USING LIDAR IN DETERMINING FOREST CANOPY PARAMETERS Alicia M. Rutledge, Graduate Research Assistant Sorin C. Popescu, Assistant Professor Spatial Sciences Laboratory Department of Forest Science Texas A&M University 5 Research Parkway, Suite B25 College Station, Texas 77845 arutledge@tamu.edu s-popescu@tamu.edu ABSTRACT The use of airborne Light Detection and Ranging (LIDAR) to evaluate forest canopy parameters is vital in order to properly address both forest management and ecological concerns. This study was conducted in the Piney Woods region of East Texas around Huntsville, Coldspring and Livingston. The overall goal of this paper is to develop the use of airborne laser methods in evaluating various canopy parameters such as percent canopy cover and Leaf Area Index (LAI). Both these parameters are of interest in determining biomass and carbon models as well as fuel loads. A model is being developed through linear regression for estimating LAI and canopy cover from scanning LIDAR and QuickBird infrared imagery. For accuracy purposes, the parameters of interest are assessed by both airborne and ground-based methods. The primary objective of this study is to determine LIDAR-derived percent canopy cover through individual tree identification and use this parameter in regression models, and then evaluate for accuracy. The methods discussed in this paper shows great potential for improving the speed and accuracy of forest inventory. INTRODUCTION This project attempts to determine the percent canopy cover and leaf area index (LAI) of a region through use of LIDAR. Percent canopy cover, also known as canopy closure, is defined as the percent of a forest area occupied by the vertical projection of tree crowns. LAI is defined as the one sided green leaf area per unit ground area. These parameters are especially important in estimating carbon models and biomass, but are also useful in studying carbon sequestration. Percent canopy cover and LAI are also helpful in examining global climate change, fuel models, and fire risk, as well as aiding in forest inventory and management. The use of airborne LIDAR to evaluate forest canopy parameters is vital in order to properly address both forest management and ecological concerns in a timely and cost-effective manner. Objectives This project attempts to meet the following goals: ) to develop scanning LIDAR methods to estimate LAI and canopy cover over both coniferous and hardwood forests; 2) to develop a frame to fuse scanning LIDAR data with multispectral imagery and improve estimates over large areas from local to regional scales; 3) to generate maps of percent canopy cover and LAI for the study region. Study Area The region to be studied is the Piney Woods near Huntsville, Texas. LIDAR data of the region was collected and in situ data is used for accuracy assessment purposes. This data is from sixty-two plots in the study area, taken from May to July, 24. These sixty-two plots were inventoried and their canopies photographed using a hemispherical lens. Figure details the study area and plot locations.
METHODS Various methods used in this study are detailed in Figure. LIDAR data is used to generate both a canopy height model (CHM) and LIDAR-derived percent canopy cover. The CHM, useful in itself, is also processed using TreeVaW software to determine local (plot-level) percent canopy cover. Airborne LIDAR has been shown to be accurate in estimating biophysical parameters of forest stands (Popescu et al, 24). The normalized difference vegetation index (NDVI) calculated from Quickbird data can be used to estimate LAI (Curren et al, 992; White et al, 997). Statistical data is extracted from all four of these parameters and is used to formulate a regression model based on ground reference data taken from the hemispherical photos. This model is then entered into ENVI image processing software by means of Band Math and used to generate regional maps of percent canopy cover and LAI. Multispectral Data (Quickbird) LIDAR Data In situ data: Hemispherical Photos NDVI Height Bin Approach Canopy Height Model (CHM) TreeVaW Hemiview Analysis Band Math, LIDARderived Canopy Cover Statistical data extraction Binary Image Filter, CC values LAI and Canopy Cover Linear Regression Band Math Final Maps Figure. Methodology used in this study Height Bins The height bin method is used in this project. The term height bin refers to a subdivision of LIDAR height returns. In the case of this project, the LIDAR returns are sectioned as follows: Bin : -.5ft Bin 2:.5-.ft Bin 3:.-.5ft Bin 4:.5-2.ft Bin 5: 2.-5.ft Bin 6: 5.-.ft Bin 7:.-5.ft Bin 8: 5.-2.ft Bin 9: 2.-25.ft Bin : 25.-3.ft Bin : >3.ft Figure 2. Height Bin. Figure 2 illustrates the height bins method while highlighting Height Bin. The lower four height bins correspond to field data (in half-meter increments) while the upper bins are spaced in five-meter increments. The entire LIDAR point cloud is broken down into small segments to be analyzed in this manner. This particular method uses the single lowest height bin to determine canopy cover. The formula used to determine LIDAR-derived canopy cover is: CC lidar = - Height Bin This LIDAR-derived canopy cover is used in conjunction with other parameters to successfully formulate linear models of percent canopy cover and LAI.
