ON THE ESTIMATION OF FOREST RESOURCES USING 3D REMOTE SENSING TECHNIQUES AND POINT CLOUD DATA Mika Karjalainen(1), Kirsi Karila(1), Xinlian Liang(1), Xiaowei Yu(1), Guoman Huang(2), Lu Lijun(2) (1)
Finnish Geospatial Research Institute, Geodeetinrinne 2, 02340 Masala, FINLAND, Email:
[email protected],
[email protected],
[email protected],
[email protected] (2) Chinese Academy of Mapping Sciences, Lianhuachixi Road No. 28, Beijing 100830, CHINA, Email:
[email protected],
[email protected]
ABSTRACT In recent years, 3D capable remote sensing techniques have shown great potential in forest biomass estimation because of their ability to measure the forest canopy structure, tree height and density. The objective of the Dragon3 forest resources research project (ID 10667) and the supporting ESA young scientist project (ESA contract NO. 4000109483/13/I-BG) was to study the use of satellite based 3D techniques in forest tree height estimation, and consequently in forest biomass and biomass change estimation, by combining satellite data with terrestrial measurements. Results from airborne 3D techniques were also used in the project. Even though, forest tree height can be estimated from 3D satellite SAR data to some extent, there is need for field reference plots. For this reason, we have also been developing automated field plot measurement techniques based on Terrestrial Laser Scanning data, which can be used to train and calibrate satellite based estimation models. In this paper, results of canopy height models created from TerraSAR-X stereo and TanDEM-X INSAR data are shown as well as preliminary results from TLS field plot measurement system. Also, results from the airborne CASMSAR system to measure forest canopy height from P- and Xband INSAR are presented.
1.
INTRODUCTION
Accurate and up-to-date forest maps are needed in wide range of applications from local forest management practices to global forest resources estimation. In particular, the amount of forest Above Ground Biomass (AGB) is under an intense international discussion, as it is related to the global climate change and carbon cycle. Earth Observing (EO) satellites, thanks to their vast spatial coverage, will obviously play a key role in the forest mapping task. However, satellites cannot solve the mapping problem alone, and ground truth data are required for training and calibrating EO data sets. We believe that by combining satellite-based and in-situ 3D forest information data, e.g., though SAR and with accurate Terrestrial Laser Scanning (TLS) would lead to
enhanced forest mapping in terms of spatial coverage, estimation accuracy and mapping frequency. TLS can collect ground truth data with increased speed compared to traditional field measurement techniques, and by combining this information with wide-area mapping capability of satellite SAR, to our knowledge, the results of the research project is the way forward in producing more accurate forest maps. [2] The all-weather and day-and-night imaging capability of satellite SAR is an advantage over optical imaging in monitoring applications of large areas. New very high resolution X-band SAR satellite data has created new possibilities in many applications. In forestry applications, temporal variability and saturation problems of radar intensity can be overcome by extracting 3D data from the imagery. There are two alternative techniques for obtaining height data from SAR imagery 1) SAR interferometry (InSAR) and 2) Radargrammetry. For the application of InSAR simultaneous data acquisition of the two SAR images is required with an appropriate baseline/geometry. Such data is currently available from the TanDEM-X satellite mission. Radargrammetry does not require simultaneous imaging and thus more satellite data are available, but similar weather conditions and an appropriate intersection angle between images are needed. SAR penetrates into the forest canopy (depending on the radar wavelength and forest structure) and there is some underestimation of tree height [6]. However, the SAR height information can be used to estimate AGB and other forest attributes when field data and digital terrain model are available. Radargrammetry has been used in [4, 9] and InSAR in [1, 8, 5]. There are also a few comparison studies of 3D remote sensing techniques including SAR based techniques [7, 10, 11]. TLS is a laser-based instrument that automatically measures its surrounding in three-dimensional (3D) space using millions to billions 3D points. The major advantage of using TLS in forest inventories lies in its capability to document the forest rapidly, automatically and in the millimetre-level detail [3]. Research on the application of TLS in forest inventories started around 2000. TLS were initially used in collecting basic tree
_______________________________________ Proc. ‘Dragon 3 Final Results & Dragon 4 Kick-Off Symposium’, Wuhan, PR China, 4–8 July 2016 (ESA SP-739, August 2016)
attributes in sample plots, such as Diameter at Breast Height (DBH) and tree position. More recently, TLS has been shown to be capable of determining high-quality tree attributes that are important and largely not directly measurable in conventional forest inventories, such as stem volume and biomass components. 2.
based canopy height model creation, the TDX image pair acquired in June 2014 was used. An example of SAR based 3D points compared to the Airborne Laser Scanning (ALS) data is given in Figure 1. In the forest change detection, image pairs of June 2014 and August 2012 were used in the comparison.
