Methodology of Synergizing Remotely Sensed Information and Biophysical and Ecophysiological Process Models

The advantage of remote sensing is that signatures over broad electromagnetic domains can be detected on remote/nondestructive, wide area, or realtime bases, while the surrounding issue is that measurements are usually instantaneous, directional, and infrequent, and must be converted to bio-physically meaningful variables. Conversely, the advantage of process modeling is that numerical models can take account of multiple variables, and can provide dynamic simulations as well as predictions under imaginary situations, while the issue is that experimental determination of model parameters and model validation are not easy, and that it is tedious or impossible to gather the necessary input data. Hence, one of the most promising approaches for effective monitoring and accurate prediction of plant production processes is the synergy of remote sensing and process models, which can reinforce each other.

A variety of approaches for relating remotely sensed signatures to plant and ecosystem variables are summarized in Fig. 2(a). One of the most widely used approaches is the simple regression of target variables on remotely sensed signatures, such as spectral reflectance, thermal temperature, and microwave backscattering coefficients (Fig. 2A). Several of these relations are reviewed in Section 2.

Nevertheless, physical processes such as spectral reflection, thermal emission, and scattering should be taken into account to extend the applicability and to improve the accuracy of relations between remotely sensed signatures and target variables. The first type of models to be linked with remotely sensed signatures are radiative transfer models, which represent physical processes such as spectral reflectance/absorption, thermal emission, and microwave scattering. Examples of such models in the optical domain are BRDF (bidirectional reflectance distribution function) models, which can take into account sun angle, sensor angle, and some other spectral parameters (Qi et al., 1995). Detailed approaches may include the effect of the complex architecture of plant stands, such as the presence of stems, fruit organs (Weiss et al., 2001) as well as the 3D distribution of canopy elements such as tree crowns and bushes (Gas-tellu-Etchegorry et al., 1999). The reflectance model SAIL is a well-known process model in the optical domain for plant canopy (Verhoef, 1984). Optical reflectance model at a leaf scale such as PROSPECT (Jacquemoud and Baret, 1990) and LIBERTY (Dawson et al., 1998) have been developed for broad and coniferous leaves, respectively. Remotely sensed signatures can be related more systematically to ecophysiological plant variables by inverting these models (see, for examples, recent studies by Combal et al. (2003)). For the thermal domain, the energy budget model and mass and energy transfer models are essential for describing thermal emission (Olioso et al., 1999). The backscat-tering process of microwaves by a plant canopy can also be described by some scattering models (Attema and Ulaby, 1978; Prevot et al., 1993; Wigneron et al., 1999), which take account of soil, plant, and sensor conditions such as LAI, leaf size, plant moisture, soil moisture, roughness, and incident angle. Plant eco-physiological variables can be estimated by inverting the model, based on the remotely sensed signatures (Fig. 2B).

Another type of process models to be linked with remote sensing information are the canopy functioning models, such as the energy budget model, plant growth model, and the soil-vegetation-atmospheric transfer (SVAT) model. The leaf or canopy temperature is determined as a part of the energy budget in

RS Measurements

- spectral reflectance

- brightness temperature

- backscattering coefficient etc.

direct input

Physical process model for RS signature"\

- Radiative transfer model

- Reflectance model

- Emission model

- Back scattering model, etc.

Regression or Empirical models

Simulation of Ecophysiological Variables

Simulation of Ecophysiological Variables

Real Ecophysiological Variables

- soil [water, C content, etc.] flux [energy, H2O, CO2, etc.]

Real Ecophysiological Variables

- soil [water, C content, etc.] flux [energy, H2O, CO2, etc.]

/ climate data and environmental process models N"ยป

plant parameters - Growth model soil parameters / I- SVAT model,etc. j

Simulation of RS signatures

- spectral reflectance

- brightness temperature

- backscattering coefficient etc. ^

RS Measurements

- spectral reflectance

- brightness temperature

- backscattering coefficient etc.

j f Physical process model for RS signature^.

