**1. Introduction**

216 Remote Sensing of Biomass – Principles and Applications

Shuman, J K, H H Shugart, and T L O'Halloran. 2011. Sensitivity of Siberian Larch forests to

Snyder, P. K., Delire, C., and Foley, J. A. 2004. Evaluating the influence of different vegetation biomes on the global climate. *Climate Dynamics*, Vol. 23 No 3/4, pp. 279-302. Soja, A J, W R Cofer, H H Shugart, A I Sukhinin, P W Stackhouse, D J McRae, S G Conard.

Soja, A J, N M Tchebakova, N H French, M D Flannigan, H H Shugart, B J Stocks, A I

Solomon, A M. 1986. Transient Response of Forests to CO2 Induced Climate Change:

Stocks, B J, M A Fosberg, T J Lynham, L Mearns, B M Wotton, Q Yang, J Z Jin, K Lawrence, G

Tchebakova, N M, G E Rehfeldt and E I Perfenova. 2005. Impacts of climate change on the

*Mitigation Adaptation Strategies for Global Change* Vol 11, No 4, pp. 861-882. Tchebakova N M, G E Rehfeldt and E I Parfenova. 2009. From Vegetation Zones to

Tilman D and J A Downing. 1994. Biodiversity and stability in grasslands. *Nature* Vol 367,

Urban, D L, M E Harmon, and C B Halpern. 1993. Potential Response of Pacific

Ustin, S. L. and Xiao, Q. F. 2001. Mapping successional boreal forests in interior central Alaska. *International Journal of Remote Sensing*, Vol. 22, No 6, pp. 1779-1797. Volney W J A and R A Fleming. 2000. Climate Change and impacts of boreal forest insects.

Vygodskaya N N, P Ya Groisman, N M Tchebakova, JA Kurbatova, O Panfyorov, E I

Watt A S. 1947. Pattern and Process in the Plant Community. *The Journal of Ecology* Vol 35,

Yan, X and H H Shugart. 2005. A forest gap model to simulate dynamics and patterns of Eastern Eurasian forests. *Journal of Biogeography* Vol 32, pp. 1641-1658. Yan, X and S Zhao. 1996. Simulating the response of Changbai Mountain forests to potential

Zhang, N, H H Shugart and X Yan. 2009. Simulating the effects of climate changes on

Ecosystems Siberian Larch Forests Springer Verlag

Composition. *Climatic Change* Vol 23, pp. 247-266.

*Agriculture, Ecosystems and Environment* Vol 82, pp. 283-294.

zone of northern Eurasia. *Environ. Res. Lett.* 2 (045033): 7 pp URL http://www.iop.org/EJ/abstract/1748-9326/2/4/045033/

climate change. *Journal of Environmental Sciences* Vol 8, pp. 354-366.

Eastern Eurasian forests. *Climatic Change* Vol 95, pp. 341-361.

*of Geophysical Research 109* (D14S06) doi:10.1029/2004JD004570

2486.2011.02417.x

pp. 274-296.

Laxenburg, Austria

pp. 363-365.

pp.1-22.

567-579.

climate change. *Global Change Biology* (Accepted Article) doi: 10.1111/j.1365-

2004. Estimating fire emissions and disparities in boreal Siberia (1998-2002). *Journal* 

Sukhinin, E I Parfenova and F S Chapin III. 2007. Climate-induced boreal forest change: Predictions *versus* current observations. *Global and Planetary Change* Vol 56,

Simulation Modeling Experiments in Eastern North America. *Oecologia* Vol 68, pp.

R Hartley, J A Mason and D W McKenney. 1998. Climate change and forest fire potential in Russian and Canadian boreal forests. *Clim Change* Vol 38, No 1, pp. 1-13. Stolbovoi, V and I McCallum. (eds) 2002. *CD-ROM Land Resources of Russia* International

Institute for Applied Systems Analysis and the Russian Academy of Science,

distribution of *Larix* spp. and *Pinus Sylvestris* and the climatypes in Siberia.

Climatypes: Effects of Climate Warming on Siberian Ecosystems .pp 427-447 in A Osawa O A Zyranova, Y Matsurura, T Kajimoto and R W Wein (eds) Permafrost

Northwestern Forests to Climatic Change, Effects of Stand Age and Initial

Parfenova, A F Sogachev. 2007. Ecosystems and climate interactions in the boreal

Remote sensing images provide very rich and useful information linked to relevant biophysical parameters such as the LAI (Leaf Area Index), fCOVER (fraction of vegetation cover) or fAPAR (fraction of Absorbed Photosynthetically Active Radiation). At the moment, several techniques for estimating such variables are available and widely used in many applications, such as estimation of the total biomass and monitoring the dynamics in canopy vegetation (Baret et al., 2007; Lecerf et al., 2008). For several years, a large number of Very High Spatial Resolution (VHSR) satellites, such as Quickbird, Geoeye and Ikonos, have been launched, and very important missions such as the Venus and the Sentinel-2 are expected in 2012 and 2013. This provides possibility of having more or less temporal consistency in VHSR observations of the land use on relevant agricultural sites. However, because of the heterogeneity of the available VHSR data, in particular due to their different wavelengths sensibility and of the intrinsic errors induced by the estimation processes, the resulting time series of biophysical parameters are more or less noisy. As a matter of fact, the estimated variables may only poorly fit their actual dynamics. The estimation of the complete sequence of such parameters is then of prime importance, in particular if one wants to analyze the evolution of the biomass.

