Visual interpretation

Dr. Helle Skånes
Department for physical geography and quaternary geology, Stockholm University
SE 106 91 Stockholm
Phone: +46 8 164773, Cellular: +46 70 561 4434, e-mail: helle.skanes@natgeo.su.se

WP T1 har tre syften: 1) att bedöma och utforska möjligheterna till visuell tolkning och särskiljande av nyckelvariabler för vegetationsegenskaper i laserdata; 2) att i samverka med WP T2 undersöka i vilken utsträckning olika vegetationstyper kan skiljas genom enkla bearbetningar av laserpunktmolnet; 3) att undersöka hur visuell tolkning av laserdata kan förbättra automatiska vegetationsklassificeringar, så att effektiva semi-automatiska produktionskedjor uppnås. Fokus ligger på habitat inom alpina miljöer, jordbrukslandskapet och kustzonen och primärt på deras klassificering och karaktärisering enligt EU:s habitatdirektivs bevarandemål. I studie 1) kommer ett antal nyckelvariabler (vertikal och horisontell struktur samt inverkan av markanvändning och ståndortsegenskaper) att studeras genom visuell tolkning av 3D laserdata. I studie 2) kommer möjligheten att med statistiska mått särskilja olika vegetationsklasser att studeras. I studie 3) kommer vi att utvärdera hur visuell tolkning i fotogrammetriska arbetsstationer kan komplettera automatiskt framställda vegetationskartor från WP T3 och i kustzonen även från WP A1. Projektets resultat kommer att publiseras vetenskapligt. I samverkan med WP C sammanställs praktiska slutsatser för en semi-automatisk arbetsprocess för tillämpad naturvårdskartering.

WP T1 has three aims: 1) to assess and explore the possibility to visually detect key vegetation properties in laser data; 2) to explore, in collaboration with WP T2, to which extent laser data derivates from the point-cloud can discriminate different key vegetation types; and 3) to combine visual and automatic vegetation classification procedures. The focus will be on habitats within alpine, agricultural, and coastal environments and deal primarily with their classification and characterisation in relation to conservation goals according to the EU Habitats Directive. In task 1 a number of key variables (vertical and horizontal structure, influence of land use, and site conditions) will be explored through visual interpretation of 3D laser data. In task 2 we will explore to which extent pre-processing of vegetation features obtained from laser data can statistically discriminate key vegetation classes. In Task 3, we will explore procedures that use a combination of automated products from WP T3 and in the coastal area also from WP A1, and visual editing in photogrammetric work stations aided by both images and laser data. The outcome of this project will be scientifically reported. In addition, practical conclusions related to semi-automated procedures targeting Swedish operational vegetation mapping will be issued with aid of WP Communication.

1    Purpose and Scope
The aim of WP T1 is to explore and document the usability of data from airborne laser systems for enhanced vegetation classification and registration of habitat qualities of importance to nature conservation and international and national environmental quality goals. WP T1 will test which environmentally relevant features can be deduced from laser data. Furthermore, it will be investigated to what extent the use of laser data can speed up current methods of vegetation mapping using photogrammetric instruments. The habitats in focus will range from the coastal transition zone between terrestrial and marine habitats (e.g. Skåne and Östergötland) through various aspects of managed and successional agricultural habitats (e.g. Halland, Västra Götaland, Öland and Dalarna) to the tree limit transition into the alpine zone (e.g. Abisko and Dalarna).

2    Contribution to the programme aim
WP T1 will serve other WPs and external applications with its initial visual interpretation tests and syntheses of key structures and vegetation types in laser data. Since WP T1 will include work on classification variables regarding the shore gradient it will require a certain overlap in variable content and spatial range with the terrestrial and marine WPs. The visual testing of laser data derivates created in WP A1-A2 and T2-T3 on top of aerial photographs and automatic segmentation from remotely sensed imagery will also benefit the progress of the overall programme goals. More detailed reference to the respective WPs will follow under tasks below.

