Monitoring using high resolution data

Dr. Johan Holmgren
Dept. of Forest Resource Management
Swedish University of Agricultural Sciences
SE-90183 Umeå, Sweden
Phone +46 90 7868602, Fax +46 13 778116
E-mail: johan.holmgren@srh.slu.se

Målet med detta arbetspaket är att utveckla metoder som kan användas för övervakning av speciella områden, som t.ex. Natura 2000 områden eller stickprovsområden som NILS, samt trädgränsen i fjällen, med stöd av i första hand tätt skannade laserdata från flera tidpunkter. Vi avser att utveckla metoder som effektivt och med objektiva kriterier kan skatta vegetationens förändringar, t.ex. i termer av täckningsgrad i olika höjdinterval. De begränsade områdena gör det möjligt att använda fjärranalysdata som samlas in oftare och med högre upplösning än som är rimligt för heltäckande data. Verktyg från WP T2 för klassificering av vegetation kommer dessutom att vidareutvecklas så att även multispektrala data kan användas för att få med fler klasser. Metoderna för att skatta förändringar kommer att utvärderas genom att använda kontrollerade försök. I den andra fasen av programmet kommer rekommendationer för ett monitoring system som bygger på upprepade fjärranalysmätningar och ett effektivt fältstickprov för kalibrering av data så att objektiva skattningar erhålls, att utvärderas.

The aim of this work package is to develop methods that can be used for monitoring specific areas such as Natura 2000 sites, NILS sample sites, or the arctic tree-line ecotone. We aim to develop efficient and objective methods for estimation of vegetation change, e.g., in terms of change of coverage in different height intervals using primarily high posting density laserscanner data from different years. The focus on limited areas allows for the acquisition and use of relatively frequent high resolution remote sensing data. In addition, tools developed within WP T2 will be extended to also include the use of multi-spectral data and to increase the number of vegetation classes. Methods for change detection will be validated by controlled experiments. Based on experience from the first phase, calibration methods will be tested in which field samples and multi-temporal remote sensing data can be combined.

1    Purpose and Scope
The aim is to implement and validate tools for multi-temporal use of remote sensing data in order to obtain efficient and objective information about vegetation changes for the purpose of vegetation monitoring in Swedish nature types of key interest. The analysis will be carried out using sub-metre resolution laser and optical data.

2    Contribution to the programme aim
This work package is central to obtaining the programme aim of developing efficient monitoring methods. Methods for monitoring sample areas of the landscape, or specific valuable areas, will be developed and validated for different landscape types (Table 1). The work package will partly build on experience from WP T1, and methods from WP T2 and WP T3. The end result will be recommended methods for monitoring specific areas such as Natura 2000 sites, or sample sites in monitoring schemes, for example NILS. Experiences from the early tests might contribute to recommendations for national monitoring, possibly before the end of the research programme.

Table 1: plan for deliverables (D) and activities (A)

Deliverable Month
A: Field inventory in cooperation with WP T1 and establishment of controlled experiments, in mountains (Abisko test site). The first round of laser data was acquired August 2008. 8
A: Second laser scanning in mountains (Abisko test site) 9
A: Algorithm developments for object recognition; continued development of tools, complementary to those developed in WP T2. 12
D: Change detection study in mountain area completed and manuscript written (PhD student article #1) 18
A: Field inventory, controlled experiments, at agriculture and forest test site 18
A: Second laser scanning at agriculture and forest test site 18
A: Acquisition of multi-spectral aerial images from low altitude at agriculture and forest test site 18
D: Object detection and classification study at agriculture and forest landscape test site (PhD student article #2) 30
D: Change detection study at agriculture and forest test site completed (PhD student article #3) 36
D: Multi-temporal calibration study completed (PhD student article #4) 48

3    Background, Theory and Methods
New high resolution remote sensing techniques, especially laser scanning, will provide new opportunities for efficient and objective monitoring of vegetation changes. One such example would be the monitoring of spreading and growth of individual trees and bushes in open lands with high natural values. In order to monitor the expected changes in the mountain vegetation, a 1500 km transect was flown in Norway during 2006 (Næsset and Nelson, 2007). Early experiences from the Norwegian study show that >91% of the trees >1m height can be detected. There is however a need to improve the knowledge about the relationship between remote sensing data, in situ calibration data, and final results, before large area operational surveys can be recommended. The main focus of the mountain studies will be to evaluate methods to monitor the changes in cover of mountain birch, spruce and pine as well as the change in cover of willows (Salix shrubs), both in and adjacent to the mountain birch forest and as a shrub layer in dwarf-shrub heath. The main focus of the agricultural landscape studies will be to monitor overgrowth, successional changes and undesired recruitment of trees and shrubs. In addition, a characterisation of the amount of valuable trees and shrubs resulting from proper management will be performed.
High posting density laser scanner data and also multi-spectral aerial images will be used to identify the vegetation type (e.g., grass, herbs or small shrubs). Studies have indicated that the combination of laser scanner data and multi-spectral data is useful for classification (e.g., Hill et al. 2002; Holmgren et al. 2008). Individual trees can be detected and measured using high resolution laser scanner data and image analysis methods (Hyyppä & Inkinen 1999; Persson et al. 2002). Multi-temporal airborne laser scanner data have proven to be powerful for change detection of vegetation (Hyyppä et al. 2003; Yu et al. 2004; Yu et al. 2008). Because the LiDAR technique changes through time, due to technical developments, the remotely sensed data should be combined with data from field plots, in order to obtain objective measures of change.
Within this project, research will focus on methods for detection of changes, calibration of the amount of change by aid on in situ data, and to the extent possible, also classification of the nature of the change. The expected results will be validated methods and knowledge about the relationship between the remote sensing data (e.g., the posting density for the laser data), the amount of in situ data, and the accuracy for the final classifications and change estimates. This knowledge will be essential for recommending future monitoring systems.

