Jukuri, open repository of the Natural Resources Institute Finland (Luke)               
   
 
 
All material supplied via Jukuri is protected by copyright and other intellectual property rights. 
Duplication or sale, in electronic or print form, of any part of the repository collections is prohibited. 
Making electronic or print copies of the material is permitted only for your own personal use or for 
educational purposes. For other purposes, this article may be used in accordance with the publisher’s 
terms. There may be differences between this version and the publisher’s version. You are advised to 
cite the publisher’s version. 
This is an electronic reprint of the original article.  
This reprint may differ from the original in pagination and typographic detail. 
 
Author(s): Aleksi Isoaho, Merja Elo, Hannu Marttila, Parvez Rana, Anssi Lensu, Aleksi Räsänen 
Title: Monitoring changes in boreal peatland vegetation after restoration with optical 
satellite imagery 
Year: 2024 
Version: Published version 
Copyright: The Author(s) 2024 
Rights: CC BY 4.0 
Rights url: https://creativecommons.org/licenses/by/4.0/  
 
Please cite the original version: 
Aleksi Isoaho, Merja Elo, Hannu Marttila, Parvez Rana, Anssi Lensu, Aleksi Räsänen, Monitoring 
changes in boreal peatland vegetation after restoration with optical satellite imagery, Science of The 
Total Environment, Volume 957, 2024, 177697, ISSN 0048-9697, 
https://doi.org/10.1016/j.scitotenv.2024.177697. 
Monitoring changes in boreal peatland vegetation after restoration with
optical satellite imagery
Aleksi Isoaho a,b,*, Merja Elo c,d,e, Hannu Marttila b, Parvez Rana a, Anssi Lensu d,
Aleksi Rasanen a,f
a Natural Resources Institute Finland (Luke), Paavo Havaksen tie 3, FI-90570 Oulu, Finland
b Water, Energy and Environmental Engineering Research Unit, Faculty of Technology, University of Oulu, FI-90014 Oulu, Finland
c Finnish Environment Institute, Survontie 9A, FI-40500 Jyvaskyla, Finland
d Department of Biological and Environmental Science, University of Jyvaskyla, FI-40014 Jyvaskyla, Finland
e School of Resource Wisdom, University of Jyvaskyla, FI-40014 Jyvaskyla, Finland
f Geography Research Unit, Faculty of Science, University of Oulu, P.O. Box 8000, Oulu, Finland
H I G H L I G H T S G R A P H I C A L A B S T R A C T
 We assess post-restoration changes in
peatland vegetation with satellite
imagery.
 Different peatland types should be
monitored with different optical
variables.
 Floristic gradients of pine mires forests
and open mires can be remotely sensed.
 Monitoring vegetation development is
feasible in peatlands with little or no
trees.
A R T I C L E I N F O
Editor: Paulo Pereira
Keywords:
Remote sensing
Machine learning
Mires
Land use
Drainage
A B S T R A C T
Restoration can initiate a succession of plant communities towards those of pristine peatlands. Field inventory-
based vegetation monitoring is labour-intensive and not feasible for every restored site. While remote sensing has
been used to monitor hydrological changes in peatlands, it has been less used to monitor post-restoration changes
in vegetation composition. We utilised vegetation inventories from Finnish peatland monitoring network con-
taining 10-year before-after-control-impact monitoring data from 150 peatland sites, representing three peatland
types (spruce mire forests, pine mire forests, open mires), and optical observations from Landsat 5–9 and
Sentinel-2 satellites. We employed non-metric multidimensional scaling (NMDS) to produce floristic gradients,
representing wetness and productivity, from the vegetation data. We constructed random forest regression
models with NMDS dimensions, i.e. floristic gradients, as response variables and satellite imagery variables as the
predictors. Our results show that the floristic gradients in different peatland types should be monitored with
different satellite imagery variables. However, midsummer NIR and red band consistently explain variation in
the gradients in all peatland types. Our results indicate that the gradients and the post-restoration changes in
them can be modelled with reasonable accuracy in open mires and sparsely treed pine mire forests but not in
* Corresponding author at: Natural Resources Institute Finland (Luke), Paavo Havaksen tie 3, FI-90570 Oulu, Finland.
E-mail address: aleksi.isoaho@luke.fi (A. Isoaho).
Contents lists available at ScienceDirect
Science of the Total Environment
journal homepage: www.elsevier.com/locate/scitotenv
https://doi.org/10.1016/j.scitotenv.2024.177697
Received 13 May 2024; Received in revised form 18 November 2024; Accepted 19 November 2024
Science of the Total Environment 957 (2024) 177697 
Available online 26 November 2024 0048-9697/© 2024 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). 
densely treed spruce mire forests. We suggest that optical satellite imagery can serve as a proxy for assessing the
post-restoration vegetation changes in peatlands with little or no trees.
1. Introduction
Peatlands provide important ecosystem services, harbour unique
biodiversity and store large carbon stocks (Chapman et al., 2003; Bonn
et al., 2014; Joosten et al., 2016). The degradation of peatlands mainly
due to land use activities has caused major negative impacts on these
functions (Marttila et al., 2020). Currently, around 11 % of the world’s
peatlands are degraded (Leifeld and Menichetti, 2018) and the situation
is particularly severe in the European Union, where half of the peatlands
are already degraded (European Commission, 2020). In the boreal zone,
drainage for forestry and agriculture is the main reason for peatland
degradation (Minkkinen et al., 2008). The drainage for forestry lowers
the water table level within the peat (Price et al., 2003; Haapalehto
et al., 2014), and leads to a shift in plant communities towards drier,
more mineral soil forest type of vegetation (Laine et al., 2006; Maana-
vilja et al., 2014).
Restoration can significantly improve the condition of degraded
peatlands and re-activate their natural functions (Rana et al., 2024).
Restoration in boreal forestry-drained peatlands aims to raise the water
table, allow regeneration of peatland vegetation and initiate formation
of new peat. This process involves rewetting the drained areas through
damming and infilling the drainage channels and in some cases,
removing trees that have grown after drainage (Andersen et al., 2017).
Restoration has fast effect on hydrological properties, such as raising the
water table (Haapalehto et al., 2011, 2014; Menberu et al., 2016; Isoaho
et al., 2024).
Restoration is also successful in initiating the recovery of vegetation
towards pristine-like vegetation (Laine et al., 2011; Haapalehto et al.,
2011; Maanavilja et al., 2014; Kareksela et al., 2015; Haapalehto et al.,
2017; Elo et al., 2024b). However, vegetation in restored sites remains
dissimilar compared to pristine sites 10 years after restoration (e.g.
Haapalehto et al., 2011; Elo et al., 2024b), and restoration may not re-
turn the peatlands fully to their pristine condition even after 30 years
(Kreyling et al., 2021). Hence, long-term monitoring is needed to cap-
ture the full spectrum of vegetation changes and challenges associated
with restoration efforts which may not be evident immediately after the
restoration. However, comprehensive vegetation monitoring is labour-
intensive and not feasible for every restored sites. Thus, the develop-
ment of new, more automated monitoring methods is needed.
Remote sensing (RS) techniques have been increasingly utilised for
upscaling the vegetation observations across spatial and temporal scale
in peatlands (Harris et al., 2015; Beyer et al., 2019; Rasanen et al., 2019,
2020; Pang et al., 2023). Usually, optical imagery has been utilised, and
the importance of near-infrared (NIR) region has been emphasized
(Harris et al., 2015; Rasanen et al., 2020; Kolari et al., 2022) due its
association with vegetation patterns, and being able to separate different
plant species, habitats and peatland types (Middleton et al., 2012; Salko
et al., 2023). Typically, it has been suggested to use fine spatial reso-
lution imagery captured by uncrewed aerial vehicles (UAVs) due to the
spatially heterogeneous structure of peatlands and because the vegeta-
tion inventories are usually conducted on small, often 1 m2 or smaller
plots (Rasanen and Virtanen, 2019; Rasanen et al., 2020; Wolff et al.,
2023; Steenvoorden and Limpens, 2023). This makes coarse resolution
(>30 m) imagery from satellites less usable. However, the limited
temporal coverage of fine-resolution imagery from UAVs, crewed air-
crafts or satellites makes their use impractical for long-term monitoring.
Instead, medium resolution satellite imagery (10–30 m pixel size), such
as Landsat 5–9 (L5–9) and Sentinel-2 (S2) has been successfully used for
vegetation change detection in peatlands (Kolari et al., 2022) and
proven effective in detecting seasonal and spatial vegetation patterns
(Bhatnagar et al., 2020; Garisoain et al., 2023; Pang et al., 2023).
