WP T3

Large area mapping

Dr. Mats Nilsson
Department of Forest Resource Management, SLU,
SE 901 83 Umeå, Sweden
Phone: +46 90-7868422, e-mail: mats.nilsson@srh.slu.se

Sammanfattning
WP T3 syftar till att utveckla metoder för automatisk kartering av terrester vegetation inom större områden m.h.a. digitala flygbilder och glest skannade laserdata. I arbetet ingår att utveckla automatiska segmenterings- och klassificerings-strategier för såväl trädklädda som öppna marker. Metoderna ska tillvarata de fjärranalysdata och fältdata som bedöms finnas tillgängliga inom några år, främst landstäckande digital flygbilder och de glesa laserskanner data som lantmäteriet planera att upphandla, samt fältdata från Riksskogstaxeringen och NILS. Med extern finansiering kommer dessutom kombination med optiska satellitdata att ingå. En ny doktorand avses delta i projektet

Abstract
WP T3 aims to develop methods for automated large area mapping of terrestrial vegetation, primarily using digital aerial photographs and sparsely sampled laser scanner data. The work includes development of automated strategies for segmentation and classification for forested and open lands. The methods should utilise the remote sensing data and field reference data that both exist currently, as well as data planned to be available in the near future. These data sources include, for example, high resolution digital aerial photos and nationwide sparsely sampled laser data, as well as field references from the National Forest Inventory and the NILS inventory. External financing will also be sought for the acquisition of satellite data. A new PhD student will be associated with this project.

1    Purpose and Scope
The main objective is to develop robust automated methods for vegetation mapping across large areas using data from digital aerial photos and the expected nationwide laser scanning. The developed products will be of two types: i) products produced with automated procedures, without visual interpretation (such as the nationwide kNN maps with forest variables produced in Finland and Sweden); ii) products that are developed to serve as a pre-processing step for further visual interpretation. The further development of suitable methods for visual interpretation will be done in WP T1.

2    Contribution to the programme aim
New methods for mapping terrestrial vegetation, ranging from alpine to coastal vegetation, will be developed with special focus on automatic segmentation methods and classification of vegetation in both open and tree-covered areas. WP T3 will provide methods for producing extensive vegetation maps that can be related to the distribution pattern of species, thereby increasing the understanding of habitat-species relations. WP T3 will deliver automatically estimated map products to WP T1, where the need for refinement and further post-editing of specific features will be identified.
 
Table 1. WP T3 Actitivites (A) and Deliverables (D)

Deliverable Time
A: Field survey in Remningstorp - establish a dense systematic grid of field plots for evaluating canopy height models and estimated vegetation characteristics as well as different segmentations and classifications. Month 9
D: Scientific article on the accuracy of canopy height models and estimated vegetation characteristics derived from low density laser data and multi-view-angle matched aerial photos (PhD student article #1). Month 18
A: Evaluation of segmentations based on different input data derived from low density laser data and data from digital aerial photos. Month 20
A: Delivery of map pre-processed map products to WP T1. Month 24
D: Scientific article on the classification accuracy for different vegetation types derived using features from laser scanner data extracted with tools developed in WP T2 as well as features from digital aerial photos (PhD student article #2). Month 30
D: Scientific article about how the viewing geometry in digital aerial photos affects the possibility to classify tree species and different types of vegetation (PhD student article #3). Month 36
D: Evaluation of the combined use of laser data and optical satellite data for vegetation mapping (PhD student article #4, not financed by EMMA). Month 48
D: Presentation of accuracy and methods for new types of vegetation maps. Scientific article in cooperation with WP T1. Month 48

