Generation of crown bulk density for Pinus sylvestris L. from lidar

Transcription

1 Remote Sensing of Environment 92 (2004) Generation of crown bulk density for Pinus sylvestris L. from lidar David Riaño a,b, *, Emilio Chuvieco b, Sonia Condés c, Javier González-Matesanz d, Susan L. Ustin a a CSTARS, Department of Land, Air, and Water Resources, University of California, One Shields Ave. Davis, CA , USA b Dept. de Geografía, Universidad de Alcalá. Colegios 2, E Alcalá de Henares, Spain c Dept. de Economía y Gestión Forestal, ETS de Ingenieros de Montes, Universidad Politécnica de Madrid, Ciudad Universitaria S/N. E Madrid, Spain d Servicio de Geodesia. Instituto Geográfico Nacional, General Ibáñez Íbero 3, Madrid 28003, Spain Received 27 July 2003; received in revised form 27 November 2003; accepted 18 December 2003 Abstract The use of lidar data to estimate critical variables needed for modeling wildfire behavior was tested on a Scots pine forest (Pinus sylvestris L.) in central Spain. Lidar data accurately estimated crown bulk density at the plot level (r 2 = 0.80). Lidar data could be used to directly estimate crown volume (r 2 = 0.92) and foliage biomass (r 2 = 0.84), which together produced better results than directly fitting the lidar data to crown bulk density. Incorporating equations that relate tree diameter at breast height and other forest parameters improved estimates of foliage biomass. Individual tree level analyses were not completely successful due to difficulty in accurately assigning laser pulses to the correct tree (r 2 = 0.14). D 2004 Elsevier Inc. All rights reserved. Keywords: Lidar; Forest canopy structure; Crown bulk density; Tree height; Crown base height; Foliage biomass; Crown volume; Wildland fire modeling 1. Introduction Crown fires move faster and are harder to suppress than ground fires. Therefore modeling crown fire behavior is important for fire management activities, such as planning prescribed burning or prediction of fire spread in ongoing fires. Crown bulk density, described as the foliage biomass divided by the crown volume is one of the most critical variables for modeling crown fire behavior (Scott, 1999), since where the trees are denser, fire easily spreads from one tree to the other. Lidar is particularly valuable for measuring crown bulk density because it does not saturate at high biomass levels and can easily provide tree crown information (Drake et al., 2002b). Several authors have shown that lidar produces better results than aerial photography, airborne hyperspectral sensors or airborne profiling radar (Hyyppa et al., 2000; Lefsky et al., 2001). * Corresponding author. CSTARS, Department of Land, Air, and Water Resources, University of California, One Shields Ave. Davis, CA , USA. Tel.: ; fax: addresses: driano@cstars.ucdavis.edu (D. Riaño), emilio.chuvieco@uah.es (E. Chuvieco), scondes@montes.upm.es (S. Condés), fjgmatesanz@mfom.es (J. González-Matesanz), slustin@ucdavis.edu (S.L. Ustin). Crown bulk density can be estimated from several statistics that are extracted from lidar data, and foliage biomass and crown volume can be independently derived. However, most biomass lidar studies do not predict foliage biomass directly, but rather, total aboveground biomass (Drake et al., 2002a; Lefsky et al., 1999). A similar problem applies to estimates of crown volume, since the variable usually obtained is the total stand volume (Naesset, 1997). Identifying the crown base height (Naesset & Økland, 2002) could improve estimates of crown volume because this measurement avoids inclusion of the understory. Forest lidar validation studies are generally performed at the plot level (Means et al., 2000). Sometimes lidar data analysis is performed at the tree level, but the comparison with field data is typically performed at the plot level (Hyyppa et al., 2001). Full-waveform large footprint systems are preferred at this level of analysis. The advantages over small footprint data are reduced data processing with only one X and Y needed to locate the entire waveform and the fact that tops of trees are not missed in the waveform (Dubayah & Drake, 2000). However, since the location of individual tree positions within the full-waveform large footprint is unknown, extending analysis to the tree level is difficult. Small footprint systems can provide information about individual trees for fire behavior models, a significant /$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi: /j.rse