RS-based vegetation studies in peatlands often focus on producing
categorical maps of land cover, habitat or vegetation types (e.g. Rasanen
and Virtanen, 2019; Steenvoorden et al., 2023). While some studies have
shown species-level upscaling possibilities with RS data (Kalacska et al.,
2013; Husson et al., 2014; Bergamo et al., 2023), more prominent ap-
proaches have been predicting groups of species such as plant functional
groups or plant communities (Harris et al., 2015; Rasanen et al., 2020;
Steenvoorden et al., 2023; Steenvoorden and Limpens, 2023; Wolff
et al., 2023). It has been argued that continuous vegetation maps offer a
more realistic depiction of vegetation patterns than categorical maps
(Rocchini, 2014; Rasanen et al., 2019). Particularly, floristic gradients
derived from vegetation ordination analysis provide insights into species
distribution along ecological and environmental gradients such as
wetness or nutrient status (Harris et al., 2015) which are often the focus
of restoration efforts (Menberu et al., 2016, 2017). Ordination analysis is
a widely used method in community ecology to simplify complex species
presence or abundance data into gradients that explain a large part of
the variation in the data (e.g. Anderson, 1971; Verniest and Greulich,
2019). Floristic gradients offer possibility to determine similarities and
differences between plots or sites in a continuous, multidimensional
space (e.g. Rasanen et al., 2020), and for restoration assessment, they
can show how plant communities in restored sites become less dissimilar
to communities in drained sites state and more similar to those of pris-
tine sites state (Elo et al., 2024b). In peatlands, the gradients that explain
most of the variation in the data are typically related to wetness and
productivity (Pakarinen and Ruuhijarvi, 1978; Middleton et al., 2012;
Rasanen et al., 2020).
In RS studies related to peatland restoration, Ball et al. (2023) have
demonstrated that on a large scale, spectral properties of restored
peatlands are approaching peatlands with a good ecological condition in
a 25-year timescale. A few studies (e.g. Kolari et al., 2022; Steenvoorden
et al., 2022; Talvitie et al., 2023) have shown that long-term changes in
peatland microforms or habitat types can be monitored with RS data.
Additionally multiple studies have indicated that hydrological changes
following restoration can be assessed with RS data (Rasanen et al., 2022;
Ikkala et al., 2022; Burdun et al., 2023; Isoaho et al., 2023, 2024).
Nevertheless, to the best of our knowledge, no studies have utilised RS
data for detecting long-term and frequently monitored post-restoration
vegetation changes.
In this study, we hypothesize that optical satellite imagery can be
effectively used to monitor post-restoration changes in peatland vege-
tation. To examine this hypothesis, we use extensive and widespread
peatland restoration vegetation inventory in Finland (Elo et al., 2024a,
2024b), utilising a before-after-control-impact scheme (Christie et al.,
2020). Our specific objectives are to: (1) identify the most suitable op-
tical satellite variables for modelling the continuous floristic gradients of
boreal spruce mire forests, pine mire forests and open mires with
different treatments, (2) assess how well the floristic gradients can be
predicted with optical satellite variables, and (3) determine the feasi-
bility of using optical satellite variables to monitor temporal changes in
floristic gradients.
2. Materials and methods
2.1. Finnish peatland monitoring network
We utilised field data from the peatland monitoring network estab-
lished by Metsahallitus Parks & Wildlife, Finland (Fig. 1). The network
comprises 150 monitoring sites across three peatland types: spruce mire
forests (n ˆ 49), pine mire forests (n ˆ 51), and open mires (n ˆ 50),
each undergoing different treatments: pristine (n ˆ 60), restored (n ˆ
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
2 
60), and drained (n ˆ 30). Additionally, the sites have been classified
with rich or poor productivity based on their nutrient level and species
composition by the botanical experts of Metsahallitus Parks & Wildlife,
Finland. The pristine and restored sites were established between 2006
and 2012, while all drained sites were established in 2012. The sites
have been used to monitor changes in vegetation, water table and
porewater variables. The data obtained from the network is compre-
hensive and enables large-scale studies on peatland vegetation, hy-
drology and biogeochemistry, and their responses to the restoration
(Menberu et al., 2016, 2017; Rasanen et al., 2022; Burdun et al., 2023;
Elo et al., 2024b). From the original 151 site network, one site (Id 2; Elo
et al., 2024a) have been removed due the site being destroyed after first
vegetation sampling.
Spruce mire forests are characterised by relatively dense canopy.
Based on the Finnish multi-source National Forest Inventory data
(Natural Resources Institute Finland, 2023), the pristine sites of the
monitoring network have average tree coverage of 68 %. Tree cover is
usually dominated by Picea abies or Betula pubescenes. The water table in
these mires can reach depth of over 50 cm and in drained sites, even up
to 1 m during midsummer (Menberu et al., 2016), resulting in ground
vegetation that includes both peatland and mineral soil forest plant
species such as Vaccinium myrtillus andMaianthemum bifolium. Pine mire
forests have sparser tree cover (average of 26 % in the pristine sites of
the network; Natural Resources Institute Finland, 2023) and the typical
tree species include Pinus sylvestris and Betula pubescens. Their ground
layer is characterised by different Sphagnum mosses and vascular plants
such as Eriophorum vaginatum and Rhododendron tomentosum. The water
table of pine mire forests is typically closer to the surface compared to
spruce mire forests, but in drained sites the water table is similar to
pristine or even drained spruce mire forests (Menberu et al., 2016).
Open mires have almost non-existent tree cover (average of 4 % in the
pristine sites of the network; Natural Resources Institute Finland, 2023)
and ground vegetation is characterised e.g. by Sphagnum and Carex
species that are adapted to wet conditions. Pristine open mires are
typically wetter than spruce or pine mire forests, and the water table can
rise on the peatland surface during spring and autumn floods. Addi-
tionally, seasonal water table variation in pristine open mires is lower
compared to the other two types (Menberu et al., 2016). However, in
drained sites, the water table can be similar to pine mire forests, and in
well-drained sites, similar to spruce mire forests (Menberu et al., 2016).
Vegetation monitoring at restored sites was conducted before
restoration, and again two, five, and ten years after restoration. For
pristine and drained sites, a similar monitoring schedule was followed,
although the actual number of years between the samplings may differ.
Each site includes ten permanent 1 m2 plots (Fig. 1) within which the
percent cover (%-cover) of every vascular plant and moss species has
been visually estimated by the botanical experts of Metsahallitus Parks
& Wildlife, Finland. The data from those ten plots were aggregated to
represent the average %-cover of the species at each of the 150 study
sites. There have been some uncertainties with identifying in particular
Sphagnum species, which is why some species have been grouped to
avoid uncertainty due to misidentification. A more detailed explanation
of the vegetation monitoring methodology is provided in Elo et al.
(2024b).
2.2. Remote sensing data
We utilised multitemporal optical S2 and L5–9 satellite imagery,
Fig. 1. Finnish peatland monitoring network. Map demonstrates locations of the monitoring sites. Second chart demonstrates number of monitored sites within
peatland types and productivity (poor and rich), colours of peatland treatments are same for map (A) and chart (B). Chart C demonstrates how the vegetation
inventory is conducted within the sites, i.e. two parallel lines with five 1 m2 plots and the distance between the plot edges is four metres.
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
3 
accessed and processed in Google Earth Engine (Gorelick et al., 2017).
First, we filtered out imagery with cloud cover exceeding 30 %. Leftover
clouds, cloud shadows and snow were masked using the Scene Land
Cover and Quality Assessment pixel classification for S2, and L5–9,
respectively. Second, we improved cross-usability by harmonising all
imagery to L8–9 Operational Land Imager sensor specifications using the
coefficients and slopes proposed by Roy et al. (2016) for L5 and L7, and
by Zhang et al. (2018) for S2. Third, we utilised six spectral bands and
calculated eleven spectral vegetation and wetness indices (Table 1),
broadly utilised in hydrological and ecological peatland studies from the
S2 and L5–9 imagery (Table S1). We extracted the data for the years
2000–2023 with area-weighted mean values for each RS band and index
for 15 m radius buffers. The centre of each buffer located in the middle
of parallel lines between the vegetation inventory plots (see Fig. 1), i.e.
only one buffer per site was used and it contained all the vegetation
inventory plots of the site. We removed all obvious outliers (STR  10,
EVI or SAVI outside of 0–1 range) from the dataset. Lastly, for dates with
multiple RS observations at one site, we calculated the mean value of
these observations.
We further processed the RS data to make it more comparable with
the vegetation data and across different sites, as well as to minimise
possible errors from residual clouds and cloud shadows. This involved
calculating annual medians for two distinct time windows: early sum-
mer (1 May–15 June) and midsummer (1 July–15 August) to capture
different hydrological (Sallinen et al., 2023; Isoaho et al., 2024) and
phenological (Juutinen et al., 2017; Pang et al., 2023) moments within
growing season. To further minimise the multiyear variation and in-
crease RS data availability, we calculated the so-called sampling me-
dians (median value from annual medians used for sampling
calculations) from annual medians utilising years specified in Table 2.
For 0-year sampling, we used satellite data from five years prior the
actual sampling year as these rely solely on older L5 and L7 products,
which have a sparser revisit time compared to L8–9 and S2. For the rest
of samplings, we used a three-year period centred around the actual
sampling year (Table 2). Total number of utilised observation dates used
per sampling medians varies between sites, but generally image avail-
ability is more limited prior sampling 10 for pristine and restored sites,
and prior sampling 5 for drained sites. These latter samplings always
include S2, unlike older samplings that rely solely to L5–8 (Fig. 2).
2.3. Statistical analyses
We determined observed floristic gradients with non-metric multi-
dimensional scaling (NMDS; Kruskal, 1964) which produces continuous
and interpretable gradients that can be visualised in a multidimensional
space (e.g. Harris et al., 2015). We calculated the NMDS axes from the
site-averaged vegetation %-cover data containing all peatland types,
thus describing the broad-scale vegetation patterns. We applied Wis-
consin double standardization and square root transformation to the
data, calculated the distance matrix with Bray-Curtis distances (Bray
and Curtis, 1957) and tested the stress level with 20 random starts.
Table 1
Used remote sensing bands, vegetation indices, and wetness indices with equations. Justifications for bands and indices are in Table S1.
