Assessing aboveground tropical forest biomass from optical very high resolution images Pierre COUTERON IRD AMAP Lab, Montpellier, France N. Barbier, J-F Bastin, V. Droissart, R. Pélissier, P. Ploton, C. Proisy, B. Sonké, G. Viennois FORESEE workshop 10/10/2014, Nancy 1
Highlights Quantifying canopy texture from VHSR optical data Correlating texture with some stand variables as to predict biomass (inversion) Building 3D stand models and simulate images to test the inversion process Towards regional studies: across several forest types and heterogeneous acquisition conditions of images 2
Tree/stands measurements In the field From space dbh - Direct measurement of dbh Height and crown: costly and of limited accuracy Canopy + tree allometries = Meeting point - Direct data on height or crowns of canopy trees Limited information on boles and understorey 3 3
What makes AGB gradients in forests? Variation in: Canopy closeness/openness Stature of canopy trees Most of AGB in: Canopy trees tree boles Need of proxies observable from space: Height => lidar : GLAS ( ), airborne small footprint (still costly) Crown dimensions? Potential of VHR optical imagery 4
RCA: GeoEye pansharpened G NIR B 5
Very high spatial resolution optical data Potentiel of VHR (pixel < 1 2 m) : Satellites Quickbird, Ikonos, GeoEye, Worldview, Pléiades, Spot6/7, Increase in affordability, agility, build up of archives (Google Earth, ) Our aim: towards operational methods for stand type and biomass mapping Three steps: Objective/automatic measuring of canopy texture Correlate with field measurements of stand parameters Benchmark the inversion process by simulating images from modelled (and known) 3D forest structures 6
Quantifying canopy image texture Senescent Mature Adult Young Pionnier «radial» spectrum Dominant frequencies 2D Fourier spectra French Guiana, mangroves Frequencies (cycles/km) Proisy et al. RSE (2007) 7
8
Textural ordination: principal axes of variation 100 m FOTO analysis on digitized photographs in French Guiana Consistent with visual interpretation Couteron et al. J Appl Ecol (2005) 9
Heterogeneous canopy Medium crowns Logging concession L ARGER CROWNS / GAPS S MALLER CROWNS Canopy texture analysis: the FOTO method From N. Barbier, unpubl. Km 0 0.5 1 2 10
818000.000000 818500.000000 819000.000000 818000.000000 818500.000000 819000.000000 819500.000000 819500.000000 820000.000000 820000.000000 820500.000000 820500.000000 10/10/2014 11 1096900.000000 1097200.000000 1097500.000000 1097800.000000 1098100.000000 1098400.000000 1096900.000000 1097200.000000 1097500.000000 1097800.000000 1098100.000000 1098400.000000 Broad-leaved forests, North-Eastern France
818000.000000 818000.000000 818500.000000 818500.000000 819000.000000 819000.000000 819500.000000 819500.000000 820000.000000 820000.000000 820500.000000 820500.000000 10/10/2014 12 1096900.000000 1097200.000000 1097500.000000 1097800.000000 1098100.000000 1098400.000000 1096900.000000 1097200.000000 1097500.000000 1097800.000000 1098100.000000 1098400.000000
Relationships with forest parameters (AGB) No saturation! No saturation Biomass prediction using Ikonos imagery in mangrove stands; French Guiana Proisy et al. RSE (2007) 13
AGB prediction in the evergreen forests, Western Ghats of Karnataka (India) Panchro Ikonos and Google Earth images (1 m) RMSE = 55 t DM ha -1 (< 15%) Good results were also obtained with India s Cartosat 1 (2 m) G Rajashekar (unpl.) AGB assessed from field plots, local allometry (t DM ha -1 ) Références: Ploton et al. (2012) Ecol. Appl. 14
AGB Mapping from Ikonos/Google Earth images Western Ghats of Karnataka, India Ploton et al. (2012) Ecol. Appl. 15
Why does it work? Allometric relationships Dbh - AGB : main way of assessing AGB Dbh - Height : high variability Dbh crown : is probably more stable can a dbh 1.27 a can dbh 1.36 India (WG) Antin et al. 2013 Panama (BCI) Muller-Landeau et al. Ecol. Lett. (2006) Crown sizes inform about dbhs of canopy trees and AGB 16
Simulating canopy images Measures 3D stand models Canopy images D1 D2 D3 D4 Radiative transfer (DART) D5 D6 D7 D8 3D mockup (Stretch) 17
Radiative transfer DART, well established physical model: Ray tracing (turbid/solid media) in 3D scene The only model to produce scene images Gastellu Etchegorry. Meteorol Atm Phys (2008) 18
TERRESTRIAL LIDAR FOR THE STUDY OF LARGE TREE ARCHITECTURE (CAMEROUN) P. Ploton, umpublished 50 m Triplochiton scleroxylon RAUH S MODEL Terminalia superba AUBRÉVILLE S MODEL Piptadeniastrum africanum TROLL S MODEL 19
Examples from succession stages in Amazonian mangroves Proisy et al., 2012, Intech 20