TEST SITES AND MATERIAL
2.1. Evo, Finland The basis of the study is a comprehensive set of field reference data. In summer 2014, approximately 100 forest test plots were measured in a field survey campaign covering the Evo test area. In the field survey campaign, standard forest resources related variables were collected: stem volume, biomass, basal area, mean forest canopy height, and mean diameter. The Evo test site is located in Southern Finland (61°13’ N, 25°6’ E). Goal was to measure the field plots based on traditional forestry field surveys, as well as with modern Terrestrial Laser Scanner (TLS) system. SAR satellite data, Evo InSAR data from the DLR TanDEM-X (TDX) satellite mission and stereo image data from the DLR TerraSARX (TSX) mission were used in this study. The dates of the images used in the study are given in Table 1. Table 1. TDX and TSX data of the study. TDX StripMap image pairs for InSAR
Baseline (HoA)
Inc. Angle
14.8.2012
192 m (45 m)
48°
5.6.2014
190 m (45 m)
48°
Orbit
Inc. Angle
9.7.2014
Descending
26°
4.7.2014
Descending
36°
29.6.2014
Descending
44°
9.7.2014
Ascending
30°
3.7.2014
Ascending
40°
8.7.2014
Ascending
47°
TSX SpotLight images for radargrammetry
In the radargrammetric 3D modelling, images from the same orbital nodes were used, i.e. combining ascending+ascending or descending+descending pairs only. In all, 6 stereo pairs were used to create radargrammetric canopy height model. In the InSAR
Figure 1. Examples of point clouds. TLS data, Evo TLS data were collected in all field sample plots using the multi-scan approach resulting in detailed point clouds that represent the sample plots. Five scans were made, one at the plot centre and others in north-east, south-east, south-west and north-west directions. Then all five scans were co-registered using artificial reference targets placed throughout the plot. 2.2. Danling, China In November 2014, CASMSAR airborne system flew over the test area of Danling to acquire single pass X band InSAR and repeat pass P band InSAR data. The Danling area selected for forest height estimation study is located in southwest China (29°55’N, 103°32’E, see Figure 2). The SAR Images were acquired by the CASMSAR system, which provides X band full polarimetric image data with middle frequency of 9.6GHz and 0.5m resolution and P-band full polarimetric image data with middle frequency of 600MHz and 1m resolution.
Figure 2. Location of the test site in China The test area is mainly covered by forest of eucalyptus. During the time of airborne system acquiring the data, 25 forest sample plots were measured in a field survey campaign covering the Danling test area. In the field
survey basic tree attributes were collected, such as mean forest canopy height, Diameter at Breast Height (DBH) and tree position.
Figure 4. TSX-stereo stem volume map (m3/ha) 29.69.7.2014
3.
Example of TDX-InSAR forest change map
RESULTS
3.1. Finland Examples of biomass maps based on TDX-InSAR and TSX-stereo In the accuracy comparison of SAR 3D data based forest biomass predictions, we used test plot of 32 by 32 meters (1024 m2). Altogether 91 test plots with reference forest inventory variables were used in the study. The results of the comparison study have been reported in Yu et al. (2015) in detail. The stem volume prediction accuracies (root mean square error %) for 3D points clouds derived from different remote sensing data were for ALS 16.7 %, aerial images 19.37%, optical stereo satellite data (WorldView-2) 15.9%, TDX SAR interferometry 22.0% and TSX radargrammetry 31.2%. Compared to the other RS techniques, both radargrammetric and InSAR processing performed well. Especially the TDX data resulted in very good estimation for the standard forest inventory variables. Examples of output biomass maps (stem volume m3/ha) for the larger area are given in Figure 3 and Figure 4, for TDX-InSAR and TSX-stereo respectively.
