**2.3 Direct and indirect methods for determining canopy structure**

Canopy structure can be determined by direct methods or estimated through indirect methods. The direct methods involve the measurement of leaf area, angle and position of leaves in the canopy or by destructive sampling. A group of methods involve the measurement of leaf area with optical area meter devices, scanners, hand-held planimeters or weighing of paper replicates (Nobel et al., 1993, Asner et al., 2003). Generally these methods allow the computation, separately, of the form, size and number of leaves (Bréda, 2003). Another way to obtain leaf area is by correlating either green or dry biomass to leaf area to find a conversion factor i.e. the *specific leaf area* (m2 kg-1). All these methods give a standard reference for the calibration or evaluation of indirect methods of determining LAI (Bréda, 2003) or vegetation cover (Schut & Ketelaars, 2003). On the other hand leaf angle distribution and leaf clumpiness determination are restricted to on site observations as they need to be carried out in undisturbed canopies.

Direct methods are cumbersome and time demanding. As a result, the indirect methods have been largely used for estimating canopy structure. These methods use the relationship with other more easily measurable parameters, such as canopy transmittance, green cover estimate or correlation with dry or green biomass. Gap-fraction inversion is amply used to estimate LAI (Welles & Norman, 1991). Gaps are obtained from devices specially designed for this (e.g., LAI-2000\*1) or hemispherical photographs (e.g., CID 110\*1). The LAI-2000 finds the gap fraction at five angles, making it possible to estimate also the average leaf angle. Ceptometers are also used to estimate LAI through gap fraction (López-Lozano et al., 2009). Inversion of light transmitted through the canopy, measured by a line quantum sensor, has also been used to estimate plant area index, a composition of leaves and stems of the plants (Cohen et al., 1997). Hemispherical photographs are usually taken looking upward (Rich, 1990) and the digital processing allows the estimation not only the total gap fraction but also to partition these gaps in different angles. However, the indirect methods based on light extinction through the canopy can be affected by canopy structure assumptions, exceptionally regarding the clumping of leaves (Larsen & Kershaw, 1996).

A sensitivity analysis of the LAI estimation using the LAI-2000, a ceptometer (AccuPAR\*1) and hemispherical photographs was carried out by Garrigues et al. (2008), over 10 crops including pasture grasses. They found that the hemispherical photographs were the most robust technique, from many standpoints of evaluation, the least sensible to illumination conditions and thus can be applied for a large range of canopy structures.

#### **2.4 Remote sensing of forage grasses**

The use of remotely sensed data for monitoring forage grasses is of great interest due to its possibility of monitoring large areas and the broad range of sensors available, with varying

<sup>\*1</sup> The mention of a trademark does not constitute an endorsement by the Federal Rural University of Rio de Janeiro of these products and does not imply approval for the exclusion of other suitable products.

spatial, spectral, temporal and radiometric resolutions. Remote sensing is also cost effective for monitoring large areas as it drastically reduces field work compared to *in situ* samplings. The rationale behind the use of remote sensing for vegetation monitoring relies on the fact that leaf reflectance varies throughout the solar spectrum (Knipling, 1970). The combination of leaf properties with the canopy structure and sun-viewing orientation lead to complex interactions that causes varying reflectance values at each spectral region, e.g., see Ollinger (2011) for a review.

Grass canopy biophysical properties and the illumination/viewing geometry affect reflectance observed at surface level (Walter-Shea et al, 1992). At the satellite level the reflectance values calculated from the radiance reaching the sensor undergoes the effects from the atmosphere, soil and litter background, canopy structure, bidirectional anisotropy, spatial heterogeneity, nonlinear mixing and topography (Myneni et al., 1995). These effects make it much more complex to establish fixed relationships between canopy parameters (like LAI) and canopy reflectance. Despite this limitation, remote sensing has been largely used for monitoring and canopy parameter estimation. Two approaches that have been largely used to estimate canopy parameters from remotely sensed data, involve the use of vegetation indices and inversion of canopy radiative transfer models.

