**4. Conclusions**

314 Remote Sensing of Biomass – Principles and Applications

Fig. 9. RMSEr differences between FRB maps obtained using the three extraction methods

Previous research has shown the utility of wetness variables obtained from Landsat TM images to estimate FRB, regardless of the image date in the summer period. This is because these kinds of variables yield high and similar estimation FRB models using June, July and August scenes. In addition, statistical differences were not found in the moisture content of the four pines studied over the 3 month period (García-Martín et al.,

In relation to results in previous sections, the model selected using the extraction of 3 x 3 fixed windows was applied to the MID57 neocanal image derived from the July 2008 Landsat image in order to inventory the FRB. This model was finally selected because it had higher predictive power (R2 of 0.711 versus 0.595 on the model selected in the visual analysis method and 0.535 in the segmentation method) and allows the development of maps with the lowest estimation error. This last point shows that the limited set of plots used in the 3 x 3 fixed windows method is as representative for FRB estimation as those used in the other two methods, although they were composed of nearly double the

Figure 10 shows the estimation cartography obtained for the entire study area. This cartography allows calculation of the total amount of FRB resource at the provincial level, which amounts to 5,449,252 tons. In addition, with the high spatial resolution (25 x 25 m), the cartography precisely reveals the richest regions and FRB distribution within

This makes this cartography especially suitable for determining optimal areas, taking into account other spatial variables that also determine the technical and economic feasibility in the harvest of this renewable energy resource, for example: (i) slope, which influences the possibility of using machinery and its efficiency; (ii) distance to forest tracks, which determines a portion of the transport costs; and (iii) area of forest stands, which is related to the necessary displacement during the working day. These three spatial factors and the quantity of FRB derived from remote sensing data at 25 m resolution can be integrated into a Geographical Information System (GIS) to identify areas more suitable for harvest of FRB, with attention to principles of sustainable ecological forest management (Pascual et al., 2007;

**3.5 Inventory of potentially available FRB** 

2008a).

number of plots.

García-Martín et al., 2011).

them.

This study demonstrates the utility of Landsat TM images and forest inventory data in estimating FRB in Mediterranean areas. The methodology employed provides a continuous and complete estimation of FRB that overcomes problems associated with the limitations of point inventory. This information can be very useful in determining the most suitable areas in which to install power stations that make use of this resource; the lack of this type of information is one of the main problems currently facing the industry. As a result, an increase in the use of this kind of biomass resource will help to achieve the stated objectives of renewable energy production in Spain. Moreover, this is especially important considering two facts: (i) the current socio-economic context, with an increase in the price of petrol due to international instability, and public concerns regarding nuclear power, owing to Fukushima incident; and (ii) biomass is currently the only renewable energy that can be used as a strategic source of energy, as it is always available independent of weather conditions and can be stored easily prior to use (Jarabo, 1999).

The three sources of information typically used for biomass estimation (data from field sampling, satellite imagery and ancillary data) (Lu, 2006) have been carefully integrated, bearing in mind the specific characteristics of the Mediterranean environment. Thus, to reduce the problem that heterogeneity introduced into estimation models, three different methods were tested to extract radiometric data from a TM image. In agreement with previous studies focused on AGB estimation (i.e. Foody, 2001; Lu et al., 2004; Lu, 2005; Lu & Batistiella, 2005; Steininger, 2000) and with previous analyses carried out with our data (García-Martín et al., 2006, 2008a, 2008b), the derived spectral variables related to wetness showed the strongest correlations to FRB, independent of the method used to extract the

Using Remote Sensing to Estimate a Renewable Resource: Forest Residual Biomass 317

data from different sensors. This will allow the execution of better predictive multivariate regression models not affected by collinearity. Lastly, it would be useful to integrate data from physical variables that can be related to FRB quantity and distribution such as elevation, slope and aspect, as well as diverse biophysical parameters such as soil, lithology

This research has been supported by a grant provided by the Ministry of Science and Technology (AP2003-3097) and the project LIGNOSTRUM (AGL2002-03917-AGR-FOR),

Alonso, E.; Asín, J. & Pascual, J. (2005). Biomasa residual forestal: regresiones para las

Anaya, J.A.; Chuvieco, E. & Palacios-Orueta, A. (2009). Aboveground biomass assessment in

Asikainen, A.; Björheden, R. & Nousiainen, I. (2002). Cost of wood energy, In: *Bioenergy from* 

Austin, J.M.; Mackey, B.G. & Van Niel, K.P. (2003). Estimating forest biomass using satellite

Borsboom, N.W.J.; Hektor, B.; McCallum, B. & Remedio, E. (2002). Social implications of

Chander, G; Markham, B.L. & Helder, D.L. (2009). Summary of current radiometric

Chavez, P.S. (1996). Image-based atmospheric corrections: Revisited and improved. *Photogrammetric Engineering and Remote Sensing*, Vol. 62, pp. 1025– 1036 Chuvieco, E. (2002). *Teledetección ambiental. La observación de la tierra desde el espacio*, Ariel,

Colby, J.D. (1991). Topographic normalization in rugged terrain. *Photogrammetric Engineering* 

Domínguez, J. (2002). *Los sistemas de información geográfica en la planificación e integración de* 

Eriksson, H.M.; Hall, J.P. & Helynen, S. (2002). Rationale for forest energy production, In:

*Bioenergy from Sustainable Forestry: Guiding Principles and Practice*, J. Richardson; R.

