Integration of laser and digital imagery for classification of aquatic habitats

Dr. Michael Tulldahl,
Div. of Sensor Systems, Dept. of Optronic Systems
Swedish Defence Research Agency FOI; Box 1165; S-581 11 Linköping, SWEDEN
Phone +46 13 378516, Fax +46 13 378287, Cellular: +46 73 444 7716
e-mail: michael.tulldahl@foi.se

Genom att extrahera mer information än bottendjup från laserdata kan robusta data skapas för fusion med multispektrala flygbilder. Dessutom har laserskanningen större djupräckvidd än flygbilder och kan därmed dra nytta av att mer information tas fram direkt ur laserdata. Vi kommer att utveckla metoder för klassificering av akvatiska miljöer direkt från laserdata samt från laserdata kombinerat med flygbilder. I båda fallen kommer vi att nyttja in situ data för träning av metoderna och för validering av klassificeringsdata. Metoderna kommer att utvecklas med data från akvatiska miljöer runt Sveriges kust och från sötvattenmiljöer.

Extraction of more information than bottom depth from laser data can provide robust data for fusion with multispectral image data. Additionally, the laser scanner produces data to a larger maximum depth range than passive airborne optical sensors, and will thus benefit from increased information extracted directly from laser data. We will develop methods for classification of aquatic environments directly from laser data and from laser data combined with digital multispectral aerial imagery. In both cases, underwater video and other in situ data will be used as training data sets and for validation of the generated classification data. The methods will be developed with data covering aquatic environmental types around the Swedish coast and at freshwater sites.

1    Purpose and Scope
The aim of WP A1 is to develop robust methods, mainly automated, for interpretation of data from airborne laser scanners, aerial digital imagery and underwater video data. The expected outcomes of the project are classification prototype tools which produce point-data (geographical positions) or maps with attached classification (e.g., sand, macrophyte) and auxiliary features (such as bottom reflectivity and roughness). WP A1 will be responsible for acquisition of aquatic laser data and aerial imagery to be used in the programme. The County Boards of Norrbotten, Västerbotten and Östergötland and Ystad municipality have agreed to provide laser data, aerial imagery (some also in situ data) for the EMMA-programme. Additional laser data, aerial imagery and in situ data will be collected within the EMMA-budget in cooperation with other data users.

2    Contribution to the programme aim
New methods for mapping aquatic vegetation will be developed. The results will be directly adapted for user cases developed in project WP A2 and the long term relevance development of WP A3. Results will also be delivered to WP T1 for the preliminary development of a new vegetation map including also the shoreline and shallow habitats of major importance to nature conservation.

3    Background, Theory and Methods
Airborne laser scanning for depth bathymetry is well documented, in operational use and can reach down to approximately 2-3 times the Secchi depth (e.g., Steinvall 1996). Initial studies by FOI on using airborne laser scanning for classification of submerged  vegetation, such as the distribution of eelgrass (Zostera marina) in southern Sweden (Tulldahl et al. 2007, 2008a) have shown promising results (>80% overall accuracy, Tulldahl et al. 2008a). The classification method utilizes estimates of bottom reflectance and bottom vegetation height, which is new as opposed to earlier studies where only estimated bottom reflectance has been used (Tuell et al. 2005). The WMAL model (Wave Model for Airborne Laser, FOI; Tulldahl and Steinvall 2004) was used for corrections of the laser scanning data and this method will be further developed in this project. Digital aerial multispectral imagery has good potential for mapping sub-surface environments down to depths smaller than the Secchi Disk depth. To obtain robust methods, a number of factors affecting the images will be addressed. These include camera calibration, atmospheric transmission, solar and sky reflections, effects from viewing angle and from attenuation in water. The method was tested in a study (Tulldahl et al. 2008b), where laser data were integrated with multispectral imagery from the QuickBird satellite (Figure 1). The image was corrected for atmospheric effects (radiative transfer code 6S, Vermote et al. 1997), depth and water quality, resulting in an image showing only the reflectance of the sea floor (Fig. 1 (b)). Subsequent image-based analysis showed that different types of vegetation and substrate were divided into separable clusters which significantly improved the classification performance. Without the water-correction algorithm such classification would be virtually impossible as the uncorrected reflectance over different bottom/vegetation types overlaps (Fig. 1 (c)-(d)).
    Task 1 - Correction of laser data and aerial imagery. The optical properties of water will be retrieved from the laser data by inversion techniques (e.g., Hoge 2006; Klett 1985; Zege et al. 2004) and subsequently used for correction of the laser and imagery data. The aerial imagery data, will with help of depth measurements (from laser data) and calibrated optical reflectance models (e.g., Gordon et al. 1975; Kirk 1981; Mobley 2001; Strömbeck 2001; Lee et al. 1998) be used for estimation of the spectral optical properties of the water column and the spectral reflectance of the bottom with attached vegetation.

Figure 1. Water-uncorrected (a) and corrected (b) high-resolution multi-spectral QuickBird satellite imagery at a site North of Arkösund in the Archipelago of Östergötland. Data from the small white rectangles in (a) and (b): scatter plots of water-uncorrected (c) and corrected (d) spectral reflectance in bands 2 (green) and 3 (red) together with classification (Tulldahl et al. 2008b). In (a), the image pixels over water and covered by the laser survey, were multiplied with a factor of two to improve the visibility. The bottom depth in the area varies from 0 m to 6 m.