Variable Abbreviation Equation Reference
Blue reflectance Blue
Green reflectance Green
Red reflectance Red
Near-infrared reflectance NIR
Shortwave infrared band 1
reflectance
SWIR1
Shortwave infrared band 2
reflectance
SWIR2
Shortwave infrared transformed
reflectance
STR …1  SWIR1†2
2*SWIR1
Sadeghi et al., 2015
Normalised Difference Vegetation
Index
NDVI NIR  Red
NIR‡ Red
Tucker, 1979
Enhanced Vegetation Index EVI
2:5*
NIR  Red
NIR‡ 6*Red  7:5*Blue‡ 1
Liu and Huete, 1995
Soil Adjusted Vegetation Index SAVI
1:5*
NIR  Red
NIR‡ Red‡ 0:5
Huete, 1988
Tasseled cap Greennessa TCGreenness Blue*…  0:2941† ‡ Green*…  0:243†‡ Red*…  0:5424†‡ NIR*0:7276‡ SWIR1*
0:0713‡ SWIR2*…  0:1608†
Kauth and Thomas, 1976; Crist and
Cicone, 1984
Normalised Difference Water
Index
NDWI Green  NIR
Green‡ NIR McFeeters, 1996
Modified Normalised Difference
Water Index
MNDWI Green  SWIR2
Green‡ SWIR2 Xu, 2006
Normalised Difference Moisture
Index
NDMI NIR  SWIR1
NIR‡ SWIR1 Gao, 1996
Normalised Difference Moisture
Index 2
NDMI2 NIR  SWIR2
NIR‡ SWIR2 Gao, 1996
Tasseled cap Wetnessa TCWetness Blue*0:1511‡ Green*0:1973‡ Red*0:3283‡ NIR*0:3407‡ SWIR1*…  0:7117†‡
SWIR2*…  0:4559†
Kauth and Thomas, 1976; Crist and
Cicone, 1984
Tasseled cap Angleb TCAngle
arctan

TCGreenness
TCBrightness

Powell et al., 2010
a Landsat OLI tasseled cap coefficients (Baig et al., 2014) are used because the data are harmonised to the OLI sensor.
b TCBrightness ˆ Blue * 0.3029 ‡ Green * 0.2786 ‡ Red * 0.4733 ‡ NIR * 0.5599 ‡ SWIR1 * 0.508 ‡ SWIR2 * 0.1872.
Table 2
Used years for calculating remote sensing sampling medians from annual
medians.
Vegetation
sampling
Remote sensing data utilised (difference from the years of
vegetation sampling)
0-yeara  1,  2,  3,  4,  5
2-year  1, 0, ‡1
5-year  1, 0, ‡1
10-year  1, 0, ‡1
a We excluded the sampling year in the 0-year sampling calculations as
sometimes the actual restoration month is unknown. Inclusion of 0-year could
have resulted in post-restoration observations to be included if vegetation
sampling was conducted in early summer and restoration conducted in
midsummer of the same year.
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
4 
Given that a two-dimensional ordination resulted in an acceptable stress
level (~0.14) while being ecologically meaningful (Figs. S1, S2), we
refrained from adding more dimensions, considering the challenges in
visualising or interpreting three or more dimensions. First floristic
gradient (NMDS1) represented wetness gradient and second floristic
gradient (NMDS2) was connected to productivity of the sites (Figs. S1,
S2).
To predict the floristic gradients with RS data, we used random forest
regression (Breiman, 2001). We utilised all the RS variables (Table 1) as
predictor variables, incorporating data from both early and midsummer
periods, while the floristic gradients (NMDS axes) were the response
variables. The simultaneous use of both periods is based on the fact that
multitemporal imagery can enhance the predictive accuracy of vegeta-
tion community models (Dudley et al., 2015; Rasanen et al., 2020), and
this also produced more accurate models based on our preliminary tests.
The modelling was conducted separately for spruce mire forests, pine
mire forests and open mires to better capture variation within peatland
type specific treatments. We did not use site productivity information for
further peatland type specification as this would have halved the peat-
land type specific data and made the models less stable. We opted for
default random forest parameter settings as hyperparameter tuning has
led only to minor improvements in RS applications (Belgiu and Dragut¸,
2016; Probst et al., 2019); thus, we set the number of trees to 500 and
the number of variables tested at each tree node to 1/3 of available
variables. Before implementing the final models, we conducted peatland
type specific (i.e. spruce mire forests, pine mire forests, open mires)
automated variable selection with Variable Selection Using Random
Forests (VSURF; Genuer et al., 2015). This step was taken to determine
the most suitable RS variables for modelling, to avoid overfitting, and to
boost the regression performance. VSURF utilises a three-step procedure
(thresholding, interpretation, prediction) with nested random forest
models and variable importance metrics to remove irrelevant and
redundant variables from the dataset. We used default parameters of
VSURF with the exception that we increased number of forests in each
step from the default value of 10 to 20 to further eliminate randomness
in the variable selection.
We assessed model performance separately for spruce mire forests,
pine mire forests and open mires to evaluate the differences in the model
performance between peatland types. This evaluation utilised out-of-bag
estimation (approximately 1/3 of data is left for validation when
building each tree), which provides a conservative estimate of model fit
in RS applications (Clark et al., 2010). We calculated the % explained
variance (Pseudo R2 ˆ 1  mean square error / variance of response
variable) and normalised root mean square error (nRMSE; RMSE divided
by the range of response variable values) to evaluate model perfor-
mances. We repeated the models 100 times and calculated the mean of
evaluation metrics for the full variable set and the VSURF determined
subset to see if variable selection improved the performance. We also
extracted variable importance from the subset models utilising increase
in mean square error statistic for each model iteration separately,
summing them, and normalising the results between 0 and 100 %.
We separately assessed temporal changes in the observed and
Fig. 2. Number of observation dates per peatland type and treatment used for calculation of sampling medians. For some observation dates, there are multiple
satellite images available. In these cases, we calculated the mean value of the observations. Early summer refers to 1 May to 15 June and Midsummer to 1 July to 15
August. Drained sites have more observation dates in sampling 5 because all of them were established in 2012, meaning that they always include Sentinel-2 images
from 2017 and 2018. In contrast, most of the pristine and restored sites rely solely on Landsat 7 and 8 for the same sampling.
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
5 
predicted NMDS axes to determine if restoration effect could be deter-
mined from the axes. Here, we calculated peatland type and treatment
specific averages over samplings 0 and 10, separately for observed (i.e.,
based on vegetation inventory data) and RS predicted (i.e., based on
random forest modelling) gradients. These were plotted in two-
dimensional space to show development of the vegetation, and to see
if predicted changes were consistent with observed ones (Fig. 3). We
conducted all statistical analyses with R version 4.2.1 using packages
vegan (Oksanen et al., 2022), randomForest (Liaw and Wiener, 2022),
and VSURF (Genuer et al., 2015).
3. Results
3.1. Peatland type-specific variable importance
The most important variables based on VSURF differed between
peatland types (Fig. 4). For spruce mire forests, the most important
variables were early summer NDWI (NMDS1) and midsummer NIR
(NMDS2). For pine mire forests, the most important variables included
midsummer red (NMDS1) and midsummer NIR (NMDS2). For open
mires, the most important variables were partly the same as for pine
mire forests. Midsummer red was the most important for NMDS1 and
two vegetation indices (early summer SAVI, TCGreenneess) were the
two most important variables before midsummer NIR for NMDS2.
Importantly, variables from both early and midsummer were found to be
important for all peatland types and some variables, such as midsummer
NIR and red, were consistently among the most important variables in
all peatland types. The number of chosen variables was the highest for
open mires (NMDS1 ˆ 10; NMDS2 ˆ 9), followed by pine (NMDS1 ˆ 5;
NMDS2 ˆ 10) and spruce mire forests (NMDS1 ˆ 4; NMDS2 ˆ 5).
3.2. Peatland type-specific modelling performance
The constructed random forest regression models could predict
floristic gradients (i.e., NMDS axes) with reasonable accuracy (%var
explained ˆ 28.62–69.44, nRMSE ˆ 11.15–18.38 %; Table 3, Fig. 5).
However, performance varied across peatland types. The highest overall
performance was for open mires and the lowest for spruce mire forests.
Fig. 3. Methodological flow chart. Spectral indices are listed in Table 1.
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
6 
Additionally, in open mires and pine mire forests, NMDS1 predictions
had remarkably higher accuracy compared to NMDS2, but in spruce
mire forests, both axes were predicted with a similar, quite low accu-
racy. Finally, variable selection improved the model performance in
every NMDS axis.
3.3. Observed and predicted temporal changes in floristic gradients
The average of peatland type and treatment specific NMDS-axes
demonstrated that the overall temporal behaviour was rather similar
between the observed and RS predicted axes in pine mire forests and
open mires, especially in restoration treatment (Fig. 6). However, in
spruce mire forests, there was opposite restoration development in the
RS predicted gradients compared to the observed ones. Additionally, the
impact of restoration was mostly observed in NMDS1 for open mires
while in pine mire and spruce mire forests, the impact was more
apparent in NMDS2.