2 346 D. Riaño et al. / Remote Sensing of Environment 92 (2004) advantage. However, results are noisy when compared to those at the plot level (Morsdorf et al., 2003; Naesset & Økland, 2002). One of the difficulties in lidar data at the tree scale is determining the correct geo-location of individual trees. Also, the tops of trees can be missed by the lidar if point density is insufficient (Dubayah & Drake, 2000). Riaño et al. (2003a) described a methodology for estimating different plot level forest parameters from lidar data that are needed for predicting fire behavior. This paper applies the equation described in Riaño et al. (2003a), to estimate crown bulk density for Scots pine (Pinus sylvestris L.) trees. Each parameter involved in the crown bulk density computation, i.e. foliage biomass, crown volume, tree height, and crown base height, was analyzed independently. Additionally, we also explored the capability of lidar data to estimate crown bulk density and its components at the tree level. 2. Materials and methods 2.1. Study site The study was conducted southwest of Canencia, in the Sierra de Guadarrama, about 50 km north of the city of Madrid, Spain (longitude, jW; latitude, jN). A naturally regenerated and planted Scots pine forest dominated the site. The forest is intensively managed for wood production and has a regular tree size and spatial distribution. The understory was generally cleared to reduce fire danger. Tree height ranged between 9 and 16 m, located at elevations between 1360 and 1535 m, and slopes ranging between 10% and 23%. units (double differences integer solution) served to locate points outside the forest. Field topography was used in locating the center of the plots. The trees were located within the plot, measuring aspect corrected for declination and distance to the center of the plot. A KB-14/360R Suunto compass (Suunto, Finland) and a Vertex III hypsometer (Haglof, Sweden) were used, respectively. The tree measurements were DBH, tree height, crown base height and crown projected surface in the cardinal North, East, South and West directions (Fig. 1). Heights were calculated with the Vertex III hypsometer. Crown base height was measured as the distance to the first live branch; DBH and crown projected surface were obtained using a measuring tape. Foliage biomass was computed for each tree based on the previously established allometric equation from DBH. Tree crown volume was calculated from tree height, crown base height and crown projected surface. We assumed an ellipsoid ( mathworld.wolfram.com/ellipsoid.html) for all trees. Table 1 shows the computed variables for each plot that were related to the lidar parameters Estimation of crown bulk density from lidar data at plot level The lidar data were acquired 16 August 2002 with Toposys II ( This system recorded the first and last return for small footprints with a diameter of 0.45 m. The lidar recorded one laser pulse every 1.73 m (distance between the center points of the footprints) for the 2.2. Generation of an allometric equation for biomass estimation Destructive sampling was used to estimate foliage biomass through an allometric equation following the methodology described in Lopez-Serrano et al. (2000). We randomly selected 10 pine trees from different size classes with diameter at breast height (DBH) ranging between 14.0 and 32.8 cm. Coinciding with logging activities, trees were cut down and all branches weighed on 14 June The foliage biomass of three branches each from the upper, middle, and lower part of the crown was weighed, providing a relationship between the weight of branches and foliage biomass. An allometric equation relating DBH and foliage biomass was also established Pine tree field measurements Pine tree sampling was done during summer A total of 10 circular plots were established in stands with a range of different size classes. We mapped all trees with DBH >7 cm within a radius of 10 m from the center point. The trees were located to an absolute accuracy of centimeters. Two GPS Fig. 1. Pine tree field measurements: tree height, crown base height, crown projected surface and diameter at breast height.