Task 1: Classification of vegetation
The objective is to develop and validate automatic methods for classification of vegetation cover in key vegetation types such as grasslands, forest and transition ecotones between forest and surrounding open vegetation. A reference dataset, a grid that covers each test site, is used to manually classify vegetation (made in cooperation with WP T1 and WP T3). The laser scanner data is first used for segmentation of homogeneous patches (evaluation of segmentation techniques made in cooperation with WP T3) and/or associated to raster cells. For each derived patch or raster cell, relevant features are extracted from laser scanner data (input from WP T2) and multi-spectral images, and these features are then used as input for supervised classification.
Task 2: Change detection
The objective is to develop and validate automatic methods for detection of vegetation changes. The cover and height of plants in different vegetation layers on the plots are measured, in as much detail as possible. For this purpose, we will use the ‘ellipse method’ by Holmgren et al. (2008) to estimate area extent and cover at a very detailed level. Since the registration is made at the individual level, it is straightforward to compare between species, as well as individuals of different size. At the same time as the inventory, experimental treatments are performed in which plants are partly or totally removed, in parcels with different treatments in an experimental design. After the removal of plants, the plots are once more laser scanned. The difference between two multi-temporal laser scanner datasets are used in order to find relevant features that can be used for the statistical analysis. The experimental treatments will show how the reliability of cover estimates from laser scanning depends on various factors. The experiments will be set up so that the interaction between the factors can also be analysed. These removal experiments will be performed primarily in the mountain region (the taiga-tundra gradient) and in the agricultural landscape (grasslands and various successional stages); the results from this should also be relevant for forest and mires.

Task 3: Multi-temporal calibration
The objective is to validate methods that combine a field sample with multi-temporal laser scanner data in order to not only detect but to also quantify changes. A network of field plots is established that cover a sample unit, e.g., a NILS square, or a Natura 2000 area. Multi-temporal laser scanner data from systems with different parameter settings are used. The task is to evaluate how automatic change detection methods, interpretation methods based on laser scanner data and aerial images, and a field sample should be combined in order to achieve cost-efficient production of maps with estimates of absolute changes. For the experiment, a denser network of field calibration plots will be used than would ordinarily be realistic for an operational case. The dense network of field plots will allow for validation of different pre-stratification strategies. The field data samples are used in order calibrate the different laser datasets. The calibrated differences will serve as input for models to estimate the level of change (e.g., change of crown cover).

4    Practical Relevance
The outcome of the project will be methods and recommendations for design of monitoring systems of sample units (e.g., NILS squares, Natura 2000 areas) or monitoring of the arctic tree-line ecotone. The multi-temporal remotely sensed data can be combined with data from field plots producing digital maps with classification of vegetation type, identification of objects, canopy structure (vertical layering), and changes (e.g., tree cover). The new high resolution maps combined with guided field samples can, in future monitoring systems, be used for ratio of means estimation, producing unbiased estimates. By using this framework there is no need to use the same laser scanner system for data acquisition from different years.

5    References

  • Brandtberg, T. 1999. Automatic individual tree-based analysis of high resolution remotely sensed data. PhD thesis. Acta Universitatis Agriculturae Sueciae, Silvestria 118. Swedish university of agricultural sciences, Uppsala, 47 p.
  • Hill, R. A., Smith, G. M., Fuller, R. M., Veitch, N. 2002 Landscape modelling using integrated airborne multi-spectral and laser scanning data, Int. J. Remote Sens., 23:2327-2334.
  • Holmgren, J., Persson, A., Söderman, U. 2008. Species identification of individual trees by combining high resolution LIDAR data with multi-spectral images. Int. J. Remote Sens., 29:1537-1552.
  • Holmgren, J., Johansson, F., Lindberg, E., Olsson, H. and Glimskär, A. Estimation of crown coverage by using airborne laser scanning. Accepted for presentation at Silvilaser 2008 conference, Edinburgh.
  • Hyyppä, J. and Inkinen, M. 1999. Detecting and estimating attributes of single trees using laser scanner. The Photogrammetric Journal of Finland, 16:27-42.
  • Hyyppä, J., Yu, X., Rönnholm, H., Kaartinen and Hyyppä, H. 2003. Factors effecting laser-derived object–oriented forest growth estimation. The Photogrammetric Journal of Finland, 18: 16-31.
  • Næsset, E. and Nelson, R. 2007. Using airborne laser scanning to monitor tree migration in the boreal–alpine transition zone. Remote Sens. Environ. 110:357–369.
  • Persson, Å., Holmgren, J. and Söderman, U. 2002. Detecting and measuring individual trees using an airborne laser scanner. Photogrammetric Engineering & Remote Sensing 68:925-932.
  • Yu X., Hyyppä, J., Kaartinen, H., Maltamo, M. 2004. Automatic detection of harvested trees and determination of forest growth using airborne laser scanning. Remote Sens. Environ. 90: 451-462.
  • Yu, X., Hyyppä, J., Kaartinen, H., Maltamo, M., Hyyppä, H. 2008. Obtaining plotwise mean height and volume growth in boreal forests using multi-temporal laser surveys and various change detection techniques. Int. J. Remote Sens. 29:1367-1386.