|
|
|
The
BIOSOS project focuses on developing a robust system for habitat
mapping and monitoring, named EO Data for Habitat Monitoring (EODHaM).
The input data sources are multi-seasonal EO measurements and
on-site data, including ancillary information and in-field
measurements
It is
based on prior generation of land cover/use (LCLU) and change
map maps and their subsequent translation to categories of
habitat appropriate to support conservation agencies and land
managers in decisions relating to protection of Natura 2000
sites.
System peculiarities
a. The system
adopts deductive learning schemes, i.e. it based on expert
knowledge elicitation from botanist, ecologists, remote sensing
experts and management authorities.to fill the gap between LCLU
and habitat domains. Ontologies are used to formally represent
the expert knowledge:
-
For LCLU and
habitat classes description. Prior knowledge of class
spectral signatures, class phenology (i.e temporal
relations), spatial relations (e.g. adjacency, distance) as
well as class attributes (e.g. shape)
-
To translate
LCLU into habitat classes as both GHC and Annex 1 maps
-
To evaluate
the impacts that anthropic pressure may have on
biodiversity.
b. The
Food and Agricultural Organisation (FAO) Land Cover Classification Scheme
(LCCS) and the General Habitat Categories (GHCs), from which
Annex I Habitats can be defined, have been proposed for
describing LC/LU and habitat categories. LCCS is more suitable
than CORINE as the land cover/use categories can be more readily
translated to habitat categories (Tomaselli at. al, 2013)
c. EODHaM has
been implemented in python (open source software)
The
EODHaM proposed system is a 3-stage processing system (Figure
che è nella prima slide che vi ho mandato l’altro giorn, vedi
qui sotto).
-
EODHaM 1st-
stage provides robust classification of bi-temporal
radiometrically calibrated EO images into LCCS Levels 1 to
2, with this based primarily on spectral data, followed by
-
a 2nd stage
that additionally utilises contextual information to
discriminate and map classes in LCCS level 3 and beyond. Based on expert knowledge of botanists, ecologists and local
site managers, land cover/use and habitat classes are
described by the experts in terms of their temporal
characteristics and/or spatial relationships which are used
in both 2nd and 3rd third EODHaM stages. The outputs of
the 2nd stage are LCLU maps.
-
The outputs of
the 3rd stage include General Habitat Categories (GHCs),
Annex 1 habitats as well as biodiversity indices and their
trends. Thematic change maps are also provided. Once LC/LU
classes and habitats are described through a semantic
language, any site can theoretically be mapped and
subsequently monitored over time.
The
expert knowledge classification approach adopted in BIO_SOS
strongly involves end users. Consequently, the products proposed
will be more familiar to the End Users since they are built on
their expertise and can be improved as they further engage with
the process.
WP2. For each of
the sites, pre-existing information and datasets were obtained
and site management authorities have been contacted and involved
in the development of the EODHaM system. The structure and
composition of the landscape at each site and also the human
pressures and threats have been identified by stakeholders and
experts (D2.2). A workshop was held to define stakeholder needs
and to select appropriate biodiversity indicators (D2.1).
WP3. The EODHaM
system architecture was defined (D3.1) and the workflow
execution environment prototyped (D3.2, D3.3)
WP4. Metadata on
pre-existing datasets were collected and an evaluation of their
quality and relevance to the project was undertaken (D4.1). A
metadata geoportal was also developed and a methodology for
evaluating the quality of new spatial datasets was implemented
(D4.5). Two new modules have been developed within the WebGIS
platform for metadata management and sharing, to enhance
communication.
Criteria were
defined for the selection of EO data considered most appropriate
for mapping and monitoring habitats (D4.4). Connections to
previous and on-going projects were also established (D4.2).
Protocols for
in-field collection of data were defined (D4.3). Training
sessions on in-field protocols relevant to the identification of
GHCs were held in Italy, Wales, Greece and Portugal involving
also local users. In-field campaigns were performed in several
sites.
WP5. The main
achievement has been the implementation of a standardised
procedure for classifying Land Cover/Use (LC/LU) classes
according to the Food and Agriculture Organisation (FAO) Land
Cover Classification System (FAO-LCCS) from very high resolution
(VHR) satellite and airborne (i.e., LiDAR) and translating
these to general habitat categories (GHCs), from which Annex I
classes can be extracted. EODHAM system was prototyped using
commercial object-orientated software but then transferred for
use within open source software (i.e. Python). The system can
now be applied more rapidly and consistently across sites
although future refinement of the programs is needed to ensure
robustness and transferability between sites. Based on the use
of spectral indices, the EODHaM 1st stage can produce
classification of vegetation/non-vegetation or
aquatic/terrestrial surfaces (i.e., FAO-LCCS level 1 to level 2
classes (D5.3). Based on context-sensitive features, the EODHaM
2nd stage modules which had been applied to several sites can
classify FAO-LCCS Levels 3 and beyond. The approach is being
developed with reference to ontologies for LC/LU class
description. Tests have also been carried out to generate
translation ontologies of LCCS into GHC classes within Protégé.
The Dempster-Shafer theory of evidence has been collated to
handle uncertainty in the data and rules (D5.1, D5.2, D5.5).