3    Background, Theory and Methods
There is a need for a new general vegetation map of Sweden, as well as for efficient methods for generating vegetation maps over special areas. For a vegetation map to be useful for further analysis in combination with other data, the mapping criteria must be consistent across all regions and over a large ecological range. It should be sufficiently detailed, both in spatial resolution and in the number of classes. The usefulness in monitoring programs places high demands on time- and cost-efficient methods, so that the map can be regularly updated. To have access to such vegetation maps in the near future, it is important that the methods developed are based on remote sensing data that can be expected to be available within a few years time, for example, high resolution digital aerial photos (resolution < 50 cm) and the low density laser scanner data expected to be acquired by the Swedish National Land Survey to produce a new DEM for Sweden. The Swedish National Land Survey has a national policy for acquiring aerial photos covering about one-third of the land area in Sweden annually using the Z/I DMC camera. Another important data source is the national satellite database (SACCESS) with a yearly nationwide coverage of Landsat and SPOT data.
It has been shown that low density laser scanner data (approx. 1 point/m2) can be used to derive canopy height models and to estimate vegetation characteristics at stand or plot level, for example tree height and biomass with high accuracy (e.g., Næsset, 2002; Holmgren et al. 2003; Maltamo et al. 2004; Thomas et al. 2006). Canopy height models and vegetation characteristics can also be derived using 3D data obtained by multi-view-angle matching of digital photos. Ofner et al. (2006) have shown that tree heights can be estimated with a nearly similar accuracy as those obtained using laser scanner data by using multi-view-angle matching of five digital photos (15 cm resolution). The accuracy of the estimated tree heights were however lower when only three photos were matched. Still, matching of only three photographs might provide valuable information for segmentation and classification of vegetation types, especially when up-to-date laser data are not available.
Tree species information classified using spectral data from aerial photos (e.g., Bohlin et al., 2006; Holmgren et al., 2008) might also be valuable for segmentation and classification of vegetation types. It has also been shown that accurate species-specific forest stand characteristics can be derived using a combination of laser scanner data and texture measures from aerial photos (Packalén and Maltamo, 2007). Altogether this indicates that the use of digital aerial photos and laser scanner data trained with field data offers new and cost effective methods for vegetation mapping. In addition, the combined use of low density laser scanner data and satellite data from SACCESS might also be an efficient approach for mapping vegetation in Sweden.
The tasks in this WP will partially be carried out as a part of a PhD project. An application for additional funding regarding the combined use of optical satellite data from sensors like SPOT HRG and laser scanner data in vegetation mapping (Task 4) will be sent to the Swedish National Space Board. Costs for attending PhD courses will be paid by SLU. 

Task 1 - Canopy height models
The aim is to derive canopy height models and to estimate vegetation characteristics such as tree height, canopy cover, and tree species using low density laser data and data from the Z/I DMC camera acquired from various flight altitudes, as well as data from SAAB C3 Technologies (www.c3technologies.com), who create 3D data from their own camera platform. Laser data from existing test areas and the anticipated national laser scanning dataset will be used. The canopy height models and estimates of vegetation characteristics will be evaluated at a plot level using field data from a dense systematic grid of geo-located sample plots collected at the Remningstorp test site, including additional areas with broad-leaved forest near Remningstorp.

Task 2 - Segmentation
An important task will be to investigate how features extracted from canopy height models derived from both multi-view-angle matching of digital photos and laser data (Task 1) can be used for automatic, or semi-automatic, segmentation of the vegetation into homogenous patches. Vegetation characteristics such as height data, canopy cover and tree species information (derived in Task 1) might also provide valuable input data for the segmentation. Different segmentation software will be tested and compared. The segmentation approaches will be validated using field data from the dense grid of sample plots to be collected at the Remningstorp test site (see Task 1).
 
Task 3 - Classification
The aim is to investigate how accurately different segments can be classified into vegetation types using low density laser data and data derived from digital aerial photos. The classification will be made using the canopy height models and features extracted from them as well as the vegetation characteristics derived in Task 1. A method similar to the one developed by Packalén and Maltamo (2007) will be tested for predicting vegetation characteristics at plot and segment levels based on a combination of data derived from aerial photos and laser scanning. This method involves a radiometric calibration of the photos using methods such as the one presented by Tuominen and Pekkarinen (2004). The evaluation will be made for a sample of segments for which species information and other vegetation characteristics have been assessed during the field survey in Remningstorp (see Task 1). The result from this task will show which vegetation types can be classified with sufficient accuracy, and which vegetation types need to be grouped into broader classes for further analysis using visual interpretation methods developed in WP T1. The classification method will be developed in cooperation with WP T1, T2 and T4.