3 D. Riaño et al. / Remote Sensing of Environment 92 (2004) Table 1 Field measurements obtained for each plot Plot Number of trees Total foliage biomass (kg/m 2 ) Max tree height (m) Min crown base height (m) Percentile 10 of crown base height (m) Crown volume (m 3 /m 2 ) Crown bulk density (kg/m 3 ) across-track direction and one laser pulse every 0.11 m in the along-track direction. The raw Lidar data gave heights (Z) above the sea level and we needed the vegetation height above the ground for each laser pulse. The data provider generated the 1-m resolution digital terrain model (DTM) based on the bisection principle (von Hansen & Vögtle, 1999). We used this DTM as true ground height above the sea level to interpolate the raw Lidar data using the software Matlab 6.0 (The MathWorks, USA). We obtained for each laser pulse (X, Y) the interpolated ground height referenced to the sea level (Z ). Therefore Z Z was the vegetation height above the ground. The algorithm used was spline function interpolation (Sandwell, 1987). The first and last laser pulses were classified as in-canopy or non-canopy using Matlab 6.0 based on the Z coordinate for each plot. Various methods were tested: a 3-m fixed limit, minimum Euclidean distance clustering (as used in Riaño et al., 2003a), K-means clustering (used for tree segmentation in Morsdorf et al., 2003) and Expectation Maximization clustering. The K-means and Expectation Maximization algorithm were downloaded from David Corney s web page ( Different cluster sizes were tested. The separation between upper canopy and understory was based on the maximum distance between mean value of clusters when more than two clusters were generated (Fig. 2). If the cluster group was too small then it was not considered; that is, when the percentage of laser pulses of a group had less than 1/(total number of clusters) 2. The initial points for the K-means clustering were defined at regular height intervals starting from zero height to the maximum tree height within the plot. Both 1 and 500 interactions were computed. Crown bulk density (CBD) at the plot level was estimated by applying the equation described in Riaño et al. (2003a) for this species: CBD kg m 3 ¼ FBðkg=m2 Þ CVðm 3 =m 2 Þ Where FB was the estimated foliage biomass, CV was the estimated crown volume. Crown volume at plot level was obtained considering the crown length and the covered ground, which is the vertically projected canopy area per unit ground area: CV m3 m 2 ¼ðTH CBHÞ*CG Where TH was the estimated tree height, CBH was the estimated crown base height and CG was the estimated covered ground. Each parameter involved in crown bulk density computation was estimated independently. Foliage biomass was calculated from mean height of all laser pulses (MLH) and from 99th percentile of canopy hits (P99); tree height from P99 and from maximum laser height (MaxLH); and crown base height from the 1st percentile of canopy hits (P1) and minimum canopy laser height (MinCLH). Covered ground was already validated in Riaño et al. (in press) using digital fish-eye photographs Fig. 2. Separation between upper canopy and understory when more than two clusters were generated. According to this figure, the maximum distance between mean value of clusters was between clusters 2 and 3, therefore clusters 1 and 2 were classified as the upper canopy and cluster 3 as the understory.

4 348 D. Riaño et al. / Remote Sensing of Environment 92 (2004) from the percentage of laser canopy hits (CH%). We also applied a correction for the effect of lower canopy shading by taller canopy described in Riaño et al. (2003a) that changes the CH%: h¼th CG ¼ X h¼cbh relchpðhþ Where relchp was the sum of the corrected relative canopy height profile between tree height and crown base height. This correction was erroneously reported in this paper since these relative values did not include the ground laser hits. Therefore the corrected term was calculated as: CG ¼ h¼th X h¼cbh Lr cor ðhþ Tot h Where Lr cor was the sum of the corrected reflected laser pulses from the canopy and Tot cor was the total number of laser pulses for that plot. We tested for ability to generate crown volume from P99, for the tree height, P1, for the crown base height, together with the uncorrected and corrected CH%, for the covered ground Estimation of crown bulk density from lidar data at tree level The first step for estimating crown bulk density at tree level was to assign each laser pulse to each tree, using a threedimensional (X, Y and Z) K-means clustering, following Morsdorf et al. (2003). The K-means algorithm was selected because it was the fastest method that consistently worked well at the plot level for the laser pulses classification as incanopy or non-canopy. The initial points for the clusters were extracted from the field tree inventory. We considered the position of each tree and four Z values, equal to zero, crown base height, half crown height and tree height, respectively. The first cluster grouped the ground laser pulses for each tree, and the other three identified canopy positions. We performed clustering with 1 and 500 interactions. The 500 interactions result was expected to be more independent from the locations of the initial cluster points. Estimators similar to those used at the plot level were applied at the tree level: MLH and P99, for foliage biomass, MaxLH and P99, tree height; and MinCLH and P1, for crown base height. Crown volume was estimated using the equation used at the plot level, i.e., considering the crown length and the covered ground. We also adjusted the crown volume estimate to match the field measurements. That is, we considered the furthest laser pulse in N, E, S and W directions, within F 20j to estimate the crown projected surface. Finally, crown volume was calculated from the P99, P1 and crown projected surface assuming an ellipsoidal shape. 3. Results The destructive sampling of foliage biomass generated the following equation, relating DBH and foliage biomass (FB): FB kg ¼ 4:48*10 4 *DBH 3:15 tree R 2 ¼ 0:95 P value < 0:001 This allometric equation was used to calculate the foliar biomass of each tree, since DBH was measured in all plots, Table 2 Correlation coefficients (r 2 ) between lidar estimators and variable used in the computation of crown bulk density at plot level (n = 10 pine plots) Estimated variable Lidar estimator Fix 3 m MinDis EM Kmeans Kmean1int Six K-means Equations Foliage biomass 1 MLH 0.84*** 0.84*** 0.84*** 0.84*** 0.84*** 0.84*** y = 0.37*e (x * 0.15) Foliage biomass P *** 0.81*** 0.81*** 0.81*** 0.81*** 0.81*** y = 0.20*e (x * 0.11) Max tree height MaxLH 0.95*** 0.95*** 0.95*** 0.95*** 0.95*** 0.95*** y = 1.03*x Max tree height P *** 0.94*** 0.94*** 0.94*** 0.94*** 0.94*** y = 1.01*x Min crown base height MinCLH 0.34* * 0.27 y = 0.22*x Min crown base height P * y = 0.24*x Percentile 10 of crown MinCLH 0.72** 0.95*** 0.40* 0.95*** 0.85*** 0.75** y = 0.57*x base height Percentile 10 of crown P1 0.95*** 0.95*** 0.74** 0.97*** 0.95*** 0.88*** y = 0.60*x base height Crown volume 2 (P99 P1)*CH% 0.45* 0.36* * 0.36* 0.92*** y = 1.11*x 1.71 Crown volume (P99 P1)*CH%cor 0.35* * ** y = 0.91*x 0.76 Crown bulk density MLH/((P99 P1)*CH%) ** y = 0.27*x 0.08 Crown bulk density Foliage biomass 1 /crown volume * 0.43* 0.44* 0.41* 0.41* 0.80*** y = 1.55*x 0.12 Different methods were used to identify canopy hits. A fixed 3-m limit, a three-group cluster using minimum distance (MinDis), Expectation Maximization (EM), K-means with 500 interactions (Kmeans), K-means with one interaction (Kmean1int), and a six-group cluster using K-means with 500 interactions (six Kmeans). Equations are presented for six Kmeans. * P < 0.1. **P < ***P <

5 D. Riaño et al. / Remote Sensing of Environment 92 (2004) Fig. 3. Crown volume estimation at plot level using three K-means clusters (right) and using six K-means clusters (left). Each crown volume estimation is presented with circles whereas the outlier plot is presented with a cross. it provided field estimates of foliage biomass at tree and plot levels. Table 2 represents the estimates at the plot level for foliar biomass, maximum tree height, minimum crown base height, 10th percentile of crown base height, crown volume and the final crown bulk density. The relationships between foliage biomass at the plot level, generated from the previous allometric equation, and lidar variables generally followed an exponential relationship. MLH provided the best fit. Tree height estimates were good for both MaxLH and P99. The equation relating tree height and lidar data had a near 1:1 relationship. However, lidar produced much higher crown base height values than the field data. P1 was better than MinCLH for the crown base height estimation. Both of these lidar variables were much worse estimators of the minimum crown base height than the 10th percentile of crown base height. The lidar CH% correction that adjusted for shading of the lower canopy by a taller canopy did not improve prediction