Ilkka Korpela University of Helsinki 1 SingleTree Remote

Ilkka Korpela University of Helsinki 1) Single-Tree Remote Sensing with Li. DAR and Multiple Aerial Images 2 A) Mapping forest plots: A new method combining photogrammetry and field triangulation 2 B) Sovellettuna MARV 1 -kurssille

Contents - Part 1 Single-tree remote sensing, STRS Coupling allometric constraints to STRS A STRS SYSTEM - Treetop positioning with template matching (TM) - Treetop positioning with multi-scale TM - Species recognition in aerial images - LS-adjustment of crown models with lidar points Results and Conclusions, Demos

Contents - Part 2 Point positioning in the Forest Existing methods: Case “Tree mapping in a forest plot” The new method: Point mapping directly into a global 3 D frame Part 2 B Soveltaminen MARV 1: llä

Single-Tree Remote Sensing (STRS) • Rationales: Forest inventory, 3 D models • Since 1930 s→ “Substitute for arduous and expensive field measurements of trees” 2 D/3 D position Species Height Crown dimensions Stem diameter

Single-Tree Remote Sensing • Airborne, active / passive • 2 D or 3 D • Direct estimation & indirect allometric estimation • Restrictions: Tree discernibility: detectable object size, occlusion and shading, interlaced crowns • Alternative or complement • Accuracy restricted by “allometric noise” → tree and standlevel bias, tree-level imprecision in dbh~10 -12 %. • Measurements subject to bias • Timber quality remain unsolved, only quantity • Unsolved issues: 1. Species recognition

Photogrammetric STRS • Scene and object variation • Occlusion & shading • Scale: h = 0. . 40 m, dcrm 0. . 10 m • BDRF → automation challenging

Manual STRS - Demo 3 D treetop, height, crown width, Species stem diameter = f(Species, height, crown width) Image matching fails for treetop positioning unless we use a feature detector for treetops Demo – Single-Scale TM in treetop positioning PFG 1/2007

Airborne Li. DAR in STRS + No texture needed + Active → no shading + Real ease of 3 D − Discrete sampling − High sampling rates are costly − Reconstructing high-frequency relief − Species recognition − Underestimation of height Algorithms that process point clouds directly or interpolated DSMs / CHMs

Coupling allometric constraints to the STRS tasks Regularities in the relative sizes of plant parts Reduce ill-posedness of STRS Does species give the shape of the “crown envelope” ?

Coupling allometric constraints to the STRS tasks Empirical data on conditional distribution of Crown width & Shape | (Sp, height) → Consistency of measurements, Rule out impossible observations → Initial approximations for iterative approaches in finding true crown shape “Short trees have small crowns” Adjust search space accordingly

A STRS system combining Li. DAR and images

Multi-scale TM – Treetop positioning Assume that the optical properties and the shape of trees are invariant to their size. I. e. small trees appear as scaled versions of large trees in the images (within one species and within a restricted area)

Multi-scale TM in 3 D treetop positioning Maxima at different scales, take global → (X, Y, Z)

Multi-scale TM – Crown width estimation Demo 2 Near-nadir views have been found best for the manual measurement of crown width in aerial images

Species recognition Spectral values Texture Variation: - Phenology - Tree age and vigor - Image-object-sun geometry => reliable automation problematic => bottleneck

LS-adjustment of a crown model with lidar points Assume that 1) Photogr. 3 D treetop position is accurate 2) Trees have no slant 3) Crowns are ± rotation symmetric 4) We know tree height and species approximation of crown size and shape → Li. DAR hits are “observations of crown radius at a certain height below the apex” Assume a rather large crown and collect Li. DAR hits in the vicinity of the 3 D treetop position. Use LS-adjustment to find a crown model.

“Li. DAR hits are observations of crown radius at a certain height below the apex? ”

Li. DAR hits are observations of crown radius at a certain height below the apex – what if crowns are interlaced? ”

Example - a 19 -m high spruce: Solution in three iterations. Final RMSE 0. 31 m Note apex! Li. DAR did not hit the apex and the “crown width at treetop” (constant term) is negative.

Example - a 22 -m high birch: Solution in six iterations. Final RMSE 0. 47 m For some reason RMSEs are larger for birch in comparison to pine and spruce. Convergence?

Conclusions and outlook A 1) Multi-scale TM works in a manual semi-automatic way for 3 D treetop positioning - Possible to automate? - Computation costs? (NCC) 2) Multi-scale TM in crown width estimation needs comprehensive testing (Image scales, required overlaps) 3) Species recognition was overlooked here, 3 D treetop positions help? 4) Use of Li. DAR points LS-adjustment of a crown model: - Aggregated crowns are problematic. - Inherent underestimation of crown extent

Conclusions and outlook B If we have a STRS system that can be operated so that a tree measurement takes 3 -6 seconds and the measurement inaccuracies (RESULTS) are : height ~ 0. 6 m crown width ~ 10% stem diameter ~ 13 -18 % XY position ~ 0. 3 m Species classification ~ 95% Is this fast and accurate enough for sample-plot based STRS? Can we afford the images and Li. DAR? Can we compete against area-based methods?

ISPRS SILVILASER 2007 WORKSHOP, ESPOO SEPTEMBER 12 -14, 2007 HUT / FGI