Communication Dans Un Congrès Année : 2020

Fast LUT Building by Combining PROSAIL and DART: Application to Gap Fraction and Leaf Pigments Estimations over a Woodland Savanna

Accélérer la construction de LUT grâce a l'utilisation conjointe de PROSAIL et de DART: application aux estimations de la fraction d'écart et des pigments de feuilles sur une savane boisée.

Résumé

Gap Fraction and leaf Chlorophyll and Carotenoid contents (Cab and Car) are recognized indicators to assess and monitor forest health. Numerous methods exist to retrieve these variables from hyperspectral imagery, often relying on Radiation Transfer Models (RTM), either 1D or 3D. 1D RTMs, although often unrealistic as they consider canopies to be infinite turbid volumes, have the advantage of being computationally undemanding. Meanwhile, 3D RTMs allow for a detailed description of the canopy that can be better suited for sparse forests, but are considerably more resource-intensive which limits the number of simulations that can be realistically achieved. In this study, we investigated the possibility that a 3D RTM (DART) could be approximated by adjusting results of a 1D RTM (PROSAIL). The aim was to quickly generate extensive Look-Up Tables (LUT) with a finer sampling scheme of the variables' space to improve the estimations of Gap Fraction and leaf pigments, in comparison with the 3D RTM alone and a coarser sampling scheme to generate LUTs. The new method derived from this 1D-3D approximation was then applied to 18 m resolution AVIRIS-C images acquired over an open-canopy woodland savanna (Tonzi Ranch, California; mean canopy cover (CC) 48%). We found that the error of this new method was minimal with regards to sensitivity to measurements of Gap Fraction and leaf biochemical content (the maximum error in the worst configuration was 21% when approximating 30% CC on synthetic forest scenes). The new method built the 300,000 entries of the fine LUT in 1.5 CPU hours, a negligible time compared to the 12,666 CPU hours necessary to compute the 21,840 entries of the calibration coarse LUT with DART. Results of the LUT-based inversions showed that the accuracy in Gap Fraction estimations did not improve when using a finer sampling scheme (Willmott model performance index (dr) going from 0.79 to 0.78; best RMSE: 0.1), while Cab and most notably Car benefited considerably (dr going from 0.65 to 0.77 and from 0.34 to 0.65 for Cab and Car, best RMSE: 4.14 µg/cm² and 1.05µg/cm², respectively). Using the fast 1D-3D approximation method permitted to account for the relationship between Cab and Car when building the LUT, which in turn allowed using vegetation indexes dedicated to Cab to estimate Car and ensured more realistic LUT entries.

Fichier non déposé

Dates et versions

hal-04970270 , version 1 (27-02-2025)

Identifiants

  • HAL Id : hal-04970270 , version 1

Citer

Thomas Miraglio, Karine Adeline, Margarita Huesca, Susan Ustin, Xavier Briottet. Fast LUT Building by Combining PROSAIL and DART: Application to Gap Fraction and Leaf Pigments Estimations over a Woodland Savanna. AGU Fall Meeting 2020, Dec 2020, Online, United States. ⟨hal-04970270⟩

Collections

41 Consultations
0 Téléchargements

Partager

  • More