ArticlePDF Available

Abstract

Developing a methodology to map the distribution of riparian forests to entire river networks and determining the main environmental factors controlling their spatial patterns. Cantabrian region, northern Spain. We mapped the riparian forests at a physiognomic and phytosociological levels by delimiting riparian zones and generating vegetation distribution models based on remote sensing data (Landsat 8 OLI and LiDAR PNOA). We built virtual watersheds to define a spatial framework where the catchment environmental information can be routed to each river reach, jointly with the vegetation map. In order to determine the drivers playing a significant role on the observed spatial patterns in the riparian forest we modelled interactions between these datasets of environmental information and riparian vegetation by using the Random Forest algorithm. The modelling results obtained reproduced a reliable variation of riparian forest structure and composition across Cantabrian watersheds. The produced maps were highly accurate, with more than a 70% overall accuracy for the forest occurrence. A clear differentiation between Eurosiberian (91E0 and 9160 habitats) and Mediterranean (92E0) riparian forests was shown on both sides of the mountain range. Topography and land use were the main drivers defining the distribution of riparian forest as a physiognomic unit. In turn, altitude, climate and percentage of pasture were the most relevant factors determining their composition (phytosociological approach). Our study confirms that the anthropic control ultimately defines the distribution of the vegetation in the riparian area at a regional to local scale. Human disturbances constrain the extension of forest patches across their potential distribution defined by topoclimatic boundaries, which establish a clear limit between Mediterranean and Eurosiberian biogeographical regions.
508
|
wileyonlinelibrary.com/journal/avsc Appl Veg Sci. 2019;22:508–521.
Applied Vegetation Science
© 2019 International Association
for Vegetation Science
Received: 16 May 2019 
|
  Revised: 4 September 2019 
|
  Accepted: 13 September 2019
DOI: 10.1111/avsc.12458
RESEARCH ARTICLE
Modelling riparian forest distribution and composition to entire
river networks
Ignacio Pérez‐Silos | José Manuel Álvarez‐Martínez | José Barquín
Environmental Hydraulics Institute “IH
Cantabria”, University of Cantabria,
Santander, Spain
Correspondence
Ignacio Pérez‐Silos, Environmental
Hydraulics Institute “IH Cantabria”,
University of Cantabria, PC TCAN. C/ Isabel
Torres 15, 39011 Santander, Spain.
Email: ignacio.perez@unican.es
Funding information
Spanish Ministry of Science, Innovation
and Universities, Grant/Award Number:
FPU‐2015‐03018; European Regional
Development Fund through the Interreg
Atlantic Area as part of the ALICE project,
Grant/Award Number: EAPA‐261/2016
Co‐ordinating Editor: Duccio Rocchini
Abstract
Aim: Developing a methodology to map the distribution of riparian forests to entire
river networks and determining the main environmental factors controlling their spa
tial patterns.
Location: Cantabrian region, northern Spain.
Methods: We mapped the riparian forests at a physiognomic and phytosociologi
cal level by delimiting riparian zones and generating vegetation distribution models
based on remote sensing data (Landsat 8 OLI and LiDAR PNOA). We built virtual
watersheds to define a spatial framework where the catchment environmental in
formation can be specified for each river reach, in combination with the vegetation
map. In order to determine the drivers that play a significant role in the observed
spatial patterns in riparian forests, based on our data sets we modelled interactions
between environmental information and riparian vegetation by using the Random
Forest algorithm.
Results: The modelling results obtained reliably reproduced the variation of ripar
ian forest structure and composition across Cantabrian watersheds. The produced
maps were highly accurate, with a more than 70% overall accuracy for forest occur
rence. A clear differentiation between Eurosiberian (habitats 91E0 and 9160) and
Mediterranean (92E0) riparian forests was shown on both sides of the mountain
range. Topography and land use were the main drivers defining the distribution of
riparian forest as a physiognomic unit. In turn, altitude, climate and percentage of
pasture were the most relevant factors determining their composition (phytosocio
logical approach).
Conclusions: Our study confirms that anthropic control ultimately defines the
distribution of vegetation in the riparian area at a regional to local scale. Human
disturbances constrain the extension of forest patches across their potential dis
tribution defined by topoclimatic boundaries, which establish a clear limit between
Mediterranean and Eurosiberian biogeographical regions.
KEYWORDS
Cantabrian Cordillera, random forest, remote sensing, river networks, riverine landscapes,
virtual watersheds
    
|
 509
Applied Vegetation Science
PÉREZ‐SILOS Et aL.
1 | INTRODUCTION
Riparian zones can be defined as heterogenic mosaics of biological
communities, land morphologies and environments that are charac
terized by sharp gradients in environmental‐biological factors and
ecological processes (Gregory, Swanson, McKee, & Cummins, 1991).
These areas constitute one of the most diverse, dynamic and com
plex habitats on earth (Naiman, Decamps, & Pollock, 1993). Riparian
forests are a fundamental component of riparian zones (Richardson
et al., 2007) and of the whole river ecosystem (Magdaleno, Blanco‐
Garrido, Bonada, & Herrera‐Grao, 2014). They play a major role in
regulating stream water temperature, trapping sediments and chem
icals from uplands, providing organic matter to the river, supporting
aquatic and riparian wildlife, serving as biological corridors, prevent
ing bank erosion and mitigating flood effects, among others (Naiman,
Decamps, & McClain, 2010). Thus, their conservation is crucial for
maintaining river ecological processes and for the delivery of key
ecosystem services to human societies (NRC, 2002). However, ri
parian forests have been subjected to human disturbances for mil
lennia (Hooke, 2006; Petts, Möller, & Roux, 1989), and their natural
functioning has been highly impaired by hydromorphological pres
sures, land use changes and other alterations of the fluvial landscape
(Logan & Furse, 2002; Poff, Koestner, Neary, & Henderson, 2011).
Nowadays, there is a large body of environmental legislation
worldwide that promotes an adequate management and conser
vation of the riparian forests (e.g. Clean Water Act in the USA,
Streamside Protection Regulation in Canada, Water Framework
Directive, Nitrates or Floods Directives in Europe). In Europe, the EU
Habitats Directive (HD, 1992/43/DC), with several habitats which
belong to riparia, and the EU Water Framework Directive (WFD,
2000/60/EC) are the most important legislation underpinning ripar
ian forest management. The last one contemplates riparian structure
as an element of hydromorphological quality and it should be re
ported along with other ecological quality indicators. For both, there
is a strong need for information on the current and potential riparian
vegetation distribution and composition patterns, together with the
identification of the main drivers determining them (Palmer et al.,
2005). The generated information should be adapted to the specific
requirements of these legal frameworks. In this regard, the WFD
needs information at the physiognomic level to derive some of the
riparian structure quality elements (length, width, land cover, etc.)
proposed by the European Commission (2003), while the HD also
requires species composition data at the level of phytosociological
associations.
Many studies have focused on characterizing interactions be
tween riparian vegetation and independent environmental factors
such as climatic conditions (e.g. Rocha et al., 2015), hydrologic reg
imens (e.g. Camporeale & Ridolfi, 20 06), river topography and mor
phodynamics (e.g. Magdaleno & Fernández‐Yuste, 2013), anthropic
pressures (e.g. Salinas & Casas, 2007) and land uses (e.g. Wasson
et al., 2010). However, these approaches do not achieve a proper
biogeographical perspective considering a whole river corridor
from its source to the sea. The lack of continuous information for
riparian forests prevents deriving a complete diagnosis of riparian
conservation status and limits our understanding of the interre
lationships between riparian conservation status and impairment
of relevant riverine ecosystem processes (Stella, Rodri, Dufour, &
Bendix, 2013). Moreover, this limitation also compromises the se
lection of the best river reaches and floodplains for restoration or
conservation of specific riparian functions and related ecosystem
ser vices (e.g., shade for controlling water temperature and thus im
proving water quality, reducing investment in water purification;
Guillozet, 2015). So, there is an increasing interest in completely
characterizing riparian forests along entire river networks while
defining the main fac tors that control their spatial patterns (A guiar,
Cerdeira, Martins, & Ferreira, 2013; Aguiar & Ferreira, 2005;
Douda, 2010). Nevertheless, studies covering large areas are still
scarce (Macfarlane et al., 2017) and they are based on particular
locations instead of continuous riverine landscapes. This might be
the consequence of methodological limitations, especially those
related to the scarcity of ground truth, validated data and coeta
neous spatial predictors of landscape structure, which hamper the
production of detailed and continuous maps of riparian vegetation
at regional scales (Jeong, Mo, Kim, Park, & Lee, 2016). As a con
sequence, traditional attempts to produce information on riparian
buffers and its condition have basically relied on interpretation of
aerial photographs and field visits (Goetz, 2006), requiring a large
effort in terms of time and resources.
Advances in remote sensing can contribute to overcoming these
limitations. Classification of satellite imagery has already been used
to map riparian buffers successfully (Goetz, 2006; Klemas, 2013),
even generating a European continental map by combining the im
ages with ancillary data such as elevation (Weissteiner et al., 2016).
However, land cover heterogeneity and the small patch sizes of ri
parian areas in human‐dominated landscapes hamper the ability to
map forests at the habitat level (i.e. different types of forests) using
only optical information (Jeong et al., 2016). Aerial Light Detection
and Ranging technology (LiDAR) yields very precise information
for the discrimination of riparian forests (Michez, Piégay, Lejeune,
& Claessens, 2017). LiDAR data have been used to characterize
the structure (Seavy, Viers, & Wood, 2009), physical attributes
(Johansen, Phinn, & Witte, 2010; Michez, Piégay, Lisein, Claessens,
& Lejeune, 2016) and ecological status (Ivits, Cherlet, Mehl, &
Sommer, 2009) of riparian vegetation. However, approaches that
combine satellite images and LiDAR data to map riparian vegetation
types are scarce and do not cover extensive areas (e.g. Jeong et al.,
2016). The combination of these remote sensing data with commu
nity‐level species distribution models could represent a significant
step forward in riparian vegetation mapping (He et al., 2015), and it
has already been succes sfully applie d to terrestria l habitats (Á lva rez
Martínez et al., 2018).
Based on these considerations, the main objective of this study
is to develop a methodology to map the distribution and composi
tion of riparian forests across entire river networks using a com
bination of remote‐sensing data and modelling techniques while
determining the main environmental factors controlling their
510 
|
    