Table 1, deliverables from WP T1

Deliverable Time
Initial report on interpretation tests of laser data in cooperation with WP A1 & T3 (Task 1) Month 12
Initial report on statistical separability of vegetation types in laser data in cooperation with T2 (Task 2) Month 18
Conference paper on interpretation tests and statistical features of laser data for key vegetation types (Task 1 and 2) Month 24
Submission of scientific paper concerning visual interpretation and statistical separability of laser data (Task 1 and 2) Month 30
Report on guidelines for key pre-classifications from laser data and automatic segmen-tation of RS data as support for visual and semi-automatic vegetation mapping (Task 3)  
Report on tests with integrated marine and terrestrial mapping in the coastal zone (in cooperation with A1) (Task 3) Month 36
Contribution to scientific paper on the use of laser data derivates and segmentation of RS in semi-automatic vegetation mapping with WP T3 (Task 3) Month 48

3    Background, Theory and Methods
Policy makers and land managers increasingly demand hard figures regarding the state and trends of biodiversity and habitat qualities. Although remote sensing and GIS methods of collecting data have greatly improved, there is still a lack of spatially detailed and consistent habitat data to meet these requirements (Bunce et al. 2007), especially those that depict the 3D structure of ecosystems providing important animal-habitat relationships (Vierling et al. 2008). WP T1 will work on the integration of high resolution aerial imagery with laser data to bridge this gap.
Important qualities of the vegetation are often hidden from visual and automatic classification in high resolution remote sensing imagery since they are typically covered by a forest canopy. Laser data can penetrate the canopy and add crucial information that will increase the degree of detail and consistency in a vegetation map. In recent studies, laser has also shown to be useful in distinguishing subtle human-made structures at the bare-earth level under the forest canopy (Bailly et al. 2008; Doneus et al. 2008; Gallagher and Josephs 2008). It would be immensely useful if recognition of former land use and human influence could be diversified, to distinguish ditches, former arable fields, and structures of former grazing, etc., and extended also to characterizing key natural site conditions. Even in open environments, such as grasslands, wetlands and coastal plains, a detailed vertical accuracy is crucial to resolve low vegetation heights and structural properties (Streutker et al. 2006, Sadro et al. 2007).

Task 1 (2009-2010): The first task will be to visually assess and explore key properties in vertical and horizontal vegetation structure and site conditions in full wave-form and point laser data (collaboration with A1-A2 and T2-T4).
A number of test sites, covering a wide range of habitat types, site conditions and conservation status will be selected and preliminarily classified and described by visual interpretation of 3D laser data, processed to enhance objects of interest. From this, a number of key indicators related to classification and conservation values are identified for further testing and evaluation by a group of interpreters. Special attention will be given to variables such as ditches, encroachments, canopy closure, vertical structure, dead wood, field layer characteristics, and site properties (e.g. boulders, fine-grained soils i.e., flat areas or geomorphologic landforms indicating high drainage properties such as eskers and glaciofluvial deposits). Indirect structural and dynamic variables which are extremely difficult to completely computerize, such as land use and land use history, will also be addressed.
The laser data and aerial imagery are at first evaluated separately, and from that, a combined method including super-imposition of laser data on CIR aerial imagery in photogrammetric stereo work stations (DAT/EM Summit Evolution) can be designed and properly evaluated to explore how well the chosen variables match structures visible in aerial photos and/or in the field. The influence of laser point density and canopy closure on the interpretation results will be assessed.

Task 2 (2009-2010): The second task (parallel with Task 1), will be an early investigation of the possibility to statistically separate different key vegetation classes, using features derived from laser scanner data (collaboration with T-T4 and NILS).
Recent studies have shown that spectral properties of laser data can be used to discriminate different types of vegetation types (cf. Díez et al 2008; Korpela 2008). Signature studies, regarding standard deviations within windows, will be performed on laser data to see if spectral properties will increase the separability. Training areas of key vegetation types will be classified in detail from air-photo interpretation and field investigations to match the signatures. The experiences from this task will be used when experiments in WP T3 and T4 are designed