    Task 2 - Separability studies. The possibility to statistically separate different key vegetation classes will be evaluated, using corrected laser and aerial imagery data. Important relations will be established between aerial data and field inventory results comprising vegetation type, stand height, and cover. The result from this task will indicate which vegetation types can be classified with sufficient accuracy, and which vegetation types need to be grouped into broader classes for further analysis. Evaluation areas and key vegetation will be classified in detail from underwater video interpretation and field investigations (collaboration with WP A2) to match the signatures.
    Task 3 - Classification. A first subtask will be the design of pre-processing methods for field inventory data, such as grouping of vegetation types and subsequently using these for calibration of group-specific models of laser and aerial imagery data. The second subtask will develop and evaluate classification schemes where laser and aerial imagery data are used together with the calibrated group-specific models. Example of classification methods that will be tested are conventional maximum likelihood, as well decision-tree classifiers. The result will be raster layers with bottom classes (e.g., sand, macrophyte) and auxiliary features such as cover estimation and error estimates. The generated data will iteratively be forwarded to and evaluated in WP A2. Data will also be generated for tests with a new integrated map type in the coastal zone in cooperation with WP T1.
    Task 4 - Multi-temporal and variable resolution classification. This task is concentrated to the latter part of phase 1 and to phase 2 of the programme. The objective is to evaluate the performance of the classification methods from Task 3, based on data collected at different times and with different resolution (collaboration with A2 and A3). A previously laser scanned site will be scanned a second time and classification data will be generated for evaluation of repeatability, change detection, and influence from spatial resolution. The site will be thoroughly documented by field inventories to allow for detailed studies of the requirements on field inventory data such as necessary amount of field data (field points or transect data) for calibration of group-specific models and the influence on the final classification result.

4    Practical Relevance
Remote sensing methods can be efficient tools for investigations of large areas. Field inventories are, due to the high costs, limited to few local areas or transects. By integrating remote sensing and in-situ data, a presentation of data for larger areas will be available and field investigations can be focused on the most important and representative areas. The project will have a large impact on cost-efficient planning, conservation and monitoring of aquatic and coastal environments for different levels of management and science, by providing relevant data on key species and ecological habitats and by enabling the utilization of remote sensing techniques for mapping of aquatic environments on a large scale. Furthermore, the results can be used in applications such as planning of land use, mitigation and adaptation to climate change.

5    References

  • Gordon, H. R., Brown, O. B., and Jacobs, M. M. 1975. Computed relationships between the inherent and apparent optical properties of a flat homogeneous ocean. Appl. Opt. 14, 417-427.
  • Hoge, F. E. 2006. Beam attenuation coefficient retrieval by inversion of airborne lidar-induced chromophoric dissolved organic matter fluorescence. I. Theory. Appl. Opt. 45, 2344.
  • Kirk, J. T. O. 1981. A Monte Carlo study of the nature of the underwater light field in, and the relationships between optical properties of turbid, yellow waters. Australian Journal of Marine and Freshwater Research 32, 517-532.
  • Klett, J. D. 1985. Lidar inversion with variable backscatter/extinction ratios. Appl. Opt. 24, 1638-1643.
  • Lee, Z. P., Carder, K. L., Steward, R. G., Peacock, T. G., Davis, C. O., and Patch, J. S. 1998. An empirical algorithm for light absorption by ocean water based on color. J. Geophysical Research 103, 27968-27978.
  • Mobley, C. D. 2001. Hydrolight 4.2 Technical Documentation: Sequoia Scientific.
  • Steinvall, O., Koppari, K., Lejdebrink, U., Winell, J., Nilsson, M., Ellsén, R., and Gjellan, E. 1996. Depth sounding lidar-performance and models. In Proceedings of Laser Radar Technology and Applications, Orlando, Florida,  1996, SPIE Vol. 2748, 18-38.
  • Strömbeck, N. 2001. Water Quality and Optical Properties of Swedish Lakes and Coastal Waters in Relation to Remote Sensing. Ph D Thesis, Department of Limnology, Evolutionary Biology Center, Uppsala University, Uppsala, Sweden.
  • Tuell, G., Park, J. Y., Aitken, J., Ramnath, V., Feygels, V., Guenther, G., and Kopilevich, Y. 2005. SHOALS-enabled 3-d benthic mapping. In: Proceedings of Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XI, Orlando, FL, 28 March–1 April 2005, SPIE Vol. 5806, 816-826.
  • Tulldahl, H. M. and Steinvall, K. O. 2004. Simulation of sea surface wave influence on small target detection with airborne laser depth sounding. Appl. Opt. 42, 2462-2483.
  • Tulldahl, H. M., Strömbeck, N., and Philipson, P. 2008b. Combining high resolution satellite images and airborne lidar for benthic mapping (Manuscript in preparation). Remote Sensing of the Environment.
  • Tulldahl, H. M., Vahlberg, C., Axelsson, A., and Janeke, H. 2008a. Sea floor characterization from airborne lidar data. In Proceedings of International Lidar Mapping Forum 08, Denver, USA, Februari 21-22 2008.
  • Tulldahl, H. M., Vahlberg, C., Axelsson, A., Karlsson, H., and Jonsson, P. 2007. Sea floor classification from airborne lidar data. In Proceedings of Lidar Technologies, Techniques, and Measurements for Atmospheric Remote Sensing III, Florence, Italy, 17-20 September 2007, SPIE Vol. 6750, 675003-675001 -- 675012.
  • Vermote, E. F., Tanré, D., Deuzé, J. L., Herman, M., and Morcrette, J. J. 1997. Second simulation of the satellite signal in the solar spectrum, 6S, an overview. IEEE Transactions on Geoscience and Remote Sensing 35, 675-686.
  • Zege, E. P., Katsev, I. L., Prikhach, A., and Malinka, A. 2004. Elastic and Raman lidar sounding of coastal waters. Theory, computer simulation, inversion possibilities., EARSeL eProceedings Vol. 3.