of crown volume compared to the uncorrected CH%. The crown bulk density estimate was better using foliage biomass and crown volume than the direct prediction of crown bulk density from other lidar variables. MLH and MaxLH estimators were independent of the clustering method (Table 2). Therefore, foliage biomass and tree height estimations were the same for all methods. P99 depended on the number of canopy hits and consequently on Fig. 4. Representation of laser pulses for one of the plots, which was an outlier due to a tree much smaller than the rest of the canopy, causing crown volume overestimation. X coordinates represent distance to plot center. Table 3 Correlation coefficients (r 2 ) between lidar estimators and variables used in the computation of crown bulk density at tree level (n = 183 pine trees) Estimated variable Lidar estimator Kmeans 500 int. Kmeans 1 int. Equations Foliage biomass MLH y = 4.55*e (x * 0.14) Foliage biomass 1 P y = 1.58*e (x * 0.17) Tree height MaxLH y = 0.96*x Tree height P y = 0.96*x Crown MinCLH y = 0.59*x base height Crown P y = 0.58*x base height Crown volume (P99 P1)* y = 19.02*x CH% Crown volume 2 Estimated crown volume y = 0.71*x Crown bulk density Foliage biomass 1 /crown volume y = 0.59*x All P < K-means with 1 and 500 interactions. Equations are presented for K-means with one interaction.

6 350 D. Riaño et al. / Remote Sensing of Environment 92 (2004) the clustering method. But it produced inappreciable variations between the methods used to estimate foliar biomass and tree height. Out of all clustering methods proposed to discriminate between upper canopy and understory, and obtain crown base height, the only method that always failed was the Expectation Maximization clustering. This clustering method also failed for the estimation of crown volume, since this approach was based on the estimation of crown base height. K-means with 500 interactions provided results similar to only one interaction, results being only slightly better for the crown volume variable. An outlier also caused a large decrease in crown volume accuracy for all methods, which disappeared only when considering six clusters (Fig. 3). A tree that was shorter than the other trees in one plot caused this outlier, which produced an overestimation of crown length. The estimated crown base height was too low (Fig. 4). Table 3 shows the results at the tree level. Foliage biomass estimation followed an exponential relationship similar to plot level analysis, and P99 performed better than MLH in this case. Tree height prediction was very good using both MaxLH and P99. Crown base height was also well estimated by MinCLH and P1. Crown volume was again more difficult, producing worse results than any other variable involved in crown bulk density generation. The estimated crown volume performed slightly better than the crown length times the covered ground [(P99 P1)*CH%], achieving a closer fit to a 1:1 relationship. The estimated crown volume was built out of the lidar data considering the furthest laser pulse to the tree location in the four cardinal directions within F 20j and assuming an ellipsoidal shape. Crown bulk density provided the worst result, although was statistically significant. Fig. 5 shows an Fig. 5. Field measured (black circles) versus estimated (gray circles) crown bulk density (kg/m 3 ) for one of the plots. A cross marks the plot center and the dotted circle the 10-m limit of the plot. X and Y coordinates represent distance to plot center. example of the estimation of crown bulk density for each tree in one of the plots. Tree height and crown base height were highly correlated with the field measurements. Foliage biomass, crown volume and crown bulk density estimations from lidar data produced a lower correlation with field measurements, although they were also statistically significant. One interaction produced results similar to 500 interactions, except for crown volume estimation for which one interaction performed better. 