Applied Vegetation Science PÉREZ‐SILOS Et aL.
distributional patterns. Riparian vegetation was firstly mapped at
two different aggregation levels in order to explore interactions
over the riparian structure (distribution of physiognomic units in
the riparian zone) and the riparian forest composition (at a phytoso
ciological level following Annex I of the HD). We hypothesized that
anthropic land use is the main force driving the organization of the
physiognomic units that appear in th e riparian zon e. In contrast, we
expect that variables controlling hydrologic conditions (e.g. type
of valley, slope, substrate, etc.) and climate would be the most im
portant for defining the composition of riparian forests (i.e. forest
habitat types).
2 | METHODS
2.1 | Study area
The study was conducted in more than 10 river catchments from
the Cantabrian region (area > 5,0 00 km2; Figure 1). The mountain
ous relief and coastal location of the region define the climatic
conditions and morphologic characteristics of main rivers (Barquín
et al., 2012), which have their source in the Cantabrian Mountains.
This mountain range, that runs from west to east parallel to the
Atlantic Ocean coast and reaches more than 2,600 m, defines a
triple division across the study area: hydrographic, climatic and
biogeographic. The northern part of the region is characterized by
short and steep rivers with high erosive power that drain into the
Cantabrian Sea (Atlantic Ocean). The largest catchments slightly
exceed 1,000 km2 and 20 m3/s of mean daily flow, wit h hig hly vari
able valley widths that could be over 1.5 km in most of the mid
dle and upper courses. The climate of this area is humid oceanic
temperate, with an average annual temperature of 14°C and an
average annual precipitation of 1,200 mm. The rivers draining the
southern face of the Cantabrian Mountains are longer and with
smoother slopes, draining into the Atlantic Ocean (Douro river
basin) and into the Mediterranean Sea (Ebro river basin). This
southern region is dominated by a continental climate with an av
erage annual temperature of 10°C and an average annual precipi
tation of 700 mm.
In relation to the vegetation, the Atlantic area is characterized
by Eurosiberian vegetation. The riparian forests are dominated by
oceanic alder groves (Alnus glutinosa) from almost sea level up to
700 m and Salix atrocinerea, which replace the former after any
FIGURE 1 (a) Study area with the representation of selected river networks in the Cantabrian region, (b) the basins at which they belong,
and (c) their location in Europe
    
|
 511
Applied Vegetation Science
PÉREZ‐SILOS Et aL.
disturbance, when soils are not deep enough or if there are large
flow fluctuations (Fernández, Barquín, Álvarez‐Cabria, & Peñas,
2012). At a higher altitude, ashes (Fraxinus excelsior) or hazelnuts
(Corylus avellana) commonly dominate riparian forests, while in
steep valleys and uplands, beech (Fagus sylvatica), oaks (Quercus
robur) and mixed Atlantic forest predominate (Lara, Garilleti, &
Calleja, 2004). On the southern face of the Cantabrian Mountains,
there is a transition from Eurosiberian to Mediterranean‐type
vegetation. A native vegetation community dominated by sub‐
Mediterranean alder groves (Alnus glutinosa) and Salix cantabrica
in the riparian areas, and holm oaks (Quercus ilex), Pyrenean oaks
(Quercus pyrenaica), bay laurels (Laurus nobilis) and strawberry trees
(Arbutus unedo) in the uplands are the main communities. The nat
ural riparian vegetation has been extensively modified by human
activity in both the northern and southern areas. This involves the
introduction of eucalyptus (Eucalyptus globulus) and pine (Pinus sp.)
plantations and a widespread land cover change to pasturelands or
shrub formations on modified banks where agrarian activities are
and have been more intensive. Rubus sp., Rosa sp., Crataegus mon
ogyna and Prunus spinosa usually dominate the riparian vegetation
of these impaired areas.
2.2 | Methods
The methodology followed in this work can be split in two main
blocks: (a) building the riparian forest maps following Fernández et
al. (2012) for delimiting riparian zones and Álvarez‐Martínez et al.,
(2018) for vegetation mapping, and (b) determining the drivers of
their spatial patterns using a Random Forest algorithm (RF; Breiman,
2001) (Figure 2).
2.2.1| Building the riparian forest maps
In this study, we considered 28 habitats of Community interest in
the Cantabrian region (92/43/EEC) as vegetation types belonging
to riparian zones and adjacent areas, besides pine and eucalyptus
plantations (see Álvarez‐Martínez et al., 2018). All these habitats
were first modelled independently and then aggregated at two
different levels (Table 1): (a) a map of riparian structure in which
the riparian vegetation was divided into six physiognomic units
(i.e. land cover types); and (b) a map of riparian forest composition
that categorizes the former physiognomic units into a phytosocio
logical classification (i.e. the 10 forest habitat types of Annex I of
FIGURE 2 General framework for mapping riparian vegetation and determining the main driving forces in the Cantabrian rivers
512 
|
    
Applied Vegetation Science PÉREZ‐SILOS Et aL.
TABLE 1 Habitat types modelled using a combination of remote sensing and environmental data (see Álvarez‐Martínez et al., 2018 for
a more detailed explanation). The resulting models were used in this study to build riparian vegetation maps following two classification
approaches: physiognomic and phytosociological (i.e. forest habitats). Anthropic structures were added to the list of habitat types to
represent non‐vegetated surfaces (see Methods). Alder–ash forest and willow–poplar–elm forest were abbreviated in the text as AAF and
WPEF, respectively
Source Code
Habitat types (Álvarez‐Martínez et al., 2018) Riparian vegetation
Habitat definition
Physiognomic
units Forest habitats
Habitats
Directive
(EEC, 1992)
4020 Temperate Atlantic wet heaths with Erica ciliaris and Erica tetralix Shrubland
4030 European dry heaths
4060 Alpine and Boreal heaths
4090 Endemic oro‐Mediterranean heaths with gorse
512 0 Mountain Cytisus purgans formations
5230 Arborescent matorral with Laurus nobilis
6160 Oro‐Iberian Festuca indigesta grasslands Pasture
6170 Alpine and subalpine calcareous grasslands
6210 Semi‐natural dry grasslands and scrubland facies on calcareous sub
strates (Festuco‐Brometalia)
6230 Species‐rich Nardus grassland, on siliceous substrates in mountain areas
(and submountain areas in continental Europe)
6410 Molinia meadows on calcareous, peaty or clayey‐silt‐laden soils
(Molinion caeruleae)
6420 Mediterranean tall humid herb grasslands of the Molinio‐Holoschoenion
6510 Lowland hay meadows (Alopecurus pratensis, Sanguisorba officinalis)
8130 Western Mediterranean and thermophilous scree Non‐vegetated
8210 Calcareous rocky slopes with chasmophytic vegetation
8220 Siliceous rocky slopes with chasmophy tic vegetation
8230 Siliceous rock with pioneer vegetation of the Sedo Scleranthion or of the
Sedo albi‐Veronicion dillenii
9120 Atlantic acidophilous beech forests with Ilex and sometimes also Ta xus
in the shrub layer (Quercion robori‐petraeae or Ilici‐Fagenion)
Hillside forest Beech forest
9150 Medio‐European limestone beech forests of the Cephalanthero‐Fagion
9230 Galicio‐Portuguese oak woods with Quercus robur and Quercus
pyrenaica
Oak forest
9240 Quercus faginea and Quercus canariensis Iberian woods Gall oak forest
9260 Castanea sativa Woods Chestnut forest
9330 Quercus suber forests Cork oak forest
9340 Quercus ilex and Quercus rotundifolia forests Holm oak forest
9380 Forests of Ilex aquifolium Holly forest
91E0 Alluvial forests with Alnus glutinosa and Fraxinus excelsior (Alno‐Padion,
Alnion incanae, Salicion albae)
Riparian forest Alder–ash forest; AAF
(VV.AA., 2009)
92A0 Salix alba and Populus alba galleries Willow–poplar–elm
forest; WPEF (V V.
AA., 2009)
9160 Sub‐Atlantic and medio‐European oak or oak‐hornbeam forests of the
Carpinion betuli
Mixed forest (V V.AA.,
2009)
Other Eucalyptus plantations Forest
plantations
Pine plantation
Anthropic structures Non‐vegetated
    
|
 513
Applied Vegetation Science
PÉREZ‐SILOS Et aL.
the HD). Five main steps could be differentiated in this process: (a)
generation of a synthetic river network; (b) delimitation of the ri
parian zone; (c) generation of distributional maps of habitat types;
(d) modelling vegetation co‐occurrence; and (e) production of final
riparian forest maps.
First, a synthetic river network was derived for the whole re
gion of Cantabria following the procedure described by Benda,
Miller, and Barquín (2011) using the analysis toolkit “NetMap”
(Benda et al., 2007). The network was delineated applying the
algorithms described by Clarke, Burnett, and Miller (2008) in a
5‐m Digital Elevation Model (DEM), obtaining a set of 1,410 river
reaches (average length, 983 m).
Second, a geomorphological criterion based on the surface
that intersects valley walls at a height of n times the river bank full
depth (BFD) was applied to delimit the riparian zone. Other stud
ies in the region pointed to values of 0.5–1.25 BFD (Fernández et
al., 2012) as the best fit for delimiting 50‐year floods, an optimal
hydrological descriptor for riparian areas because it roughly cor
responds with the first terrace or other upward‐sloping surface
(Ilhardt, Verry, & Palik, 2000). However, a height of 5 BFD was
selected in this study in order to incorporate the Cuaternarian al
luvial terraces of open valleys. These alluvial terraces still have
a high hydrological influence, suffering important water table
oscillations in relation to river flow changes, and riparian species
mainly dominate the vegetation. The BFD value was estimated for
each segment of the obtained river network through a regional
regression of drainage area and mean annual precipitation to bank
full depths measured in the field (for more detailed information,
see Benda et al., 2011). The delineation was also extended apply
ing a 30‐m buffer to ensure matching all pixels obtained from veg
etation models.
Third, we used 30 habitat s types previously modelled in Álvarez‐
Martínez et al. (2018) for the whole region of Cantabria (Table 1).
The models produce independent probabilistic maps of habitat dis
tribution with suitability values at the pixel level ranging from 0 to
1 — 1 being the maximum probability for a particular habitat at any
point in geographical space. The maps are supported by a conceptual
framework based on modelling community‐level entities using veg
etation occurrence (presence/absence) data and a set of predictors.
Among the predictors, environmental layers (climate, topography
and soil properties) and remote‐sensing data (Landsat and LiDAR)
were used (Appendix S1). The Maximum Entropy Method (MaxEnt;
Phillips, Aneja, Kang, & Arya, 2006) was selected as the modelling
algorithm.
Fourth, each probabilistic map of habitat distribution was in
tersected with the riparian polygon to extract the area of the map
belonging to the riparian zone. Subsequently, a vegetation co‐occur
rence model was built using the Highest Position algorithm (ESRI,
2011) over the whole data set of riparian maps. The algorithm pro
duces a multi‐class map in which each pixel is represented by the
habitat with the highest suitability value. The co‐occurrence model
had to be calibrated because of commission and omission errors
among habitat types in the first stages of application of the algorithm
on the base of ground‐truthing and expert knowledge of the study
area. These problems were mainly located in mixed pixels without
any dominant habitat (i.e. poor–pure ecotones or immature vegetal
formations) in which the probabilistic competition occurred between
the suitability histogram tiles (low suitability) of several probabi
listic maps of habitat distribution, triggering a high uncertainty in
final predictions (Álvarez‐Martínez, Suárez‐Seoane, Palacín, Sanz, &
Alonso, 2015). In these pixels, the criteria for habitat selection by the
co‐occurrence model is weak, so the output may not be representa
tive of reality (Álvarez‐Martínez, Stoorvogel, Suárez‐Seoane, & de
Calabuig, 2010). The calibration was done by iteratively applying the
Highest Position algorithm over different sets of probabilistic maps
of habitat distribution in which the expression of oversized habitats
had been limited through expert‐based thresholds of their suitability
histograms. Finally, the resulting co‐occurrence model was reclassi
fied into physiognomic units and forest habitat types according to
Table 1 in order to produce the two riparian forest maps (i.e. phys
iognomic and phytosociological) described above. The former map
identified the distribution of the riparian forest classes, while the
latter defined the composition of these physiognomic units at the
habitat level.
Individual riparian vegetation models and the final co‐occurrence
models were all locally validated by visual interpretation with coeta
neous aerial photographs and Google Street View (Olea & Mateo‐
Tomás, 2016) and field campaigns in selected locations. We made a
particular effort in areas of high uncertainty in which low probability
values co‐occur for different habitats. Additionally, we performed a
regional comparison of both riparian forest maps with two indepen
dent, readily available, land use and land cover maps: the National
Forest vegetation Map at the 1:50,000 scale (NFM, 2006) and the
Cantabrian Riparian Cover map at 1:5,000 (CRC, 2010), which are
able to produce overall accuracy scores and omission–commission
errors.
2.2.2 | Determining the drivers of the spatial
patterns of riparian forests
We used RF models to identify the main factors driving the struc
ture and composition of riparian vegetation in the Cantabrian region.
First, the two vegetation maps were adjusted to each river reach by
using virtual watershed algorithms (i.e. addData; Benda et al., 2007).
By means of these, raster information was transferred to each river
reach as the percentage of riparian area that is covered by each
physiognomic class and forest habit at type. Secondly, a group of abi
otic–anthropogenic variables describing environmental attributes
important for the riparian vegetation (see Lara et al., 2004; Naiman
et al., 2010) were also integrated into the synthetic river network
using the same procedure. The contribution of the available GIS data
sets to each river reach in the network was calculated using different
spatial scales: (a) the catchment draining to the reach; (b) local wings
draining to the reach; and (c) 50‐m buffer polygons draining to the
reach. In total, four group of variables (86 variables; see Appendix
S2) were used as predictors:
514 
|
    
