Sara Jurdao^a,, Marta Yebra^a,b and Emilio Chuvieco^a
a Department of Geography, University of AlcalÌÁ. Calle Colegios 2, 28801. AlcalÌÁ de Henares. Spain.
^bCSIRO Land and Water, GPO Box 1666, Canberra ACT 2601, Australia.

Abstract

Live fuel moisture content (LFMC) is a key variable in fire danger assessment. Recent studies have developed reasonably accurate models to determine LFMC in Mediterranean ecosystems by the inversion of radiative transfer models within ecologically based parameters. However, areas with temperate climate have received less attention. The objective of this study was to estimate LFMC for temperate grassland and shrubland located in the Eurosiberian region of Spain. To achieve this, we first assessed the adequacy of already published Mediterranean models to the Eurosiberian region. Secondly, we recalibrated the Mediterranean models to better resample temperate ecosystems by using ecological data collected in field. Finally, we proposed an alternative inversion procedure based on the look up table (LUT) technique. Reflectance from the first seven bands of MODIS and the NDII₆ vegetation index were used to achieve this.

The last approach was the one that efficiently estimated LFMC, mainly for higher danger situations (RMSE equals 30.6åÊ and 18.81percent for grassland and shrubland with LFMC of less than 200åÊ and 105 percent, respectively). This approach can be used together with previous models developed for Mediterranean grassland and shrubland to monitor LFMC of the Iberian Peninsula with a standardized methodology.

1. Introduction

Climate change alters fire regimens and increases fire danger [1, 2]. Fuel moisture content of live vegetation (LFMC) is one of the most dynamic variables of wildfires, so there is a growing interest in accurately estimating this variable in our warming world.

Satellite images constitute a great tool in this regard, since they periodically cover vast areas. The latest tendencies in the use of remote sensing data to determine LFMC are based on one of two approaches: (i) empirical models which relate satellite products with field LFMC observations [3, 4] and (ii) radiative transfer models (RTM) åÊshown to estimate LFMC more accurately over large and geographically distant areas [5]. In the context of RTM, LFMC can be determined by look-p table (LUT) inversion methodologies that allow the identification of ambiguous situations involving several sets of input parameters. This approach produces a modeled result that agrees with the observations within a tolerance [6]. The estimation of LFMC using LUT inversion methodologies is achieved in three steps: 1) simulation scenario establishment, 2) running RTM in ‰ÛÏforward mode‰Û within the established simulation scenario, and 3) finding which portion of the modeled spectrum is most similar to an observed one using a merit function of spectral similarity (‰ÛÏbackward mode‰Û or inversion). Alternatively, a multivariate linear regression analysis can be also carried out to establish a theoretical relationship between LFMC and spectral information stored in the LUT. As a result, an equation is fitted with the most explanatory variables.

Yebra and Chuvieco [5, 7] established ecological simulation scenarios for Mediterranean shrublands and grasslands that obtained accurate results (RMSE of 19.77 åÊand 24.57 percent, respectively). As far as we are aware, these ecological approaches have not been applied to temperate ecosystems. Consistent with this, the present study estimates LFMC of grassland and shrubland located in the Eurosiberian region of Spain.

2. Data and methods

2.1. Field data

A field campaign was carried out in the Eurosiberian region of Spain from April to September of the years 2006, 2009 and 2010.åÊ A total of 12 plots were selected to collect samples of two vegetation types: grassland (eight plots) and shrubland (four plots) (Figure 1).

Three samples per plot (of around 80 g) were collected and weighed in the field with a field balance so as to get the fresh weight. Then, they were carried to the laboratory and dried in an oven for 48 hours at 60 degrees C [10] to obtain the dry weight. LFMC was computed following Equation 1.

 

 

 

 

 

W_w is the wet weight and W_d is the dry weight.

Average values of the three samples per plot were assigned to each plot for each sampling date. We followed the rigorous standard protocol described in Chuvieco et al. [11]. The obtained data (35 samples of grassland and 28 samples of shrubland) can be freely accessed though the database developed by the Department of Geography of the University of AlcalÌÁ.

2.2. Satellite information

The input satellite information was the 500-meter moderate-resolution imaging spectroradiometer reflectance product (MCD43A4, [12]). The data for the study areas in the sampling dates were downloaded from NASA and reprojected from the sinusoidal system to the UTM coordinate system (Datum European 1950) using nearest neighbor interpolation resampling. Higher spatial resolution images from Landsat were used to select homogeneous pixels from the plots inside aåÊ three-by-three meter window and avoid mixed pixels. Then, following Yebra et al. [7], the median value of the pixels assigned to each plot was computed to estimate the final LFMC value.