4. Discussion Foliage biomass was predicted at the plot level (r 2 = 0.84) quite accurately providing results similar to studies where total aboveground biomass was estimated (Drake et al., 2002a; Lefsky et al., 1999). Additionally, we also estimated foliage biomass at tree level (r 2 = 0.36), which was noisier but based on a large number of samples (n = 183 pine trees). MLH (r 2 = 0.84) performed slightly better than P99 (r 2 = 0.81) at plot level, but the order was reversed at tree level, where MLH (r 2 = 0.26) performed much worse than P99 (r 2 = 0.36). MLH was independent of the clustering at plot level, since all laser pulses within each plot were considered to obtain the mean laser height value. But MLH depended on the clustering at tree level, since the mean value varied according to the number and height of laser pulses assigned to one tree or the other. P99 also depended on this but to a lesser extent, since it is always closer to the maximum laser height. Clustering is difficult, as was shown in the lower accuracy of crown base height compared to tree height. Therefore, it was better to base the estimation of foliage biomass on a variable as independent as possible of clustering, such as the case of P99 at tree level. Tree height estimate at the tree level after applying a K- means clustering was accurate. In contrast, crown base height was better estimated than that obtained by Naesset and Økland (2002), probably due to better tree locations and our method for identifying the space between the upper canopy and the understory. The first live branch is a more relevant measure of crown base height for fire behavior models since it acts as a fuel ladder. However, a minimum crown base height was not consistently well estimated at the plot level, with some significant relationships but low correlation coefficients. The reason for this is that the lidar data did not capture the location of the first living branch. Even using the 10th percentile of crown base height, the predicted equation ( y = 0.60*x ) had a slope less than 1, overestimating crown base height. A change in field measurement protocol is required to identify the first live branch that is big enough for the lidar beam to hit if a 1:1 relationship is to be found. A similar concept for matching field and lidar data must be developed for estimating crown volume at the tree level. Crown volume from lidar data was best estimated when

7 D. Riaño et al. / Remote Sensing of Environment 92 (2004) using the same or similar methodology as used to estimate crown volume from field data. The CH% correction for shading of the lower canopy by taller vegetation is useful to correct understory cover, but it did not improve the crown volume estimation, possibly because this effect is compensated by the larger number of laser pulses returned from the taller canopy. The K-means 500 interactions proved slightly better at the plot level than the one-interaction method for crown volume. The 500-interactions method was able to find the space between upper canopy and understory that was independent of the initial cluster points, since we assumed the gap position was unknown. On the other hand, one interaction worked better at the tree level since we wanted to group data according to the initial point or position of each tree. Another handicap for producing accurate estimates at the tree level was that only one laser pulse was recorded every 1.73 m in the across-track direction while the average crown radius size was 2.11 m. Therefore the point density made it difficult for the cluster analysis to find laser pulses returned from each tree. This difficulty could explain the low correlations for crown volume and also foliage biomass or crown bulk density at tree level. If it were not possible to obtain higher pulse density to improve an estimate, a possible solution would be to combine lidar data with very high resolution aerial photography. A combined cluster analysis could better decide from which tree each laser pulse was returned. Lidar data provided accurate crown bulk density estimation at the plot level. In addition, lidar data could be used to provide a direct crown volume estimation. This approach produced better results than directly fitting any of the lidar variables to crown bulk density. Even more holistic approaches could be developed to improve crown bulk density estimation, such as inverting the GORT model (Ni-Meister et al. 2001). Clustering to identify the gap between the upper canopy and the understory at the plot level was good at a 3-m fixed limit probably because the understory in our plots was always less than 3 m. Nonetheless, clustering would be useful when processing data with large height differences that occupy the volume between the upper canopy and the understory. Clustering also could be used to remove outliers that were not representative of the overall plot height. Analysis at the tree level was not completely successful due to the difficulty in assigning laser