Applied Vegetation Science PÉREZ‐SILOS Et aL.
1. Climatic variables. Data were derived from monthly climate
variables calculated on a 1‐km grid map by means of an in
terpolation procedure based on data recorded at more than
5,000 weather stations of the Spanish network and rescaled
to 30 m. They were obtained from the Integrated System
for Rainfall‐Runoff modelling (SIMPA model) developed by the
Centre for Hydrographic Studies (CEDEX, Ministry of Public
Works and Ministry of Agriculture and Environment, Spanish
Government).
2. Topographic and geomorphic variables. Catchment topography
and bed morphology were derived from the 5‐m DEM using some
of NetMap's analysis tools. In turn, geological catchment vari
ables were obtained from the 1:50,000 National Geological Map
(MAGNA50) developed by the Geological and Mining Institute of
Spain (IGME).
3. Land use and land cover variables. Data were obtained from the
Soil Occupancy Information System of Spain (SIOSE 1:25,000),
created by the National Geographic Institute (CNIG) of the
Spanish Government.
4. Anthropogenic pressures. We defined four groups of variables
that consisted of the distance of each river to the closest up/
downstream impact (effluents, embankments, dams or weirs and
other transversal obstacles). ESRI's ArcPy Python module (ESRI,
2011) was used to calculate these distances.
A preliminary exploration of the whole dataset was carried out with
Spearman's correlation coefficients, choosing as final predictor vari
ables those that: (a) were related to the occurrence of at least three
riparian vegetation classes (ρ > |0.3|) from each of the approaches; (b)
were independent in relation to the rest of the environmental variables
(ρ < |0.7|). After that, RF was used to model the interaction between
the selected predictors and the percentage cover of each riparian
vegetation type. This technique can handle thousands of input vari
ables (Archer & Kimes, 2008) and accounts for non‐linear relation
ships and interactions among predictors (Olson & Hawkins, 2012). The
“RandomForest” package (Liaw & Wiener, 2015) within the R environ
ment (R Core Team, 2011) was used in the regression mode and default
values. To assess the model fit, R2 b etween the obser ved and pred icted
variable values was calculated. The structure of each RF model was ex
amined using the Increment Node Purity index (IncNodePurity), which
defines the importance of predictor variables, and partial dependence
plots that evaluate the marginal effects of each predictor in the final
models (Kuhn, Egert, Neumann, & Steinbeck, 2008).
3 | RESULTS
3.1 | Riparian forest maps
Riparian forests are the most common vegetation formation in the
riparian areas of the Cantabrian region, followed by pastures and
hillside forests (Table 2). Specifically, mixed forest and alder–ash for
est (A AF) are the two most abundant habitats: 23.12 and 13.94 km2
respectively. The proportion of willow–poplar–elm forest (WPEF)
is lower than that of other non‐riparian forests like beech and oak
formations.
By visual interpretation of vegetation models based on expert
knowledge we determined that the performance of riparian forest
maps was highly suitable for the scale of the analyses (1:50,000–
1:75,000). In turn, regional comparisons against the NFM and the
CRC provided a global accuracy of 79% and 76%, respectively, for
forest occurrence within the riparian zone, with major omission er
rors related to pasture formations and commission affecting pasture
and shrubland vegetation types in both maps (Appendix S3). At these
scales, the resulting maps perfectly collate the longitudinal gradient
of vegetation alon g the flu vial course and the lateral pat te rn of vege
tation change, although this last one is more limited by the pixel size.
3.2 | Drivers of riparian forest distribution and
composition
3.2.1| Physiognomic units
Riparian and hillside forests showed the best performance among
all modelled vegetations (R2 = 0.77 and R2 = 0.76 respectively;
Figure 3a), followed by non‐vegetated and pasture (R2 = 0.55 and
R2 = 0.57, respectively; Figure 3a). Shrubland and forest plantations
did not perform well and they were excluded from further analyses
(R2 = 0.42 and R2 = 0.14, respectively; Figure 3a).
Pasture land use in the riparian zone (PAS) and mean altitude
(ALT) were both optimal predictor variables of riparian and hillside
forests (Appendix S4). Increases in PAS relative cover resulted al
most lineally in decreases of riparian and hillside forests (Figure 4h,l),
TABLE 2 Surface area (in km2) and percentage of occupation
of vegetation types (at both physiognomic and phytosociological
levels) across the entire riparian zone of Cantabria
Surface area (km2)%
Riparian forest 41.34 36.34
Mixed 23.12 20.33
AAF 13.94 12.25
WPEF 4.28 3.76
Hillside forest 20.05 17. 62
Beech 5.11 4.48
Oak 11.92 10 .47
Gall oak 0.23 0.19
Cork oak 00.08
Holm oak 1.53 1.33
Chestnut 0.44 0.37
Holly 0.82 0.71
Non‐vegetated 8.93 7.85
Pasture 30.26 26.21
Shrubland 12.21 10.74
Forest plantation 0.95 0.84
Abbreviations: AAF: Alder–ash forest; WPEF: willow–poplar–elm forest.
    
|
 515
Applied Vegetation Science
PÉREZ‐SILOS Et aL.
FIGURE 3 Predicted values (x‐axis) vs observed values (y‐axis) for: (a) physiognomic unit models and (b) forest (phytosociological)
habitats models. AAF: Alder – ash forest; WPEF: Willow – poplar ‐ elm forests
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Non−vegetated
R2 = 0.55
0.00.2 0.40.6 0.81.0 0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Pasture
R2 = 0.57
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.8
Shrubland
R2 = 0.52
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Forest Plantation
R2 = 0.14
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Hillside Forest
R2 = 0.76
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Riparian Forest
R2 = 0.77
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Beech Forest
R2 = 0.65
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Oak Forest
R2 = 0.67
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Gall oak Forest
R2 = 0.34
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Cork oak Forest
R2 = −0.00
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Holm oak Forest
R2 = 0.04
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Chestnut Forest
R2 = 0.16
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Holly Forest
R2 = 0.34
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
Mixed Forest
R2 = 0.77
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
AAF
R2 = 0.72
0.00.2 0.40.6 0.81.0
0.00.2 0.40.6 0.81.0
WPEF
R2 = 0.74
(a)
(b)
516 
|
    
Applied Vegetation Science PÉREZ‐SILOS Et aL.
FIGURE 4 Partial Dependence Plots for physiognomic units. The x‐axis shows the predictor value, while the y‐axis gives the occupancy
percentage of the physiognomic units in the riparian area. Plots are ordered according to the importance of each variable in the models
(IncNodePurity)
0.00.2 0.40.6 0.81.0
0.10.2 0.30.4
Urban land use in the riparian zone [URB]
º/1
0.0 0.2 0.40.6 0.81.0
0.07 0.09 0.11 0.13
Pasture land use in the riparian zone [PAS]
º/1
020406080
0.07 0.10 0.13
Valley width index [VWI]
n.d
020406080
0.20 0.30 0.40
Valley width index [VWI]
n.d
0.22 0.26 0.30
Mean incidental solar radiation accumulated [RAD]
W/m2.yr
0.00.2 0.4 0.6 0.81.0
0.20 0.24 0.28
Non−vegetated land use in the riparian zone [NVG]
º/1
0.0 0.20.4 0.60.8
0.22 0.26 0.30
Mean slope [GRA]
º/1
0.00.2 0.40.6 0.81.0
0.10 0.20
Pasture land use in the riparian zone [PAS]
º/1
0 500 10001500
0.10 0.20
Mean elevation [ALT]
m
10 20 30 40
0.15 0.17 0.19 0.21
Spring precipitation coefficient of variation [cvPREsp
n.d
0.0 0.20.4 0.60.8
0.16 0.18 0.20
Mean slope [GRA]
º/1
0.00.2 0.40.6 0.81.0
0.25 0.35 0.45
Pasture land use in the riparian zone [PAS]
º/1
0 500 10001500
0.30 0.35 0.40
Mean elevation [ALT]
m
400,000600,000 800,0001,000,000 1,200,000
400,000600,000 800,0001,000,000 1,200,000
0.34 0.38
Mean incidental solar radiation accumulated [RAD]
W/m2.yr
0.00.2 0.40.6 0.81.0
0.25 0.30 0.35 0.40
Non−vegetated land use in the riparian zone [NVG]
º/1
(a)
(b)
(c)
(e)
(d) (f) (g)
(h) (i) (j) (k)
(l) (m) (n) (o)
Non-vegetated
Pasture
Hillside forest
Riparian fores
t
FIGURE 5 Partial Dependence Plots for forest habitats (phytosociological approach). The x‐axis shows the predictor value, and the y‐axis
the occupancy percentage of the forest habitats in the riparian area. Plots are ordered according to the importance of each variable in the
models (IncNodePurity). AAF: Alder–ash forest; WPEF: Willow–poplar–elm forest
0.00.2 0.40.6 0.8 1.0
0.03 0.05 0.07
Pasture land use in the riparian zone [PAS]
º/1
0 500 1,000 1,500
0 500 1,000 1,500
0.04 0.08
Mean elevation [ALT]
m
0510 15 20 25
0.04 0.06
Summer precipitation coefficient of variation [cvPREs]
n.d
200 220 240 260
0.03 0.05 0.07 0.09
Summer maximum temperature average [avTMXs]
d.Cº
0.04 0.08 0.12 0.16
Mean elevation [ALT]
m
0.00.2 0.40.6 0.81.0
0.06 0.10 0.14
Pasture land use in the riparian zone [PAS]
º/1
0.09 0.11
Mean annual precipitation [PRE]
mm
200 220 240 260
0.07 0.10 0.13
Summer maximum temperature average [avTMXs]
d.Cº
400,000 600,000 800,000 1,000,000 1,200,000
400,000 600,000 800,000 1,000,000 1,200,000
0.20 0.30
Mean incidental solar radiation accumulated [RAD]
W/m2.yr
0 500 1,000 1,500
0.14 0.18 0.22
Mean elevation [ALT]
m
600 8001,000 1,200 1,400 1,600 1,800
600 8001,000 1,200 1,400 1,600 1,800
600 800 1,000 1,2001,400 1,600 1,800
0.20 0.24 0.28
Mean annual precipitation [PRE]
mm
10 20 30 40
0.16 0.19 0.22
Spring precipitation coefficient of variation [cvPREsp
n.d
0 500 1,000 1,500
0.05 0.15
Mean elevation [ALT]
m
0.1000.115 0.130
Mean annual precipitation [PRE]
mm
0.00.2 0.40.6 0.81.0
0.08 0.12 0.16
Pasture land use in the riparian zone [PAS]
º/1
0 500 1,000 1,500
0.05 0.10 0.15
Mean Elevation [ALT]
m
10 20 30 40
0.04 0.06 0.08
Spring Precipitation Coefficient of Variation [cvPREsp
n.d
0.04 0.06
Mean incidental solar radiation accumulated [RAD]
W/m2.yr
600 800 1,000 1,200 1,4001,600 1,800
0.0350.045 0.055
Mean annual precipitation [PRE]
mm
(a) (b) (c) (d)
(e) (f) (g) (h)
(i) (j) (k) (l)
(m) (n) (o)
(p) (q) (r) (s)
Beech forest
Oak forest
Mixed forest
AAF
WPEF
    
