WP leader: CEH, France Gerard, Wallingford Research Station
In WP 5 we are elaborating the processing of remotely sensed images for biodiversity and their integration with habitat data linked to biodiversity indicators.
Different EO efforts have aimed at characterising either the compositional, structural or functional aspect of the landscape using a wide range of approaches. WP5 is pursuing options highlighted by WP1 in the compositional (i.e. land cover), structural (i.e. landscape metrics and diversity indices) and functional (i.e. phenology) and is incorporating coverage of non-European Mediterranean regions in line with the overall objectives of developing and testing the biodiversity observation system to ensure compatibility with developing world requirements.
Integration of in situ with EO
In the context of biodiversity monitoring, the idea of integrating in situ with EO is that the combination of the two data set types will deliver more accurate or reliable information on biodiversity than either of the two data sets used independently.
We have identified different roles for EO when it is considered for enhancing in situ observations. It can be used as a vehicle for interpolation and generalisation by delivering full coverage, or it could be used to increase the number of in situ samples in space and time. Here, the key for success is a good link between the EO derived thematic map and the in situ habitat observations. EO could also be used to search for and highlight hotspot areas of sudden or gradual change, or to provide context where it delivers additional information on, for example, land cover composition, landscape structure or phenology, complementing the in situ species and habitat data. In all cases the concept of linking EO derived information with field data to enhance observations on biodiversity is based on the premises that a relationship exists between the composition and structure of the landscape and the diversity of habitats and the species and genotypes that may be present within.
Land cover and habitats
WP5 is planning to provide clear statements on the added value of EO by testing if data integration delivers improved estimates of biodiversity measures, in particular the SEBI indicators: (i) Trends in extent of selected ecosystems and habitats and (ii) Trends in abundance and distribution of selected species.
The observation and recording of land cover and habitats require different classification systems. Their design results from a compromise between scope of use, level of detail and spatial application. EO introduces not only full area and frequent coverage, but also a new and unique set of classification parameters. The degree in which a relationship can be established between electromagnetic signals and the thematic classes (e.g. physiognomic, floristic or ecological) required by the biodiversity monitoring community, will determine the usefulness of the EO derived thematic maps. The work of Paradella et al. [1994] suggested that physiognomy may be the most important attribute which influences the EO response of vegetation. Whilst Jakubauskas et al. [2002], Moody and Johnson [2001] and Hill et al. [submitted] have reported successful crop, vegetation and species classifications when using time series of EO to exploit differences in phenology. Many have shown that when working regionally or locally, and using EO data types and classification approaches appropriate for the local scenario, accurate and reliable and therefore useful results can be achieved [Hill and Thomson 2005, Thomson et al 2003, Bock et al 2005]. However, when continental and global biodiversity monitoring requires consistency in methodology, the variety of EO data types and approaches available is greatly reduced. As a result, the global land cover maps produced from EO have been limited to reporting the extent of major vegetation types (total number of vegetation classes ranges between 7 and 18, Table 1) at pixel sizes ranging from 1km to 300m. The class number and type and the spatial detail of these products suggest these are not suitable for detailed biodiversity or habitat monitoring.
 |
| Figure 1. Schematic showing the steps required to carry out an inter-calibration between an EO derived land cover map and in situ observations. In this example, the original EO derived land cover map has a spatial resolution of 25m and the in situ sites are 1kmx1km areas. The resulting calibrated land cover product shows the proportion of coverage of habitats within 1 km grid cells. |
A second approach which we are looking at is the inter-calibration of EO land cover/habitat maps of sample sites produced to increase the in situ samples in space and/or time. The advantage of the second approach is that it could allow for the introduction of strata specific EO mapping methods and the use of more expensive imagery (airborne, high spatial resolution imagery). In this context EBONE is looking at the role of LIDAR, Landsat TM, EO time-series analysis and high spatial resolution imagery. Key questions here are:
-
What is the optimal ratio between in situ and EO based sample observations?
- Can region specific EO mapping approaches deliver enough thematic detail (number and types of habitat classes - GHC) reliably and consistently across regions for it to be used to increase the number of samples?
A third approach which we are considering is the use of EO land cover maps to post-stratify in situ derived estimates. In this case we do not need a good link between the EO land cover data and the habitat categories mapped in situ and the quality of the land cover map does not have to be very good. We do need full coverage. The land cover map is solely used to characterise within and between zone variability and heterogeneity. Our questions are:
-
How big an estimate improvement can be achieved?
-
Does the effectiveness of post-stratification vary with landscape and environmental zone?
Landscape structure: Fragmentation and connectivity
The SEBI indicator ‘Fragmentation’ is one of the few indicators which would be difficult to measure without the bird’s eye view of EO. But to derive structural indicators such as fragmentation from an EO derived land cover map which relate to the observation made in situ several conditions have to be met:
-
The EO land cover map needs to have a link with the habitat classes identified in situ.
-
The spatial resolution of the land cover map has to reflect the scale at which groups of species (flora and fauna) operate.
Within WP5 we are developing approaches to derive measures of fragmentation and connectivity with an initial focus on forest habitats and looking at alternative options for reporting the indicator at regional and national levels (example in Figure 2). We are also assessing:
-
The implications of mismatches between EO derived land cover classes and in situ habitat classes.
-
The impact of scale determined by the original EO data.
-
The impact of the size of the area considered at any one time (i.e. from which an estimate is derived) and its relationship with in situ sample.
 |
| Figure 2. Structural indicators for forest habitat reported for each 10 km grid for an environmental zone in Southern France. The location of the 1 km2 in situ sites are shown as red dots (Source Christine Estreguil, JRC) |
Phenology
Time series of EO derived Vegetation Indices (VIs) are used to characterise the phenology of vegetation and VIs or VI derived variables such as NPP have been correlated to species diversity. WP5 believes there is a role for time-series of VIs in biodiversity monitoring and decided to pursue two lines of inquiry:
-
How can such data be exploited to improve habitat mapping.
-
How can we link observed trends in derived phenometrics (e.g. start of growing season and scenescence, length of growing season) to sample based in situ observation and thus enhance monitoring.
Species distribution models
Species distribution models that incorporate in situ and EO derived information is another form of integration that could potentially deliver improved measures of the SEBI indicator ‘Trends in abundance and distribution of selected species’ [Gillespie et al. 2008]. Here both thematic (land cover maps calibrated to GHC observations) and quantitative EO derived information such as fragmentation and phenology metrics are being considered. One caveat is that distribution models highlight areas with high probabilities of a specific species occurring which does not necessarily mean the species in question will be found in all of the areas identified.