- Radiative transfer model

- Reflectance model

- Emission model

- Backscattering model, etc.

tuning

Simulation of Ecophysiological Variables

Simulation of Ecophysiological Variables

Real Ecophysiological Variables

Real Ecophysiological Variables

I climate data plant parameters soil parameters

^Plantand environmental process models ^

- Growthmodel

N tuning

Figure 2: Synergistic linkage of remote sensing with biophysical and ecophysiological process models for estimation of ecophysiological variables. RS: remote sensing; LAI: leaf area index; Tr: transpiration; Pn: photosynthesis; Chl: chlorophyll content; SVAT: soil-vegetation-atmosphere transfer.

the soil-plant-atmosphere system, where energy, water, CO2, and ecophysio-logical conditions all fluctuate. Hence, temperature data from infrared imagery provides more quantitative and reliable information when they are used as inputs into ecophysiological models or stress indices (Fig. 2C). Inoue (1987) and Inoue et al. (1990a, b) estimated leaf transpiration and stomatal resistance remotely, using infrared leaf temperatures. Canopy transpiration has been well estimated by a combination of remotely sensed canopy temperatures and an energy- and mass-transfer model under various soil-water conditions (Inoue et al., 1994).

Canopy functioning models can simulate plant growth or fluxes in vegetated surfaces dynamically using meteorological inputs without remote sensing data (Fig. 2E). In addition, these functioning process models can utilize remotely sensed signatures (Fig. 2C), or plant parameters estimated by remote sensing (Fig. 2D), as inputs; they can also be calibrated using remote sensing information to provide more realistic estimates (Fig. 2F, G). Further deep linkage between remote sensing and models is presented in Fig. 2(b). Remote sensing signatures can be simulated by radiative transfer models using output from the functioning models (Fig. 2H-J). The remotely sensed measurements are then compared with simulated signatures to fine-tune the functioning models so that they can simulate more realistic signatures (Fig. 2K-N), which in turn yield more realistic estimates of ecophysiological variables (Fig. 2M, O). Maas (1988) and Bouman (1992) conducted the first practical studies of this type of synergy. They showed the need for within-season calibration of simulation models, and demonstrated that this method effectively reduced model complexity, simplified input requirements, and made the model more operational. This synergistic approach may be a practical and effective method for linking instantaneous remote sensing data with continuous growth simulation, because process models are based on biological foundations and yield robust patterns of growth and development. This approach has an advantage over the direct use of remotely sensed data as inputs and the use of correlations between accumulated remotely sensed indices (e.g., SNDVI) and productivity, because both the latter two require frequent remote sensing observations (e.g., Wiegand et al., 1986).

In general, there are two levels of synergy for this approach; one approach uses remote sensing data to estimate a few key plant variables (e.g., LAI and evapotranspiration), which are then used to recalibrate the model (e.g., Moran et al., 1995) (Fig. 2B or A with EFG). This first approach can utilize a wide range of vegetation indices, regression models, and model inversions to estimate plant variables such as LAI and above-ground biomass, which can then be used to recalibrate simulation models. This recalibration may be done by tuning model parameters, so that simulated variables such as LAI are in agreements with their remote sensing estimates. Another possibility may be to simply correct the time course of simulated variables by reinitializing the model each time remote sensing estimates are available (Olioso et al., 2002).

Another approach is to use the outputs of a plant growth model, such as LAI and leaf-angle distribution (LAD), to calculate the radiative features (e.g., spectral reflectance and microwave backscatter) of the plant canopy by using a radiation transfer model such as SAIL (Verhoef, 1984; Olioso et al., 2001; Weiss et al., 2001). Then, simulated and measured radiative features are compared to recalibrate the plant growth model (Bouman, 1992; Clevers et al., 1994; Olioso et al., 2002) (Fig. 2H-K, N, O). This second approach would be more attractive, provided a plant model is able to produce the output of several geometrical and spectral variables for a canopy that are required by radiation transfer models, and also that the spectral model is well calibrated to yield accurate spectral features of the canopy. To simulate canopy reflectance, Moulin et al. (1998) used the output LAI from a growth model for sugar beet (SUCROS) as the input to a canopy reflectance model (SAIL). The parameters of SUCROS were then optimized to minimize the difference between the simulated and measured reflectance. This approach proved useful for predicting sugar beet production at a regional scale. The recalibration approach may be more robust and operational than others, because the process model can provide some normal simulation results based on weather and plant inputs, and because intermittent or infrequent remote-sensing observations can be used efficiently to adjust the model parameters and to rerun the simulations using modified parameters. The tuning of model parameters is usually based on iterative optimization toward the minimized residual difference between observed and simulated data (Moulin et al., 1998; Olioso et al., 2002).

Was this article helpful?

0 0

Post a comment