|
 517
Applied Vegetation Science
PÉREZ‐SILOS Et aL.
while ALT was positively related to hillside forests and negatively
to riparian zones (Figure 4i,m). Additionally, the valley width index
(VWI), mean incidental solar radiation accumulated over a year
(RAD), non‐vegetated land cover in the riparian zone (NVG) and
mean slope (GRA) appeared as the most important predictors for
the relative cover of pasture (Appendix S4), the relation with VWI
and RAD being positive (Figure 4d,e), while it was negative with NVG
(Figure 4f). Finally, relative cover of URB was the main predictor for
the unproductive class (Appendix S4), maintaining a strong positive
relation with the response variable (Figure 4a).
3.2.2 | Forest habitat types
The three habitats that belong to the riparian forest physiognomic
unit showed a good model fit (mixed forest: R2 = 0.76, AAF: R2 = 0.72
and WPEF: R2 = 0.74; Figure 3b). On the other hand, only beech and
oak forest showed acceptable model fit in the hillside forest physi
ognomic unit (R2 = 0.65 and R2 = 0.67 respectively; Figure 3b). The
other RF models corresponding to gall oak forest, cork oak forest,
holm oak forest, chestnut forest and holly forest were therefore not
considered in further analyses.
RF results showed an opposite behaviour of the two strictly
riparian forests across environmental gradients. AAF was related
negatively to ALT (first variable in importance, Appendix S4), domi
nating below 600 m (Figure 5m), while WPEF increases its presence
above 500 m (Figure 5p). AAF maintained a positive relation with
PRE (Figure 5n), being the second and fourth variable in importance
(Appendix S4), while WPEF had a strong negative relation with this
variable (Figure 5s). The relation of mixed forest to ALT and PRE was
similar to that of AAF, although it appeared at higher values of these
variables (Figure 5j,i). The main variable controlling mixed‐forest rel
ative cover, maintaining a negative concave semi‐exponential rela
tion, is R AD (Appendix S4, Figure 5i). The other two non‐riverine
forests modelled showed a positive relationship to ALT (Table 3),
although beech forests showed a higher preference for increasng al
titude than oak forests (Figure 5b,e). PAS showed a marked negative
relation with both forests types, but was stronger for beech forests
(Figure 5a,f). Summer maximum temperature average (avTMXs) was
an important factor producing a clear separation between the rela
tive cove rs of these two habitats, oak forest presenting a positive re
lation when this variable exceeds 22°C (Figure 5d), and beech forest
a negative trend when avTMXs was below 24°C (Figure 5h).
4 | DISCUSSION
In this work, we have integrated a variety of modelling procedures
that allowed not only mapping the current distribution of riparian
vegetation, but also determining the main factors controlling the
observed patterns at both physiognomic and phytosociological lev
els. This study is the first region‐wide effort to map riparian forests
across multiple drainage networks using an approach based on com
munity distribution models. At the regional scale, the vegetation
pat tern is in line with other studies and mapping approaches of a dif
ferent nature, which established the adequacy and accuracy of the
proposed method. Our first hypothesis is confirmed by the results
since land use variables appear as the main drivers of physiognomic
unit distribution. Similarly, climatic variation across the area shows
a great importance in defining riparian forest composition together
with other variables related to the hydrological regime.
4.1 | Riparian forest mapping
We applied a consistent geomorphology‐based method to create the
drainage network and delimitate the riparian zone. Contrary to using
fixed buffers, this method reduces inaccuracies in the estimation of
the riparian forest extent because it does not include other areas of
vegetation that are not associated with fluvial and phreatic dynam
ics (Aguiar, Fernandes, & Ferreira, 2011). Within this area, riparian
maps allow differentiating vegetation structure together with habi
tat types, in contrast to general mapping exercises that focus only
on land use and land cover typologies (Klemas, 2013; Tiede, Lang,
Albrecht, & Hölbling, 2010). The strength of our approach relies on
the combination of predictors from different sources, from environ
mental limiting factors to remote sensing, for building the models
of habitat distribution (Álvarez‐Martínez et al., 2018). It allows to
overcome the high spatial variability and reduced width of riparian
areas (Corbane et al., 2015; Fernandes, Aguiar, & Ferreira, 2011) and
avoid the “salt and pepper” effect, common in pixel‐based mapping
approaches covering complex landscape mosaics and based on high
to medium resolution data (Ke, Quackenbush, & Im, 2010).
Validation results demonstrated the adequacy of the methods
applied for mapping riparian forests and associated vegetation
types at both physiognomic and phytosociological levels. In Álvarez‐
Martínez et al. (2018), an independent fieldwork validation campaign
by expert botanists reached over 70% overall accuracy when defin
ing vegetation composition at the habitat level. At a regional scale,
comparisons made against the NFM and CRC maps highlighted the
discrimination between forest and non‐forest areas, with classifica
tion errors being almost negligible. Nevertheless, individual forest
habitat maps achieved different accuracy levels. Large validation er
rors of some types were noted, mainly related to the low amount of
ground reference data.
The 30‐m resolution of forest models is unable to define ac
curately riparian forests in narrow riparian belts (i.e. headwaters;
Kl ema s, 2013). In this sense, it wou ld be inte res ting to con duct a sen
sitivity analysis to determine at which thresholds of riparian width
the co‐occurrence model overestimates or underestimates the for
est cover at the pixel level. In some Mediterranean environments,
remnant riparian forests as well as other formations distributed
dispersedly in small patches (e.g. invasive species) frequently have
widths below 15–20 m, which could be a limiting factor in applying
this methodology properly. Despite this, the good results obtained
in highly modified open valleys as well as on the southern face of
the Cantabrian Mountains back up the use of this methodology in
other locations. Next steps should focus on increasing the spatial
518 
|
    
Applied Vegetation Science PÉREZ‐SILOS Et aL.
(and temporal) resolution of vegetation maps by using e.g. Sentinel
2 imagery, with pixel sizes of 10 m and available worldwide every
five days, and even other VHR sensors with pixels smaller that 1 m,
for opening a wide spectrum of mapping capabilities in these com
plex ecosystems. Additionally, dominant species of habitat types can
be grouped on the base of functional traits to provide a mechanis
tic understanding of the spatial variation of vegetation assemblages
(McGill, Enquist, Weiher, & Westoby, 2006). One step further, in
corporation of the lateral and subsurface dynamics of the riparian
zone into the modelling procedure at the pixel level (e.g. hydrological
pulses, soil composition, flooding frequency) may increase its pre
dictive capabilities, being also particularly useful when prescribing
restoration actions, as we explain below.
4.2 | Assessing the driving forces of riparian forests
Several studies have shown that topographic–geomorphological
(Magdaleno et al., 2014) and, at a broader scale, climatic factors
(Douda et al., 2016) are the main drivers that define the potential
distribution of vegetation composition in the riparian area. Our
results support this abiotic control at the level of composition of
forest types, identifying climatic variables and altitude as impor
tant drivers for potential vegetation within the riparian zone. RF
models clearly allocate the two main riparian habitats in the two
biogeographic domains that converge in the Cantabrian Cordillera:
AAF in the Eurosiberian region and WPEF in the Mediterranean.
The partial dependence plots were opposite between both do
mains, explaining the observed pattern through three essential
variables: PRE, ALT and cvPREsp (Figures 5m,n,p,s). Other studies
related to floodplain forests in the Iberian Peninsula have demon
strated that this biogeographic division is driven mainly by sum
mer ar idity at the phytosoci ological level , while continentalit y and
precipitation generate the division at the alliance level (Biurrun,
Campos, García‐Mijangos, Herrera, & Loidi, 2016). Moreover, the
composition of hillside forests found in the riparian area was also
expl ained by climatic variabl es. In this reg ard , oak fore st was more
abun dant in river reac hes unde r more ther mophilic and less humid
conditions (Figure 5h,g), while beech forest was more abundant
in zones with a higher precipitation and lower summer tempera
tures (Figure 5d). This oak/beech differentiation is a well‐known
pattern recognized in the Cantabrian Mountains (Díaz‐González
& Penas, 2017).
A conceptual model based on ALT explains the changes in for
est composition along riparian corridors because it drives climatic
variables (Figure 5b,e,j,n,p) or is a proxy of competence phenomena
(Aguiar & Ferreira, 2005; Karrenberg, Kollmann, Edwards, Gurnell,
& Petts, 2003). While in low–middle river reaches riparian forest
prevails in Cantabrian riparian areas (Figure 4m), in the headwater
reaches there is a predominance of hillside forest including oaks and
beeches (Figure 4i). AAF and mixed forest disappear at the highest
altitudes (Figure 5p,j) in favour of hydrophilic montane forests (oak
and beech forests; Figure 5b,e). This has been explained by Lara et
al. (2004) as due to the high capacity of these forests to compete
for water and the closed shadow generated by them (e.g. in narrow
courses they inhibit the light necessary for typical riparian commu
nities to become established). Nevertheless, this natural effect could
be intensified by the limited spatial resolution of the remote‐sensing
data (Klemas, 2013).
In relation to the spatial variability of physiognomic units, our re
sults also demonstrate that land cover types related to human man
agement (i.e. non‐vegetated and pasture land), are mainly related to
a number of variables that come together in wide (Figure 4c,d), flat
(Figure 4g) and well‐illuminated (Figure 4e) fluvial valleys (i.e. alti
tude, slope and width of the fluvial terraces). These characteristics
favour the establishment of human populations and agrarian uses
that benefit from deep and wet soils (Puerto, 1991). In turn, AAF
frequently appears together with mixed forest (Figure 5j), above all
in closer valleys or more shadowy and wetter reaches (Figure 5i)
that are less favourable for farming and grazing activities (Terradas,
2001). Therefore, our study confirms that anthropic control ulti
mately drives the distribution of riparian vegetation at a local to re
gional scale. It constrains the extension of forest patches inside the
limits established by bioclimatic patterns and drives the occurrence
of successional states (i.e. shrublands, pastures and non‐vegetated
land; Fernández, Barquín, & Raven, 2011; Kauffman & Krueger,
1984). This model of functional structuring of the riparian vegetation
has also been reported by Aguiar and Ferreira, (2005) along the river
Tagus. In this study, am ong non‐fore sted areas, ant hropogenic st ruc
tures and pastureland (Figure 4a) were found to be the most preda
tory for riparian forests (Figure 4o). Other studies conducted in the
Cantabrian (Fernández, Barquín, Álvarez‐Cabria, & Peñas, 2014) and
river Tagus regions (Portugal; Fernandes, Aguiar, & Ferreira, 2011)
highlighted a similar pattern.
4.3 | Applications in riparian forest management
Most of the qualitative elements for assessing riparian structure in
the WFD are related to the extension of the physiognomic units that
appear in the riparian zone (e.g. length, width or continuity of the
riparian forest, land cover, etc.; European Commission, 2003). Our
models produce a spatially explicit native forest distribution in the
given riparian zone of every reach, so they can be applied to assess
the hydromorphological status according to WFD. In fact, several
hydromorphological methods and/or assessment criteria applied in
different European countries for the implementation of the WFD
have used metrics or features that could be derived from our physi
ognomic unit map (e.g. percentage of tree cover in Munné & Prat,
1998; lack of riparian forest in Chandesris et al., 2008). Furthermore,
using the same data from the models but at a more specific level of
aggregation (i.e. phytosociological map), we could also estimate the
area occupied by the HD habitats. So, our approach allows achieving
systematic mapping and a long‐term monitoring system of riparian
zones, covering the needs of WFD and HD monitoring programs al
together and being replicable in any geographical area and any mo
ment in time in retrospective analyses or under future scenarios of
global change.
    