In this chapter, we propose to explore the possibilities of using tools issued from tracking techniques, in particular particle smoothing, to recover time-consistent series of LAI (Doucet et al., 2001; Kitagawa, 1996) from noisy and incomplete observations. Such techniques, based on Monte-Carlo strategies, allow performing the estimation of an unknown state function, LAI in current case, according to a given dynamical model and to possibly corrupted measurements. The dynamical model on which we rely on is GreenLab model, a functional-structural plant model simulating plant development and growth (Yan et al., 2004). Given model parameters, GreenLab can compute the evolution of LAI, the biomass production and partitioning, the organ size and biomass. Inverse method can be applied to estimate hidden model parameters by fitting model output with measured data (Kang et al., 2008). We suppose that from remote sensing data observing agricultural parcels, the type of

Reconstructing LAI Series by Filtering Technique and a Dynamic Plant Model 219

resolution. In this case, the variability of the image luminance inside a given culture is high, and the pixel reflectance is not informative. Texture analysis strategies can be used to characterize and label the different crops, which use 1st or 2nd order statistical criteria or more advanced techniques like wavelets (Lefebvre et al., 2010). Many commercial software

Estimating the biophysical variables, such as LAI, fCOVER or fAPAR, from satellite observations can provide crucial information for numbers of applications, for instance, monitoring changes in canopy vegetation at global or regional scales, identifying bare soils, or detecting grassland areas. Among the different techniques available, there is a technique based on the inversion of the SAIL+PROPSPECT radiative transfer model (Verhoef, 1984; Jacquemoud and Baret, 1990) using training samples and neural networks, as introduced in (Baret et al., 2007). The SAIL model deals with light scattering by leaf layers with application to canopy reflectance model, and PROSPECT is about leaf optical properties spectra. It has been proved in (Baret et al., 2007; Lecerf et al., 2008) that this approach performs efficiently for low, medium, high and very high-resolution data and is therefore adapted to the

GreenLab model simulates the two basic processes of plant: development (organogenesis) and growth (organ expansion). In GreenLab, the organogenesis is simulated by an automaton, which gives the dynamics of number, age and type of organs in plant architecture (Yan et al., 2004). The organ expansion is simulated by a source-sink approach. At each time step *t*, the source function gives the biomass production of a plant, *Qt*, as a

1-exp - *<sup>t</sup> tt P*

In Eqn. (1), *S*P represents the projection area of an individual plant, which is equivalent to the inverse of planting density *d* when crop canopy is closed, i.e, *S*P=1/*d*. *r* is a model parameter that can be estimated inversely (Kang et al., 2008; Guo et al., 2006; Dong et al., 2008). In case that *E*t represents the intercepted light by crop, *r* means light use efficiency.

The produced biomass is shared among all growing organs in proportion to their current sink strength, based on common pool hypothesis. For an organ of type *O* and age *j*, its

, / *O OO*

In GreenLab, each type of organs (e.g. blade, sheath, internode, female organ) has certain relative sink strength *PO*, which may vary during the expansion of an individual organ, described by an empirical function *fOj*. Total plant demand *Dt* is sum of sink strength from all growing organs. The organ biomass, which is the accumulation of biomass increment during its life time, is dependent on the ratio between biomass production and demand (*Qt/Dt*), called source-sink ratio. According to the appearance time of each individual organ given by the automaton, and its increment in biomass since appearance, the biomass of all

*<sup>S</sup> Q E rS <sup>S</sup>*

*P*

(1)

*<sup>t</sup> j j t t q Pf Q D* (2)

(e.g. Idrisi, ENVI or eCognition) allow this kind of classification.

function of plant leaf area *St* and environmental factor *E*t, see Eqn. (1).

variability of satellite images available.

The ratio *S*t/*S*P gives a LAI series.

increment is biomass in computed as in Eqn. (2).

**2.2 GreenLab model** 

crops can be identified, on which some GreenLab hidden parameters can be initialized. Besides, we suppose that the noisy time series of LAI has been estimated from remote sensing images, which have occluded area due for instance to cloud coverage, aerosols, etc. The objective is to construct a continuous LAI series by re-estimating GreenLab model parameters. This finally enables to simulate the complete plant growth and to output 3D evolution of the observed crops, using empirical geometrical parameters for the given crop. The overall strategy is illustrated in Fig. 1. The different steps of the methodology are presented in following sections.

Fig. 1. Illustration of overall process of estimating biomass dynamics from remote sensing images
