Utility of an image-based canopy reflectance modeling tool for remote estimation of LAI and leaf chlorophyll content at regional scales

Martha C. Anderson
2009 Journal of Applied Remote Sensing  
This paper describes a novel physically-based approach for estimating leaf area index (LAI) and leaf chlorophyll content (C ab ) at regional scales that relies on radiance data acquirable from a suite of aircraft and operational satellite sensors. The REGularized canopy reFLECtance (REGFLEC) modeling tool integrates leaf optics (PROSPECT), canopy reflectance (ACRM), and atmospheric radiative transfer (6SV1) model components, facilitating the direct use of at-sensor radiances in green, red and
more » ... ar-infrared wavelengths. REGFLEC adopts a multi-step LUT-based inversion approach and incorporates image-based techniques to reduce the confounding effects of land cover specific vegetation parameters and soil reflectance. REGFLEC was applied to agricultural and natural vegetation areas using 10 m and 20 m resolution SPOT imagery, and variable environmental and plant development conditions allowed for model validation over a wide range in LAI (0 -6) and C ab (20 -75 μg cm -2 ). Validation against in-situ measurements yielded relative root-mean-square deviations on the order of 13% (0.4) for LAI and between 11 -19% (4.9 -9.1 μg cm -2 ) for C ab . REGFLEC demonstrated good utility in detecting spatial and temporal variations in LAI and C ab without requiring site-specific data for calibration. The physical approach presented here can quite easily be applied to other regions and has the potential of being more universally applicable than traditional empirical approaches for retrieving LAI and C ab . Remote sensing is a cost-effective means for monitoring the terrestrial biosphere and vegetation dynamics at a range of spatial and temporal scales. The success of the remote sensing approach depends on the nature and quality of the radiometric data and on our ability to relate the satellite signal (at-sensor radiance) to characteristics of the land surface vegetation. A critical step is the conversion of at-sensor radiances to surface reflectances by removing the atmospheric effect of gaseous absorption, molecular scattering and aerosols. Accurate atmospheric correction of satellite data is particularly important for physically-based retrieval algorithms that require accurate quantitative reflectance observations as input [15] . While the empirical-statistical approach that links vegetation indices (VI) and vegetation variables using experimental data is less affected by e.g. radiometric calibration accuracy and atmospheric factors, relationships derived using this approach tend to be specific to the study region and the atmospheric and experimental conditions at the time of the satellite acquisition. Nevertheless, the empirical approach has been widely adopted for retrieving vegetation variables [16] [17] [18] [19] due to its simplicity and low computational demand. However empirical relationships typically lack generality; there is no unique relationship between a sought vegetation variable and a VI of choice but rather a family of relationships, each a function of canopy characteristics, soil background effects and external conditions (i.e. atmospheric state, view-sun geometry) [20] [21] [22] [23] . Radiative transfer models based on physical laws that describe the transfer and interaction of radiation within the atmospheric column and canopy provide an explicit connection between the vegetation biophysical variables and the radiance signal received at the satellite sensor. Physical insight into radiative transfer mechanisms is needed to develop more flexible retrieval schemes that correct for the confounding influence of internal and external factors and assure applicability in diverse geographic locations with widely varying environmental conditions and species compositions. The physical approach has become a promising alternative given the high radiometric quality of current satellite sensors [24] , advances in atmospheric radiative transfer modeling [25, 26] , and enhanced capabilities for describing atmospheric scattering and absorption characteristics in space and time [27, 28] . The physically-based retrieval of canopy variables requires the inversion of a CR model. In this process, satellite reflectance observations are matched with simulated reflectance spectra to identify the combination of soil and vegetation variables providing the best reflectance fit. Commonly used inversion strategies include iterative numerical optimization methods [29, 30] , look-up table approaches [31-33] and artificial neural network methods [34] [35] [36] . Irrespective of the strategy, the inversion process is ill-posed by nature due to measurements and model uncertainties (i.e. different combinations of model parameters may correspond to almost identical spectra) [31] . The use of a priori knowledge (e.g. canopy type and architecture, model parameter ranges) has been suggested as an efficient way to solve illposed inverse problems [31, 37, 38] , but this 'regularization' technique typically relies on the existence of experimental data collected at the site of interest. Using multiple MODIS images, Ref. 23 demonstrated how the temporal evolution of LAI derived from EVI -LAI relationships could be used to constrain the inverse retrieval of canopy characteristics. Ref. 39 suggested that adjacent pixels belonging to the same crop field contain supplementary spectral information, and he demonstrated that confounding effects between LAI and leaf inclination angle were reduced when incorporating the radiometric information from neighboring pixels during model inversion. Ref. 40 took this a step further and reported good LAI and C ab retrieval accuracies using an image-based regularization strategy that assumed spatial and temporal invariance of dry matter content, vegetation clumping and leaf angle distribution within well-defined land cover classes. Recently Ref. 10 developed the REGularized canopy reFLECtance (REGFLEC) tool that combines atmospheric radiative transfer, canopy reflectance and leaf optics modules for direct image-based retrievals of LAI and C ab from at-sensor radiance observations. REGFLEC requires radiometric information from only 3 spectral bands (green, red and near-infrared)
doi:10.1117/1.3141522 fatcat:mu6av5cdxfdznfn6ch57kowlbu