By Abebe Ali
Abstract:
Leaf Area Index (LAI) is a key canopy descriptor that is used to determine foliage cover, and predict photosynthesis and evapotranspiration and is an input to many ecological models. Its estimation from remote sensing data has been the focus of many investigations in recent years. In this context, we have used a three dimensional Discrate Anisotropic Radiative Transfer (DART) model to invert LAI of a mature Norway spruce forest stand from Airborne Imagining Spectroradiometer (AISA) image data. The AISA image was acquired on September 18th 2004 while the Ground truth LAI measurements and DART input parameters were collected in September 2004 and 2005 in a study area at Bily Kriz near Beskydy Mountains in Czech Republic. The DART model has been run to simulate the BRF of the forest stand under different canopy closures, which was used to build Look-up table. An Artificial Neural Network (ANN) was trained with the look-up table and employed to retrieve the LAI of the forest stand from the hyperspectral image. The retrieved LAI was validated against the ground truth LAI estimated by combination of the Hemispherical photography (HP) and TRAC (Tracing Radiation and Architecture Canopies) optical instruments.
The inversion of the model showed that the mean LAI of the stand is 5.62 with RMSE of 1.84. Overestimation in lower LAI values and underestimation at high LAI values were generally observed most probably due to the influence of scene size, crown shape, Understory, saturation, limited sample size and error in ground truth LAI measurements. This indicates a need for improvements in the parameterization of the model and selection of model out puts to be used for inversion. The addition of noise to the look up table as prior information improved the RMSE of the estimate. Evaluation of predictions revealed certain level of agreement with ground based LAI estimation (R2 = 0.44 and 0.54 for ANN trained with and with out noise): The ground based LAI estimation by combination of hemispherical photography and TRAC optical instruments reduced the gap fraction saturation problem and enabled better estimation.
Keywords: LAI; DART; ANN; AISA; Norway spruce