Task 4 – Combining laser data and optical satellite data
The aim is to investigate if a combination of low density laser scanner data and optical satellite data trained with field data from the Swedish NFI and NILS inventories can be used to map vegetation across Sweden with a sufficient accuracy. Low density laser data will be provided by the National Land Survey and satellite data is available through SACCESS. The classification accuracy will be assessed using field data from existing test sites. Some of the mapped vegetation classes will be mixtures of important vegetation types that will have to be further separated using visual interpretation methods developed in WP T1. This task is scheduled for the second phase of the project, and it will be made in cooperation with WP T1.

4    Practical Relevance
A nationwide vegetation map gives information on the occurrence and spatial distribution of different vegetation types, and it can, for example, be used for stratification in surveys where valuable and threatened species or habitats are to be monitored. For NILS, the new mapping approach makes it possible to analyze landscape patterns on the 5*5 km squares. The mapping of tasks in this WP are linked to the EBONE project (European Biodiversity Observation Network: Design of a plan for an integrated biodiversity observing system in space and time) which aims to develop a data collection system for biodiversity linked with extant data at national, regional and European levels, in which SLU participates.

5    References

  • Bohlin, J., Olsson, H, Olofsson, K., and Wallerman, J. 2006. Tree species discrimination by aid of template matching applied to digital air photos. In: proceedings of the ISPRS - EARSeL Workshop on 3D Remote sensing of forests. Vienna, February 2006. http://www.rali.boku.ac.at/3drsforestry.html
  • Holmgren, J., Nilsson, M., and Olsson, H. 2003. Estimation of tree height and stem volume on plots using airborne laser scanning. Forest Science, 49:419−428.
  • Holmgren, J., Persson, Å. and Söderman, U. 2008. Species identification of individual trees by combining high resolution LiDAR data with multi-spectral images. International Journal of Remote Sensing, 29:1537-1552.
  • Maltamo, M., Eerikäinen, K., Pitkänen, J., Hyyppä, J. and Vehmas, M. 2004. Estimation of timber volume and stem density based on scanning laser altimetry and expected tree size distribution functions. Remote Sensing of Environment. 90:319-330.
  • Næsset, E. 2002. Predicting forest stand characteristics with airborne scanning laser using a practical two-stage procedure and field data. Remote Sensing of Environment, 80:88−99.
  • Ofner, M., Hirschmugl, M., Raggam, H., and Schardt, M. 2006. 3D Stereo mapping by means of Ultracam data. In: Proceedings of the Workshop on 3D Remote Sensing in Forestry, 14-15´th February 2006, Vienna.
  • Packalén, P. and Maltamo, M. 2007. The k-MSN method for the prediction of species-specific stand attributes using airborne laser scanning and aerial photographs. Remote Sensing of Environment, 109:328-341.
  • Thomas, V., Treitz, P., McCaughey, J. H., and Morrison, I. 2006. Mapping stand-level forest biophysical variables for a mixedwood boreal forest using lidar: An examination of scanning density. Canadian Journal of Forest Research, 36:34−47.
  • Tuominen, S., and Pekkarinen, A. 2004. Local radiometric correction of digital aerial photographs for multi source forest inventory. Remote Sensing of Environment, 89:72-82.
  • Waser, L.T., Eisenbeiss, H., Kuechler, M., and Baltsavias, E. 2008. Potential and limits of airborne remote sensing data for extraction of fractional canopy cover and forest stands and detection of tree species. The international Archives of the Photogrammetry, Remote Sensing, and Spatial Information Sciences, Vol. XXXVII. Part B8. Beijing.