Recent experience with automated habitat mapping and possible ways forward to produce an operational monitoring procedure integrated with in situ data.

This paper emphasizes that monitoring of biodiversity should be based on real data and not only modelling, because statistically robust real life figures are required. An overriding consideration is that automated methods must be linked to in situ data to ensure validation of results. Reliable figures of change need to be obtained with robust protocols because otherwise, real changes cannot be separated from background noise.
Habitats are a measure of biodiversity that are convenient to record and also hold correlations with species (Bunce et al (2013). There are many general habitat mapping methods without strict rules, eg mapping of vegetation associations but, as Hearn et al (2011), have shown, such maps cannot be repeated reliably for monitoring. Also, despite the extensive literature on mapping using Remote Sensing (RS), limited examples have been found where monitoring has been linked to in situ data. Although land cover and non -complex habitats have been monitored successfully using RS eg Levin et al (2006), diverse habitats have not been covered.
Lang et al (2015) extrapolated from in situ data to RS (Figure 1). Data from a standardized procedure for habitat mapping (Bunce et al (2008)) from a central 1 km square were extrapolated to the eight surrounding squares but some habitats were missing, although otherwise the accuracy was acceptable. Change was not estimated. EODHAM (Lucas et al (2015), Mucher et al (2015)) was developed to map and monitor habitats with high resolution RS and optional LiDAR data. Extensive habitat maps were produced, but these were not validated in the field.
The following are possible ways forward:
1. Record control points in the field for habitats in 1 km squares and then use those to map the whole area using Very High Resolution (VHRS) imagery. Dispersed random 1 km squares from strata could then be used for estimates
2. Due to the spatial complexity of many landscapes, VHRS imagery is required, with resolution of between 0.5 and 2.0 m, to detect small changes. Tests are required for integration with in situ data. Costs are about 10 euro per km square when obtained from archives.
3. Use images from drones at different time intervals to map 1 km squares, using control points only, to reduce mapping time. Results from dispersed samples could be linked using stratification.
The Dutch case has shown that LiDAR measurements can be used to detect changes in vegetation structure.