clearly, satellite remote sensing cannot replace detailed in situ measurements, but it can provide a new level of multiscale monitoring of ecosystems (Corbane et al., 2015;Reddy, 2021).Openly available Sentinel-2 imagery and the methods used here allow observation over the whole of Finland, or any EU country, almost continuously and with reasonable accuracy, and can contribute to repeated systematic ecosystem condition assessments across spatial scales.

Methodological appraisal
RS-based metrics and their ability to describe surface moisture conditions need to be considered in terms of potential shortcomings.A key uncertainty is that correlation between spectral reflectance and actual moisture levels is not perfect or straightforward.In previous studies, optical satellite-based variables explained 46-70% of the WTL fluctuations in open mires, with large betweensite variation (Burdun et al., 2020;Räsänen et al., 2022).Surface moisture is generally easier to estimate with RS than underground WTL (Meingast et al., 2014) although moisture-RS index relationships may be complicated due to effects of vegetation properties on moisture-sensitive wavelengths.In homogenous environments, vegetation moisture indices usually apply another wavelength or index to control for bias effects.
In our case, the heterogeneity of the aapa mire environment brings about a challenge: a moisture index should be able to reflect the moisture status on various surfaces, ranging from bryophyte and vascular plant vegetation to open water pools and sparsely vegetated mudbottom flarks.A single moisture-sensitive wavelength (SWIR) was used here for calculating the STR index, which has a linear relationship to soil and vegetation water content (on a given vegetation coverage) and a continuing response to water increase after the moisture content reaches saturation (Sadeghi et al., 2017).STR is not only mainly controlled by moisture content, but it is also sensitive to vegetation properties (Ceccato et al., 2001;Sadeghi et al., 2017).Such effects are difficult to control due to mire heterogeneity, hindering the development of accurate measures of actual moisture content.However, the STR index can reflect relative changes and anomalies in moisture status over time.Thus, interannual comparison within a site or region can be considered reliable, as well as the comparison of relative changes between sites and regions (cf.Lees et al., 2020).However, actual reliable predictors of surface moisture would certainly provide an optimal solution for mire monitoring.There is a clear need to develop and test moisture indices that might be suitable also to heterogenous aapa mire environments, as these are currently lacking.For example, OPTRAM might be a good alternative mire moisture indicator for STR, if automated solutions for its parameterization and reliable global application are developed further (Burdun et al., 2020).Moreover, testing and developing models for predicting mire surface moisture status alongside with WTL may provide useful future applications.Surface (vegetation) moisture is more easily predicted with RS methods than WTL (Meingast et al., 2014), and from ecological point of view it is a very relevant measure of the drought stress encountered by mire plant species.
The pixel-based classification method used here, however, had a very high overall classification accuracy (93.6%), like in the corresponding classifications of inundated surfaces in wetland ecosystems (Lefebvre et al., 2019).Interpretation is more straightforward compared to STR, although the classification process may be subjected to similar vegetation-related uncertainties in SWIR-band reflectance responses.The main factor limiting the accuracy of classification approaches is the spatial resolution of imagery, which becomes particularly important in patterned mires characterized by microform topography.For these reasons, high-resolution imagery has been recommended to be used in classification applications in this kind of heterogenous environments (Räsänen & Virtanen, 2019).Flark pools smaller than 10-20 m Sentinel-image pixels are not necessarily recognized in estimated wet surface coverage, causing potential underestimation of the wet area in densely patterned mires.Moreover, our classification tree is largely based on SWIR-bands of 20 m spatial resolution originally.We tried to control this expected bias by including deliberately some wet-surface-dominated mixed pixels to the training data, as described in Text S1.Despite of the uncertainties, a classification approach provides valuable information that is not achieved using only moisture index meansit enables focusing on the actual flark habitats for which spatially explicit data is very sparse.Moreover, although Sentinel-2 pixel size is rough compared to higher resolution alternatives provided by aerial and drone imagery, the advantages of satellite-derived information are undisputable when systematic frequent monitoring across large areas is the target.Indeed, this study provides the first spatially explicit nationwide estimation of permanently wet flark area across the N2K mires in the aapa mire zone.For species and ecosystems processes, the estimation of wet area coverage is as important as the magnitude of moisture change in the area.
Employing different RS approaches in observation enhances the validity of conclusions.Here, simultaneous changes both in STR and wet surface area support our interpretations of significant changes in wetness status.As our classification data reveals, in most of the sites the areal delineation to flark-dominated mires still also contains habitats other than flark surfaces.This information supports the interpretation of averaged STR values.The drought event in 2018 decreased the STR moisture index values in mires, and a simultaneous decrease in the extent of wet surfaces confirmed that this decrease was not caused only by moisture changes at the hummock level but included changes in actual flarks.Increased reliability emerging from the multi-method approach would be particularly useful in large-scale semi-automated monitoring, where a wholesale validation of results is not possible.

