Tools to estimate vegetation structure from laser scanner data

MSc. Eva Lindberg, Department of Forest Resource Management, SLU,
SE-901 83 Umeå,
Phone: +46 (0)90-7868536, e-mail: eva.lindberg@srh.slu.se

I detta arbetspaket kommer vi att utveckla och utvärdera automatiska metoder för att detektera vegetationsstruktur från laserdata. Vertikal vegetationsstruktur kan härledas antingen från punktlaserdata eller från den returnerade laserpulsens fulla vågform. Arbetspaketet ska jämföra olika metoder för att skatta höjdprofiler från den vertikala fördelningen av laserpunkter på rastercellsnivå samt från vågformslaserdata. Vårt mål är att skatta höjd och täckning för fältskikt, buskskikt och olika trädskikt i skog, samt att extrahera data som kan bidra till klassning av olika vegetationstyper. Det förväntade resultatet av arbetspaket T2 är ett verktyg som ger rasterdata med tillhörande höjdprofiler och klassificering utifrån höjd och täthet.

Since airborne laser scanning measures the height of vegetation elements and the ground, it is to a certain extent possible to obtain information about the height and density of different vegetation layers such as field layer, shrub layer and tree layers. This work package will develop and evaluate automatic methods to detect vegetation structure from laser scanner data. Vertical vegetation structure can be derived either from point laser data or waveform laser data. We will compare different methods to estimate vegetation height profiles from the vertical distribution of laser points on the raster cell level as well as from waveform laser data. Our goal is to estimate the plant coverage in different height intervals of the field layer, shrub layer or forest; we also aim to determine if the forest is single-story or multi-storied based on the vertical vegetation structure. An additional step will be to make some inference about the type of vegetation, based on the laser scanner data. The expected outcome of work package T2 is a tool which produces raster data with a vegetation height profile and a classification based on height and density.

1    Purpose and Scope
The aim of this work package is to develop and evaluate automatic methods to detect vegetation structure and elementary terrain characteristics such as slope and aspect from laser scanner data. This is of interest in order to detect shrubs and small trees in mountain areas, to identify multi-story forest and to identify grassland with overgrowth.

2    Contribution to the programme aim
The results from this work package will be used in all other terrestrial work packages, to estimate vegetation structure in areas valuable for nature conservation and changes in vegetation structure over time.
Table 1, deliverables from WP T2

Deliverable Time
Scientific article about estimation of vertical vegetation structure. The developed methods will also serve as input to vegetation interpretation and statistical analysis in WP T1, classification in WP T3 and change detection in WP T4 Month 12
Scientific article about classification of vegetation based on full waveform laser scanner data Month 48

3    Background, Theory and Methods
Vertical vegetation structure can be derived either from point laser data or waveform laser data. The shape of the distribution of point laser data may be used to separate single-story and multi-story stand structures (Maltamo et al. 2005). An alternative is to analyze the variance of laser-derived tree height data (Zimble et al. 2003). A quantitative measure of the vertical vegetation structure can be derived by dividing the vertical canopy into key layers and fitting a Weibull distribution to each layer (Coops et al. 2007).

Waveform laser data can be used to estimate a canopy height profile (CHP) which quantitatively represents the relative vertical distribution of a canopy surface area. Harding et al. (2001) have developed a method to account for occlusion of the laser energy by canopy surfaces, transforming the backscatter signal to a CHP. The estimated CHP was highly correlated with a ground based CHP, defined as a relative distribution of plant area as a function of height. The range, amplitude, width, and back-scatter cross section may be estimated for each echo by modelling the waveform as a series of Gaussian pulses to provide more accurate vegetation classification (Wagner et al. 2008).

Horizontal vegetation structure is of interest for identifying overgrowth in grassland. One way to do that is to estimate the fraction of a grassland area which is covered by trees or shrubs. This has not been investigated with laser data before but an approach would be to use a classification scheme similar to Rees (2007) to classify points as shrubs and trees or open land. The fraction of open land may then be estimated as well as the average size of each patch of shrubs and trees or open land. One outcome of this work package will be a tool to estimate vegetation height and density. This may also eventually be used to estimate horizontal vegetation structure, but that will not be a part of this work package.

The use of laser data to study lower vegetation such as grass or the herbaceous layers has not been thoroughly investigated yet. Su et al. (2007) determined the height of the herbaceous layer, under-story shrub and over-story tree layer by calculating the average height of all laser points falling in each height interval. Straatsma et al. (2007) used the 95th percentile of the laser reflections to predict height and the percentage index (the percentage of laser hits that fall within the height range of the vegetation) to predict density for herbaceous vegetation in the lower Rhine floodplain. Cobby et al. (2001) used segmentation based on the standard deviation of laser data in 10 m side windows to classify pixels as short or tall vegetation. The short vegetation height (grass and cereal crops) was estimated as proportional to the logarithm of the standard deviation of laser data. The RMSE was 14 cm for short vegetation height and 17 cm for underlying topography.