2.3. LFMC retrieval

LFMC was retrieved using MODIS data and PROSAILH model. This model consists of combining two RTM: PROSPECT [13] and SAILH [14] models. The former simulates leaf reflectance and transmittance as a function of the number of layers inside the leaf (N), the chlorophyll concentration (C_a+b), the equivalent water thickness (EWT), (3) and the dry matter content (DMC) (4). The SAILH model simulates canopy reflectance using the output reflectance and transmittance from PROSPECT, the leaf area index (LAI), the leaf inclination distribution function (LIDF),soil reflectance, viewing and illumination conditions, and the hot-spot parameter (h).

Three alternative approaches were tested to retrieve LFMC (Figure 2): (1) implementation of the Mediterranean models in the Eurosiberian region, (2) recalibration of the models to resample the Eurosiberian region, and (3) testing a new inversion approach.

2.3.1. Implementation of Mediterranean models for the Eurosiberian region

We selected the grassland and the shrubland models as explained in Yebra et al. [7] and Yebra and Chuvieco [5], respectively. The simulated spectra were convolved to the first seven MODIS reflectance bands. Afterwards, a linear relation between LFMC and LAIåÊ was used to filter out unrealistic combinations between these parameters since the lowest LAI values were considered unlikely to combine with the highest LFMC [15]. Finally, shrublands spectra were linearly mixed with normal and dry soils in a proportion of 40- 60 percent to account for the heterogeneity in the fraction of cover.

The inversion procedure was different for each vegetation type. In the case of grassland, the authors performed a MLR analysis comparing the LFMC, computed as Equation 2, and other spectral and biophysical information stored in the LUT.

 

Equation 5 was then applied to obtain the estimated LFMC. The MOD15 product [17] was used as the source of LAI, and the NDII₆ was computed from the MODIS reflectance product.

The inversion procedure for shrublands was based on the LUT technique. The merit function used to compare simulated and observed spectra was the spectral angle (SA, [18], Eq. 7).

2.3.2. Re-calibration of the Mediterranean models: Eurosiberian models

The second approach tested consisted in a recalibration of the aforementioned models (Figure 3). To achieve this, a field campaign was carried out in Asturias (northern Spain) following the protocol described in Yebra et al. [7]. Field measurements were different from these used to obtain the LFMC.

Based on the measured data, we increased the upper end of the range of C_a+b, EWT and DMC of grasslands (Figure 3).

A MLR analysis was carried out to fit a LFMC model for grassland (Eq. 8).

In the case of shrubland, a set of 49 simultaneous measurements of leaf parameters obtained in Asturias was used to parameterize the RTM. Another set of six combinations of leaf parameters from the Caba̱eros National Park (used to calibrate the Mediterranean models) was included to increase the range of low LFMC

values. At canopy level, the LAI was increased and the simulated spectra were linearly combined with the soil in a proportion of 60- 40 percent.

For both models, the solar zenith angle (ëü_s) was fixed at 0 in consonance with our reference reflectance product (MCD43), which models the values as if they were taken from the nadir view. Finally, a reference soil measured using the GER2600 (GER Corp., Millbrook, N.Y.) spectroradiometer was used to represent the soil reflectance.

2.3.3. New inversion proposal

We proposed a new inversion approach using the LUT inversion technique for both grassland and shrubland.

Several empirical studies have demonstrated that NDII₆ provide better estimations of LFMC than other spectral indices (RMSE ‰äö 11 percent, [4]; RMSE=24.57 percent, [7]). Consequently, we decided to keep NDII₆ in the inversion in addition to the reflectance bands. However, LAI was discarded according to Jurdao et al.[19].

2.4. Assessment

The comparison between observed and estimated LFMC was performed according to the root mean square error (RMSE), the systematic RMSE (RMSEs), and the unsystematic RMSE (RMSEu) [20]. An appropriate model should present an RMSEu higher than the RMSEs as the model estimates errors should be random.