|
 519
Applied Vegetation Science
PÉREZ‐SILOS Et aL.
Other similar approaches offer metrics and integrative indices
to help design management actions and select sites for restoration
(Macfarlane et al., 2017; Rohde, Hostmann, Peter, & Ewald, 2006).
The integration of such indicators of conservation status obtained
at the reach scale could improve the ability of our models to inform
about riparian habitat quality (i.e. fulfilling HD requirements).
Our approach therefore provides a baseline for developing more
detailed studies of vegetation composition and trends, and can be
useful to inform watershed planning such as selecting areas for res
toration or conservation at both regional and local scales. Vegetation
maps make it possible, for example, to find areas that may favour forest
expansion in order to guide connectivity analyses when designing a
Blue/Green Infrastructures Networks (Kerr & Ostrovsky, 2003) or for
reducing diffuse pollution from livestock activities (Garcia et al., 2018).
5 | CONCLUSION
The joint application of modelling techniques based on remote sens
ing and the definition of riparian zones through Virtual Watersheds
algorithms allow understanding riparian landscape patterns and the
related driving forces of change. At a regional scale, the potential
distribution of riparian forests is controlled by the existence of a
biogeographic gradient in which climate is primarily responsible, set
ting a clear differentiation between Eurosiberian (AAF/mixed forest)
and Mediterranean (WPEF) riparian forest composition. However,
anthropic uses are the main agent of change of riparian landscapes,
shaping the real distribution of current land cover types in these
areas. In this regard, we have seen that riparian forests are more im
pacted in floodplains of lowlands around human settlements, mainly
due to conversions from forestry into grazing and farming uses. Our
maps therefore provide useful information that can be translated
into indicators for WFD and HD for supporting, among others, a
multi‐scale evaluation of the conservation status of habitat types,
reach‐level assessments of current conditions and watershed‐scale
restoring planning activities under global change.
AUTHOR CONTRIBUTIONS
JB and JMAM conceived the ideas and designed methodology; IPS
implemented the database and conducted the analyses, with sub
stantial input from JMAM; IPS led the writing; all authors discussed
the results and commented on the manuscript.
DATA AVAILAB ILITY STATEMENT
The primary datasets used in this study are available for the
whole study area as online data support: https ://doi.org/10.5281/
zenodo.3444285
ORCID
Ignacio Pérez‐Silos https://orcid.org/0000‐0001‐6183‐4752
José Manuel Álvarez‐Martínez https://orcid.
org/0000‐0002‐8150‐0802
José Barquín https://orcid.org/0000‐0003‐1897‐2636
REFERENCES
Aguiar, F. C., Cerdeira, J. O., Martins, M. J., & Ferreira, M. T. (2013).
Riparian forests of Southwest Europe: Are functional trait and spe
cies composition assemblages constrained by environment? Journal
of Vegetation Science, 24(4), 628–638. https ://doi.org /10.1111/
jvs.12009
Aguiar, F. C., Fernandes, M. R., & Ferreira, M. T. (2011). Riparian vegeta
tion metrics as tools for guiding ecological restoration in riverscapes.
Knowledge and Management of Aquatic Ecosystems, 402, 21. https ://
doi.org/10.1051/kmae/2011074
Aguiar, F. C., & Ferreira, M. T. (2005). Human‐disturbed landscapes:
Effects on composition and integrity of riparian woody vegetation
in the Tagus River basin. Portugal. Environmental Conservation, 32(1),
30–41. https ://doi.org/10.1017/S0376 89290 5001992
Álvarez‐Martínez, J. M., Jiménez‐Alfaro, B., Barquín, J., Ondiviela,
B., Recio, M., Silió‐Calzada, A., & Juanes, J. A. (2018). Modelling
the area of occupancy of habitat types with remote sensing.
Methods in Ecology and Evolution, 9(3), 580–593. https ://doi.
org /10.1111/2041‐210X .12925
Álvarez‐Martínez, J. M., Stoorvogel, J. J., Suárez‐Seoane, S., & de
Calabuig, E. L. (2010). Uncertainty analysis as a tool for refining
land dynamics modelling on changing landscapes: A case study in a
Spanish Natural Park. Landscape Ecology, 25(9), 1385–1404. https ://
doi.org/10.10 07/s10980 010‐9492‐z
Álvarez‐Martínez, J. M., Suárez‐Seoane, S., Palacín, C., Sanz, J., & Alonso,
J. C. (2015). Can Eltonian processes explain species distributions at
large scale? A case study with great bustard (otis tarda). Diversity and
Distributions, 21(2), 123–138. https ://doi.org/10.1111/ddi.12256
Archer, K. J., & Kimes, R. V. (2008). Empirical characterization of ran
dom forest variable importance measures. Computational Statistics
and Data Analysis, 52(4), 2249–2260. https ://doi.org/10.1016/j.
csda.2007.08.015
Barquín, J., Ondiviela, B., Recio, M., Álvarez‐Cabria, M., Peñas, F. J.,
Fern ández, D., … Jua nes, J. A. (2012). Asse ssing the cons ervati on st a
tus of alder‐ash alluvial forest and Atlantic salmon.pdf. In P. J. Boon,
& P. J. Raven (Eds.), River conservation and management (pp. 191–208).
Singapore: John Wiley & Sons.
Benda, L., Miller, D., Andras, K., Bigelow, P., Reeves, G., & Michael, D.
(2007). NetMap: A new tool in support of watershed science and
resource management. Forest Science, 53(2), 206–219. https ://doi.
org/10.1093/fores tscie nce/53.2.206
Benda, L., Miller, D., & Barquín, J. (2011). Creating a catchment scale per
spective for river restoration. Hydrology and Earth System Sciences,
15(9), 2995–3015. https ://doi.org/10.5194/hess‐15‐2995‐2011
Biurrun, I., Campos, J. A., García‐Mijangos, I., Herrera, M., & Loidi, J.
(2016). Floodplain forests of the Iberian Peninsula: Vegetation clas
sification and climatic features. Applied Vegetation Science, 19(2),
336–354. https ://doi.org/10.1111/avsc.12219
Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. https
://doi.org/10.1023/A:10109 3340 4324
Camporeale, C., & Ridolfi, L. (2006). Riparian vegetation distribution
induced by river flow variability: A stochastic approach. Water
Resources Research, 42(10), 1–13. https ://doi.org/10.1029/2006W
R004933
Chandesris, A ., Mengin, N., Malavoi, J. R., Souchon, Y., Pella, H., &
Wasson, J. G. (2008): Systeme Relationnel d’Audit de l’Hydromorpholo
gie des Cou rs d’Eau. Principes et methodes, v3.1 (P. 81). Cemagref, Lyon
Cedex, France.
520 
|
    
