The effect of digital hemispherical photograph binarization methods on the estimation of plant area index with LiDAR data

SHAOHUI ZHANG; LAURI KORHONEN; MAIT LANG; GASTÓN MAURO DÍAZ; ILKKA KORPELA

Berlín

Conferencia; ForestSAT 2022; 2022

ForestSAT

Plant area index (PAI) is often used as a proxy of leaf area index to describe the characteristics offorest canopy structure. Although it includes the contributions of woody materials and canopyclumping, it is easier to measure and can be converted to leaf area index if needed. Digitalhemispherical photography (DHP), as an indirect method of measuring PAI, has gained popularitythanks to recent technological advance over the last two decades. It offers high-resolutionpermanent directional records of forest canopy structure and can be used for the estimation of PAIat multiple zenith and azimuth angles.Airborne laser scanning (ALS) provides a great opportunity for large-area PAI mapping as laser pulsespenetrate forest canopy in a similar way as solar beams. As a result, canopy gap fraction can bederived from LiDAR data and PAI can subsequently be mapped using the area-based approach,which involves the construction of a model to predict PAI from LiDAR metrics. Field measurementfrom DHP provides reference data to calibrate Lidar-derived PAIs. However, the accuracy of theresults depends on the reliability of the binarization procedure that distinguishes sky from plantpixels, which is a crucial step in the estimation of PAI from DHPs. Multiple algorithms have beenproposed so far. This study compares the accuracies of ALS-based PAI models derived using threeestablished methods that can binarize DHPs: the Hemispherical Project Manager (HSP) software(Lang et al., 2017), R-package ‘rcaiman’ (Díaz and Lencinas, 2018) and the thresholding method ofNobis and Hunziker (2005). Binary images from these three methods are then used to calculate PAIusing the formulas employed by the commonly used LAI-2000 plant canopy analyzer. Lastly, theDHP-derived PAIs are regressed against an ALS based canopy density index using the semi-physicalmodel form: PAI = β × log(1–density) (Solberg et al. 2009). We used n=81 field plots measured inHyytiälä, southern Finland.The results showed that the PAIs obtained from the three methods have good correlations with theALS based canopy density index. The model obtained with HSP generated slightly more accurateresults (RMSE=0.36) compared with those from rcaiman (RMSE=0.43) and the method of Nobis &Hunziker (RMSE=0.49). All methods provided comparable model coefficients (-2.4