Because of the importance of giving the maximum accuracy for lower LFMC conditions, we presented the results of pooling data by different LFMC thresholds. For shrublands, the Mediterranean model did not include situations with LFMC higher than 135.7 percent[5], so the results were presented below this threshold. The moisture of extinction , defined as the moisture threshold above which •Âre cannot be sustained [21], also was used as a threshold for shrubs (LFMC = 105 percent [22]). For grasslands, we established a threshold of 200 percent åÊto try to avoid samples highly affected by the characteristic dew of this region.

Finally, the model that provided the highest accuracy was analyzed by the temporal evolution similitude between observed and estimated LFMC.

3. Results and discussion

3.1. Mediterranean models assessment

Mediterranean models could not adequately be applied to the Eurosiberian region (Figure 4). The Mediterranean grassland model obtained a RMSE higher than 200 percent. For the Mediterranean shrubland model, the systematic errors were much higher than the unsystematic errors. The climate, soil, geography, anthropic factors, etc., determine the type of vegetation found in a region [23]. The Mediterranean and the Eurosiberian regions show dramatic climatic differences, most notably the summer drought that characterizes the Mediterranean region. Thus the main groups of vegetation found in each region differ. For this reason, specific models that reflect the ecological peculiarity of each region wereåÊ necessary.

3.2. Eurosiberian models assessment

Similar results were obtained when applying the Eurosiberian grassland model. The RMSE for both thresholds was around 200 percent (Figure 5). The model produced important overestimations for high LAI or NDII₆ values, reaching a LFMC above 500 percent since the estimation was based on a linear equation. Therefore, another inversion technique was considered indispensable.

The Eurosiberian shrubland model showed better results (RMSE was 28.38 percent for all cases and RMSE was 26.55 percent for LFMC below 105 percent) and presented RMSEs greater than RMSEu.

3.3. New inversion approach assessment

The Eurosiberian models applied to grass with the new LUT inversion approach greatly improved the results (Figure 6), especially when only the cases with LFMC less than 200 percent were considered (RMSE notably decreased from 41.9åÊ to 30.6 percent, and the RMSEs was close to 0 while RMSEu approximated RMSE.). Part of the error came from the field data used for the validation. Observed LFMC was computed as the average of three samples of each species per plot, yielding an average standard deviation of 22.81 percent for grassland. There were many cases in which it was higher than 40 percent, reflecting the uncertainty of the field measurements since all samples were randomly collected within the plot. Additionally, it was remarkable that, although we avoided sampling when it rained 48 hours or less before the scheduled sampling date, the samples were occasionally wet because of the dew and fog typical of the mountain grasslands of this region.

In the case of shrubland, the RMSE was lower than the RMSE obtained with the other inversion approach, and the RMSEs was much lower than the RMSEu (RMSEs was 6.38 percent and RMSEu, 17.69 percent). When only the cases below the moisture of extinction were considered, the estimates slightly improved (RMSE was 18.81 percent).

The temporal evolution of the observed and estimated LFMC presented similar patterns (Figure 7). The greater susceptibility of grassland to environmental conditions in comparison to shrubland implied rapid LFMC changes in the former vegetation type. Eurosiberian grasslands could reach LFMC values exceeding 300 percent even in summer, due to the moisture contained within the plant tissues because of the dew, but it might drop to below 50 percent when the period was particularly dry. Shrubs have better developed mechanisms to resist summer drought [15], and they have a greater capacity to extract water from the soil [3]. Hence, due to the water available in this region, shrubs tended to show a more uniform behavior with LFMC values above 50 percent. Both shrublands and grasslands models obtained some overestimations and underestimations (Figure 7). From the fire prevention point of view, overestimation is considered less desirable since it would tend to reduce the fire risk rating, although false alarms also are undesirable [7]. However, the overestimations occurred for LFMC above 200åÊ and 106 percent for grassland and shrubland respectively, so it was not considered an important limitation since these LFMC intervals are considered low fire risk. Conversely, the small number of underestimations or false alarms occurred mainly for the lowest LFMC.

4. Conclusion

Through this research, a field campaign was designed for Eurosiberian grasslands and shrublands to collect LFMC samples and other ecological information. We found out that the Mediterranean models were not accurate when applied to the Eurosiberian region (RMSE exceeded 200 percent for grassland and RMSEs exceeded RMSEu for shrubland). Therefore, we re-parameterized the models with the collected ecological data and proposed inverting the models using the LUT technique, taking into account the reflectance bands and the NDII₆ index. The results showed the ability of the models to estimate LFMC, mainly for the higher fire danger situations, i.e., those with lower LFMC (RMSEequalled 30.6 percent for grassland with LFMC of less than 200 percent; RMSE åÊequalled 18.81 percent for shrubland with LFMC of less than 105 percent). Part of the errors in high LFMC values for grasslands came from the field measurements since the average standard deviation of the samples collected was in some cases higher than 20 percent.