Applied Vegetation Science PÉREZ‐SILOS Et aL.
Clarke, S. E., Burnett, K. M., & Miller, D. J. (2008). Modeling streams and
hydrogeomorphic attributes in Oregon from digital and field data.
Journal of the American Water Resources Association, 44(2), 459–477.
https ://doi.org /10.1111/j.1752‐1688.20 08.0 0175. x
Corbane, C., Lang, S., Pipkins, K., Alleaume, S., Deshayes, M., García
Millán, V. E., … Michael, F. (2015). Remote sensing for mapping
natural habitats and their conservation status ‐ New opportunities
and challenges. International Journal of Applied Earth Observation
and Geoinformation, 37, 7–16. htt ps ://doi.org/10.1016/j.
jag.2014.11.005
Díaz‐González, T. E., & Penas, Á. (2017). The High Mountain Area of
Northwestern Spain: The Cantabrian Range, the Galician‐Leonese
Mountains and the Bierzo Trench. In J. Loidi (Ed.), The vegetation
of the Iberian Peninsula (pp. 251–321). Cham, Switzerland: Springer
International Publishing.
Douda, J. (2010). The role of landscape configuration in plant com
position of floodplain forests across different physiographic
areas. Journal of Vegetation Science, 21(6), 1110–1124. https ://doi.
org /10.1111/j.1654‐1103 .2010.01213.x
Douda, J., Boublík, K., Slezák, M., Biurrun, I., Nociar, J., Havrdová, A.,
… Zimmermann, N. E. (2016). Vegetation classification and bioge
ography of European floodplain forests and alder carrs. Applied
Vegetation Science, 19(1), 147–163. http s ://doi .org/10.1111/
avsc.12201
ESRI (2011). Arc GIS desk top: Release 10. Redlands, CA, USA: Environmental
Systems Research Institute.
European Commission (2003). Monitoring under the water framework di‐
rective. Common implementation strategy for the water framework di‐
rective (2000/60/EC). Luxembourg: European Communities.
Fernandes, M. R., Aguiar, F. C., & Ferreira, M. T. (2011). Assessing
riparian vegetation structure and the influence of land use
using landscape metrics and geostatistical tools. Landscape and
Urban Planning, 99(2), 166–177. https ://doi.org/10.1016/j.landu
rbplan.2010.11.001
Fernández, D., Barquín, J., Álvarez‐Cabria, M., & Peñas, F. J. (2012).
Quantifying the performance of automated GIS‐based geomorpho
logical approaches for riparian zone delineation using digital eleva
tion models. Hydrolog y and Earth System Sc iences, 16(10), 3851–3862.
ht tps ://doi .or g/10 .5194/h ess‐16‐38 51‐2012
Fer nández , D., Ba rq uín, J., Álvare z‐Cabria, M., & Peñas, F. J. (2014). La nd‐
use coverage as an indicator of riparian quality. Ecological Indicators,
41, 165–174. https ://doi.org/10.1016/j.ecoli nd.2014.02.0 08
Fernández, D., Barquín, J., & Raven, P. J. (2011). A review of river habi
tat characterisation methods: Indices vs. characterisation protocols.
Limnetica, 30(2), 217–234.
Garcia, A., Fernandez, H., Rolandi, M., Gultemirian, L., Sanchez, N.,
Pla, L., & Fernandez, H. R. (2018). Effect of diffuse pollution on
water quality in mountain forest streams. Forestry Research and
Engineering. International Journal, 1(1), 1–7. https ://doi.org/10.15406/
freij.2017.01.00001
Goetz, S. J. (2006). Remote sensing of riparian buffers: Past progress and
future prospects. Journal of th e American Water Reso urces Associati on,
42(1), 133–143. https ://doi.org/10.1111/j.1752‐1688.2006.tb038
29. x
Gregory, S. V., Swanson, F. J., McKee, W. A., & Cummins, K. W. (1991). An
ecosystem perspective of riparian zones. BioScience, 41(8), 540–551.
htt ps ://doi.org/10. 2307/1311607
Guillozet, K. (2015). Shade trading: An emerging riparian forest‐
based payment for ecosystem services market in Oregon, USA .
Environmental Management, 56(4), 957–970. https ://doi.org/10.1007/
s00267‐015‐0563‐4
He, K. S., Bradley, B. A., Cord, A. F., Rocchini, D., Tuanmu, M.‐N.,
Schmidtlein, S., … Pettorelli, N. (2015). Will remote sensing shape
the next generation of species distribution models? Remote Sensing
in Ecology and Conservation, 1, 4–18. https ://doi.org/10.1002/rse2.7
Hooke, J. M. (2006). Human impacts on fluvial systems in the
Mediterranean region. Geomorphology, 79(3–4), 311–335. https ://
doi.org/10.1016/j.geomo rph.2006.06.036
Ilhardt, B. L ., Verr y, E. S., & Palik, P. (200 0). Defining riparian areas. In E.
S. Verry, J. W. Hornbeck, & C. A. Dollof (Eds.), Riparian management in
forests of the continental Eastern United States. Boca Raton, FL: Lewis
Publishers.
Ivits, E., Cherlet, M., Mehl, W., & Sommer, S. (2009). Estimating the eco
logical status and change of riparian zones in Andalusia assessed by
multi‐temporal AVHHR datasets. Ecological Indicators, 9(3), 422–431.
https ://doi.org/10.1016/j.ecoli nd.2008.05.013
Jeong, S. G., Mo, Y., Kim, H. G., Park, C. H., & Lee, D. K . (2016). Mapping
riparian habitat using a combination of remote‐sensing techniques.
International Journal of Remote Sensing, 37(5), 1069–1088 . ht tps ://doi.
org /10.1080/014 31 161.2016.1142685
Johansen, K., Phinn, S., & Witte, C. (2010). Mapping of riparian
zone attributes using discrete return LiDAR, QuickBird and
SPOT‐5 imagery: Assessing accuracy and costs. Remote Sensing
of Environment, 114(11), 2679–2691. https ://doi.org/10.1016/j.
rse.2010.06.004
Karrenberg, S., Kollmann, J., Edwards, P. J., Gurnell, A. M., & Petts, G. E.
(20 03). Patterns in woody vege tation along the active zone of a near‐
natural Alpine river. Basic and Applied Ecology, 4(2), 157–166. https ://
doi.org/10 .1078/1439‐1791‐00123
Kauffman, J. B., & Krueger, W. C. (1984). Livestock riparian impacts on
ecosystems and streamside management implications..a review.
Journal of Range Management, 37(5), 430–438.
Ke, Y., Quackenbush, L. J., & Im, J. (2010). Synergistic use of QuickBird
multispectral imagery and LIDAR data for object‐based forest spe
cies classification. Remote Sensing of Environment, 114(6), 1141–1154.
https ://doi.org/10.1016/j.rse.2010.01.002
Kerr, J. T., & Ostrovsky, M. (2003). From space to species: Ecological ap
plications for remote sensing. Trends in Ecology and Evolution, 18 (6),
299–305. https ://doi.org/10.1016/S0169‐5347(03)00071‐5
Klemas, V. (2013). Remote sensing of coastal and wetland biomass: An
overview. Journal of Coastal Research, 29(5), 1016–1028. https ://doi.
org /10.2112/JCOAS TRES‐D‐11‐00197.1
Kuhn, S., Egert, B., Neumann, S., & Steinbeck, C. (2008). Building blocks
for automated elucidation of metabolites: Machine learning meth
ods for NMR prediction. BMC Bioinformatics, 9, 400. https ://doi.
org /10.1186/1471‐2105‐9‐4 00
Lara, F., Garilleti, R., & Calleja, J. A . (200 4). La vegetación de ribera de la
mitad norte española (Monografía). Madrid, Spain: Centro de Estudios
y Experimentación de Obras Públicas.
Liaw, A., & Wiener, M. (2015). Title Breiman and Cutler’s Random Forests
for Classification and Regression Description Classification and regres
sion based on a forest of trees using random inputs.
Logan, P., & Furse, M. (2002). Preparing for the European Water
Framework Directive ‐ Making the links between habitat and aquatic
biota. Aquatic Conservation: Marine and Freshwater Ecosystems, 12(4),
425–437. https ://doi.org/10.1002/aqc.535
Macfarlane, W. W., Gilbert, J. T., Jensen, M. L., Gilber t, J. D., Hough‐Snee,
N., McHugh, P. A ., … Bennett, S. N. (2017). Riparian vegetation as
an indicator of riparian condition: Detecting departures from historic
condition across the North American West. Journal of Environmental
Management, 202, 447–460. https ://doi.org/10.1016/j.jenvm
an.2016.10.054
Magdaleno, F., Blanco‐Garrido, F., Bonada, N., & Herrera‐Grao, T. (2014).
How are riparian plants distributed along the riverbank topographic
gradient in Mediterranean rivers? Application to minimally altered
river stretches in Southern Spain. Limnetica, 33(1), 121–138.
Magdaleno, F., & Fernández‐Yuste, J. A. (2013). Evolution of the riparian
forest corridor in a large Mediterranean river system. Riparian Ecology
and Conservation, 1(November), 36–45. https ://doi.org/10.2478/
remc‐2013‐0005
    