Task 3 (2010-2012): The third task, divided into two subtasks, will be to combine and validate laser data derivates which have been generated within the programme in order to facilitate and strengthen visual as well as semi-automatic and automatic vegetation classification procedures, covering terrestrial and coastal habitats and their key structural qualities (collaboration with A1-A2 and T2-T4 as well as NILS, Natura 2000, Swedish vegetation Land Survey).
The focus of the first subtask will be to test how pre-processing of vegetation features obtained from laser data and high resolution aerial imagery can improve visual interpretation of vegetation in CIR stereoscopic imagery. Today, enhanced vegetation classification is hampered since interpreters use derivates from various existing topographic and thematic maps, using similar visual methods but created for different purposes and on different scales. Major improvement will be achieved if interpreters have consistent and objectively derived pre-classifications representing key properties of the vegetation, such as canopy height and density, vertical layering, detailed bare-earth model indicating soil surface structures, etc., as support in delineation of vegetation types and their properties. The combination of automatic and manual methods increases the flexibility and allows more complex analyses on cause and effect, which is especially important in a phase of method development.
The focus of the second subtask will be to set up practical tests in collaboration with WP T3 to explore how a semi-automatic mapping workflow can be improved through visual editing of known key issues with low accuracy from the automated procedures. The current methods would preferably be replaced by this approach, where visual and manual editing is restricted to decisions regarding complex vegetation types and structures. Here, the detection of ditches, low-growing shrubs and shrubs under the tree canopy is emphasized, because of the difficulty to detect those visually from aerial imagery. Tests with a new integrated map type in the coastal zone will be carried out in cooperation with WP A1.

4    Practical Relevance
The outcome of WP T1 will be implemented both as support to visual classification of aerial imagery and as components in semi-automated procedures targeting Swedish operational environmental monitoring. WP T1 will extend the work to include coastal issues of high relevance to nature conservation. The coastal transition zone is currently poorly dealt with in vegetation mapping and yet, many of its habitats are included in environmental monitoring efforts. The intersection between water and land is dynamic and ambiguous and a more precise and common delineation of the shoreline and improved classification of shore habitats would benefit terrestrial, coastal, and marine applications. Detailed needs of stakeholders within nature conservation and environmental monitoring applications, such as NILS and Natura 2000, among others, will be identified during the programme.

5    References

  • Bailly, J.S., Lagacherie, P., Millier, C., Puech, C. and Kosuth, P. 2008. Agrarian landscapes linear features detection from LiDAR: application to artificial drainage networks. International Journal of Remote Sensing 29:3489-3508.
  • Bunce, R.G.H, Metzger, M.J., Jongman, R.H.G., Brandt, J.,de Blust, G., Elena-Rossello, R., Groom, G.B., Halada, L., Hofer, G., Howard, D.C., Ková?, P., Mücher. C.A., Padoa-Schioppa, E., Paelinx, D., Palo, A., Perez-Soba, M., Ramos, I., Roche, P., Skånes, H. and Wrbka, T, 2007. A standardized procedure for surveillance and monitoring European habitats and provision of spatial data. Landscape Ecology 23:11-25.
  • Díez, A., Arozarena, A., Ormeño, S., Aguirre, J., Rodríguez, R. and Sáenz, A. 2008. Fusion and optimization of lidar and photogrammetric technologies and methodologies for cartographic production. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1. Beijing 2008, pp 349-355.
  • Doneus, M., Briese, C., Fera, M. and Janner M. 2008. Archaeological prospection of forested areas using full-waveform airborne laser scanning. Journal of Archaeological Science 35:882-893.
  • Gallagher, J.M. and Josephs, R.L. 2008. Using LiDAR to Detect Cultural Resources in a Forested Environment: an Example from Isle Royale National Park, Michigan, USA. Archaeol. Prospect. DOI: 10.1002/arp.333.
  • Korpela, I.S., 2008 Mapping understory lichens with airborne discrete-return LiDAR data. Remote Sensing of Environment. in press. Doi:10.1016/j.rse.2008.06.007.
  • Sadro, S., Gastil-Buhl, M. & Melack, J., 2007. Characterizing patterns of plant distribution in a southern California salt marsh using remotely sensed topographic and hyperspectral data and local tidal fluctuations. Remote Sensing of Environment 110 (2007) 226-239.
  • Streutker, D.R. and Glenn, N.F. 2006. LiDAR measurement of sagebrush steppe vegetation heights. Remote Sensing of Environment 102:135-145.
  • Vierling, K.T., Vierling, L.A., Gould, W.A., Martinuzzi, S. and Clawges, R.M., 2008 LIDAR: shedding new light on habitat characterization and modelling. Front Ecol Environ 6:90-98.