Hemispherical Photography Hemispherical photography of the forest canopy from the ground is an accurate method of evaluating percent canopy cover and LAI (Riaño et al, 24). These photographs are analyzed using HemiView, a Delta-T Devices software application, in order to determine local percent canopy cover values and LAI values (Hemiview, 25). An example is shown in Figure 3. Figure 3. Hemi. photo. TreeVaW Processing Local values of percent canopy cover can be determined by using TreeVaW, an IDL executable program that uses a continuously varying filter window to detect tree locations and crown radii. These tree crowns are superimposed on our plot locations and sizes to determine the percentage of crown area overlaying the plot area of 38m 2. An example of local percent canopy cover is shown in Figure 4. Figure 4. TreeVaW-derived percent canopy cover. Plot shown in blue, trees in pink. These local values are then used along with previously discussed parameters to formulate a linear model. A regional estimation of TreeVaW-derived canopy cover is generated by assigning every tree area values of and all non-tree covered areas value of. This binary image is then filtered with a window of size 7.5m x 7.5m approximately the same size as a ground plot. The resulting image is a TreeVaW model of percent canopy cover. Linear Regression Linear regressions are performed using SAS software in order to relate various statistical properties of the LIDAR and MSS data to in situ data. The following parameters are used in the regression equations: x cc_lidar_tvw = mean, LIDAR-derived canopy cover s cc_lidar_tvw = standard deviation, LIDAR-derived canopy cover x ndvi = mean, NDVI s ndvi = standard deviation, NDVI X chm = maximum value, CHM x chm = mean, Canopy Height Model (CHM) s chm = standard deviation, CHM cc tvw = TreeVaW-derived canopy cover CC = percent canopy cover LAI = leaf area index
RESULTS Various regressions are performed for both observed (in situ) percent canopy cover and LAI. These regressions are done once using only LIDAR-based variables, then once more using all available parameters. Several statistical outliers were removed from the dataset during the regression process. The models with the greatest accuracy are as follows: Model : Percent canopy cover using all available variables CC =.987 +.9466 x cc_lidar.449 x ndvi. x chm +.4525 cc tvw, R 2 =.8748 Model 2: LAI using all available variables LAI = -.4374 + 3.83 x cc_lidar + 5.97 s ndvi.5954 x chm +.49882 cc tvw, R 2 =.8475 All variables left in the models are significant at the.5 level. Significant variables included statistics from NDVI, indicating that the vegetation index as well as LIDAR is important in predicting both LAI and percent canopy cover. All variables were checked for variance inflation using SAS, and all variance inflation factors are well under the cutoff value of. LIDAR-derived canopy cover and TreeVaW-derived canopy cover are compared to observed values of canopy cover in Figures 5 and 6. Figure 7 shows observed (in situ) values of percent canopy cover compared to predicted (regression model) values and Figure 8 details a comparison of observed LAI values to predicted LAI values..9.8 y =.988x +.973 R 2 =.884.9.8 LIDAR-derived percent canopy cover, mean.7.6.5.4.3 TreeVaW-derived percent canopy cover values.7.6.5.4.3 y =.499x +.54 R 2 =.66.2.2....2.3.4.5.6.7.8.9 Figure 5. LIDAR-derived canopy cover compared to observed values...2.3.4.5.6.7.8.9 Figure 6. TreeVaW-derived canopy cover compared to observed values. 4.9 3.5.8 3.7 Predicted percent cover values.6.5.4 R 2 =.8748 Predicted LAI values 2.5 2.5 R 2 =.8475.3.2..5..2.3.4.5.6.7.8.9.5.5 2 2.5 3 3.5 4 Observed LAI values Figure 7. Figure 8. Observed LAI values compared to compared to predicted values (Model ). predicted values (Model 2).
The correlation coefficient for the percent canopy cover regression using both LIDAR and NDVI statistics (Model ) is approximately.87 and the correlation coefficient for the LAI regression (Model 2) is approximately.85. Thus Models and 2 will be used to generate regional maps of percent canopy cover and LAI. DISCUSSION AND CONCLUSIONS As one can see from the correlation coefficients and Figures 7 and 8, the predicted regression results for LAI (LAI) are highly correlated with the actual results, and the predicted results for canopy cover (CC) are highly correlated as well. These results strongly suggest that the height bin and TreeVaW methods show potential for being a future standard method of LIDAR analysis in determining percent canopy cover and LAI. These methods could one day be used in analysis packages as it allows one to depict the vertical distribution of a forest canopy. Regional maps of percent canopy cover and LAI have not yet been generated because of difficulty of processing large amount of TreeVaW data. Methods for more streamlined TreeVaW data integration are under review. REFERENCES Curren, P.J., Dungan, J.L., Gholz, H.L. (992). Seasonal LAI in Slash Pine Estimated with Landsat TM. Remote Sens. Environ, 39:3-3. HemiView Online Documentation. (25). Dynamax Inc., Houston, Texas. Popescu, S.C., and Wynne, R. H. (24). Seeing the trees in the Forest: Using Lidar and Multispectral Data Fusion with Local Filtering and Variable Window Size for Estimating Tree Height. Photogrammteric Engineering & Remote Sensing, 7(5):589-64. Riaño, D., Valladares, F., Condés, S., Chuvieco, E. (24). Estimation of leaf area index and covered ground from airborne laser scanner (Lidar) in two contrasting forests. Agricultural and Forest Meteorology, 24:269-275. White, J.D., Running, S.W., Nemani, R., Keane, R.E., Ryan, K.C. (997). Measurement and remote sensing of LAI in Rocky Mountain montane ecosystems. Can. J. Remote Sensing, 27:74-727.