#### **2.4.1 Vegetation indices**

226 Solar Radiation

radiation transmission in a turbid medium, in which K is the attenuation coefficient, the concept as applied here is different from Beer's law, since this equation defines only the amount (or fraction) of direct beam left after passing through a canopy layer with a defined

Canopy structure can be determined by direct methods or estimated through indirect methods. The direct methods involve the measurement of leaf area, angle and position of leaves in the canopy or by destructive sampling. A group of methods involve the measurement of leaf area with optical area meter devices, scanners, hand-held planimeters or weighing of paper replicates (Nobel et al., 1993, Asner et al., 2003). Generally these methods allow the computation, separately, of the form, size and number of leaves (Bréda, 2003). Another way to obtain leaf area is by correlating either green or dry biomass to leaf area to find a conversion factor i.e. the *specific leaf area* (m2 kg-1). All these methods give a standard reference for the calibration or evaluation of indirect methods of determining LAI (Bréda, 2003) or vegetation cover (Schut & Ketelaars, 2003). On the other hand leaf angle distribution and leaf clumpiness determination are restricted to on site observations as they

Direct methods are cumbersome and time demanding. As a result, the indirect methods have been largely used for estimating canopy structure. These methods use the relationship with other more easily measurable parameters, such as canopy transmittance, green cover estimate or correlation with dry or green biomass. Gap-fraction inversion is amply used to estimate LAI (Welles & Norman, 1991). Gaps are obtained from devices specially designed for this (e.g., LAI-2000\*1) or hemispherical photographs (e.g., CID 110\*1). The LAI-2000 finds the gap fraction at five angles, making it possible to estimate also the average leaf angle. Ceptometers are also used to estimate LAI through gap fraction (López-Lozano et al., 2009). Inversion of light transmitted through the canopy, measured by a line quantum sensor, has also been used to estimate plant area index, a composition of leaves and stems of the plants (Cohen et al., 1997). Hemispherical photographs are usually taken looking upward (Rich, 1990) and the digital processing allows the estimation not only the total gap fraction but also to partition these gaps in different angles. However, the indirect methods based on light extinction through the canopy can be affected by canopy structure assumptions,

A sensitivity analysis of the LAI estimation using the LAI-2000, a ceptometer (AccuPAR\*1) and hemispherical photographs was carried out by Garrigues et al. (2008), over 10 crops including pasture grasses. They found that the hemispherical photographs were the most robust technique, from many standpoints of evaluation, the least sensible to illumination

The use of remotely sensed data for monitoring forage grasses is of great interest due to its possibility of monitoring large areas and the broad range of sensors available, with varying

\*1 The mention of a trademark does not constitute an endorsement by the Federal Rural University of Rio de Janeiro of these products and does not imply approval for the exclusion of other suitable products.

exceptionally regarding the clumping of leaves (Larsen & Kershaw, 1996).

conditions and thus can be applied for a large range of canopy structures.

**2.3 Direct and indirect methods for determining canopy structure** 

leaf area, LAD and clumpiness.

need to be carried out in undisturbed canopies.

**2.4 Remote sensing of forage grasses** 

Vegetation indices are band combinations of remotely sensed data that have some relation with canopy parameters. The first introduced was the *simple ratio vegetation index* (SR or RVI) between the reflectance of near infrared band (ρNIR) and the reflectance of the red band (ρred). Many other indices have been introduced after the SR. The idea behind the indices was that they could remove soil background effects from remotely sensed data while keeping the high sensitivity of NIR reflectance to canopy LAI. The commonly used vegetation index is the *normalized difference vegetation index* (NDVI), which is given by:

$$NDVI = \frac{\rho\_{NIR} - \rho\_{rad}}{\rho\_{NIR} + \rho\_{rad}} \tag{8}$$

One index that has been used is the *soil adjusted vegetation index* (SAVI), which is calculated through the following equation (Huete, 1988):

$$SAVI = \frac{(1+L)\*(\rho\_{NIR} - \rho\_{read})}{\rho\_{NIR} + \rho\_{read} + L} \tag{9}$$

where L is the canopy background adjustment factor. If L is disregarded SAVI reduces to NDVI. Vegetation indices are affected by the atmosphere (Myneni & Asrar, 1994). Thus Huete et al. (2002) introduced the *enhanced vegetation index* (EVI) which is calculated as:

$$EVI = G \frac{\rho\_{NIR} - \rho\_{red}}{\rho\_{NIR} + \mathcal{C}\_1 \* \rho\_{red} - \mathcal{C}\_2 \* \rho\_{blue} + L} \tag{10}$$

where G is a gain factor, C1 and C2 are the coefficients of the aerosol resistance term and ρblue is the reflectance in the blue band. The G, C1 and C2 have been set to be 2.5, 6 and 7.5, respectively (Huete et. al. 2002). The authors also used a value of L equal to 1. The C1 and C2 terms use the blue band to correct for aerosol influences in the red band. The EVI was developed to optimize the vegetation signal by improving the sensitivity in high biomass conditions, taking out the background effects (leaf litter or soil) and by reducing the effects of the atmosphere. EVI has been mostly used for images of the Moderate-Resolution Imaging Spectroradiometer (MODIS) sensor for monitoring vegetation.