pulses to correct the tree since across-track laser density was not high enough. The analysis would become even more complicated if the geographic location of each tree was not used to run the segmentation process. Acknowledgements We wish to acknowledge the Autonomous Region of Madrid (Spain) for financing project number: 07M/0067/ 2001 and to Fulbright Grant and the financial sponsorship of the Spanish Ministry of Education, Culture and Sports that supported David Riaño. Additional funding was obtained from the European Spread project (contract number EVG1-CT ). We also thank all the people in the Department of Geography, University of Alcalá, the Division of Forest Fires of the Autonomous Region of Madrid (especially Álvaro Sánchez), for helping with the fieldwork. References Drake, J. B., Dubayah, R. O., Clark, D. B., Knox, R. G., Blair, J. B., Hofton, M. A., Chazdon, R. L., Weishampel, J. F., & Prince, S. D. (2002a). Estimation of tropical forest structural characteristics using large-footprint lidar. Remote Sensing of Environment, 79, Drake, J. B., Dubayah, R. O., Knox, R. G., Clark, D. B., & Blair, J. B. (2002b). Sensitivity of large-footprint lidar to canopy structure and biomass in a neotropical rainforest. Remote Sensing of Environment, 81(2 3), Dubayah, R. O., & Drake, J. B. (2000). Lidar remote sensing of forestry. Journal of Forestry, 98(6), Hyyppa, J., Hyyppa, H., Inkinen, M., Engdahl, M., Linko, S., & Zhu, Y. H. (2000). Accuracy comparison of various remote sensing data sources in the retrieval of forest stand attributes. Forest Ecology and Management, 128(1 2), Hyyppa, J., Kelle, O., Lehikoinen, M., & Inkinen, M. (2001). A segmentation-based method to retrieve stem volume estimates from 3-D tree height models produced by laser scanners. IEEE Transactions on Geoscience and Remote Sensing, 39(5), Lefsky, M. A., Cohen, W. B., & Spies, T. A. (2001). An evaluation of alternate remote sensing products for forest inventory, monitoring, and mapping of Douglas-fir forests in western Oregon. Canadian Journal of Forest Research, 31(1), Lefsky, M. A., Harding, D., Cohen, W. B., Parker, G., & Shugart, H. H. (1999). Surface lidar remote sensing of basal area and biomass in deciduous forests of eastern Maryland, USA. Remote Sensing of Environment, 67(1), Lopez-Serrano, F. R., Landete-Castillejos, T., Martínez-Millán, J., & Cerro- Barja, A. (2000). LAI estimation of natural pine forest using a nonstandard sampling technique. Agricultural and Forest Meteorology, 101, Means, J. E., Acker, S. A., Fitt, B. J., Renslow, M., Emerson, L., & Hendrix, C. J. (2000). Predicting forest stand characteristics with airborne scanning lidar. Photogrammetric Engineering and Remote Sensing, 66(11), Morsdorf, F., Meier, E., Koetz, B., Itten, K., & Allgöwer, B. (2003). Deriving geometric properties of single tress from high resolution airborne laser scanning data. Proceedings of the 4th International Workshop on Remote Sensing and GIS Applications to Forest Fire Management: Innovative Concepts and Methods in Fire Danger Estimation ( pp ). Ghent, Belgium: EARSEL. Naesset, E. (1997). Estimating timber volume of forest stands using airborne laser scanner data. Remote Sensing of Environment, 61(2), Naesset, E., & Økland, T. (2002). Estimating tree height and tree crown properties using airborne scanning laser in a boreal nature reserve. Remote Sensing of Environment, 79(1), Ni-Meister, W., Jupp, D. L. B., & Dubayah, R. (2001). Modeling lidar waveforms in heterogeneous and discrete canopies. IEEE Transactions on Geoscience and Remote Sensing, 39(9), Riaño, D., Meier, E., Allgöwer, B., Chuvieco, E., & Ustin, S. L. (2003a). Modeling airborne laser scanning data for the spatial generation of critical forest parameters in fire behavior modeling. Remote Sensing of Environment, 86,

8 352 D. Riaño et al. / Remote Sensing of Environment 92 (2004) Riaño, D., Valladares, F., Condés, S., & Chuvieco, E. (2003b). Estimation of effective leaf area index, tree height, and covered ground from airborne laser scanner (Lidar) in two contrasting forests. Agricultural and Forest Meteorology (in press). Sandwell, D. T. (1987). Biharmonic spline interpolation of GEOS-3 and SEASAT altimeter data. Geophysical Research Letters, 2, Scott, J. H. (1999). NEXUS: A system for assessing crown fire hazard. Fire Management Notes, 59(2), von Hansen, W., & Vögtle, T. (1999). Extraktion der Geländeoberfläche aus flugzeuggetragenen Laserscanner-Aufnahmen. Photogrammetrie, Fernerkundung, Geoinformation, 4,