Mapping Three Dimensional Spatiotemporal Distributions Of Forest Photosynthetically Active Radiation Using UAV-based Laser Scanning Data

The three dimensional?3-D?spatiotemporal variations of forest photosynthetically active radiation?PAR?dictate the exchange rates of matter and energy in the carbon and water cycle processes between the plant-soil system and the atmosphere.It is still challenging to explicitly simulate spatial PAR values at any specific position within or under a discontinuous forest canopy.The rapid and accurate acquisition of 3-D canopy structure information is the basis for building the quantitative simulation model of spatiotemporal distribution of forest PAR.The angular gap fraction?AGF?of forest canopy is an important structural parameter affecting the transmission and scattering process and distribution of PAR in the canopy,however,the conventional methods for observing and estimating AGF are limited in space and can not depict its 3-D hemispherical distribution effectively.With the development of light detection and ranging?lidar?technology,based on the 3-D structure of the forest captured by the point cloud data,the spatiotemporal distribution and variation rule of PAR in the forest scene are obtained after the realization of the 3-D quantitative estimation of the forest canopy AGF.In this study,unmanned aerial vehicle laser scanning system?UAV-LS?was used to obtain the 3-D point cloud data of three forest plots in Baima Experimental Forest Farm,Lishui District,Nanjing.Two different estimation methods were used to estimate the AGF of forest canopy using the 3-D point cloud data.The AGF obtained by TLS was used as the standard value to verify and quantitatively compare the two results,and the error source and adaptability of two methods were analyzed.Secondly,combined with the segmentation-statistics strategy with double parameters for AGF estimation and the anisotropic sky diffuse PAR distribution,the simulation model of forest canopy PAR was established which considers the direct PAR,sky diffuse PAR and multiple scattering PAR reaching any point in the canopy respectively.And then the 3-D spatiotemporal distribution of forest PAR was obtained.Finally,the influence of laser data point density and 3-D PAR radiation model parameters on the results was explored.The main conclusions of this study could be drawn as follows:?1?Both the two estimation methods can accurately estimate the AGF for three forest plots,but the accuracy of the segmentation-statistics strategy with double parameters?SSDP?is higher than that of the binarization strategy with single parameter?BSSP?,and the estimation error is more uniform in each annulus ring with different zenith angle interval.When TLS data including trunks was used as standard data,the mean RMSE of BSSP and SSDP were 0.1083 and 0.0654,respectively.Under SSDP,the correlation coefficient?R²?between the estimated AGF and the reference value of all observation points?0.91?is 0.11 higher than the corresponding value?0.80?of BSSP.?2?SSDP is more flexible and applicable than BSSP.SSDP is a better choice than BSSP when hemispherical photographs or original TLS data are taken as the standard.In addition,the application range of SSDP to UAV-LS data point density?NPD?is about twice that of BSSP.?3?In a discontinuous broad-leaved forest canopy,the effective path length with gaps removed and effective path length weighted by point density and point distribution factors along a light transmission path were feasible and powerful parameters to capture the spatiotemporal variations of total PAR,the R² of the results based on two parameters were 0.92 and 0.97,and the mean bias was-53.04?mol·m^-2·s^-1?-6.8%?and-40.99?mol·m^-2·s^-1?-5.2%?,respectively.The total path length could not effectively track the variation of forest PAR on 5-min time scale and the R² of the estimated results was 0.54 with the mean bias of-203.76?mol·m^-2·s^-1?-26.12%?.?4?The estimated sky diffuse PAR is up to 17%higher at noon and 10%lower at sunrise or sunset by assuming the anisotropic sky diffuse distribution rather than the isotropic sky diffuse distribution.In addition,the forest PAR estimation model has good applicability to point cloud data with a large range of point density.Therefore,satisfactory accuracy can be guaranteed for data with various point density through the adjustment of the model parameters to the optimal values.