Figure 3. INSAR stem volume map (m3/ha) 5.6.2014
TDX image pairs from 2012 and 2014 were used to derive canopy height models using the ALS terrain model as the ground elevation. Finally, the 2014 CHM was subtracted from the 2012 CHM and a difference map was obtained. A preliminary result of forest 3D change map between 2012 and 2014 is given in Figure 5. The areas of forest clear cut and logging activities are clearly observable from the 3D difference map (red areas in the map). As a comparison, an aerial photo taken in 2014 is also shown in the Figure, and the clearcut areas are easily detectable from the optical image as well.
Figure 5. Top: TDX CHM difference 2012-2014 (red areas indicate negative values), Bottom: aerial image 2014 Example of tree stem modelled with TLS data TLS data were automatically processed using a robust modelling method. Stem points were first identified from the TLS data based on spatial distribution characteristics. A stem point was identified if it was on a vertical planar structure. The selected points were further processed to build stem models. A stem was modelled through series of 3-D cylinders along the tree growth direction. Example of a modelled tree stem is shown in Figure 6. The tree stems models can be, for example, used in estimating the plot level forest inventory variable accurately and objectively in the
future to enhance the collection of required ground truth data for satellite based forest mapping.
Figure 7. Digital Surface Model derived from X-band SAR
Figure 6. Tree stem (brown) modelled from the TLS data. 3.2. China
Figrue 8. Digital Elevation Model derived from P-band SAR
The Digital Surface Model (DSM) (Figure 7) is derived from the full polarimetric X band data using SARPlore developed by CASM with independent intellectual property rights. The full polarimetric P band data generates Digital Elevation Model (DEM) (Figure 8) based on RVoG (Random Volume over Ground) model. Finally, the difference DSM-DEM considered as forest tree height (Figure 9) extracted from the DSM (X-band) and DEM (P-band). There are 25 sample plots of 10 by 10 meters with tree inventory variables were used in the study. When comparing the mean tree height measured in the field survey and the tree height from X- and Pband, the RMSE (Root Mean Square Error) is 2.47m (Figure 10). When comparing to the option using only P band data (result, see Figure 11), the combination of Xand P- band data performed better. Figure 9. The tree height estimation result with Xand P- band data
TLS is a very accurate and efficient tool for forest field plot measurements and very close to the commercial breakthrough as a data collection tool in forestry. TLS measurements can automatically provide accurate field reference data to train, for example, satellite-based 3D information from InSAR and radargrammetry. By combining TLS field plots with satellite based 3D forest canopy height estimations we will move forward to enhanced forest mapping in terms of spatial coverage, estimation accuracy and mapping frequency.
Figure 10. The accuracy of tree height estimation result with X- and P- band data (RMSE:2.47m)
The result of forest tree height has appeared feasible to extract forest tree height by combining X band and P band SAR Images acquired by airborne CASMSAR system. The DEM generated from the full polarimetric P band data based on RVoG model is better than the DEM derived from single polarization InSAR data. The merit of forest tree height estimation with two bands of data source of X- and P- band is taking full advantage of the different penetration characteristics, i.e., the weak penetration for X-band data and the strong penetration for P-band. The result of test area also shows that the tree height derived from the X- band and P- band data has higher accuracy than the tree height generated by only one band data. 5.
Figure 11. The accuracy of tree height estimation result with P- band data (RMSE:2.85m) 4.
DISCUSSION AND CONCLUSIONS
The results based on TSX-stereo and TDX-InSAR have shown that these techniques have good potential in forest biomass mapping, as long as a good quality Digital Terrain Model is available from the target area and there are forest reference plots for the training and calibrating the estimation model. In addition, detection of forest biomass changes from multi-temporal TDX data and their 3D model differences show good potential as well. InSAR based estimates are better than those obtained by radargrammetry. However, appropriate InSAR data is not always available and then radargrammetry could be used. Even though, the accuracy of ALS is not reached in biomass estimation, the accuracy of SAR based point clouds may be adequate for forest inventories of large areas. In addition to good quality SAR data, a terrain model and field data are needed for accurate biomass estimation. For change detection these are not as essential and change maps could be derived based on SAR data only. However, in order to relate the height change to biomass change auxiliary data is required similar to forest biomass prediction.
