Utility of an image-based canopy reflectance modeling tool for remote estimation of LAI and leaf chlorophyll content at the field scale

Abstract

This paper presents a physically-based approach for estimating critical variables describing land surface vegetation canopies, relying on remotely sensed data that can be acquired from operational satellite sensors. The REGularized canopy reFLECtance (REGFLEC) modeling tool couples 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 near-infrared wavelengths for the inverse retrieval of leaf chlorophyll content (C_ab) and total one-sided leaf area per unit ground area (LAI). The inversion of the canopy reflectance model is constrained by assuming limited variability of leaf structure, vegetation clumping, and leaf inclination angle within a given crop field and by exploiting the added radiometric information content of pixels belonging to the same field. A look-up-table with a suite of pre-computed spectral reflectance relationships, each a function of canopy characteristics, soil background effects and external conditions, is accessed for fast pixel-wise biophysical parameter retrievals. Using 1 m resolution aircraft and 10 m resolution SPOT-5 imagery, REGFLEC effectuated robust biophysical parameter retrievals for a corn field characterized by a wide range in leaf chlorophyll levels and intermixed green and senescent leaf material. Validation against in-situ observations yielded relative root-mean-square deviations (RMSD) on the order of 10% for the 1 m resolution LAI (RMSD = 0.25) and C_ab (RMSD = 4.4 μg cm^− 2) estimates, due in part to an efficient correction for background influences. LAI and C_ab retrieval accuracies at the SPOT 10 m resolution were characterized by relative RMSDs of 13% (0.3) and 17% (7.1 μg cm^− 2), respectively, and the overall intra-field pattern in LAI and C_ab was well established at this resolution. The developed method has utility in agricultural fields characterized by widely varying distributions of model variables and holds promise as a valuable operational tool for precision crop management. Work is currently in progress to extend REGFLEC to regional scales.

Introduction

Remotely sensed data in the reflective optical domain function as a unique cost-effective source for providing spatially and temporally distributed information on key biophysical and biochemical parameters of land surface vegetation. Leaf area index (LAI), defined as the single sided leaf area per unit horizontal ground area, is a critical structural variable for understanding biophysical processes of vegetation canopies and for quantifying exchange processes of energy and matter between the land surface and lower atmosphere (M et al., 1995, No et al., 1995, Anderson et al., 2005, Do et al., 2004). Measurements of total leaf chlorophyll content (C_ab), defined as the sum of the contents of chlorophyll a and chlorophyll b per unit leaf area, can assist in determining photosynthetic capacity and productivity (Boe et al., 2002, Gitelson et al., 2006, Ni et al., 1995). Leaf chlorophyll is a good indicator of vegetation stress (Carter, 1994, Pe and Filella, 1998, Gitelson and Merzlyak, 1997), it is strongly related to leaf nitrogen content (Filella et al., 1995, Da et al., 2000, Yoder and Pettigrew-Crosby, 1995) and could therefore prove valuable for precision crop management (Moran et al., 1997).

Remote sensing techniques for estimating vegetation characteristics from reflective optical measurements have either been based on the empirical–statistical approach that links vegetation indices (VI) and vegetation parameters using experimental data, or on the inversion of a physical canopy reflectance (CR) model. The empirical approach is simple and computationally efficient, and the potential of empirical VI relationships for the determination of crop parameters has been demonstrated in numerous studies (e.g. Br and Mortensen, 2002, Co et al., 2003, Gitelson et al., 2005, Tu, 1980). However, a fundamental problem with the VI approach is its lack of generality. The shape and form of canopy reflectance spectra depend on a complex interaction of several internal (e.g. vegetation structure, leaf biochemical composition, soil background) and external (e.g. view-sun-target geometry, atmospheric state) factors (Baret, 1991) that may vary significantly in time and space and from one crop type to another. As a consequence, there is no unique relationship between a sought vegetation parameter and a VI of choice, but rather a family of relationships, each a function of canopy characteristics, soil background effects and external conditions (Baret and Guyot, 1991, Co et al., 2003, Gobron et al., 1997, Ha et al., 2004, Ho et al., 2007a, Za et al., 2003).

Physically-based models have proven to be a promising alternative as they describe the transfer and interaction of radiation inside the canopy based on physical laws and thus provide an explicit connection between the biophysical variables and the canopy reflectance. Different strategies have been proposed for the inversion of these models including iterative numerical optimization methods (e.g. Ja et al., 1995, Ja et al., 2000), look-up table approaches (e.g. Co et al., 2002b, Kn et al., 1998, W et al., 2000) and artificial neural network methods (e.g. Bac et al., 2006, Fa and Liang, 2005, W et al., 2004, W and Baret, 1999); each associated with specific advantages and disadvantages (Kimes et al., 2000). The iterative optimization approach facilitates a direct retrieval of biophysical parameters from observed reflectances without the use of calibration or training data of any kind, and Houborg and Boegh (2008) presented a way to make this kind of inversion computationally feasible at local to regional scales.