Test of model inversion from simulated images The potential of inversion is established for: Mean crown width Quadratic mean dbh (predictor of biomass) PCA 2 <D crown > 4 35 3.9 3.8 30 3.7 mean DBH [cm] 3.6 3.5 25 3.4 0.33 0.5 1 2 R²=0.8 3.3 20 3.2-5 0 5-6 -4-2 0 2 4 6 PCA1 Texture index (PCA1) of simulated forests PCA 1 Simulated images automatically ordinated along texture gradients Barbier et al. Ann. For. Sci., 2012 21
10/10/2014 Effect of image acquisition conditions (scene geometry) 0 45 90 vz 2 4 1 3 9 8 7 5 6 15.07558 24.15956 10.09665 13.56898 27.00608 24.97276 23.22279 17.16577 17.17448 sv 162.8582 42.2848 92.1472 120.1528 7.2147 45.9589 119.7254 87.9187 14.2949 sz 34.72709 28.16477 33.16828 37.70374 31.49029 31.6663 33.18998 21.56988 26.0551 year 2000 2000 2002 2002 2006 2007 2008 2009 2009 180 270 Meters 360 Santarem, Brazil 22
Combining images in forward configuration for diachronic and/or regional studies 180 30 24 18 12 6 0 Sun zenith = 30 150 2 7 3 1 120 5 90 180 30 24 18 12 6 0 38 Sun zenith = 20 55 62 150 15 13 40 5 65 24 120 GeoEye Ikonos Quickbird 90 180 30 24 18 12 6 0 64 Sun zenith = 30 150 54 61 11 14 35 45 26 29 33 18 120 43 52 34 39 90 31 9 60 Santarem, Brazil (diachronic) 60 60 6 30 0 30 Central Africa 0 30 0 Ikonos vs GeoEye (+10 years interval) No bias in the main FOTO index (PCA1) for matching areas between images N. Barbier, unpubl. 23
Mitigating instrumental bias 24
Biomass texture calibration by forest types RDC several forest types, Quickbird + GeoEye Bastin et al. (2014), Ecological Applications Predicted biomass by 2D-FFT (t/ha) 500 400 300 200 100 0 Without clustering 0 100 200 300 400 500 500 400 300 200 100 0 With clustering 0 100 200 300 400 500 Observed biomass (t/ha) Observed biomass (t/ha) R² = 0.327 RMSE = 80.6 (~30%) R² = 0.79 (after cross val) RSME = 42.4 (~15%) 25
Biomass calibration All plots, R²=0.68843 RMSE=51.8922 Images Predicted biomass 700 600 500 400 300 200 49 51 3 36 42 50 34 52 65 35 62 38 64 40 Work in progress (more plots needed) Samples in 3 countries Including the main forest types: Mixed evergreen Evergreen monodominant Mixed evergreen/ deciduous Open canopy with Marantaceae 200 300 400 500 600 700 Biomass tdm /ha N. Barbier, unpubl. 26
Conclusion Potential of VHSR optical images Canopy aspect information is relevant to predict stand structure variables (including AGB) via texture metrics Effects of varying acquisition parameters can be controlled and even mitigated (to someextent avoid«hot spots») Upscaling from stand to landscape Field > airborne (lidar) > VHSR > High SR > Moderate SR Affordability thanks to the nested sampling Prospects/ ongoing Better understanding of noise and bias across scales via 3D stand/radiative transfer modelling Calibrating and cross validating for diverse African forests, but more data (plots + images) are needed 27
A matter of team and field work Thank you! Merci! pierre.couteron@ird.fr http://amap.cirad.fr 28
References Bastin J-F, Barbier N, Adams B, Shapiro A, Couteron P, Bogaert J, De Cannière C 2014 Aboveground biomass mapping of African forest mosaics using canopy texture analysis from contrasted acquisitions: towards a regional approach. Ecological Applications,. Couteron, P., Barbier, N., Proisy, C., Pélissier, R., Vincent, G., 2012. Linking remote-sensing information to tropical forest structure: the crucial role of modelling. Earthzine, : 1-4. / Barbier, N., Couteron, P., Gastellu-Etchegorry, J. P., Proisy, C., 2012. Linking canopy images to forest structural parameters: potential of a modeling framework. Annals of Forest Science, 69 (2): 305-311. Ploton, P., Pélissier, R., Proisy, C., Flavenot, T., Barbier, N., Rai, S. N., Couteron, P., 2012. Assessing above-ground tropical forest biomass using Google Earth canopy images. Ecological Applications, 22 (3): 993-1003. Barbier, N., Proisy, C., Vega, C., Sabatier, D., Couteron, P., 2011. Bidirectional texture function of metric resolution optical images of tropical forest: an approach using LiDAR hillshade simulations. Remote Sensing of Environment, 115 (1): 167-179. Barbier, N., Couteron, P., Proisy, C., Malhi, Y., 2010. The variation of apparent crown size and canopy heterogeneity across lowland Amazonian forests. Global ecology and biogeography, 19 (1): 72-84. Proisy, C., Couteron, P., Fromard, F., 2007. Predicting and mapping mangrove biomass from canopy grain analysis using Fourier-based textural ordination (FOTO) of IKONOS images. Remote Sensing of Environment, 109 (3): 379-392. Couteron, P., Pélissier, R., Nicolini, E., Paget, D., 2005. Predicting tropical forest stand structure parameters from Fourier transform of very high resolution canopy images. Journal of Applied Ecology, (42): 1121-1128. 29