The rationale behind the vegetation indices is that increasing vegetation LAI increases reflectance in the NIR due to a low leaf absorption in this spectral region, while in the red region a decrease occurs. It has been shown that most of the indices saturate for LAIs above around 3 (Tucker, 1979). However, the shape of the curve of fAPAR (the fraction of absorbed PAR) is the same as NDVI versus LAI. Sellers (1987) showed that the SRVI has a nearly linear relationship to fAPAR. Myneni et al. (1992) showed a nearly linear relationship between fAPAR and NDVI and a simple linear model relating fAPAR to top of canopy NDVI has been proposed (Myneni & Williams, 1994). Global datasets of time series of NDVI and EVI are available (e.g, Lhermitte et al., 2011), which make them highly attractive for monitoring grasslands over large areas and for estimating canopy conditions.

#### **2.4.2 Canopy radiative transfer model inversion**

Canopy radiative transfer models have been developed to simulate bidirectional reflectance factor of vegetation canopies (BRF). One simulated BRF (BRFj(S)) can be represented by a function or algorithm, f, of subsystem characteristics (aj, bj, cj, dj, ej) (Goel & Strebel, 1983, Goel, 1988):

$$\text{BRF}\_{\vec{\gamma}}(\mathbf{S}) = \text{f}(\mathbf{a}\_{\forall}, \mathbf{b}\_{\forall}, \mathbf{c}\_{\forall}, \mathbf{d}\_{\forall}, \mathbf{e}\_{\mathbf{i}}) \tag{11}$$

where aj, bj, cj, dj and ej define the source, the atmosphere, the vegetation, the soil and the sensor subsystems, respectively. The source is characterized by the solar zenith and azimuth angles and the total flux intensity (normalized to one), the atmosphere is characterized by the direct plus diffuse radiation, the canopy is characterized by the leaf reflectance (ρ) and transmittance (τ), LAI, LAD and the leaf spatial distribution parameter (λ0), the soil is characterized by the soil reflectance (ρs) and the sensor is characterized by the view zenith (VZA) and view azimuth angles.

The inversion process consists of deriving a function or algorithm, g, that will yield the set of canopy parameters {cj}, as a function of the observed canopy BRF (BRF(O)) and the other subsystem characteristics (Goel & Strebel, 1983):

$$\{\mathbf{c}\} = \operatorname{g}(\text{BRF}\_{\!\!\!/\!\!/\!\/\!\/ \!\/ \!\/ \!\/ \!\/ \!\/ \/ \/ \text{a}\_{\!\!\/\/ \/\/ \/ \text{a}\_{\!\!\/} \text{b}\_{\!\!\/} \text{a}\_{\!\!\/} \text{b}\_{\!\!\/} \text{c})\tag{12}$$

Soil reflectance {dj} may also be derived through the inversion process along with the vegetation canopy parameters. The numerical inversion of a canopy radiative transfer model involves the minimization of differences between a set of simulated and observed BRF values acquired under different illumination/viewing geometries. Canopy parameter values that give the lowest difference between BRF(S) and BRF(O) are the estimated canopy values.

One limitation of the inversion process is that the number of observed values must be at least equal to the number of canopy parameters to be retrieved. This makes the process difficult for satellite image applications as most of the sensors collect single illumination/viewing geometry. This also limits the number of estimated parameters, which in most applications is the LAI. The validity of the inversion process has been validated for estimating canopy parameters of grasses using field radiometry observations (Privette et al., 1996). They have demonstrated the feasibility of estimating LAI, LAD, leaf reflectance and transmittance, total canopy *albedo* (the reflected radiation integrated over the hemisphere and the solar spectrum) and the fraction of absorbed photosynthetically active radiation. Inversion has also been used to estimate LAI and leaf chlorophyll for grassland (Darvishzadeh et al., 2008) and for sugar beet canopies (Jacquemoud et al., 1995) using spectroradiometer data and the PROSAIL model developed by Jacquemoud & Baret (1990).
