The Full Picture

[

] 260

spectral resolution mapping of biodiversity related activities. However,

alone, even systematic satellite-based biodiversity monitoring is not

able to reveal changes in biodiversity (past data records are partially

incomplete, discontinuous, or simply not long enough to monitor rele-

vant changes). On the predictive side, Earth observation is well suited

for now-casting applications, but cannot predict the future evolution

of ecosystems.

Continental, or even global, scale Earth observation at larger spatial

resolution (typically 0.25 to 1 kilometre) using instruments such as

MERIS on ENVISAT, MODIS on Aqua/Terra, or NOAA/AVHRR, cannot

be directly used for individual and indicator species identification.

However, these data are well suited to derive dominant species compo-

sition or depict abundances of plant functional types (PFTs).

Spatio-temporal disaggregation schemes, evidential reasoning and

data assimilation are often part of such applications. In addition, given

that sufficient temporal resolution data are available, products such

as vegetative change cover and species-specific phenological indicators

may be further derived. These products are well suited to initialize

and parameterize ecological models that predict vegetation develop-

ment or dynamics. However, not many of the existing vegetation

models are ready to directly assimilate Earth observation data.

Regional scale biodiversity assessment, which has a typical spatial

resolution between 0.5 and 30 metres, has been pioneered by Landsat

sensors. In this category, instruments such as Landsat ETM+, CHRIS

on PROBA, EO-1 Hyperion, HyMap, and NASA JPL AVIRIS are named

most frequently. Semantic interoperability and proper interfacing

sampling design, as well as schemes between ecological and Earth

observation activities are the most important factors driving the success

of this bottom-up scaling-based approach.

Indicator species mapping is only possible if the spatial resolution

is at a fraction of the species size, otherwise only dominant species

mapping can be performed. Recent imaging spectrometers have

contributed to the mapping of quantitative vegetation biochemical

and structural parameters (such as, concentrations of leaf biochem-

istry, assessment of foliar pigments), disturbance occurrence, and

invasive species. These applications have the potential to contribute

to vegetation modelling. However, many of these models (for example,

SMART/SUMO/NTM model suite to predict floristic diversity) have

been developed without Earth observation input in mind.

Combined disaggregation/aggregation scheme: towards habitat

abundance mapping

Large-scale characterization of ecosystems, landscape diversity and

functions is in increasing demand, particularly at finer spatial resolu-

tion and temporal scales. While large-scale stratification using climatic

data may be sufficient to define broad eco-regions, combining it with

landscape spatial information allows a variety of subclasses to be iden-

tified at a finer scale. Such landscapes consist of a multitude of habitats

and plant communities of coexisting biological species. A method

combining disaggregation of remotely sensed data with evidential

reasoning can produce probability maps of dominant plant species

habitats. Such maps can be re-aggregated to produce fractional abun-

dances of PFTs.

The PFTs represent important input into dynamic vegetation models

(Sitch et al., 2003),

because they have the capacity to forecast floris-

tic diversity under various scenarios. The PFT maps allow the

evaluation of results from scenarios initiated in the past (for example,

using a historical pollen distribution database) as well as to assess the

potential accuracy of future projections.

A current disaggregation scheme in development is

based on a pan-European dataset. The approach relies on

identifying all major habitats in Europe defined in accor-

dance with Annex I of the Habitat Directive, which

specifies 198 distinct habitats (Mücher et al., 2005).

The

data set is compiled using current state-of-the-art land

cover databases (namely CORINE, PELCOM and

GCL2000). Harmonization of the data sets is imple-

mented as far as possible according to the CORINE land

cover nomenclature to avoid loss of information.

Additional stratification and evidential reasoning is

used by integrating datasets such as biogeographic

regions, digital elevation, soil information and other

geographic and topographic data to finally arrive at a

probabilistic species distribution of a particular habitat.

These maps can support better management of protected

areas, allowing monitoring of the long-term stability of

ecosystems, identification of potential species reintro-

duction sites and finally, protection of the original

ecosystem species against invasive species by predicting

their potential colonization areas.

Towards dominant species mapping

Using a remote sensing-based approach of vegetation

sampling in the field may lead to significantly differing

results compared to a more traditional vegetation mapping

scheme, such as the method of Braun-Blanquet. Remote

sensing approaches will always spatially integrate infor-

mation over the instantaneous field of view (IFOV). This

results in spectrally mixed radiometric quantities affected

by the most dominant species and the fraction of non-

photosynthetic vegetation and soil in vertical projection.

Ecological sampling schemes, however, frequently

focus on indicator species sampling, where these indica-

tor species may have no significant spatial abundance in

the sampled area. Recent advances in measurement of leaf

optical properties combined with advanced radiative

transfer (RT) modelling (for example, PROSPECT/SAIL),

allow in forward mode, to-scale leaf biochemistry and

structural parameters up to higher scales. They also allow

an inverted mode to retrieve vegetation biochemical and

structural properties from the reflectance signal measured

within the sensor IFOV. Spectral libraries of leaf optical

properties can be used to spectrally unmix large areas and

derive abundances of dominant species, given that all

dominant species are represented well in the library.

Further analysis of dominant species abundance results

in solid estimates of vegetation net primary productivity

(NPP). This is approximated from remotely sensed light

use efficiency (proxied by the photochemical reflectance

index (PRI), photosynthetically active radiation (PAR),

and the fraction of PAR absorbed by photosynthetic

tissues. The latter, being basically a function of leaf area

index (LAI), is again retrievable from spectrometric data

by inversion of RT models.

These advanced remote sensing product combinations

can serve as an input to dynamic vegetation models, such

as the SMART (soil processes), SUMO (vegetation

processes and succession), and NTM (potential floristic

OCIETAL

ENEFIT

REAS

– B

IODIVERSITY