4. Discussion
4.1. Importance of peatland type-specific remote sensing variables
Our results indicate that vegetation in different peatland types,
namely boreal spruce mire forests, pine mire forests and open mires,
should be monitored with different satellite-based optical variables
(Fig. 4). This is in line with previous studies (e.g. Rasanen et al., 2020;
Steenvoorden et al., 2023) which have shown the differences in the
importance of RS variables in predicting vegetation in different peat-
lands. However, our result is based on larger and more heterogeneous
set of peatland sites than the previous studies, which makes our results
more generalizable. Differences in important variables are probably
caused by divergent vegetation composition in the different peatland
types. Moreover, in densely treed spruce mire forests, satellites are
mostly observing tree cover and shadows, while in open mires, ground
vegetation is mostly observed. In pine mire forests, the RS signature is
composed of trees, shadows and ground vegetation. Additionally, in
hydrological monitoring of peatlands, the importance of variables and
functionality of one variable differ between the sites and peatland types
(Rasanen et al., 2022; Burdun et al., 2020, 2023) which further high-
lights the site and type specificity in RS studies.
Interestingly, our results suggest that SWIR bands are not important
for predicting changes in vegetation patterns even though they have
played a key role in monitoring peatland surface moisture and water
table (Burdun et al., 2020; Rasanen et al., 2022; Burdun et al., 2023;
Jussila et al., 2023; Isoaho et al., 2024). In our results, NIR band was
important in every peatland type, which is in line with earlier studies
depicting peatland vegetation and land cover (Middleton et al., 2012;
Harris et al., 2015; Rasanen et al., 2020; Kolari et al., 2022; Steen-
voorden and Limpens, 2023). NIR’s ability to separate e.g. different
Sphagnum species (Salko et al., 2023) might be one of the reasons behind
Fig. 4. Mean relative variable importance over 100 model fits for NMDS axes of spruce mire forests, pine mire forests and open mires. Abbreviations are explained in
Table 1 and suffixes “_es” and “_ms” refer to early summer and midsummer, respectively. The models for calculating variable importance were run with the variables
that were chosen after the VSURF prediction step.
Table 3
Mean percentage of variance explained (%var) and normalised root mean square error (nRMSE) of 100 fitted models for both axes. Full set refers to models constructed
before variable selection and subset refers to models constructed with the variables selected by VSURF algorithm.
Type Axis %var full set %var subset nRMSE (%) full set nRMSE (%) subset
Spruce mire forests NMDS1 25.38 28.62 18.33 17.99
Spruce mire forests NMDS2 28.37 36.27 19.48 18.38
Pine mire forests NMDS1 51.59 52.86 18.23 17.99
Pine mire forests NMDS2 30.02 33.10 17.19 16.81
Open mires NMDS1 69.02 69.44 11.23 11.15
Open mires NMDS2 34.10 37.26 17.11 16.70
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
7 
its importance. Furthermore, the importance of red band in peatland
wetness prediction has been lately emphasized (Kolari et al., 2022;
Isoaho et al., 2023). In our analysis, it is an important variable for
predicting the wetness-related NMDS1 axis in pine mire forests and open
mires (Figs. S1, S2). Nevertheless, there seems to be no one-size-fits-all
solution but on average NIR and red band seemed to be rather impor-
tant across the peatland types and floristic gradients.
As the most important variables include both early and midsummer
observations, our results highlight the importance of multitemporal RS
data that covers phenological and seasonal variation within the sites (e.
g. Dudley et al., 2015; Halabisky et al., 2018; Rasanen et al., 2020). Early
summer imagery shows the situation during spring floods and emerging
vegetation (Rasanen et al., 2020) while midsummer imagery represents
the vegetation phenological peak (Juutinen et al., 2017; Pang et al.,
Fig. 5. Scatterplots between predicted (mean over 100 fits) and observed NMDS axes values, dashed line indicates 1:1 line.
Fig. 6. Two-dimensional plot of observed and RS predicted floristic gradients with NMDS1 in x-axis and NMDS2 in y-axis. Points indicate average NMDS-values of
peatland type and treatment over 0- and 10-year samplings. Arrow start from sample 0 and end in sample 10 showing the average development of the gradients over
time. Observed NMDS development is very similar to Elo et al. (2024b) who used model-based ordination for a similar analysis with the same field data.
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
8 
2023). The importance of multitemporal approach was also a key
finding of recent study (Christiani et al., 2024) which successfully used
early and midsummer time windows identical to ours for detecting
peatland greenhouse gas sinks and sources of entire Finland with sat-
ellite imagery.
Additional RS and environmental datasets could further improve the
floristic gradient predictions in the future studies. For instance, peatland
vegetation predictions can be improved with high-resolution topo-
graphical data (Beyer et al., 2019; Harris and Baird, 2019; Rasanen
et al., 2020; Korpela et al., 2020; Kaneko et al., 2024) or radar satellite
imagery (Wijaya et al., 2010; Merchant et al., 2017; Khakim et al.,
2020). In addition to RS variables, also other field-observation based
variables, such as water table, can be used for explaining the floristic
gradients (Pellerin et al., 2009; Maanavilja et al., 2014). The main
challenge with the multisource approach lies in data availability. In our
study, some sites have been monitored since 2007 and only optical
satellite data is consistently available for the whole monitoring period.
Topographic datasets are available only for specific time points and not
consistently for pre- and post-restoration situations while radar imagery
such as Sentinel-1 or ENVISAT offer limited temporal coverage of our
monitoring period. Moreover, hydrological monitoring of the peatland
monitoring network has been conducted only at a subset of sites
(Rasanen et al., 2022).
4.2. Monitoring changes in peatland vegetation with remote sensing
Our results show that temporal changes in peatland vegetation can
be monitored with optical satellite imagery. We have demonstrated that
floristic gradients, which reflect community-level vegetation patterns
and which are in our case quantified with NMDS axes, are predictable
with optical satellite imagery in pine mire forests and open mires but not
as validly in densely treed spruce mire forests. Therefore, in-line with
the previous studies (Rasanen et al., 2022; Pang et al., 2023; Burdun
et al., 2023), high tree cover seems to be an insurmountable obstacle for
optical sensors. In sites where the ground vegetation is mostly hidden
under tree canopy in coarser spatial resolution data, fine resolution
sensor data could be tested in the future studies.
While long-term changes in microforms and habitat types have been
previously monitored with RS techniques (Steenvoorden et al., 2022;
Kolari et al., 2022), to the best of our knowledge, our results provide the
first more frequent link between post-restoration vegetation changes
and RS data in boreal peatlands. Previous studies havemostly focused on
e.g. changes in peatlands’ hydrological properties with optical data
(Rasanen et al., 2022; Burdun et al., 2023; Isoaho et al., 2024) or on the
spectral properties of restored and pristine sites (Ball et al., 2023). While
hydrological changes can be drastic and quick (Menberu et al., 2016,
2017, 2018), changes in vegetation are slower (Elo et al., 2024b). This
layout is difficult because drastic hydrological changes can be seen with
RS data (Isoaho et al., 2024), even if changes in vegetation composition
have presumably not occurred. Similarly, changes in spectral informa-
tion are caused by tree cutting which could explain changes in restored
treatments of pine mire forests and open mires. This raises the question
of how much of the predicted change is actually due to changes in
vegetation and how much is due to the other factors such as increased
wetness and change in tree cover.
We used 10 1 m2 vegetation plots for constructing training and
validation data at each site. The plots were located within 15 m radius
circular area from which we extracted satellite data. In a way, each site
in our data resembles plots in the previous studies that have utilised
higher spatial resolution RS data for one or few sites (Harris et al., 2015;
Rasanen et al., 2020). We acknowledge that the 10 m2 describes only
around 1.5% of the 15m radius area causing a possible spatial mismatch
that can affect the modelling performance. Because peatlands are
spatially heterogenous (Rasanen and Virtanen, 2019; Steenvoorden and
Limpens, 2023), it is possible that the vegetation sample does not
describe the area fully. Additionally, particularly spruce mire forests
have a high variation in species composition between plots of a single
site and between the sites (Elo et al., 2024a, 2024b). This furthermore
challenges our approach in spruce mire forests as NMDS may not be able
to capture the within-site heterogeneity which hampers our ability to
model them with satellite imagery.
One of the restoration targets is to recover target species (Haapalehto
et al., 2011) which often are used as indicators of restoration success
(Kyrkjeeide et al., 2024). However, species-level predictions in heter-
ogenous areas with environmental or RS data are very difficult to
implement (e.g. Maanavilja et al., 2014; Saarimaa et al., 2019; Simpson
et al., 2024). Bayesian joint species distribution modelling (Ovaskainen
et al., 2017) provides a potential solution to find relationships between
individual species and environmental variables derived e.g. from RS
data (e.g. Palmroos et al., 2023). Therefore, merging optical and radar
RS variables, as well as other relevant environmental variables with
vegetation data and conducting Bayesian modelling could be a future
research avenue.
Vegetation communities in the Finnish peatland monitoring network
sites have not fully recovered in 10 years (Elo et al., 2024b), which can
also be seen with our RS-based approach (Fig. 6). Additionally there is a
large variation in restoration effect (Elo et al., 2024b), but on average, a
shift towards pristine-like vegetation has been achieved. We emphasize
the need for the longer field monitoring while recognising that RS has
potential to complement the field work. Future work should develop
indicators for peatland restoration that are ecologically meaningful and
detectable with RS.
Also other types of northern peatlands should be included into
restoration monitoring schemes. These include e.g. rich fens, treeless
eutrophic peatlands with unique vegetation, and flark fens, wet and
treeless boreal peatlands with wet flarks altering with drier strings.