The new approach proposed in this work to estimate LFMC of Eurosiberian grassland and shrubland can be used together with previous models developed for the Mediterranean regionåÊ to monitor LFMC of the whole Iberian Peninsula with a standardized methodology.

Finally, radiative transfer models are physical models, so they can be extrapolated to different study areas than the calibration ones when they share similar climatic conditions. Therefore, future research will be focused on testing if the models obtained can adequately estimate LFMC of temperate shrublands and grasslands in other areas.

6. Acknowledgements

This research has been funded by the Spanish Ministry of Education and Science and supported by the FIREGLOBE project (CGL2008-01083). The authors would like to thank the assistance of Juan Pablo Guerschman (CSIRO Land and Water, Canberra, Australia) and the technical team of the deputations of Asturias and the Basque country.

7. References
[1]åÊåÊ åÊD. M. J. S. Bowman, J. K. Balch, P. Artaxo, W. J. Bond, J. M. Carlson, M. A. Cochrane, C. M. D’Antonio, R. S. DeFries, J. C. Doyle, S. P. Harrison, F. H. Johnston, J. E. Keeley, M. A. Krawchuk, C. A. Kull, J. B. Marston, M. A. Moritz, I. C. Prentice, C. I. Roos, A. C. Scott, T. W. Swetnam, G. R. Van der Werf, and S. J. Pyne, “Fire in the Earth System,” Science, vol. 324, pp. 481-484, 2009.
[2]åÊåÊ åÊB. J. Stocks, A. A. Fosberg, T. J. Lynham, L. Mearns, B. M. Wotton, Q. Yang, J. Z. Jin, K. Lawrence, G. R. Hartley, J. A. Mason, and D. W. McKenney, “Climate Change and Forest Fire Potential in Russian and Canadian Boreal Forests,” Climatic Change, vol. 38, pp. 1-13, 1998.
[3]åÊåÊ åÊM. Garcia, I. Aguado, and E. Chuvieco, “Combining AVHRR and meteorological data for estimating live fuel moisture content in forest fire danger rating,” Remote Sensing of Environment, vol. 112, pp. 3618-3627, 2008.
[4]åÊåÊ åÊG. Caccamo, L. A. Chisholm, R. A. Bradstock, M. L. Puotinen, and B. G. Pippen, “Monitoring live fuel moisture content of heathland, shrubland and sclerophyll forest in south-eastern Australia using MODIS data,” Interational Journal of Wildland Fire, vol. Published online, 2011.
[5]åÊåÊ åÊM. Yebra and E. Chuvieco, “Linking ecological information and radiative transfer models to estimate fuel moisture content in the Mediterranean region of Spain: Solving the ill-posed inverse problem,” Remote Sensing of Environment, vol. 113, pp. 2403-2411, 2009.
[6]åÊåÊ åÊN. Gobron, B. Pinty, M. M. Verstraete, J. V. Martonchik, Y. Knyazikhin, and D. J. Diner, “Potential of multiangular spectral measurements to characterize land surfaces: Conceptual approach and exploratory application,” Journal of Geophysical Research, vol. 105, pp. 17539-17549 2000.
[7]åÊåÊ åÊM. Yebra, E. Chuvieco, and D. Ria̱o, “Estimation of live Fuel Moisture Content from MODIS images for fire risk assessment,” Agricultural and Forest Meteorology, vol. 148, pp. 523-536, 2008.
[8]åÊåÊ åÊM. Yebra, E. Chuvieco, and I. Aguado, “ComparaciÌ_n de modelos empÌ_ricos y de transferencia radiativa para estimar contenido de humedad en pastizales: Poder de generalizaciÌ_n.,” Revista de TeledetecciÌ_n, vol. 29, pp. 73-90, 2008.
