Document Type : Research Paper
1 Expert of general bureau of natural resource and watershed management organization, Chaharmahal and Bakhtiari, Shahrekord
2 Zahra Azizi* Assistant professor, Department of remote sensing and GIS, Science and Research Branch, Islamic Azad University
3 Associate professor, Shahid Beheshti University
LiDAR (Light Detection and Ranging) employs pulse models which penetrates vegetation cover easilyand provides the possibility of retrieving data related to Digital Terrain Model (DTM).Pulses sent by the Lidar sensorhitdifferent geographical features on the surfaceground and scatter inall directions. Distance to the object is determined by recording the time between transmitted and backscattered pulses and by using the speed of light to calculate the distance traveled by the small portion of pulses backscattered. Most LiDAR receivers at least record the first and last backscattered pulses. The first backscattered pulses are used to produce Digital Surface Models (DSMs) and the last ones are used to produce DTMs. Despite the fact that these data can provide a valuable source for DTM generation, the volume of vegetation (vegetation density) in forest areas reducesthe accuracyof DTMs. Onthe other hand, ground surveying of forest areas is rather expensive and time consuming, especially in largerforests. Aerial images are also used as a source for DTM generation, but this approach requires a 60–80% overlap between images which along with canopy height reduce the potential of this method for DTM generation. Also, low spatial resolution of satellite images collected from forest areas increases errors in DTM generation to a large degree. The present study investigates the accuracy and precision of DTMsproduced from LiDAR data in a forest area. Furthermore, the effect of different methods of filtering and DTM interpolation was explored. Different methods of DTM generation were also closely analyzed and evaluated.
Materials & Methods
The case study area is located in Doroodforests, a part of Zagros forests, in the southeastern regions of Lorestan province in Iran (48°51’19’’E to 48°54’31’’E and 33°19’21’’N to 33°21’15’’N). Minimum and maximum altitude above sea level were 1143 and 2413m, respectively. The study area covers 100 hectares of mountains with an average slope of 38%. Approximately 50% of the area is covered by forests in which Brant’s oak (Quercusbrantii Lindley) is the most frequent species. LiDAR data were collected by the National Cartographic Center of Iran (NCC) in 2012 using a Laser scanner system (Litermapper 5600) fixed on an aircraft flying at an average altitude of 1000m. LiDAR data consisted of the first and last returns (backscattered pulses), distance and their intensity value. Collected data had an irregular structure and included an average of more than four points per square meter. A DTM was produced using a two-step filtering. First, a morphological filter removed most of non-ground points, and then a slope-based filter detected remaining points. Inforest areas with rough-surface, DTM was producedthrough processing ofLiDAR data with statistical methods likekriging and inverse distance weighting (IDW). These methods apply third and fourth power to detect and remove non-ground points. To assess the accuracy of DTMs produced by different approached, 5 percent of the LiDAR point cloudswererandomly separated as the test data. Amongst these data sets, 62 points with a suitable dispersion were selected and measured using a GPS-RTK. An error matrix, along with accuracy indices (including correlation and Root Mean Square Error (RMSE)) were calculated based on these data.
Results & Discussion
Results indicated that 44-degree slope is the best threshold for isolation of non-ground points and inverse distance weighting (IDW) is the best third power interpolation method with the highest correlation (0.9986) and the lowest RMSE (0.204 meter). Amongst the filtering methods, slope-based filter used for separation of ground and non-ground points had the best performance. Since this filter combines two parameters of slope and radius, it can remove cloud points related to the vegetation cover and results in high efficiency for steep forest areas. Slope-based filter is suitable for processing of near-surface vegetation, whilst statistical filter is well-suited for vegetation cover consisting of tall trees.
The present study proposed and investigated different scenarios for the production offorest areas’ DTM using LiDAR data and two interpolation methods. These algorithms were practicallyassessed using LiDAR data collected from Dorood forest areas. The best scenario was slope-based filter with inverse distance weighting (IDW) interpolation. Based on accurate assessment, this approach can produce reliable DTM in forest areas.