Importantly, the highest areal need for peatland restoration in Finland is
in large aapa mire complexes (Rasanen et al., 2023), and flark fens
within the northern aapa mires are in transition from fens to bogs
(Sallinen et al., 2019; Granlund et al., 2022; Kolari et al., 2022; Kolari
and Tahvanainen, 2023). To conduct large-scale monitoring of the flark
fen vegetation changes, there is a need to develop RS-based monitoring
methods but also field-based calibration data is needed. Nevertheless,
the developed approach could also be tested in sites lacking field-based
monitoring. As different peatland types should be modelled with
different RS variables (Fig. 4) and as the approach functions the best in
peatlands with few or no trees, the approach requires existing knowl-
edge about the peatland type for which there is geospatial information
available for several countries (e.g. Minasny et al., 2023; Pontone et al.,
2024). Additionally, visual interpretation of high-resolution satellite
imagery or aerial imagery also enables definition of coarse level iden-
tification of peatland types when needed.
4.3. Implications for management
Our study suggest that vegetation of restored and pristine sparsely
treed boreal peatlands can be monitored with optical satellite imagery,
particularly by utilising spectral regions of red and NIR reflectance.
However, there is a high variability between types in the modelling
performance and in the variables that are sensitive to vegetation
changes. Therefore, comprehensive peatland type or even site-specific
field inventories are required when RS is used for predicting detailed
changes in peatland vegetation. In other words, RS can be used to sup-
plement and upscale field inventory-based observations of peatland
plant community change. While we did not explore possibilities for
detecting spatial patterns of vegetation, the floristic gradients can be
spatially predicted if there is spatially comprehensive field inventory
based training data available (see e.g. Harris et al., 2015). Spatial as-
sessments using RS datasets are important because the spatial impacts
on restoration are not uniform, as shown in relation to the hydrological
impacts (Isoaho et al., 2024) and as field inventory data cover only a
small proportion of the peatland sites under interest. Finally, current
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
9 
monitoring schemes should be continued to further capture the longer-
term restoration impacts because currently, the sites have not fully
returned to their pristine-like state.
5. Conclusion
We demonstrated that optical satellite imagery can be used for large-
scale monitoring of floristic gradients and post-restoration vegetation
changes in boreal peatlands. The most important explanatory RS vari-
ables consist of NIR and red band captured during early and midsummer
periods. However, the most important RS variable combinations differ
between peatland types. Therefore, peatland-type specific regression
models are necessary for effectively monitoring the changes in peatland
vegetation. Our results also show that vegetation changes in open mires
and sparsely treed pine mire forests can be modelled with reasonable
accuracy, whereas the explanatory capacity in densely treed spruce mire
forests is lower. Future work could explore the use of RS data to monitor
species-level changes and develop RS-based ecological indicators for
peatland restoration monitoring. Overall, we suggest that optical satel-
lite imagery can serve as a proxy for assessing the vegetation response to
peatland restoration in peatlands with little or no trees.
CRediT authorship contribution statement
Aleksi Isoaho: Writing – original draft, Validation, Software,
Methodology, Investigation, Formal analysis, Conceptualization. Merja
Elo:Writing – review& editing, Investigation, Funding acquisition, Data
curation. Hannu Marttila: Writing – review & editing, Supervision,
Funding acquisition. Parvez Rana: Writing – review & editing. Anssi
Lensu:Writing – review& editing. Aleksi Rasanen:Writing – review &
editing, Supervision, Project administration, Investigation, Funding
acquisition, Conceptualization.
Declaration of competing interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
Acknowledgements
This study was funded by the Ministry of Environment, Finland (VN/
14352/2022). We thank numerous individuals for their contribution to
the vegetation inventories.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.scitotenv.2024.177697.
Data availability
Data will be made available on request.
References
Andersen, R., Farrell, C., Graf, M., Muller, F., Calvar, E., Frankard, P., Caporn, S.,
Anderson, P., 2017. An overview of the progress and challenges of peatland
restoration in Western Europe: peatland restoration in Western Europe. Restor. Ecol.
25, 271–282. https://doi.org/10.1111/rec.12415.
Anderson, A.J.B., 1971. Ordination methods in ecology. J. Ecol. 59, 713. https://doi.org/
10.2307/2258135.
Baig, M.H.A., Zhang, L., Shuai, T., Tong, Q., 2014. Derivation of a tasselled cap
transformation based on Landsat 8 at-satellite reflectance. Remote Sens. Lett. 5,
423–431. https://doi.org/10.1080/2150704X.2014.915434.
Ball, J., Gimona, A., Cowie, N., Hancock, M., Klein, D., Donaldson-Selby, G., Artz, R.R.E.,
2023. Assessing the potential of using Sentinel-1 and 2 or high-resolution aerial
imagery data with machine learning and data science techniques to model peatland
restoration progress – a northern Scotland case study. Int. J. Remote Sens. 44,
2885–2911. https://doi.org/10.1080/01431161.2023.2209916.
Belgiu, M., Dragut¸, L., 2016. Random forest in remote sensing: a review of applications
and future directions. ISPRS J. Photogramm. Remote Sens. 114, 24–31. https://doi.
org/10.1016/j.isprsjprs.2016.01.011.
Bergamo, T.F., De Lima, R.S., Kull, T., Ward, R.D., Sepp, K., Villoslada, M., 2023. From
UAV to PlanetScope: upscaling fractional cover of an invasive species Rosa rugosa.
J. Environ. Manag. 336, 117693. https://doi.org/10.1016/j.jenvman.2023.117693.
Beyer, F., Jurasinski, G., Couwenberg, J., Grenzdorffer, G., 2019. Multisensor data to
derive peatland vegetation communities using a fixed-wing unmanned aerial
vehicle. Int. J. Remote Sens. 40, 9103–9125. https://doi.org/10.1080/
01431161.2019.1580825.
Bhatnagar, S., Gill, L., Regan, S., Naughton, O., Johnston, P., Waldren, S., Ghosh, B.,
2020. Mapping vegetation communities inside wetlands using Sentinel-2 imagery in
Ireland. Int. J. Appl. Earth Obs. Geoinf. 88, 102083. https://doi.org/10.1016/j.
jag.2020.102083.
Bonn, A., Reed, M.S., Evans, C.D., Joosten, H., Bain, C., Farmer, J., Emmer, I.,
Couwenberg, J., Moxey, A., Artz, R., Tanneberger, F., Von Unger, M., Smyth, M.-A.,
Birnie, D., 2014. Investing in nature: developing ecosystem service markets for
peatland restoration. Ecosyst. Serv. 9, 54–65. https://doi.org/10.1016/j.
ecoser.2014.06.011.
Bray, J.R., Curtis, J.T., 1957. An ordination of the upland forest communities of southern
Wisconsin. Ecol. Monogr. 27, 325–349. https://doi.org/10.2307/1942268.
Breiman, L., 2001. Random forests. Mach. Learn. 45, 5–32. https://doi.org/10.1023/A:
1010933404324.
Burdun, I., Bechtold, M., Sagris, V., Komisarenko, V., De Lannoy, G., Mander, Ü., 2020.
A comparison of three trapezoid models using optical and thermal satellite imagery
for water table depth monitoring in Estonian Bogs. Remote Sens. 12, 1980. https://
doi.org/10.3390/rs12121980.
Burdun, I., Bechtold, M., Aurela, M., De Lannoy, G., Desai, A.R., Humphreys, E.,
Kareksela, S., Komisarenko, V., Liimatainen, M., Marttila, H., Minkkinen, K.,
Nilsson, M.B., Ojanen, P., Salko, S.-S., Tuittila, E.-S., Uuemaa, E., Rautiainen, M.,
2023. Hidden becomes clear: optical remote sensing of vegetation reveals water
table dynamics in northern peatlands. Remote Sens. Environ. 296, 113736. https://
doi.org/10.1016/j.rse.2023.113736.
Chapman, S., Buttler, A., Francez, A.-J., Laggoun-Defarge, F., Vasander, H., Schloter, M.,
Combe, J., Grosvernier, P., Harms, H., Epron, D., Gilbert, D., Mitchell, E., 2003.
Exploitation of northern peatlands and biodiversity maintenance: a conflict between
economy and ecology. Front. Ecol. Environ. 1, 525–532. https://doi.org/10.1890/
1540-9295(2003)001[0525:EONPAB]2.0.CO;2.
Christiani, P., Rana, P., Rasanen, A., Pitkanen, T.P., Tolvanen, A., 2024. Detecting spatial
patterns of peatland greenhouse gas sinks and sources with geospatial environmental
and remote sensing data. Environ. Manag. https://doi.org/10.1007/s00267-024-
01965-7.
Christie, A.P., Abecasis, D., Adjeroud, M., Alonso, J.C., Amano, T., Anton, A., Baldigo, B.