|
 521
Applied Vegetation Science
PÉREZ‐SILOS Et aL.
McGill, B. J., Enquist, B. J., Weiher, E., & Westoby, M. (20 06).
Rebuilding community ecology from functional traits. Trends in
Ecology and Evolution, 21(4), 178–185. https ://doi.org/10.1016/j.
tree.2006.02.002
Michez, A., Piégay, H., Lejeune, P., & Claessens, H. (2017). Multi‐temporal
monitoring of a regional riparian buffer network (>12,000 km) with
LiDAR and photogrammetric point clouds. Journal of Environmental
Management, 202, 424–436. https ://doi.org/10.1016/j.jenvm
an.2017.02.034
Michez, A., Piégay, H., Lisein, J., Claessens, H., & Lejeune, P. (2016).
Classification of riparian forest species and health condition using
multi‐temporal and hyperspatial imagery from unmanned aerial sys
tem. Environmental Monitoring and Assessment, 188(3), 146. https ://
doi.org/10.10 07/s10661‐015‐4996‐2
Munné, A., & Prat, N. (1998). QBR: Un índice rápido para la evaluación
de la calidad de los ecosistemas de ribera. Tecnología Del Agua, 175,
20 37.
Naiman, R. J., Decamps, H., & McClain, M. E. (2010). Rip aria: Ecology, con‐
servation, and management of streamside communities. Amsterdam,
The Netherlands: Elsevier. Retrieved from https ://books.google.es/
b o o k s ? x m l : i d = N a Q 7 n w E A C A A J
Naiman, R. J., Decamps, H., & Pollock, M. (1993). Ecological Society
of America the role of riparian corridors in maintaining regional
biodiversity. Ecological Applications, 3(2), 209–212. https ://doi.
org /10. 2307/1941822
NRC (National Research Council) (2002). Riparian areas: Functions and
strategies for management. Washington DC: National Research
Council.
Olea, P., & Mateo‐Tomás, P. (2016). Exploiting virtual globes for ecology
and conservation in the digital era. Frontiers in Ecology, 14, 11–12.
https ://doi.org/10.1002/FEEOlea
Olson, J. R., & Hawkins, C. P. (2012). Predicting natural base‐flow stream
water chemistry in the western United States. Water Resources
Research, 48(2), 1–19. https ://doi.org/10.1029/2011W R011088
Palmer, M. A., Bernhardt, E. S., Allan, J. D., Lake, P. S., Alexander, G.,
Brooks, S., … Sudduth, E. (2005). Standards for ecologically success
ful river restoration. Journal of Applied Ecology, 42(2), 208–217. https
://doi.org/10.1111/j.1365‐2664.2005.01004.x
Petts, G. E., Möller, H., & Roux, A. L. (1989). Historical change of large allu‐
vial rivers. Chichester, UK: Western Europe. (John Wiley).
Phillips, S. B., Aneja, V. P., Kang, D., & Arya, S. P. (2006). Maximum en
tropy modeling of species geographic distributions. International
Journal of Global Environmental Issues, 6(2–3), 231–252. https ://doi.
org/10.1016/j.ecolm odel.2005.03.026
Poff, B., Koestner, K. A., Neary, D. G., & Henderson, V. (2011). Threats to
riparian ecosystems in Western North America: An analysis of existing
literature. Journal of the American Water Resources Association, 47(6),
1241–1254. https ://doi.org/10.1111/j.1752‐1688 .2011.0 0571.x
Puerto, A. (1991). Síntesis ecológica de los productores primarios
del sistema de dehesas en el espacio y en el tiempo. In J. Gómez
Gutiérrez (Ed.), El libro de la dehesa salamantina (pp. 583‐632).
Salamanca, Spain:Junta de Castilla y León.
R Development Core Team. (2011). R: A language and environment for
statistical computing. Vienna, Austria: R Foundation for Statistical
Computing.
Richardson, D. M., Holmes, P. M., Esler, K. J., Galatowitsch, S. M.,
Stromberg, J. C., Kirkman, S. P., … Hobbs, R. J. (2007). Riparian
vegetation: Degradation, alien plant invasions, and restoration
prospects. Diversity and Distributions, 13, 126–139. https ://doi.
org /10.1111/j.1472‐46 42.200 6.00314.x
Rocha, J., Hughes, S. J., Almeida, P., Garcia‐Cabral, I., Amich, F., & Crespí,
A. L. (2015). Contemporary and future distribution patterns of fluvial
vegetation under different climate change scenarios and implica
tions for integrated water resource management. Ecological Research,
30(6), 989–1003. https ://doi.org/10.1007/s11284‐015‐1300‐4
Rohde, S., Hostmann, M., Peter, A., & Ewald, K. C. (2006). Room for rivers:
An integrative search strategy for floodplain restoration. Landscape
and Urban Planning, 78(1–2), 50–70. https ://doi.org/10.1016/j.landu
rbplan.2005.05.006
Salinas, M. J., & Casas, J. J. (2007). Riparian vegetation of two semi‐arid
Mediterranean rivers: Basin‐scale responses of woody and herba
ceous plants to environmental gradients. Wetlands, 27(4), 831–845.
https ://doi.org/10.1672/0277‐5212(2007)27[831:RVOTS M]2.0.CO;2
Seavy, N. E., Viers, J. H., & Wood, J. K. (2009). Riparian bird response
to vegetation structure: A multiscale analysis using LiDAR measure
ments of canopy height. Ecological Applications, 19(7), 1157–1167.
https ://doi.org/10.1890/08‐1124.1
Stella, J. C., Rodrı, P. M., Dufour, S., & Bendix, J. (2013). Riparian vegetation
research in Mediterranean‐climate regions: Common patterns, eco
logical processes, and considerations for management. Hydrobiologia
719, 291–315. ht tps ://doi.org/10.1007/s10750‐012‐1304‐9
Terradas, J. (2001). Ecología de la vegetación: de la ecofisiología de las plan‐
tas a la dinámica de comunidades y paisajes.
Tiede, D., Lang, S., Albrecht, F., & Hölbling, D. (2010). Object‐based
class modeling for cadastre‐constrained delineation of geo‐objects.
Photogrammetric Engineering & Remote Sensing, 76(2), 193–202. https
://doi.org/10.14358/ PERS.76. 2.193
VV.AA. (2009). Bases ecológicas preliminares para la conservación de los
tipos de hábitat de inte‐ rés comunitario en España. Madrid: Ministerio
de Medio Ambiente, y Medio Rural y Marino.
Wasson, J.‐G., Villeneuve, B., Iital, A., Murray‐bligh, J., Dobiasova, M.,
Bacikova, S., Chandesris, A. (2010). Large‐scale relationships be
tween basin and riparian land cover and the ecological status of
European rivers. Freshwater Biology, 55(7), 1465–1482. https ://doi.
org /10.1111/j.1365‐2427.2010.0244 3.x
Weissteiner, C., Ickerott, M., Ott, H., Probeck, M., Ramminger, G., Clerici,
N., … de Sousa, A. (2016). Europe’s green ar teries‐A continental
dataset of riparian zones. Remote Sensing, 8(11), 1–27. https ://doi.
org /10.3390/r s811 0925
SUPPORTING INFORMATION
Additional supporting information may be found online in the
Supporting Information section at the end of the article.
Appendix S1 Predictor variables used in the modelling of habitat
types in the Cantabria region.
Appendix S2 Predictor variables integer in the synthetic river net
work to determine the drivers of the spatial pattern of riparian
forests.
Appendix S3 Independent validation of the riparian vegetation map.
Appendix S4 Table with the Increment Node Purity index values for
each random forest models.
How to cite this article: Pérez‐Silos I, Álvarez‐Martínez JM,
Barquín J. Modelling riparian forest distribution and
composition to entire river networks. Appl Veg Sci.
2019;22:508–521. https ://doi.org/10.1111/avsc.12458
... Results showed that 3D LiDAR point clouds have high potential to create indicators with good accuracy. Perez-Silos et al. (2019) mapped and modeled riparian forest distribution and composition around rivers in Cantabrian watersheds in Spain using remote sensing data (Landsat 8 OLI and LiDAR PNOA) and random forest algorithm. To identify the most important factors determining the spatial patterns of riparian forests, relation between environmental factors and riparian vegetation was investigated using random forest algorithm. ...
Article
Full-text available
Riparian forests in Iran are valuable ecosystems which have many ecological values. Because of destruction of these forests in recent years, providing spatio-temporal information from area and distribution of these ecosystems has been receiving much attention. This study was performed for mapping distribution, area and density of riparian forests in southern Iran using Sentinel-2A, Google Earth, and field data. First Sentinel-2A satellite image of the study area was provided. The field work was performed to take the training areas and to assess the forest density of riparian forests in Khuzestan province. In the first part of this study, after selecting training areas as pixel-based samples on the Sentinel-2A satellite image, supervised classification of image was performed using support vector machine (SVM) algorithm to classify the distribution of riparian forests. After classification of Sentinel-2A satellite image, the boundary of riparian forests map was checked and corrected on Google Earth images. In the second part of this study, field data, Normalized Difference Vegetation Index (NDVI), and regression model were used to assess the density of riparian forests. Finally, the accuracy of the final riparian forest map (showing both distribution and density of riparian forests) was assessed using Google Earth images. Results showed that the final riparian forest map (showing both distribution and density of riparian forests) with overall accuracy 89% and kappa index 0.81 had a good accuracy for classifying the distribution and density of riparian forests in Khuzestan province. These results demonstrate the accuracy of SVM algorithm for classifying the distribution of riparian forests and also capability of NDVI for classifying the density of riparian forests in this study. Results also showed that regression model (R 2 = 0.97) is reliable for estimating riparian forest density. The results demonstrated that there are 68447.18 ha of riparian forest around the main rivers in Khuzestan province, mainly distributed in the northwest and southeast of the province. From this area, 54694.15 ha have been covered by dense forests and 13753.03 ha by sparse forests. Results of this research have created the useful data of area, distribution and density of riparian forests in 10-m spatial resolution which is necessary for conservation and management of these forests in southern Iran. It is suggested that mapping area, distribution and density of these forests would be performed using SVM algorithm and NDVI in the certain temporal periods for protective management of these ecosystems in time series.
... Short and steep rivers with great erosive power [83] and frequent floods of varying magnitude [84] characterise the area's hydrology. The biodiversity of aquatic systems in the region, including river networks and lentic elements such as lagoons, lakes, springs and peat bogs (i.e., wetlands from now on), show a high richness of diatoms, macroinvertebrates and amphibians [76,78]. ...
Article
Full-text available
Human activities have caused a significant change in the function and services that ecosystems have provided to society since historical times. In mountainous landscapes, the regulation of services such as water quality or erosion control has been impacted by land use and land cover (LULC) changes, especially the loss and fragmentation of forest patches. In this work, we develop a Remote Sensing (RS)-based modelling approach to identify areas for the implementation of nature-based solutions (NBS) (i.e., natural forest conservation and restoration) that allow reducing the vulnerability of aquatic ecosystems to siltation in mountainous regions. We used time series Landsat 5TM, 7ETM+, 8OLI and Sentinel 2A/2B MSI (S2) imagery to map forest dynamics and wetland distribution in Picos de Europa National Park (Cantabrian Mountains, northern Spain). We fed RS-based models with detailed in situ information based on photo-interpretation and fieldwork completed from 2017 to 2021. We estimated a forest cover increase rate of 2 ha/year comparing current and past LULC maps against external validation data. We applied this forest gain to a scenario generator model to derive a 30-year future LULC map that defines the potential forest extent for the study area in 2049. We then modelled the distribution of wetlands to identify the areas with the greatest potential for moisture accumulation. We used an S2 mosaic and topography-derived data such as the slope and topographic wetness index (TWI), which indicate terrain water accumulation. Overall accuracy scores reached values of 86% for LULC classification and 61% for wetland mapping. At the same time, we obtained the potential erosion using the NetMap software to identify potential sediment production, transport and deposition areas. Finally, forest dynamics, wetland distribution and potential erosion were combined in a multi-criteria analysis aiming to reduce the amount of sediment reaching selected wetlands. We achieved this by identifying the most suitable locations for the conservation and restoration of natural forests on slopes and in riparian areas, which may reduce the risk of soil erosion and maximise sediment filtering, respectively. The results show a network pattern for forest management that would allow for controlling erosion effects across space and time at three levels: one, by reducing the load that originates upslope in the absence of forest cover; two, by intersecting runoff at watercourses related to sediment transport; and three, by a lack of former barriers, by trapping erosion near to the receiving wetland systems, main river axes and contributing streams. In conclusion, the proposed methodology, which could be transferred to other mountain regions, allows to optimise investment for erosion prevention and wetland conservation by using only very specific areas of the landscape for habitat management (e.g., for NBS implementation).
... Satellite-based Earth Observation (EO) has been explored as a very promising tool to support biodiversity mapping and monitoring and thus make advancements in the Aichi Biodiversity Targets (Álvarez-Martínez et al., 2018;Pérez-Silos et al., 2019). The major benefits of using EO data include repeatable coverage allowing for consistent and synoptic monitoring of otherwise inaccessible areas (O'Connor et al., 2015). ...
Article
Full-text available
Serpentine soils are globally rare, covering less than 1% of the land surface, but frequently harbouring large numbers of narrow endemic species. Many serpentine areas are endangered and would benefit from actions informed by species distribution models (SDMs). Here we assessed, for the first time, the potential use of remotely sensed descriptors of water and carbon cycles derived from Copernicus Sentinel‐2 images for fine‐scale habitat characterization and mapping of narrow endemic serpentine‐tolerant species. We focused on three endemic species with a very limited distribution in the Serra de Careón ultramafic outcrop, a regional plant biodiversity hotspot in Spain. We developed SDMs based solely on remotely sensed descriptors of plant water content and habitat dynamics estimated from multitemporal Sentinel‐2 images. The predictive capacity of the models was very high (AUC > 0.9, sensitivity > 90 and specificity > 84), even for projections beyond the range of data used for calibration (AUC > 0.8, sensitivity > 77 and specificity > 81). Overall, the suitable habitat areas predicted by the models were similar for all three species (Schoener’s D metric > 0.7). The annual mean and coefficient of variation of the Normalized Difference of Water Index (NDWI) provided a good representation of the hydromorphic soils with serpentine clay minerals and characterized by the formation of temporary pools during winter months. The annual mean and coefficient of variation of the Modified Soil Adjusted Vegetation Index (MSAVI) were also important predictors owing to their capacity to capture the serpentine habitat conditions, characterized by open vegetation and highly exposed soil. These findings confirm the potential of ecosystem functioning descriptors derived from spectral Sentinel‐2 indices as a cost‐effective means of characterizing the ecological niche of endemic plant species in serpentine habitats and as a promising tool to support their conservation and monitoring.