Hyperspectral data
have been analysed in Italy and Wales. Whilst hyperspectral data
may be beneficial for classification of particular species
types/groups, the amount of information available from
time-series of VHR data allows equivalent or better
classifications to be generated (D5.4).
WP6. A comparison
of Habitat classification systems and LC/LU taxonomies (CORINE
and IGBP, FAO-LCCS), without precedent for Mediterranean, was
carried out. FAO-LCCS was considered the most appropriate for
subsequent LC/LU into habitat maps translation
(D6.1).Disambiguation rules were implemented for LC/LU to GHC
habitats translations for the implementation of the EODHaM 3rd
stage (D6.10).
Quantitative
landscape pattern analysis framework was developed to produce
composite, site/scale specific indicator set for monitoring and
buffer area identification (D6.3). Protocols were defined for
comparative habitat and landscape modelling across sites to
relate biodiversity to habitat quality, linking in-situ and EO
data (D6.2, D6.4). Based on Ecological Niche modelling (ENM),
GHCs as environmental variables appeared better than LC/LU to
explain the distribution of the target species (D6.6, D6.7).
Threat analysis
framework was established to use EO data to extract pressure
trends and impacts(D6.8).
A new runoff
connectivity model was proposed with R-script to calculate
runoff connectivity. A new spatially implicit model for studying
the effects of migration corridor size reduction on cyclic
population and provide predictive biological information at a
reduced computational cost with respect to explicit models was
designed.
WP7. Archive and
new EO data were pre-processed for several sites and fed as
input to WP5.
WP8. The web page
of the project has been set and continuously updated. Quality
Assessment Plan (QAP) is established and implemented. An
exploitation team was appointed. The first policy brief was
produced. Several papers to conferences and journals have been
submitted, some already published.
Description of
the expected final results and their potential impacts and use
The BIOSOS project
focuses on developing a robust system for habitat mapping and
monitoring, named EODHaM, with this based on prior generation of
land cover/use (LC/LU) and change map maps and their subsequent
translation to categories of habitat appropriate to support
conservation agencies and land managers in decisions relating to
protection of Natura 2000 sites. The input data sources are
multi-seasonal EO measurements and on-site data, including
ancillary information and in-field measurements. For this
purpose, the Food and Agricultural Organisation (FAO) Land Cover
Classification Scheme (LCCS) and the General Habitat Categories
(GHCs), from which Annex I Habitats can be defined, have been
proposed for describing LC/LU and habitat categories. Key
criteria in the design of the system included a) ease of use by
end users, b) use with a defined range of satellite and, in some
cases, airborne (e.g., LiDAR) data and c) low reliance on
existing datasets (e.g., LC/LU maps, cadastral (if updated) and
infrastructure layers). The key benefits of the EODHAM System
are:
-
The use
of the LCCS provides a standardized framework for LC/LU
classification that is globally recognised.
-
The LCCS
and subsequent GHC classes differ when change (e.g., in height,
cover, frequency of inundation) occurs, and hence can be used
for monitoring.
-
For end
users, the building blocks of the system can be generated as
background and the final LCCS categories then easily defined
with reference to the LCCS components.
-
The approach is applicable to any area across a range of scales and,
in the absence of some components from EO data, other
information (e.g., thematic layers) can be included.
-
The EODHaM classification and the range of indices used in its
generation can also be used as input to species distribution
models, thereby supporting the long-term monitoring of indicator
species, which can potentially be mapped from EO data.
Impacts. The products and systems generated by the BIO_SOS
project will be made available for policy decision making and
planning (e.g., scenario building) and, more specifically,
evaluating the consequences of changes within Natura 2000 sites
and their surroundings. EODHaM is compliant with on-going GEOSS,
GMES and INSPIRE initiatives and its outputs will be used for:
-
Following up impact of existing and new policy.
-
Defining
a buffer area for each Natura 2000 site and preventing habitat
fragmentation in its surrounding area. When regional authorities
elaborate and approve a plan for a protected area, a buffer zone
needs to be defined around the protected area where rules
different from the ones to be adopted within the site have to be
defined. When they enter in negotiation with local authorities
and local people, they need to support their decision with
scientific evidence of the impacts that such rules may have on
the areas and the importance of the buffer zone;
-
The use
of BIO_SOS results is not limited to the management of Natura
2000 sites but to the integration of the sustainable management
of the regional Natura 2000 network within the ordinary planning
system.
-
Innovative planning activity at a local level must try to
modifying a static municipal planning system into a dynamic
planning system. That means that indicators are needed to
establish whether the goals associated with the approved
municipal plan were met during the implementation phase. Such
indicators should provide a dynamic monitoring of the different
planning processes and an evaluation of the effectiveness of the
policies implemented.
-
Outside
Europe as well, in mega-diverse countries such as Brazil and
India, the operational flexibility that is required given the
diversity of environments, habitats, species and threats is
supported by the BIO-SOS project in that the approach is
multi-resolution and multi-scale for mapping out the land cover.
EODHaM system would enable managers to understand the pressure
issues in multiple ecological and human environments.
|
