Nationwide forest estimates in Sweden using satellite data and airborne LiDAR (Or: going from 2 D to 3D) Håkan Olsson Mats Nilsson, Johan Fransson, Henrik Persson and Mikael Egberth (SLU) Umeå Svante Larsson Swedish Forest Agency (Skogsstyrelsen)
Medverkande i förstudien Från Skogsstyrelsen Svante Larsson (projektledare) Anders Persson Thomas Jonsson Marcus Larsson Peter Blombäck (projektägare) Johan Eriksson (ordf styrgrupp) Från SLU Mats Nilsson Håkan Olsson Mikael Egberth Jörgen Wallerman Peder Axensten Jonas Jonzen
National laser scanning being made 2009 about 2015 primarily for a new nationwide DEM 387 blocks with size 25 * 50 km. 0.5 1.0 returns / m 2 Max scan angle 20 degrees
The scannings have been made during different seasons
And with different laser scanners
Background 1. Assignment from the goverment to the Forest Agency: Work with SLU and other relevant agencies in order to utilise the national laser scanning for the forest sector. A pre-study was made during January - March 2013.
Background 2. SLU perspective: nationwide satellite data based forest data produced every 5 th year 2000 Landsat from 1997-2001 2005 SPOT from 2005-2006 2010 SPOT from 2008-2010 2015 Landsat 8 + laser scanning?
Nationwide prediction of woody biomass with 2D optical satellite data is used in the Nordic countries, but the accuracy is limited. Shadows provide the tree size related signal.
New possibilities for improved biomass estimates by use of 3D surface models and high accuracy DEM CHM = DSM DEM To be calibrated with field data Sensors that can provide (a semi?) DSM: - laser, - Point clouds from air photo - multi view angle optical satellite, - interferometric SAR, - radargrammetry DSM [m a.s.l.] Δ = CHM [m a. g.] DEM [m a.s.l.]
Data sources for planned nationwide product Airborne laser scanning data Satellite image National forest inventory sample plots
Number of NFI plots within a 100 * 100 km area. In total about 30 000 NFI plots are available for training 22 M ha productive forest land = half the land area.
Case study laser + satellite data trained with NFI plots 3 SPOT 5 images from 2010-06-04, same date and instrument settings 452 plots from the NFI, ranging from 2006 2010 LiDAR data from 8 blocks scanned 2010 and 2011. 50 km
Stem volume at estate level
RMSE% Stem volume accuracy as function of k 40 35 30 25 euclidean mahalanobis Random Forest 20 15 10 0 2 4 6 8 10 12 14 16 18 20 k value
RMSE% Mean height accuracy as function of k 40 35 30 25 20 15 euclidean mahalanobis msn Random Forest 10 5 0 0 2 4 6 8 10 12 14 16 18 20 k value
RMSE% Decidous volume with and without satellite data 350 300 250 200 150 No satellite Satellite included 100 50 0 0 2 4 6 8 10 12 14 16 18 20 k value
Contribution from optical satellite data: - Tree species - Age for young plantation - Uppdating of clear felled areas But satellite scenes might not cover the same area as laserblocks useful for finding training areas for a specific block
Other tested 3D techniques TandDEM-X + TerraSAR X Terra ASTER SPOT HRS ALOS PRISM DMC air photo camera
Tested along track stereo satellite sensors ASTER: 15 m pixels, 0 and -28 SPOT-5 HRS: 10 m pixels, +20 and -20 ALOS PRISM: 2.5 m, 0, and +24-24
Canopy height models versus field measured tree heigths SPOT HRS DMC camera Still, error from SPOT HRG colour based estimates of stem volume reduced from 31 % to 23 % when this CHM data was added Photogrammetry point cloud reduced estimation error from 31 % to 19 %.
Point cloud from laser scanner data Point cloud from digital photogrammetry over same area.
Height estimated from TandDEM-X versus validation data RMSE = 6.2%
Comparison canopy heights from TanDEM-X versus LiDAR
Early Tandem-X results Results estimated at plot level (202 training plots and 25 validation plots with 10 m radius) Biomass (tons ha -1 ) and height (m) estimation RMSE, adjusted coefficient of determination (R 2 adj ), regression coefficients (α 0 -α 1 ) and the number of plots Estimated RMSE RMSE% 2 R adj α 0 α 1 n Biomass 43.9 23.1% 0.577 37.70 0.320 25 Height 1.3 6.2% 0.707 3.42 0.84 25 Note: all regression coefficients are significant at the 0.1% significance level (p 0.001).