Article
Full-text available
Agriculture is the main source of surface and groundwater pollution. It increases suspended solids and organic matter concentration with the subsequent eutrophication. Citrus and sugarcane are the predominant crops in the subtropical area under study and the agrochemicals used in different periods of the year. This study was carried out taking into consideration main regional drivers as well as precipitation regime (seasonal rains) and ground conditions. Two monitoring sites were established: one that reflected conditions not affected by agricultural activities (upper watershed) and another affected by agricultural diffuse pollution (downstream) in the middle of the watershed. The Seco and Famaillá rivers show the effects of diffuse pollution caused by land use. There is a differential impact in river water quality that depends on the proportion of altered basins, riparian forest and kind of crop. Even though the measured physicochemical variables do not go beyond the established limits to be classified as contaminants, these monitoring stations are useful in the det
Article
Full-text available
Riparian buffers are of major concern for land and water resource managers despite their relatively low spatial coverage. In Europe, this concern has been acknowledged by different environmental directives which recommend multi-scale monitoring (from local to regional scales). Remote sensing methods could be a cost-effective alternative to field-based monitoring, to build replicable “wall-to-wall” monitoring strategies of large river networks and associated riparian buffers. The main goal of our study is to extract and analyze various parameters of the riparian buffers of up to 12,000 km of river in southern Belgium (Wallonia) from three-dimensional (3D) point clouds based on LiDAR and photogrammetric surveys to i) map riparian buffers parameters on different scales, ii) interpret the regional patterns of the riparian buffers and iii) propose new riparian buffer management indicators. We propose different strategies to synthesize and visualize relevant information at different spatial scales ranging from local (<10 km) to regional scale (>12,000 km). Our results showed that the selected parameters had a clear regional pattern. The reaches of Ardenne ecoregion have channels with the highest flow widths and shallowest depths. In contrast, the reaches of the Loam ecoregion have the narrowest and deepest flow channels. Regional variability in channel width and depth is used to locate management units potentially affected by human impact. Riparian forest of the Loam ecoregion is characterized by the lowest longitudinal continuity and mean tree height, underlining significant human disturbance. As the availability of 3D point clouds at the regional scale is constantly growing, our study proposes reproducible methods which can be integrated into regional monitoring by land managers. With LiDAR still being relatively expensive to acquire, the use of photogrammetric point clouds combined with LiDAR data is a cost-effective means to update the characterization of the riparian forest conditions.
Article
Full-text available
One of the major challenges in river restoration is to identify the natural fluvial landscape in catchments with a long history of river control. Intensive land use on valley floors often predates the earliest remote sensing: levees, dikes, dams, and other structures alter valley-floor morphology, river channels and flow regimes. Consequently, morphological patterns indicative of the fluvial landscape including multiple channels, extensive floodplains, wetlands, and fluvial-riparian and tributary-confluence dynamics can be obscured, and information to develop appropriate and cost effective river restoration strategies can be unavailable. This is the case in the Pas River catchment in northern Spain (650 km<sup>2</sup>), in which land use and development have obscured the natural fluvial landscape in many parts of the basin. To address this issue we used computer tools to examine the spatial patterns of fluvial landscapes that are associated with five domains of hydro-geomorphic processes and landforms. Using a 5-m digital elevation model, valley-floor surfaces were mapped according to elevation above the channel and proximity to key geomorphic processes. The predicted fluvial landscape is patchily distributed according to hillslope and valley topography, river network structure, and channel elevation profiles. The vast majority of the fluvial landscape in the main segments of the Pas River catchment is presently masked by human infrastructure, with only 15% not impacted by river control structures and development. The reconstructed fluvial landscape provides a catchment scale context to support restoration planning, in which areas of potential ecological productivity and diversity could be targeted for in-channel, floodplain and riparian restoration projects.
Article
Full-text available
Riparian zones represent ecotones between terrestrial and aquatic ecosystems and are of utmost importance to biodiversity and ecosystem functions. Modelling/mapping of these valuable and fragile areas is needed for improved ecosystem management, based on an accounting of changes and on monitoring of their functioning over time. In Europe, the main legislative driver behind this goal is the European Commission’s Biodiversity Strategy to 2020, on the one hand aiming at halting biodiversity loss, on the other hand enhancing ecosystem services by 2020, and restoring them as far as is feasible. A model, based on Earth Observation data, including Digital Elevation Models, hydrological, soil, land cover/land use data, and vegetation indices is employed in a multi-modular and stratified approach, based on fuzzy logic and object based image analysis, to delineate potential, observed and actual riparian zones. The approach is designed in an open modular way, allowing future modifications and repeatability. The results represent a first step of a future monitoring and assessment campaign for European riparian zones and their implications on biodiversity and on ecosystem functions and services. Considering the complexity and the enormous extent of the area, covering 39 European countries, including Turkey, the level of detail is unprecedented. Depending on the accounting modus, 0.95%–1.19% of the study area can be attributed as actual riparian area (considering Strahler’s stream orders 3–8, based on the Copernicus EU-Hydro dataset), corresponding to 55,558–69,128 km2. Similarly, depending on the accounting approach, the potential riparian zones cover an area about 3–5 times larger. Land cover/land use in detected riparian areas was mainly of semi-natural characteristics, while the potential riparian areas are predominately covered by agriculture, followed by semi-natural and urban areas.
Article
Full-text available
Light detection and ranging (lidar) and object-oriented classification (OOC) can be used to overcome the shortcomings of the traditional pixel-based classification (PBC) of coarse spatial resolution data, such as Landsat data, for habitat mapping in riparian zones. The purposes of this study were to investigate methods to classify multispectral data and lidar for riparian habitat mapping, and to identify major habitat components for two target species. The mapping of riparian habitat based on OOC and Decision Tree Classification (DTC) was carried out by merging vertical data from lidar and spectral data of high-resolution imagery. Our results showed an overall classification accuracy of 88.2%. In particular, small and continuous habitat types, such as short and tall grasses, rock outcrop and gravel, and riffles, improved the classification accuracy compared with the pixel-based methods. The habitat patches and paths for each target species were identified by incorporating the point data from the field survey and the outcomes of image classification. Our study demonstrated that the proposed methodology can be successfully used for the identification and restoration of fragmented riparian habitats, and can offer an opportunity to obtain high classification accuracies for microhabitat components in dynamic riverine areas.
Article
A current challenge of biodiversity and conservation is the estimation of the spatial extent of habitat types across broad territories. In the absence of fine-resolution maps, predictive modelling helps in assessing the spatial distribution of vegetation cover. However, such approaches are still uncommon in regional planning and management. Here, we present a framework for mapping the area of occupancy (AOO) of habitat types that allows highly suitable estimates at different scales. We model the potential AOO with abiotic variables related to topography and climate, resulting in broad AOO estimates that are subsequently downscaled to the local AOO with remote sensing. The combination of individual local AOO estimates allows the defining of the realized AOO, comprising locations with a high suitability and low uncertainty for each habitat. We applied this framework to mapping 24 protected habitat types of Natura 2000 sites in northern Spain. Local and realized AOO were highly accurate, with a 70% overall accuracy for the realized AOO. Remote sensing data, and especially LiDAR, were the most important predictors in habitat types related to forests and shrubs, followed by rock outcrops and pastures. Environmental variables were also relevant for specific habitats subject to abiotic constraints. The combination of ecological modelling with remote sensing offers multiple advantages over traditional field surveys and image interpretation, allowing the harmonization of habitat maps across large regions and through time. This is particularly useful for implementing conservation actions under Natura 2000 principles or assessing IUCN criteria for ecosystems.
Chapter
The Cantabrian Range, the Galician-Leonese mountains and the Bierzo trench comprise a territory in the north of the Iberian Peninsula on a wide variety of rocks and substrates and with significant climatic biodiversity. The area is covered by the two macrobioclimates existing in the whole of the Iberian Peninsula and the Balearic Islands: Mediterranean and Temperate, and is part of the Eurosiberian and Mediterranean biogeographical regions This variability in geology, soil and bioclimate has given rise to a highly diverse vegetation that we synthesise as follows. In the Cantabrian Range: (1) Cryorotemperate grasslands. (2) Oro-cryorotemperate vegetation on rock fields and screes. (3) Supra-orotemperate grasslands and high-mountain pastures. (4) Cantabrian high-mountain scrubland. (5) Peat-bog vegetation. (6) Forests and woodlands dominated by Fagus sylvatica, Quercus petraea, Betula celtiberica, Quercus orocantabrica, Quercus rotundifolia, Quercus pyrenaica, Quercus faginea, Juniperus thurifera, Ilex aquifolium, Tilia platyphyllos and Tilia cordata, oro-Cantabrian acer woods with ash, and riparian forests and woodlands. (7) Meso-supratemperate scrublands (broom communities, spiny scrublands, heathlands, broom scrublands). (8) Meso-supratemperate meadows and grasslands. (9) Other vegetation types (aquatic vegetation, herbaceous ruderal vegetation, and others).
Article
Mediterranean fluvial ecosystems are subjected to ever-increasing water demands and to a wide variety of other human impacts with potential negative effects on riparian vegetation. Nevertheless, few studies have compared the importance of human versus natural factors in structuring riparian vegetation in this region, particularly in semi-arid areas. We examined basin-scale responses of riparian vegetation to the main environmental gradient extracted by canonical correspondence analysis in two semi-arid Mediterranean rivers in southern Spain, considering different community attributes (woody, herbaceous, functional groups, and exotics species) and using different metrics (species composition, richness, and cover). The results show two main environmental gradients for riparian vegetation: a prime altitudinal gradient of increasing salinity towards the lower sites, and a second gradient of increasing water pollution and human physical impact parallel to the hydroperiod shortening. Species richness of woody and herbaceous life forms, and cover of woody species, reacted negatively to the increasing salinity downstream, a pattern that was driven by most functional groups. This is consistent with the key role of salinity for vegetation. Furthermore, as for other semi-arid riparian ecosystems, the decline in rainfall coupled with higher temperatures and evaporative stress at the lower elevation open-canopy sites may have constrained herbaceous species. Woody richness and cover declined with shorter hydroperiod and increasing human impacts, a response that was driven by mesic and hydric species. These results agree with the idea that surface flow permanence is an outstanding limiting factor for woody vegetation in semi-arid rivers, possibly exerting more influence on species richness than flood disturbance. However, herbs showed very little response to drought or human impacts compared to woody species, which may be related to the higher recovery rate from disturbance of herbaceous life forms, due to their shorter life-span and higher colonization rates. Generally, exotic species were positively affected by dissolved nutrients (N and P), especially exotic herbaceous richness and cover by nitrogen. As in other Mediterranean rivers, percentage richness of exotics in our rivers was substantially lower compared to fluvial ecosystems in other regions.
Article
Floodplain riparian ecosystems support unique vegetation communities and high biodiversity relative to terrestrial landscapes. Accordingly, estimating riparian ecosystem health across landscapes is critical for sustainable river management. However, methods that identify local riparian vegetation condition, an effective proxy for riparian health, have not been applied across broad, regional extents. Here we present an index to assess reach-scale (500 m segment) riparian vegetation condition across entire drainage networks within large, physiographically-diverse regions. We estimated riparian vegetation condition for 53,250 km of perennial streams and rivers, 25,685 km in Utah, and 27,565 km in twelve watersheds of the interior Columbia River Basin (CRB), USA. We used nationally available, existing land cover classification derived from 30 m Landsat imagery (LANDFIRE EVT) and a modeled estimate of pre-European settlement land cover (LANDFIRE BpS). The index characterizes riparian vegetation condition as the ratio of existing native riparian vegetation cover to pre-European settlement riparian vegetation cover at a given reach. Roughly 62% of Utah and 48% of CRB watersheds showed significant (>33%) to large (>66%) departure from historic condition. Riparian vegetation change was predominantly caused by human land-use impacts (development and agriculture), or vegetation change (native riparian to invasive or upland vegetation types) that likely resulted from flow and disturbance regime alteration. Through comparisons to ground-based classification results, we estimate the existing vegetation component of the index to be 85% accurate. Our assessments yielded riparian condition maps that will help resource managers better prioritize sites and treatments for reach-scale conservation and restoration activities.