P., Barrientos, R., Bicknell, J.E., Buhl, D.A., Cebrian, J., Ceia, R.S., Cibils-Martina, L.,
Clarke, S., Claudet, J., Craig, M.D., Davoult, D., De Backer, A., Donovan, M.K.,
Eddy, T.D., França, F.M., Gardner, J.P.A., Harris, B.P., Huusko, A., Jones, I.L.,
Kelaher, B.P., Kotiaho, J.S., Lopez-Baucells, A., Major, H.L., Maki-Petays, A.,
Martín, B., Martín, C.A., Martin, P.A., Mateos-Molina, D., McConnaughey, R.A.,
Meroni, M., Meyer, C.F.J., Mills, K., Montefalcone, M., Noreika, N., Palacín, C.,
Pande, A., Pitcher, C.R., Ponce, C., Rinella, M., Rocha, R., Ruiz-Delgado, M.C.,
Schmitter-Soto, J.J., Shaffer, J.A., Sharma, S., Sher, A.A., Stagnol, D., Stanley, T.R.,
Stokesbury, K.D.E., Torres, A., Tully, O., Vehanen, T., Watts, C., Zhao, Q.,
Sutherland, W.J., 2020. Quantifying and addressing the prevalence and bias of study
designs in the environmental and social sciences. Nat. Commun. 11, 6377. https://
doi.org/10.1038/s41467-020-20142-y.
Clark, M.L., Aide, T.M., Grau, H.R., Riner, G., 2010. A scalable approach to mapping
annual land cover at 250 m using MODIS time series data: a case study in the Dry
Chaco ecoregion of South America. Remote Sens. Environ. 114, 2816–2832. https://
doi.org/10.1016/j.rse.2010.07.001.
Crist, E.P., Cicone, R.C., 1984. A physically-based transformation of thematic mapper
data—the TM Tasseled Cap. IEEE Trans. Geosci. Remote Sens. GE-22, 256–263.
https://doi.org/10.1109/TGRS.1984.350619.
Dudley, K.L., Dennison, P.E., Roth, K.L., Roberts, D.A., Coates, A.R., 2015. A multi-
temporal spectral library approach for mapping vegetation species across spatial and
temporal phenological gradients. Remote Sens. Environ. 167, 121–134. https://doi.
org/10.1016/j.rse.2015.05.004.
Elo, M., Kareksela, S., Ovaskainen, O., Abrego, N., Niku, J., Taskinen, S., Aapala, K.,
Kotiaho, J., 2024a. Data For: A Large-scale and Long-term Experiment to Identify
Effectiveness of Ecosystem Restoration. https://doi.org/10.5281/
ZENODO.10906942.
Elo, M., Kareksela, S., Ovaskainen, O., Abrego, N., Niku, J., Taskinen, S., Aapala, K.,
Kotiaho, J.S., 2024b. A Large-scale and Long-term Experiment to Identify
Effectiveness of Ecosystem Restoration. https://doi.org/10.1101/
2024.04.02.587693.
European Commission, 2020. Peatlands for LIFE.
Gao, B., 1996. NDWI—a normalized difference water index for remote sensing of
vegetation liquid water from space. Remote Sens. Environ. 58, 257–266. https://doi.
org/10.1016/S0034-4257(96)00067-3.
Garisoain, R., Delire, C., Decharme, B., Ferrant, S., Granouillac, F., Payre-Suc, V.,
Gandois, L., 2023. A study of dominant vegetation phenology in a Sphagnum
mountain peatland using in situ and Sentinel-2 observations. J. Geophys. Res.
Biogeosci. 128, e2023JG007403. https://doi.org/10.1029/2023JG007403.
Genuer, R., Poggi, J.-M., Tuleau-Malot, C., 2015. VSURF: an R package for variable
selection using random forests. R J. 7, 19–33.
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
10 
Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R., 2017.
Google Earth Engine: planetary-scale geospatial analysis for everyone. Remote Sens.
Environ. 202, 18–27. https://doi.org/10.1016/j.rse.2017.06.031.
Granlund, L., Vesakoski, V., Sallinen, A., Kolari, T.H.M., Wolff, F., Tahvanainen, T.,
2022. Recent lateral expansion of sphagnum bogs over central fen areas of boreal
aapa mire complexes. Ecosystems 25, 1455–1475. https://doi.org/10.1007/s10021-
021-00726-5.
Haapalehto, T., Vasander, H., Jauhiainen, S., Tahvanainen, T., Kotiaho, J.S., 2011. The
effects of peatland restoration on water-table depth, elemental concentrations, and
vegetation: 10 years of changes. Restor. Ecol. 19, 587–598. https://doi.org/
10.1111/j.1526-100X.2010.00704.x.
Haapalehto, T., Kotiaho, J.S., Matilainen, R., Tahvanainen, T., 2014. The effects of long-
term drainage and subsequent restoration on water table level and pore water
chemistry in boreal peatlands. J. Hydrol. 519, 1493–1505. https://doi.org/10.1016/
j.jhydrol.2014.09.013.
Haapalehto, T., Juutinen, R., Kareksela, S., Kuitunen, M., Tahvanainen, T., Vuori, H.,
Kotiaho, J.S., 2017. Recovery of plant communities after ecological restoration of
forestry-drained peatlands. Ecol. Evol. 7, 7848–7858. https://doi.org/10.1002/
ece3.3243.
Halabisky, M., Babcock, C., Moskal, L., 2018. Harnessing the temporal dimension to
improve object-based image analysis classification of wetlands. Remote Sens. 10,
1467. https://doi.org/10.3390/rs10091467.
Harris, A., Baird, A.J., 2019. Microtopographic drivers of vegetation patterning in
blanket peatlands recovering from erosion. Ecosystems 22, 1035–1054. https://doi.
org/10.1007/s10021-018-0321-6.
Harris, A., Charnock, R., Lucas, R.M., 2015. Hyperspectral remote sensing of peatland
floristic gradients. Remote Sens. Environ. 162, 99–111. https://doi.org/10.1016/j.
rse.2015.01.029.
Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 25,
295–309. https://doi.org/10.1016/0034-4257(88)90106-X.
Husson, E., Hagner, O., Ecke, F., 2014. Unmanned aircraft systems help to map aquatic
vegetation. Appl. Veg. Sci. 17, 567–577. https://doi.org/10.1111/avsc.12072.
Ikkala, L., Ronkanen, A.-K., Ilmonen, J., Simila, M., Rehell, S., Kumpula, T., Pakkila, L.,
Klove, B., Marttila, H., 2022. Unmanned aircraft system (UAS) structure-from-
motion (SfM) for monitoring the changed flow paths and wetness in minerotrophic
peatland restoration. Remote Sens. 14, 3169. https://doi.org/10.3390/rs14133169.
Isoaho, A., Ikkala, L., Marttila, H., Hjort, J., Kumpula, T., Korpelainen, P., Rasanen, A.,
2023. Spatial water table level modelling with multi-sensor unmanned aerial vehicle
data in boreal aapa mires. Remote Sens. Appl. Soc. Environ. 32, 101059. https://doi.
org/10.1016/j.rsase.2023.101059.
Isoaho, A., Ikkala, L., Pakkila, L., Marttila, H., Kareksela, S., Rasanen, A., 2024. Multi-
sensor satellite imagery reveals spatiotemporal changes in peatland water table after
restoration. Remote Sens. Environ. 306, 114144. https://doi.org/10.1016/j.
rse.2024.114144.
Joosten, H., Sirin, A., Couwenberg, J., Laine, J., Smith, P., 2016. The role of peatlands in
climate regulation. In: Bonn, A., Allott, T., Evans, M., Joosten, H., Stoneman, R.
(Eds.), Peatland Restoration and Ecosystem Services. Cambridge University Press,
Cambridge, pp. 63–76. https://doi.org/10.1017/CBO9781139177788.005.
Jussila, T., Heikkinen, R.K., Anttila, S., Aapala, K., Kervinen, M., Aalto, J., Vihervaara, P.,
2023. Quantifying wetness variability in aapa mires with Sentinel-2: towards
improved monitoring of an EU priority habitat. Remote Sens. Ecol. Conserv.,
rse2.363 https://doi.org/10.1002/rse2.363.
Juutinen, S., Virtanen, T., Kondratyev, V., Laurila, T., Linkosalmi, M., Mikola, J.,
Nyman, J., Rasanen, A., Tuovinen, J.-P., Aurela, M., 2017. Spatial variation and
seasonal dynamics of leaf-area index in the arctic tundra-implications for linking
ground observations and satellite images. Environ. Res. Lett. 12, 095002. https://
doi.org/10.1088/1748-9326/aa7f85.
Kalacska, M., Arroyo-Mora, J., De Gea, J., Snirer, E., Herzog, C., Moore, T., 2013.
Videographic analysis of Eriophorum vaginatum spatial coverage in an ombotrophic
bog. Remote Sens. 5, 6501–6512. https://doi.org/10.3390/rs5126501.
Kaneko, K., Yokochi, M., Inoue, T., Kato, Y., Fujita, H., 2024. Topographic conditions as
governing factors of mire vegetation types analyzed from drone-based terrain model.
J. Veg. Sci. 35, e13226. https://doi.org/10.1111/jvs.13226.
Kareksela, S., Haapalehto, T., Juutinen, R., Matilainen, R., Tahvanainen, T., Kotiaho, J.S.,
2015. Fighting carbon loss of degraded peatlands by jump-starting ecosystem
functioning with ecological restoration. Sci. Total Environ. 537, 268–276. https://
doi.org/10.1016/j.scitotenv.2015.07.094.
Kauth, R.J., Thomas, G.S., 1976. The tasselled cap—a graphic description of the spectral-
temporal development of agricultural crops as seen by Landsat. In: Proceedings of
the Symposium on Machine Processing of Remotely Sensed Data. Purdue University,
West Lafayette.
Khakim, M.Y.N., Bama, A.A., Yustian, I., Poerwono, P., Tsuji, T., Matsuoka, T., 2020.
