Understorey identification through the generation of canopy base - PowerPoint PPT Presentation

Understorey identification through the generation of canopy base height models based on LiDAR data Luka Jurjevic Croatian Forest Research Institute, Croatia

1. Introduction Forest three-dimensional (3D) distribution is transcendent in ecosystem management Knowledge understorey information is a challenge Most of the research focus on the tree layer or shrubs Only a few studies evaluate understorey structural variables Depiction of possible understory components that airborne LiDAR pulses can intersect (Wing et al. 2012)

1. Introduction These researches considered an initial threshold of shrub heights and evaluated LiDAR metrics in a defined range This approach does not work in complex forest environment Our goal is to find a methodology to define the limit between overstorey and understorey in complex forest environment. An ideal irregular forest (Roth, 1935)

2. Study area Study area covers 411.30 ha of pedunculate oak forest. Study area terrain is flat (elevations in range 105m to 118m a.s.l). Presence of other species e.g. common hornbeam, black alder and narrow-leaved ash. Two understorey species: common hazel and common hawthorn

2. Study area Field data Data collection in 2017.g. Q. Robur Stands 165 measured / 112 LiDAR coverage circular plots r = 8 m or 15 m Measured and recorded: h (50% of the trees), crown base height (cbh), species, dbh.

2. Study area LiDAR data LiDAR h = 112-144 m a.s.l. Optech ALTM Gemini 167 laser scanner mounted on the Pilatus P6. Acquisition took time between 29 June and 25 August 2016. Density per m2 All returns – 13,64 points·m2 Last return – 9,71 points·m2 7% of points were classified as “ground”.

3. Methodology Understory hcb HCB Field Tree HCB value data variables model model evaluation Optimal Individual Tree heightbreak Identification LiDAR Filtering cloud data Metrics Thinning cloud per plot

3. Methodology Hypothetical heightbreak from hcb model Hypothetical heightbreak from HCB model

3. Methodology One model per species • Walters and Hann (1986) Height to H hcb= crown base Logistic function: K • model 1+EXP ∑ bixi i=1 Where: K = number of parameters to be estimated bi = parameter estimates xi = independent variables Height to Canopy Base model SPECIES TREES Alnus glutinosa 284 Carpinus betulus 876 Fraxinus angustifolia 237 Individual Quercus robur 1549 Tree Identification Tilia sp. 140 Other 70

3. Methodology Height to Original plot with 14 returns/m 2 5 returns/m 2 crown base model Thinning cloud with • Buján et al. (2019) algorithm to 5 pulse per square meter Height to Canopy Base model LiDAR metrics with • FUSION v3.80 program (McGaughey 2018) Individual Tree Identification

3. Methodology Height to crown base The variability of the hcb values measured in • model the field was studied Several metrics were calculated for each plot • (minimum, maximum, mean, median, mode, percentiles) Height to Canopy Base model These were the dependent variables of the • HCB models Independent variables were LiDAR metrics • Individual Tree Regression Models were applied • Identification

3. Methodology 3617 trees measured in the field, classified according to their visibility (No visible, • Height to Probably visible, Visible and Unclassified) crown base model Trees identification with Li et al. (2012) algorithm • In the identification snag, tree top broken, cut and fallen trees were removed • Height to Canopy Base model Individual Tree Identification

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results Height to R 2 (%) crown base hcb MODEL RMSE RMSE (%) BIAS BIAS (%) EFFICIENCY (%) model Alnus glutinosa 1.88 12.75 -0.01163 0.08 83.30 69.05 Carpinus betulus 2.56 35.17 0.01655 0.23 65.82 43.11 Height to Canopy Fraxinus angustifolia 2.73 18.69 -0.05919 0.41 79.40 62.53 Base model Quercus robur 2.76 18.05 -0.00743 0.05 78.37 61.39 Tilia sp. 2.11 24.03 -0.05307 0.60 88.13 77.06 Individual Other Tree 1.93 22.93 -0.01757 0.21 90.28 80.67 Identification

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results hcb MODEL Equation Height to crown base A. glutinosa hcb = H / (1 + exp(-1.71 + 0.0406*DG - 0.0417*dbh + 1.154*dhratio - 0.837*balmod_g)) model mean min max sd Height to Canopy hcb 14.77 0.30 21.70 3.38 Base model DG 26.58 17.00 58.20 8.15 dbh 23.58 10.25 42.70 7.24 dhratio 1.12 0.63 1.87 0.27 balmod_g 0.67 0.00 1.00 0.22 Individual Tree Identification

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results hcb MODEL Equation Height to crown base C. betulus hcb = H / (1 + exp(-1.095 + 0.0312*DM - 0.0118*G + 1.308*dhratio - 2.114*bar)) model mean min max sd Height to hcb 7.29 0.50 20.80 3.40 Canopy Base model DM 26.37 15.70 49.50 5.53 G 33.98 15.40 67.00 9.80 dhratio 1.09 0.54 3.18 0.28 bar 0.12 0.01 0.58 0.10 Individual Tree Identification