The laser data intensity provides an additional possibility of recovering ecological variables. Miura et al. (2008) found a high negative correlation between mean canopy cover and First Return Intensity in Canopy stratum (FRI_C) and a high positive correlation between mean grass cover and FRI_C. First Return Intensity in the Ground stratum (FRI_G) was highly correlated with mean canopy cover and fallen trees. Mean canopy cover and fallen trees were inversely proportional to the standard deviation of FRI_G. Korpela (2008) used intensity data from laser data (1064 nm) for separating lichens from other types of ground vegetation. The classification result was improved by compensating the intensity data for range differences. Waveform laser data provides the intensity for laser reflection at different heights above ground and should be related to ecological variables in the same way.

In work package T2, we will compare the vertical distribution of laser points with the vegetation height profiles on a raster cell level and fit functions (e.g., Weibull distributions) to the distribution of laser points to estimate the vegetation height profiles. Another alternative is to use the laser point height percentiles as independent variables in a regression model to estimate the vegetation height percentiles. We will also use waveform laser data and model it as a series of Gaussian pulses. Special care must be taken to compensate for the shielding effect of higher vegetation layers on reflections from lower layers.

Vegetation height profiles will be derived from laser data as the amount of vegetation material at different heights above ground. Detailed field plots with position, crown size, height and species for all trees and shrubs are necessary for calibration. The project has access to a dataset consisting of such field plots and high resolution laser data for one area which may be used for initial tests.

The tasks of this work package are to
•    develop methods to estimate the vertical vegetation structure, i.e., height and density of the vegetation.
•    compare the results from the different methods and evaluate them with cross validation to find the methods that produce most accurate results.
•    investigate if there are some variables, for example vegetation height, that indicate the most accurate method which can be derived from laser scanner data; it is possible that different methods will work better for different landscape types.
•    determine if the vegetation in a raster cell is field layer, shrub layer or forest and, in the case of forest, find criteria such as mean vegetation height, vegetation density or presence of more than one peak in the vertical vegetation structure to determine if the forest is single-story or multi-story.
The expected outcome of work package T2 is a classification tool which produces raster data with attached vegetation height profile and classification (e.g., field layer, shrub layer, and single-story or multi-story forest) based on laser scanner data. These results will be evaluated and further tested in WPT1.

4    Practical Relevance
Remote sensing data which can provide information about vegetation below the tree canopy is useful, for example, in species-rich semi-natural grasslands. Overgrowth is an important topic in many habitats but it is almost impossible to get a detailed view of that from aerial photographs. Vertical and horizontal vegetation structure can be used as indices for landscape structure, habitat modelling to predict species occurrence and dispersal, and automatic detection of changes in vegetation structure. In particular it is of interest for national landscape monitoring (e.g., NILS) and to delimit valuable forest stands.

5    References

  • Cobby, D.M., Mason, D.C. and Davenport, I.J. 2001. Image processing of airborne scanning laser altimetry data for improved river flood modelling. ISPRS Journal of Photogrammetry and Remote Sensing, 56:121-138.
  • Coops, N.C., Hilker, T., Wulder, M.A., St-Onge, B., Newnham, G., Siggins, A. and (Tony) Trofymow, A. 2007. Estimating canopy structure of Douglas-fir forest stands from discrete-return LiDAR. Trees, 21:295-310.
  • Harding, D. J., Lefsky, M. A., Parker, G. G. and Blair J. B. 2001. Laser altimeter canopy height profiles: methods and validation for closed-canopy, broadleaf forests. Remote Sensing of Environment, 76:283-297.
  • Korpela, S.I. 2008. Mapping of understory lichens with airborne discrete-return LiDAR data. Remote Sensing of Environment. In press.
  • Maltamo, M., Packalen, P., Yub, X., Eerikäinen, K., Hyyppä, J. and Pitkänen J. 2005. Identifying and quantifying structural characteristics of heterogeneous boreal forests using laser scanner data. Forest Ecology and Management, 216:41-50.
  • Miura, N. and Jones, S. D. 2008. The utility Of LiDAR for recovering ecological variables. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXVII, Part B8, Beijing 2008, 1393-1396.
  • Rees, W.G. 2007. Characterization of Arctic tree lines by LiDAR and multispectral imagery. Polar Record, 43:345-352.
  • Straatsma, M., and Middelkoop, H. 2007. Extracting structural characteristics of herbaceous floodplain vegetation under leaf-off conditions using airborne laser scanner data. International Journal of Remote Sensing, 28:2447-2467.
  • Su, J.G. and Bork, E.W. 2007. Characterization of diverse plant communities in Aspen Parkland rangeland using LiDAR data. Applied Vegetation Science, 10:407-416.
  • Wagner, W., Hollaus, M. Briese and C. Ducic, V. 2008. 3D vegetation mapping using small-footprint full-waveform airborne laser scanners. International Journal of Remote Sensing, 29:1433-1452.
  • Zimble, D.A., Evan, D.L., Carlson, G.C., Parker, R.C., Grado, S.C. and Gerard, P.D. 2003. Characterizing vertical forest structure using small-footprint airborne LiDAR. Remote Sensing of Environment, 87:171-182.