Peatland subsidence and vegetation cover degradation as impacts of the 2015 El ni~no
event revealed by Sentinel-1A SAR data. Int. J. Appl. Earth Obs. Geoinf. 84, 101953.
https://doi.org/10.1016/j.jag.2019.101953.
Kolari, T.H.M., Tahvanainen, T., 2023. Inference of future bog succession trajectory from
spatial chronosequence of changing aapa mires. Ecol. Evol. 13, e9988. https://doi.
org/10.1002/ece3.9988.
Kolari, T.H.M., Sallinen, A., Wolff, F., Kumpula, T., Tolonen, K., Tahvanainen, T., 2022.
Ongoing Fen–Bog transition in a boreal aapa mire inferred from repeated field
sampling, aerial images, and landsat data. Ecosystems 25, 1166–1188. https://doi.
org/10.1007/s10021-021-00708-7.
Korpela, I., Haapanen, R., Korrensalo, A., Tuittila, E.-S., Vesala, T., 2020. Fine-resolution
mapping of microforms of a boreal bog using aerial images and waveform-recording
LiDAR. Mires Peat 26, 1–24. https://doi.org/10.19189/MaP.2018.OMB.388.
Kreyling, J., Tanneberger, F., Jansen, F., Van Der Linden, S., Aggenbach, C., Blüml, V.,
Couwenberg, J., Emsens, W.-J., Joosten, H., Klimkowska, A., Kotowski, W.,
Kozub, L., Lennartz, B., Liczner, Y., Liu, H., Michaelis, D., Oehmke, C.,
Parakenings, K., Pleyl, E., Poyda, A., Raabe, S., Rohl, M., Rücker, K., Schneider, A.,
Schrautzer, J., Schroder, C., Schug, F., Seeber, E., Thiel, F., Thiele, S., Tiemeyer, B.,
Timmermann, T., Urich, T., Van Diggelen, R., Vegelin, K., Verbruggen, E.,
Wilmking, M., Wrage-Monnig, N., Wołejko, L., Zak, D., Jurasinski, G., 2021.
Rewetting does not return drained fen peatlands to their old selves. Nat. Commun.
12, 5693. https://doi.org/10.1038/s41467-021-25619-y.
Kruskal, J.B., 1964. Nonmetric multidimensional scaling: a numerical method.
Psychometrika 29, 115–129. https://doi.org/10.1007/BF02289694.
Kyrkjeeide, M.O., Jokerud, M., Catriona Mehlhoop, A., Marie Foldnes Lunde, L.,
Fandrem, M., Lyngstad, A., 2024. Peatland restoration in Norway – evaluation of
ongoing monitoring and identification of plant indicators of restoration success.
Nord. J. Bot. 2024, e03988. https://doi.org/10.1111/njb.03988.
Laine, J., Laiho, R., Minkkinen, K., Vasander, H., 2006. Forestry and boreal peatlands. In:
Wielder, K., Vitt, D. (Eds.), Boreal Peatland Ecosystems, Ecological Studies. Springer
Berlin, Heidelberg.
Laine, A.M., Leppala, M., Tarvainen, O., Paatalo, M.-L., Seppanen, R., Tolvanen, A.,
2011. Restoration of managed pine fens: effect on hydrology and vegetation:
restoration of managed pine fens. Appl. Veg. Sci. 14, 340–349. https://doi.org/
10.1111/j.1654-109X.2011.01123.x.
Leifeld, J., Menichetti, L., 2018. The underappreciated potential of peatlands in global
climate change mitigation strategies. Nat. Commun. 9, 1071. https://doi.org/
10.1038/s41467-018-03406-6.
Liaw, A., Wiener, M., 2022. randomForest: Breiman and Cutler’s random forests for
classification and regression [WWW document]. URL. https://cran.r-project.or
g/web/packages/randomForest/index.html.
Liu, H.Q., Huete, A., 1995. A feedback based modification of the NDVI to minimize
canopy background and atmospheric noise. IEEE Trans. Geosci. Remote Sens. 33,
457–465. https://doi.org/10.1109/TGRS.1995.8746027.
Maanavilja, L., Aapala, K., Haapalehto, T., Kotiaho, J.S., Tuittila, E.-S., 2014. Impact of
drainage and hydrological restoration on vegetation structure in boreal spruce
swamp forests. For. Ecol. Manag. 330, 115–125. https://doi.org/10.1016/j.
foreco.2014.07.004.
Marttila, H., Lepisto, A., Tolvanen, A., Bechmann, M., Kyllmar, K., Juutinen, A.,
Wenng, H., Skarbøvik, E., Futter, M., Kortelainen, P., Rankinen, K., Hellsten, S.,
Kløve, B., Kronvang, B., Kaste, Ø., Solheim, A.L., Bhattacharjee, J., Rakovic, J., De
Wit, H., 2020. Potential impacts of a future Nordic bioeconomy on surface water
quality. Ambio 49, 1722–1735. https://doi.org/10.1007/s13280-020-01355-3.
McFeeters, S.K., 1996. The use of the Normalized Difference Water Index (NDWI) in the
delineation of open water features. Int. J. Remote Sens. 17, 1425–1432. https://doi.
org/10.1080/01431169608948714.
Menberu, M.W., Tahvanainen, T., Marttila, H., Irannezhad, M., Ronkanen, A.-K.,
Penttinen, J., Kløve, B., 2016. Water-table-dependent hydrological changes
following peatland forestry drainage and restoration: analysis of restoration success.
Water Resour. Res. 52, 3742–3760. https://doi.org/10.1002/2015WR018578.
Menberu, M.W., Marttila, H., Tahvanainen, T., Kotiaho, J.S., Hokkanen, R., Kløve, B.,
Ronkanen, A., 2017. Changes in pore water quality after peatland restoration:
assessment of a large-scale, replicated before-after-control-impact study in Finland.
Water Resour. Res. 53, 8327–8343. https://doi.org/10.1002/2017WR020630.
Menberu, M.W., Haghighi, A.T., Ronkanen, A., Marttila, H., Kløve, B., 2018. Effects of
drainage and subsequent restoration on peatland hydrological processes at
catchment scale. Water Resour. Res. 54, 4479–4497. https://doi.org/10.1029/
2017WR022362.
Merchant, M.A., Adams, J.R., Berg, A.A., Baltzer, J.L., Quinton, W.L., Chasmer, L.E.,
2017. Contributions of C-band SAR data and polarimetric decompositions to
subarctic boreal peatland mapping. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.
10, 1467–1482. https://doi.org/10.1109/JSTARS.2016.2621043.
Middleton, M., Narhi, P., Arkimaa, H., Hyvonen, E., Kuosmanen, V., Treitz, P.,
Sutinen, R., 2012. Ordination and hyperspectral remote sensing approach to classify
peatland biotopes along soil moisture and fertility gradients. Remote Sens. Environ.
124, 596–609. https://doi.org/10.1016/j.rse.2012.06.010.
Minasny, B., Adetsu, D.V., Aitkenhead, M., Artz, R.R.E., Baggaley, N., Barthelmes, A.,
Beucher, A., Caron, J., Conchedda, G., Connolly, J., Deragon, R., Evans, C.,
Fadnes, K., Fiantis, D., Gagkas, Z., Gilet, L., Gimona, A., Glatzel, S., Greve, M.H.,
Habib, W., Hergoualc’h, K., Hermansen, C., Kidd, D.B., Koganti, T., Kopansky, D.,
Large, D.J., Larmola, T., Lilly, A., Liu, H., Marcus, M., Middleton, M., Morrison, K.,
Petersen, R.J., Quaife, T., Rochefort, L., Rudiyanto, Toca, L., Tubiello, F.N.,
Weber, P.L., Weldon, S., Widyatmanti, W., Williamson, J., Zak, D., 2023. Mapping
and monitoring peatland conditions from global to field scale. Biogeochemistry.
https://doi.org/10.1007/s10533-023-01084-1.
Minkkinen, K., Byrne, K., Trettin, C., 2008. Climate impacts of peatland forestry. In:
Strack, M. (Ed.), Peatlands and Climate Change. International Peatland Society.
Natural Resources Institute Finland, 2023. The Multi-source National Forest Inventory
Raster Maps of 2021.
Oksanen, J., Simpson, G.L., Blanchet, F.G., Kindt, R., Legendre, P., Minchin, P.R.,
O’Hara, R.B., Solymos, P., Stevens, M.H.H., Szoecs, E., Wagner, H., Barbour, M.,
Bedward, M., Bolker, B., Borcard, D., Carvalho, G., Chirico, M., De Caceres, M.,
Durand, S., Evangelista, H.B.A., FitzJohn, R., Friendly, M., Furneaux, B.,
Hannigan, G., Hill, M.O., Lahti, L., McGlinn, D., Ouellette, M.-H., Cunha, E.R.,
Smith, T., Stier, A., Ter Braak, C.J.F., Weedon, J., 2022. Vegan: community ecology
package [WWW document]. URL. https://cran.r-project.org/web/packages/vegan/
vegan.pdf.
Ovaskainen, O., Tikhonov, G., Norberg, A., Guillaume Blanchet, F., Duan, L., Dunson, D.,
Roslin, T., Abrego, N., 2017. How to make more out of community data? A
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
11 
conceptual framework and its implementation as models and software. Ecol. Lett. 20,
561–576. https://doi.org/10.1111/ele.12757.