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results hcb MODEL Equation Height to crown base F. angustifolia hcb = H / (1 + exp(-0.593 + 0.708*bar - 0.0623*rbad)) model mean min max sd Height to hcb 14.59 2.11 25.80 4.45 Canopy Base model bar 0.30 0.03 2.29 0.30 rbad 0.38 0.07 18.29 1.32 Individual Tree Identification

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results hcb MODEL Equation Height to crown base Q. robur hcb = H / (1 + exp(1.692 - 0.0918*HL - 0.000741*N + 0.0284*dbh - 0.5499*bar)) model mean min max sd Height to hcb 15.30 2.00 32.10 4.44 Canopy Base model HL 26.14 17.40 37.90 4.47 N 546.12 55.70 1691.00 279.93 dbh 38.86 10.15 120.10 17.70 bar 0.46 0.02 3.65 0.45 Individual Tree Identification

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results hcb MODEL Equation Height to crown base Tilia sp. hcb = H / (1 + exp(-2.786 - 0.101*DM + 0.223*HL - 0.009102*AGE + 0.497*dhratio)) model mean min max sd Height to hcb 8.78 1.50 19.90 4.41 Canopy Base model DM 26.33 20.80 31.70 3.10 HL 25.81 23.10 32.80 1.79 AGE 85.57 63.00 163.00 17.56 dhratio 1.17 0.59 3.27 0.31 Individual Tree Identification

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results hcb MODEL Equation Height to crown base Other hcb = H / (1 + exp(-1.207 + 0.0695*IH - 0.00379*AGE - 0.0398*dbh + 0.957*dhratio)) model mean min max sd Height to Canopy hcb 8.42 1.80 17.30 4.39 Base model IH 17.89 12.30 34.15 4.27 AGE 98.86 33.00 163.00 47.94 dbh 18.49 10.00 38.30 7.33 dhratio 1.16 0.65 2.15 0.30 Individual Tree Identification

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results Model RMSE RMSE (%) BIAS BIAS (%) R 2 adj (%) EFFIENCY (%) Height to crown base HCB_MEAN 1.05 9.23 0.00 0.02 46.01 47.35 model HCB_MIN 1.43 38.03 -0.06 1.68 14.46 16.65 HCB_MEDIAN 1.68 15.07 -0.01 0.07 38.89 39.11 HCB_P05 1.48 29.99 -0.03 0.63 14.49 15.95 Height to Canopy HCB_P10 1.31 23.04 -0.02 0.29 16.07 18.90 Base model HCB_P20 1.02 14.93 0.00 0.03 22.46 24.42 HCB_P25 1.05 14.17 0.00 0.00 28.08 34.50 Individual HCB_P30 1.30 16.21 0.00 0.00 27.34 33.34 Tree Identification HCB_P40 1.73 17.34 0.00 0.02 44.47 46.55 HCB_P50 1.65 14.35 -0.01 0.11 55.52 54.71

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results Model Equation Height to crown base HCB_MEAN 3.8728*exp(2.2135*LH_L_CV)*exp(-0.7908*LH_L_SK)*exp(0.0239*LH_P40) model HCB_P50 249.7569*LH_KUR^-0.1629*LH_P30^-0.355*TR_LH_L3^-1.6901 ………………………………….. Height to Canopy Base model Individual Tree Identification

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results Height to crown base Individual Tree Identification model Not Wrongly Correctly 80 identified identified identified 70 Identification success (%) Not Visible 36.23 26.23 33.13 60 Height to Canopy Probably 35.86 13.47 49.83 50 Base model 40 Visible 18.57 5.64 74.15 30 Unclassified 36.81 20.39 40.43 Individual 20 0 5 10 15 20 25 Tree Identification Total 30.16 16.93 50.14 Pulse per square meter

PRELIMINARY RESULTS PRELIMINARY RESULTS 4. Results Digital Canopy Model, DCM Height to Canopy Base, HCB ´ ´ Legend (m) Legend (m)

5. Discussion Goodness of fit statistics show that hcb models work correctly • Further variables should be tested for HCB modeling • The ITI method does not yield the expected results, high LiDAR return densities • introduce noise in the results Using the species as covariates instead of taking a model by species could be • tested Building a crown diameter model to weight the calculation of HCB based on the • surface of each crown would improve the predictive capacity of the model It is mandatory to contrast the processing of the point cloud under the predicted • optimum heightbreak obtained with understorey field data

6. Conclusions  The stand method works better than the individual tree method  It is feasible to assess the structural variables of understorey by filtering the point cloud  It would be essential:  to validate with the field work  to identify a more correct method to delineate crowns in irregular forests  to test different densities of LiDAR returns to check the accuracy increase in the estimations