ACKNOWLEDGEMENTS
The Tandem-X data was provided by the DLR through the DLR TDX scientific project, ID XTI_VEGE0360. Research by Kirsi Karila has been supported by the Dragon3 young scientist project “Forest canopy height models from radargrammetric and interferometric SAR processing”, ESA contract NO. 4000109483/13/I-BG. The research leading to these results has received funding from the European Community’s Seventh Framework Programme ([FP7/2007–2013]) under grant agreement n° 606971. The work was also supported by the National Natural Science Foundation of China [Grant No. 41401530].
6.
REFERENCES
1. Askne, J.I., Fransson, J.E., Santoro, M., Soja, M.J., Ulander, L.M., 2013. Model-Based Biomass Estimation of a Hemi-Boreal Forest from Multitemporal TanDEMX Acquisitions. Remote Sensing, 5:5574-5597. 2. Hyyppä, J., Karjalainen, M., Liang, X., Jaakkola, A., Yu, X., Wulder, M., Hollaus, M., White, J.C., Vastaranta, M. and Karila, K., 2015. Remote Sensing of Forests from Lidar and Radar. In Land Resources Monitoring, Modeling, and Mapping with Remote Sensing, ed. Thenkabail, P.S., (pp. 397-427). CRC Press.
3. Liang, X., Kankare, V., Hyyppä, J., Wang, Y., Kukko, A., Haggrén, H., Yu, X., Kaartinen, H., Jaakkola, A., Guan, F. and Holopainen, M., 2016. Terrestrial laser scanning in forest inventories. ISPRS Journal of Photogrammetry and Remote Sensing, 115, pp.63-77. 4. Karjalainen, M., Kankare, V., Vastaranta, M., Holopainen, M., Hyyppa, J., 2012. Prediction of plotlevel forest variables using TerraSAR-X stereo SAR data, Remote Sens. Environ., vol. 117, pp. 338–347, 2012. 5. Karila, K., Vastaranta, M., Karjalainen, M., Kaasalainen, S., 2015. Tandem-X interferometry in the prediction of forest inventory attributes in managed boreal forests, Remote Sens. Environ., 159:259-268. 6. Praks, J., Antropov, O., & Hallikainen, M.T., 2012. LIDAR-aided SAR interferometry studies in boreal forest: Scattering phase center and extinction coefficient at X- and L-band. IEEE Trans. Geosci. Remote Sens., 50, 3831–3843. 7. Rahlf, J., Breidenbach, J., Solberg, S., Næsset, E., Astrup, R., 2014. Comparison of four types of 3D data for timber volume estimation, Remote Sens. Environ., 155, 325-333. 8. Solberg, S., Astrup, R., Breidenbach, J., Nilsen, B., &Weydahl, D. (2013). Monitoring spruce volume and biomass with InSAR data from TanDEM-X. Remote Sens. Environ., 139, 60–67. 9. Solberg, S., Riegler, G., Nonin, P., 2015. Estimating Forest Biomass From TerraSAR-X Stripmap Radargrammetry, IEEE Trans. Geosci. Remote Sens., Vol. 53, No. 1, Jan 2015. 10. Vastaranta, M., Holopainen, M., Karjalainen, M., Kankare, V., Hyyppä, J., Kaasalainen, S., 2014. TerraSAR-X Stereo Radargrammetry and Airborne Scanning LiDAR Height Metrics in Imputation of Forest Aboveground Biomass and Stem Volume, in IEEE Trans. Geosci. Remote Sens., vol. 52, no. 2, pp. 1197-1204. 11. Yu, X., Hyyppä, J., Karjalainen, M., Nurminen, K., Karila, K., Vastaranta, M., Kankare, V., Kaartinen, H., Holopainen, M., Honkavaara, E., Kukko, A., Jaakkola, A., Liang, X., Wang, Y., Hyyppä, H., Katoh, M., 2015. Comparison of Laser and Stereo Optical, SAR and InSAR Point Clouds from Air- and Space-Borne Sources in the Retrieval of Forest Inventory Attributes. Remote Sens. 2015, 7, 15933-15954.