Pakarinen, P., Ruuhijarvi, R., 1978. Ordination of northern Finnish peatland vegetation
with factor analysis and reciprocal averaging. Ann. Bot. Fenn. 15.
Palmroos, I., Norros, V., Keski-Saari, S., Mayra, J., Tanhuanpaa, T., Kivinen, S.,
Pykala, J., Kullberg, P., Kumpula, T., Vihervaara, P., 2023. Remote sensing in
mapping biodiversity – a case study of epiphytic lichen communities. For. Ecol.
Manag. 538, 120993. https://doi.org/10.1016/j.foreco.2023.120993.
Pang, Y., Rasanen, A., Juselius-Rajamaki, T., Aurela, M., Juutinen, S., Valiranta, M.,
Virtanen, T., 2023. Upscaling field-measured seasonal ground vegetation patterns
with Sentinel-2 images in boreal ecosystems. Int. J. Remote Sens. 44, 4239–4261.
https://doi.org/10.1080/01431161.2023.2234093.
Pellerin, S., Lagneau, L.-A., Lavoie, M., Larocque, M., 2009. Environmental factors
explaining the vegetation patterns in a temperate peatland. C. R. Biol. 332, 720–731.
https://doi.org/10.1016/j.crvi.2009.04.003.
Pontone, N., Millard, K., Thompson, D.K., Guindon, L., Beaudoin, A., 2024.
A hierarchical, multi-sensor framework for peatland sub-class and vegetation
mapping throughout the Canadian boreal forest. Remote Sens. Ecol. Conserv.,
rse2.384 https://doi.org/10.1002/rse2.384.
Powell, S.L., Cohen, W.B., Healey, S.P., Kennedy, R.E., Moisen, G.G., Pierce, K.B.,
Ohmann, J.L., 2010. Quantification of live aboveground forest biomass dynamics
with Landsat time-series and field inventory data: a comparison of empirical
modeling approaches. Remote Sens. Environ. 114, 1053–1068. https://doi.org/
10.1016/j.rse.2009.12.018.
Price, J.S., Heathwaite, A.L., Baird, A.J., 2003. Hydrological processes in abandoned and
restored peatlands: an overview of management approaches. Wetl. Ecol. Manag. 11,
65–83. https://doi.org/10.1023/A:1022046409485.
Probst, P., Wright, M.N., Boulesteix, A., 2019. Hyperparameters and tuning strategies for
random forest. WIREs Data Min. Knowl. Discov. 9, e1301. https://doi.org/10.1002/
widm.1301.
Rana, P., Christiani, P., Ahtikoski, A., Haikarainen, S., Stenberg, L., Juutinen, A.,
Tolvanen, A., 2024. Cost-efficient management of peatland to enhance biodiversity
in Finland. Sci. Rep. 14, 2489. https://doi.org/10.1038/s41598-024-52964-x.
Rasanen, A., Virtanen, T., 2019. Data and resolution requirements in mapping vegetation
in spatially heterogeneous landscapes. Remote Sens. Environ. 230, 111207. https://
doi.org/10.1016/j.rse.2019.05.026.
Rasanen, A., Juutinen, S., Tuittila, E., Aurela, M., Virtanen, T., 2019. Comparing ultra-
high spatial resolution remote-sensing methods in mapping peatland vegetation.
J. Veg. Sci. 30, 1016–1026. https://doi.org/10.1111/jvs.12769.
Rasanen, A., Aurela, M., Juutinen, S., Kumpula, T., Lohila, A., Penttila, T., Virtanen, T.,
2020. Detecting northern peatland vegetation patterns at ultra-high spatial
resolution. Remote Sens. Ecol. Conserv. 6, 457–471. https://doi.org/10.1002/
rse2.140.
Rasanen, A., Tolvanen, A., Kareksela, S., 2022. Monitoring peatland water table depth
with optical and radar satellite imagery. Int. J. Appl. Earth Obs. Geoinf. 112,
102866. https://doi.org/10.1016/j.jag.2022.102866.
Rasanen, A., Kekkonen, H., Lehtonen, H., Miettinen, A., Wejberg, H., Kareksela, S.,
Tzemi, D., Aro, L., Kuningas, S., Louhi, P., Ruuhijarvi, J., 2023. Euroopan unionin
ennallistamisasetusehdotuksen luontotyyppi- ja turvemaatavoitteiden vaikutukset
Suomessa (No. 1/2023), Luonnonvara- ja biotalouden tutkimus.
Rocchini, D., 2014. Fuzzy species distribution models: a way to represent plant
communities spatially. J. Veg. Sci. 25, 317–318. https://doi.org/10.1111/jvs.12152.
Roy, D.P., Kovalskyy, V., Zhang, H.K., Vermote, E.F., Yan, L., Kumar, S.S., Egorov, A.,
2016. Characterization of Landsat-7 to Landsat-8 reflective wavelength and
normalized difference vegetation index continuity. Remote Sens. Environ. 185,
57–70. https://doi.org/10.1016/j.rse.2015.12.024.
Saarimaa, M., Aapala, K., Tuominen, S., Karhu, J., Parkkari, M., Tolvanen, A., 2019.
Predicting hotspots for threatened plant species in boreal peatlands. Biodivers.
Conserv. 28, 1173–1204. https://doi.org/10.1007/s10531-019-01717-8.
Sadeghi, M., Jones, S.B., Philpot, W.D., 2015. A linear physically-based model for remote
sensing of soil moisture using short wave infrared bands. Remote Sens. Environ. 164,
66–76. https://doi.org/10.1016/j.rse.2015.04.007.
Salko, S., Juola, J., Burdun, I., Vasander, H., Rautiainen, M., 2023. Intra- and
interspecific variation in spectral properties of dominant Sphagnum moss species in
boreal peatlands. Ecol. Evol. 13, e10197. https://doi.org/10.1002/ece3.10197.
Sallinen, A., Tuominen, S., Kumpula, T., Tahvanainen, T., 2019. Undrained peatland
areas disturbed by surrounding drainage: a large scale GIS analysis in Finland with a
special focus on aapa mires. Mires Peat 1–22. https://doi.org/10.19189/MaP.2018.
AJB.391.
Sallinen, A., Akanegbu, J., Marttila, H., Tahvanainen, T., 2023. Recent and future
hydrological trends of aapa mires across the boreal climate gradient. J. Hydrol. 617,
129022. https://doi.org/10.1016/j.jhydrol.2022.129022.
Simpson, G., Nichol, C.J., Wade, T., Helfter, C., Hamilton, A., Gibson-Poole, S., 2024.
Species-level classification of peatland vegetation using ultra-high-resolution UAV
imagery. Drones 8, 97. https://doi.org/10.3390/drones8030097.
Steenvoorden, J., Limpens, J., 2023. Upscaling peatland mapping with drone-derived
imagery: impact of spatial resolution and vegetation characteristics. GIScience
Remote Sens. 60, 2267851. https://doi.org/10.1080/15481603.2023.2267851.
Steenvoorden, J., Limpens, J., Crowley, W., Schouten, M.G.C., 2022. There and back
again: forty years of change in vegetation patterns in Irish peatlands. Ecol. Indic.
145, 109731. https://doi.org/10.1016/j.ecolind.2022.109731.
Steenvoorden, J., Bartholomeus, H., Limpens, J., 2023. Less is more: optimizing
vegetation mapping in peatlands using unmanned aerial vehicles (UAVs). Int. J.
Appl. Earth Obs. Geoinf. 117, 103220. https://doi.org/10.1016/j.jag.2023.103220.
Talvitie, P., Rasanen, A., Silvan, N., 2023. Changes in the open water hollows in
Haadetkeidas and Kauhaneva mires during 1947–2017 based on remote sensing. Suo
- Mires Peat 74, 71–96.
Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring
vegetation. Remote Sens. Environ. 8, 127–150. https://doi.org/10.1016/0034-4257
(79)90013-0.
Verniest, F., Greulich, S., 2019. Methods for assessing the effects of environmental
parameters on biological communities in long-term ecological studies - a literature
review. Ecol. Model. 414, 108732. https://doi.org/10.1016/j.
ecolmodel.2019.108732.
Wijaya, A., Reddy Marpu, P., Gloaguen, R., 2010. Discrimination of peatlands in tropical
swamp forests using dual-polarimetric SAR and Landsat ETM data. Int. J. Image Data
Fusion 1, 257–270. https://doi.org/10.1080/19479832.2010.495323.
Wolff, F., H. M. Kolari, T., Villoslada, M., Tahvanainen, T., Korpelainen, P., A. P.
Zamboni, P., Kumpula, T., 2023. RGB vs. multispectral imagery: mapping aapa mire
plant communities with UAVs. Ecol. Indic. 148, 110140. https://doi.org/10.1016/j.
ecolind.2023.110140.
Xu, H., 2006. Modification of normalised difference water index (NDWI) to enhance open
water features in remotely sensed imagery. Int. J. Remote Sens. 27, 3025–3033.
https://doi.org/10.1080/01431160600589179.
Zhang, H.K., Roy, D.P., Yan, L., Li, Z., Huang, H., Vermote, E., Skakun, S., Roger, J.-C.,
2018. Characterization of Sentinel-2A and Landsat-8 top of atmosphere, surface, and
nadir BRDF adjusted reflectance and NDVI differences. Remote Sens. Environ. 215,
482–494. https://doi.org/10.1016/j.rse.2018.04.031.
A. Isoaho et al. Science of the Total Environment 957 (2024) 177697 
12