Early ranking of some 3 D data for forest biomass retrieval in Sweden from best to worst Sensor Platform Type of sensor and data Laserscanning (e.g. Leica) Aircraft 0,5 1 returns / m 2 Digital Photogrammetry (DMC) Aircraft 4800 m, 60% overlap, point cloude TanDEM-X Satellite X-band interferrometry ALOS PRISM Satellite Optical 3 line puchbroom 2.5 m pixels SPOT HRS Satellite Optical 2 line puchbroom 5 * 10 m pixels Terra Aster Satellite Optical 2 line puchbroom 15 m pixels
Conclusions Forest biomass retrieval from nationwide laser scanning trained with national forest inventory plots is feasible. But the production is a sensitive issue for the commercial sector Roles for optical satellite data: Division into broad species classes Age of plantations Change detection In addition to laser are there several more techniques for obtaining 3D data related to the forest canopy (including also radargrammetry, to be studied in a planned EU FP 7 project)
Contributors Swedish National Space Board - funding Swedish National Land Survey ALS and satellite data CNES and EC for data from the ISIS program SPOT Image for permission to use SPOT HRS raw data JAXA fors ALOS Prism data FOI for field data Joanneum Research, Graz, for permission to use the RSG software for 3D matching, Chalmers University of Technology radar remote sensing group
Questions? Medverkande i förstudien Från Skogsstyrelsen Svante Larsson (projektledare) Anders Persson Thomas Jonsson Marcus Larsson Peter Blombäck (projektägare) Johan Eriksson (ordf styrgrupp) Från SLU Mats Nilsson Håkan Olsson Mikael Egberth Jörgen Wallerman Peder Axensten Jonas Jonzen
Additional material
Biomass Strong candidate for ESA Earth Explorer 7: Launching a P-band radar satellite 2019 After the User Consultation Meeting in Graz on 5-6 April 2013, the mission candidate biomass was recommended by ESAC to become ESA s 7th Earth Explorer Final decision will be taken by PBEO in May 2013
Kartering av skogstyper med satellit + 3D data Skogsklasser som i Lantmäteriets Vegetationskartor och CORINE Indata Kart Noggrannhet satellitdata 67% satellitdata + höjd från laser 77% satellitdata + höjd från 3D flygbilder 76% Satellitdata = SPOT-4 (20 m pixlar) Nordkvist et al, 2012. Remote Sensing Letters Hygge Barr > 15 m Ungskog Lövkog Barr 5-15 m Blandskog
JAXA s Kyoto & Carbon Initiative: 2004- ALOS PALSAR mosaic over Scandinavia and Finland Norway ALOS PALSAR data used Fine Beam Dual (FBD34) 63 strips from 43 orbital tracks June October 2009 Finland Other data sources Digital Elevation Model Denmark Sweden
Resultat skattning av trädhöjd 1. Krycklan, Västerbotten 2. Remningstorp, Västergötland 3. Remningstorp, Västergötland Fjärranalysdata SPOT HRG SPOT HRG + HRS SPOT HRG + DMC SPOT HRG SPOT HRS ALOS PRISM SPOT HRG + HRS SPOT HRG + ALOS PRISM SPOT HRG ALOS PRISM SPOT HRG + ALOS PRISM RMSE 13% 10% 7% 16,1% 21,6% 15,3% 16,4% 12,9% 13,6% 13,1% 10,5% SPOT HRG = Vanlig SPOT bild i färg Grönt = enbart ytmodell Rött = Vanlig SPOT bild + någon ytmodell
Artikel och presentationer Artikel Persson, H. Wallerman, J, Olsson, H och Fransson, J.E.S. 2013. Estimating forest biomass and height using optical stereo satellite data and DEM from laser scanning data. Artikel inskickad till Canadian Journal of Remote Sensing. Konferenspresentationer Wallerman, J., Fransson, J.E.S, Reese, H., Bohlin, J., and Olsson, H. 2010. Forest mapping using 3D data from SPOT-5 HRS and Z/I DMC. In proceedings from IGARSS, Honolulu, Hawaii, July 25-30, pp. 64-67. + Muntlig presentation. Persson, H., Wallerman, J., Olsson, H. and Fransson, J.E.S. 2012. Estimating biomass and height using DSM from satellite data and DEM from high-resolution laser scanning data. In: Proceedings from IGARSS 2012, Munich, Germany, July 22-27. + Muntlig presentation. Olsson, H., Henrik Persson, Jörgen Wallerman, Jonas Bohlin, Johan Fransson. 2013. 3D data från optiska satelliter - Skogliga tillämpningar. Rymdstyrelsens fjärranalysdagar, Solna, 9-10 april, 2013.