Étude de l’estimation de paramètres structuraux de la forêt à l’aide d’un cadre de modélisation, de la photographie hémisphérique et du LiDAR.

The availability of experimental data from forest sites is limited and may also have large measurement error. In addition, attributes such as foliage density, are difficult to estimate and validate. To address this problem, we designed a study in five parts: the first part was the development of a modeling framework for three-dimensional simulation of forest scenes, rendered with the freeware POV-Ray (www.povray.org). For this project, 75 simulated stands were created with angular distributions of random leaves, and erectophile and planophile. The framework allows the realization of scenes with the knowledge of architectural attributes of the forest that are sometimes difficult to measure. As a second component, the modeling framework is used to evaluate existing methods for estimating the total plant and leaf area index (PAI, LAI), and canopy closure from hemispherical photography (HP). Ten HPs are simulated per scene and were analyzed to extract the PAI which includes leaves and woody material (eg trunks and branches). The results have shown that the truncated integration Miller (10º to 65º in view zenith), and estimated from 55º to 60º are very similar with coefficient of determination (R[superscript 2])of more than 0.9. The results also show that PAI and LAI are better estimated with respect to the input when the grouping of the leaves is considered. In addition, the grouping method considered with the CLX method (for viewing angles of 55º to 60º), was estimated with an R[superscript 2] of 0.66 and an RMSE of 1.0 units (23 %). Canopy closure estimates near the zenith with 10 HPs were very good with an R[superscript 2] of 0.83 and an RMSE of 9%. As third component, the modeling framework was used to simulate terrestrial and airborne LiDAR data. A new LAI algorithm (LF), based on the CLX method adapted to LiDARs data was presented and tested on these virtual data. The new algorithm has been tested with simulated terrestrial LiDAR (TLS) data for the same 75 virtual scenes. The results show a great potential using TLS data and the best LF method combinations of parameters gave an R[superscript 2] of 0.83 and an RMS error of 0.7 (20 %), thus improving the R[superscript 2] by 0.17 and a reduction of the RMSE of 0.3 units (3 %). The same algorithm was also tested on simulated airborne LiDAR (ALS) data on the same 75 virtual scenes and gave an R[superscript 2] of 0.85 and an RMSE of 0.6 (20 %) at viewing angle of 55° but only an R[superscript 2] of 0.18 and an RMSE of 2.2 (58%) at the viewing angle of 5º. Then, in the fourth part, the new algorithm was tested on ALS and TLS data from the Matane Reserve, Quebec. As the fifth and final installment, we explore in detail the evaluation of the IST methods, and especially the grouping of foliage to better understand the causes and limits of measured errors. The density of foliage within crowns is an important factor in the underestimation of the LAI with the CCL method and the size of the segments with the LX based methods determines the number of segments without holes where the logarithmic mean is a problem. In addition, the difficulty in measuring the LAI at nadir has been demonstrated with ALS, suggesting that the LAI should be estimated at angles near 57.3º for both field data (HP and TLS) and airborne data (ALS).

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.