[9]åÊåÊ åÊM. Yebra, M. E. Beget, P. Oricchio, and C. Di Bella, “InversiÌ_n de modelos de simulaciÌ_n de la reflectividad para la estimaciÌ_n del estado hÌ_drico del combustible vivo en matorrales y pastizales de la Argentina,” Serie GeogrÌÁfica., vol. 16, pp. 51-59, 2010.
[10]åÊåÊ åÊD. X. Viegas, T. P. Viegas, and A. D. Ferreira, “Moisture content of fine forest fuels and fire occurrence in central Portugal,” The International Journal of Wildland Fire, vol. 2, pp. 69-85, 1992.
[11]åÊåÊ åÊE. Chuvieco, M. Yebra, S. Jurdao, I. Aguado, J. Salas, M. GarcÌ_a, H. Nieto, A. de Santis, D. Cocero, D. Ria̱o, S. MartÌ_nez, E. Zapico, C. Recondo, J. MartÌ_nez-Vega, M. P. MartÌ_n, J. Riva, F. PÌ©rez, and F. RodrÌ_guez-Silva, “Field fuel moisture measurements on Spanish study sites. Department of Geography, University of AlcalÌÁ,”, 2011.
[12]åÊåÊ åÊC. B. Schaaf, F. Gao, A. H. Strahler, W. Lucht, X. Li, T. Tsang, N. C. Strugnell, X. Zhang, Y. Jin, J.-P. Muller, P. Lewis, M. Barnsley, P. Hobson, M. Disney, G. Roberts, M. Dunderdale, C. Doll, R. P. d’Entremont, B. Hug, S. Liang, J. L. Privette, and D. Roy, “First operational BRDF, albedo nadir reflectance products from MODIS,” Remote Sensing of Environment, vol. 83, pp. 135‰ÛÒ148, 2002.
[13]åÊåÊ åÊS. Jacquemoud and F. Baret, “PROSPECT: a model of leaf optical properties spectra,” Remote Sensing of Environment, vol. 34, pp. 74-91, 1990.
[14]åÊåÊ åÊW. Verhoef, “Light scattering by leaf layers with application to canopy reflectance modeling: the SAIL model,” Remote Sensing of Environment, vol. 16, pp. 125-141, 1984.
[15]åÊåÊ åÊF. Valladares, EcologÌ_a del bosque mediterrÌÁneo en un mundo cambiante. Madrid, Spain: Ministerio de Medio Ambiente, EGRAF, S. A, 2004.
[16]åÊåÊ åÊM. A. Hardisky, V. Klemas, and R. M. Smart, “The influence of soil salinity, growth form, and leaf moisture on the spectral radiance of Spartina alterniflora canopies,” Photogrammetric Engineering and Remote Sensing, vol. 49, pp. 77-83, 1983.
[17]åÊåÊ åÊR. Myneni, R. S.W, J. Glassy, and P. Votava, “FPAR, LAI Userå«s Guide. (ESDT: MOD15A2) 8-day CompositeNASA MODIS Land Algorithm,” Terra MODIS Land Team Doc v1.0-09.14.2000-jmg, 2000.
[18]åÊåÊ åÊF. A. Kruse, A. B. Lefkoff, J. B. Boardman, K. B. Heidebrecht, A. T. Shapiro, P. J. Barloon, and A. F. H. Goetz, “The Spectral Image Processing (SIPS) – Interactive Visualization and Analysis of Imaging Specrometer Data,” Remote Sensing of Environment, vol. 44, pp. 145-163, 1993.
[19]åÊåÊ åÊS. Jurdao, M. Yebra, J. P. Guerschman, and E. Chuvieco, “Regional estimation of woodland moisture content by inverting Radiative Transfer Models,” Remote Sensing of Environment, vol. 132, pp. 59-70, 2013.
[20]åÊåÊ åÊC. J. Willmott, “Some Comments on the Evaluation of Model Performance,” Bulletin of the American Meteorological Society, vol. 63, pp. 1309-1313, 1982.
[21]åÊåÊ åÊR. C. Rothermel, “A Mathematical Model for Predicting Fire Spread in Wildland Fuels,” USDA, Forest Service, Intermountain Forest and Range Experiment station, Ogden, Utah, USA, Research Paper INT-115, January 1972.
[22]åÊåÊ åÊA. Dimitrakopoulos and K. K. Papaioannou, “Flammability assessment of Mediterranean forest fuels,” Fire Technology, vol. 37, pp. 143-152, 2001.
[23]åÊåÊ åÊS. Rivas MartÌ_nez, Memoria del mapa de series de vegetaciÌ_n de Espa̱a 1:400000. Madrid, Spain: ICONA. Ministerio de agricultura, pesca y alimentaciÌ_n, 1987.