Conclusions
We utilized Sentinel-2 satellite imagery and cloud computing to retrieve wetness information relevant for assessing the ecosystem condition of aapa mires across the aapa mire zone.Based on two RS-derived metrics, we created the first national-level representation of wetness variability in pristine aapa mires in Finland.Regardless of data uncertainties in providing accurate estimations of moisture, conclusions could be drawn by measuring relative changes in RS indicators.The applied metrics revealed remarkable spatiotemporal variation in wetness and sensitivity to meteorological drought, enabling the detection of mires that are potentially vulnerable to climatic changes.The approach is applicable to aapa mires outside the N2K network and mires located in other parts of Fennoscandia, enabling extensive and harmonized monitoring of this priority habitat.This might benefit EU Habitat Directive reporting, for example, which is obligatory to Member States but where previous methods have varied considerably.Our study provides a step towards systematic ecosystem monitoring using RS methods and adds a new large-scale source of information to support the effective conservation management of EU priority habitats.

Figure 1 .
Figure 1.Location of Finland in northern Europe (upper right corner), the three aapa mire vegetation zones (sensu Ruuhijärvi, 1988) and study sites (n = 2169) (left-hand side) and an aapa mire (lower right corner) covered by spring floods showing the characteristic patterned structure with elongated hummock strings and inundated flark depressions between them.Patvinsuo National Park, Eastern Finland.Photo: Maarit Similä.

Figure 2 .
Figure 2. Raw RGB and processed Sentinel imagery from Pesäneva aapa mire illustrating conditions during the drought in summer 2018 (30/07/ 2018) (A, C, E) and approximately normal late summer conditions in 2020 (01/08/2020) (B, D, F).The figures show Sentinel RGB true colour images of the mire (A, B), the variation in STR moisture index values (C, D), and classification into wet and dry mire surfaces based on supervised classification of pixels (E, F).The white line delineates the flark-dominated part of the mire for which wetness measures are calculated, determined based on the category 'Open, difficult to traverse mire', as extracted from the topographic database produced by the National Land Survey of Finland.STR, SWIR transformed reflectance; SWIR, short-wave infrared.

Figure 3 .
Figure3.Seasonal and interannual variation of climatic water balance (WAB) (A), average mire surface moisture condition indicated by the STR index (B), percentage of mire surfaces classified as wet (C), and total area of wet surfaces (D) in the studied flark fens embedded in N2K aapa mires ('flark mires') in the three aapa mire zones.Climatic WAB (precipitation-evaporation) was extracted from ClimGrid data developed by FMI and shown as a cumulative value starting from May, with the dashed line representing the 60-year average and shadings representing the range of normal values, values occurring more than once in 10 years and more than once in 30 years: middle 33%, middle 80% and middle 93.3% of 60-year observations.In the STR graphs, low values indicate dry surface moisture conditions and high values wet conditions.Data gaps due to cloudiness in monthly satellite indicator data of individual mire sites are filled with monthly averages calculated from available years.STR, SWIR transformed reflectance; SWIR, short-wave infrared.

Figure 4 .
Figure 4. Monthly and interannual variation of SWIR transformed reflectance (STR) values in the N2K network flark mires, and the monthly difference (lowest row) in STR values between the exceptionally dry year of 2018 and the monthly average of the other 3 years.Results were aggregated to a 10-km grid.Low STR values indicate dry conditions and high values wet surface conditions, respectively.No-data values occur in months when there is snow cover or abundant cloudy weather.SWIR, short-wave infrared.

Figure 5 .
Figure 5. Monthly and yearly variation of wet surface extent, expressed as a percentage with respect to the local maximum extent of wet surfaces (average spring maximum extent) in 2017-2020.The lowest row shows the monthly difference between the exceptionally dry year of 2018 and the monthly average of the other 3 years.Results are shown for the flark mires belonging to the N2K network, aggregated to a 10-km grid.No-data values occur in months when there is thick snow cover or abundant cloudy weather.

Figure 6 .
Figure 6.Monthly between-site wetness variation in the studied flark mires in the three aapa mire zones, shown as boxplots.The variation in wetness is indicated using the two remote-sensing metrics: STR and the proportion of wet surface.The values for the drought year of 2018 are compared to the variation within the climatically normal years (2017, 2019, 2020).

ª
2023 The Authors.Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London.

Table 1 .
Median wetness indicator statistics for the studied aapa mires in the three aapa mire zones.The statistics were calculated from the observations in late summer (July, August) when interannual variation is greatest.Here, first the mean STR and wet area values for individual mires across the climatically normal study years(2017, 2019, 2020)were calculated, and then the median value statistics based on the individual mire values were calculated for this table.Min and max values were calculated similarly and are provided both as absolute values and as percentage differences to the mean.Wet area is represented as a percentage value in relation to the reference point, that is, the maximum extent of the wet area in May.Minimum values for the exceptionally dry year of 2018 are provided in coloured columns.