(Eds.), 265-297, Kluwer Academic Publishers, Dordrecht, Netherlands Brown, S.L.; Schroeder, P. & Kern, J.S. (1999). Spatial distribution of biomass in forest of the

eastern USA. *Forest Ecology and Management*, Vol. 123, pp. 81-90

especies del género Pinus existentes en la provincia de Teruel, *La ciencia forestal: respuestas para la sostenibilidad. 4º Congreso Forestal Español*, CD-ROM, Zaragoza,

Colombia: A remote sensing approach. *Forest Ecology and Management*, Vol. 257, pp.

*Sustainable Forestry: Guiding Principles and Practice*, J. Richardson; R. Björheden; P. Hakkila, A.T. Lowe & C.T. Smith, (Eds.), 125-157, Kluwer Academic Publishers,

radar: an explanatory study in a temperature Australian Eucalyptus forest. *Forest* 

forest energy production, In: *Bioenergy from Sustainable Forestry: Guiding Principles and Practice*, J. Richardson; R. Björheden; P. Hakkila, A.T. Lowe & C.T. Smith,

calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. *Remote* 

financed also by the Spanish Ministry of Science and Technology.

*Ecology and Management*, Vol. 176, pp. 575-583

*Sensing of Environment*, Vol. 113, pp. 893-903

*and Remote Sensing*, Vol. 57, pp. 531-537.

*energías renovables*, CIEMAT, Madrid, Spain

Spain, September 26-30, 2005

Dordrecht, Netherlands

Barcelona, Spain.

and climate.

**6. References** 

**5. Acknowledgment** 

1237-1246

radiometric data. Analyses undertaken revealed statistically significant correlations with the three methods, but the use of 3 x 3 pixel windows and CV were indicated as the most appropriate to isolate a homogeneous group of plots that allow the most accurate regression equations (R2>0.7). In addition, despite the fact that these models were performed using half of the plots than those derived from the other two extraction methods, it was representative of the entire study area, as was shown in the validation results obtained at pixel level (RMSEr was similar in the cartographies obtained from the three methods). The visual analysis method did not yield such good results probably because of inaccuracies in the delineation of the homogeneous areas and because of the assumption of a constant value of BRF for the entire area from one precise location. The segmentation method did not improve the results either. This fact can be related to the lack of success to model effectively the spatial distribution pattern of BRF in the study area using RGB clustering. Finally, it is important to note that any of the methodologies tested to extract radiometric data successfully completed multivariate regressions models. This was because of the high autocorrelation among the well-correlated radiometric variables with FRB (wetness variables and vegetation indices). The use of these variables together will run a model affected by collinearity.

Validation cartography was based at the pixel level using plots that had not been previously used in the regression equation, removing only those with higher probability of error. This approach ensured the independence of the validation. RMSEr obtained in the three cartographies (above 65%) were better or at the same level than those produced in previous experiments conducted in boreal environments to estimate forest parameters using Landsat images. This is a positive result considering the higher complexity of Mediterranean forests. The remaining errors in the estimates of FRB can be related to the factors highlighted in García-Martín et al. (2008a,b): (i) inaccuracies in the fieldwork undertaken to establish the allometric equations and statistical analyses of the equations; (ii) problems involved in relating NFI plots to satellite data, mainly related to inaccuracies in the plot field placement (the three methodologies applied in the present study in linking ground data and remotely sensed data helped to reduce this problem, but it is still possible that errors were present in the final sample data); (iii) inaccuracies related to heterogeneity of the sample, because different pine species were considered and they are distributed in different regions of the study area; and (iv) limitations related to the spectral, radiometric, and spatial resolution of the TM in highly heterogeneous environments. However, despite these limitations, the size of the scenes and ease of distribution at no cost, make Landsat TM the most suitable in terms of achieving the objective of developing a useful methodology for estimations of FRB at regional-scales.

Finally, in order to improve outcomes, different lines of research should be considered. Firstly, different segmentation methods other than RGB clustering should be explored to determine if they can better model the forest spatial pattern and, as a result, obtain more accurate estimation models. Concerning this, the eCognition and the Definiens Developer segmentation procedures offer the possibility of considering additional features together with spectral data, based on similarities in shape and size. Secondly, focusing only on Landsat data, it is necessary to explore the use of other statistical methods that allow highly auto-correlated dependent variables to be considered jointly. In addition, it would be useful to explore the capability of hyperespectral sensors to identify narrow spectral bands highly correlated to FRB and poorly correlated between them and to examine the capability of SAR data from different sensors. This will allow the execution of better predictive multivariate regression models not affected by collinearity. Lastly, it would be useful to integrate data from physical variables that can be related to FRB quantity and distribution such as elevation, slope and aspect, as well as diverse biophysical parameters such as soil, lithology and climate.