The inversion process is ill-posed by nature due to measurements and model uncertainties and because different combinations of model parameters may correspond to almost identical spectra (Co et al., 2002b, Atz, 2004). As a result, additional information is needed to accurately estimate the vegetation parameters. While the use of a priori knowledge (e.g. canopy type and architecture, model parameter ranges) has been shown to be an efficient way to solve ill-posed inverse problems (Co et al., 2002b, Co et al., 2002a, Ko et al., 2005, Qu et al., 2007), this regularization technique typically relies on the existence of experimental data collected at the site of interest. Using multiple MODIS images, Houborg et al. (2007a) demonstrated how the temporal evolution of LAI could be utilized as another way of regularizing the inverse problem. Atzberger (2004) proposed an entirely image-based regularization technique that incorporates radiometric information from neighboring pixels during model inversion. It was shown that the so-called “object signatures” contained real supplementary information, which helped to regularize the inverse problem (Atzberger, 2004). Houborg and Boegh (2008) reported good LAI and C_ab retrieval accuracies using a related image-based regularization strategy that assumed invariance of dry matter content, vegetation clumping and leaf angle distribution within well-defined land cover classes.

This paper proposes a number of important refinements to the biophysical parameter retrieval scheme described in Houborg and Boegh (2008). The focus here is on the field-scale applicability of the refined model whereas a future paper will evaluate the model utility at regional scales. In the original model, the atmospheric correction of the sensor radiances was performed independent of the CR modeling assuming Lambertian reflectance of the land surface. In reality, surface anisotropic effects can be significant (Kimes & Sellers, 1985) and the atmospheric correction should take into account these non-Lambertian surface boundary conditions to be consistent with the directional canopy reflectance spectra required as input to a multidirectional CR model (Verhoef & Bach, 2003). This is addressed here by coupling the PROSPECT leaf optics model (Bar and Fourty, 1997, Ja and Baret, 1990) and the turbid medium Markov chain canopy reflectance model, ACRM (Kuusk, 1995, Ku, 2001) with a vector version of the 6S (Second Simulation of the Satellite Signal in the Solar Spectrum) atmospheric radiative transfer model (6SV1) (Ko et al., 2006, Vermote et al., 1997). Another improvement accommodates partially senescent vegetation. Canopies of exclusively green leaves were assumed in Houborg and Boegh (2008) which will lead to large inaccuracies when applied to a canopy with intermixing of green and senescent leaf material (Bacour et al., 2002a). In this study the spectra of green and senescent leaves are modeled using PROSPECT. The correction for background effects has also been improved; the methodology presented in Houborg and Boegh (2008) required at least two satellite scenes to represent conditions of dense green vegetation and bare soil, respectively for each land cover class. In the refined model only an approximate first estimate of the background reflectance signal from a nearby bare soil or partially vegetated field is required as input to a novel pixel-wise soil correction scheme, which makes the new model also applicable to mono-temporal imagery. Additionally, prior information about the leaf inclination angle is no longer a prerequisite.

The key objective of this study is to develop a method for reliable biophysical parameter retrievals that 1) is entirely image-based, 2) does not rely on difficult to obtain in-situ measurements, 3) can be applied at a range of scales using radiometric spectral information from different airborne and operational satellite sensor systems, and 4) can be used for agricultural fields characterized by widely varying distributions of model variables. Following Atzberger (2004) and Houborg and Boegh (2008), the radiometric information content of pixels belonging to the same field or land cover class is exploited assuming small intra-field variability of leaf inclination angle distribution, leaf mesophyll structure and Markov clumping characteristics. These field-specific parameters are estimated by an iterative inversion technique using reflectance observations characteristic of intermediate to dense vegetation coverage (Houborg & Boegh, 2008). With the determination of the field-specific parameters completed using regularized inverse modeling, pre-computed look-up tables are accessed for fast pixel-wise parameter estimations. New image-based techniques are introduced to further regularize the inversion and to avoid confounding effects between the soil background reflectance and the canopy variables.

The atmospheric radiative transfer and canopy reflectance models are described in the next two sections followed by a detailed description of the integrated biophysical parameter retrieval tool. To demonstrate the feasibility and reliability of the approach, the tool is applied to a stressed corn field in Maryland, USA using 1 m resolution aerial imagery and 10 m resolution SPOT-5 data. Finally the accuracy of LAI and C_ab retrievals is evaluated using in-situ data.