Extended Abstract
Introduction
Accurate and up-to-date data on the structural characteristics of forests, particularly at the individual tree level, are critical for sustainable forest management. Traditional field methods for forest inventory, such as 100% sampling or random sampling, are time-consuming and labor-intensive. Remote sensing, on the other hand, provides a more efficient way to estimate the biophysical and biochemical parameters of plants on a large scale. Satellite imagery has been widely used for this purpose, but it has certain limitations, such as cloud coverage and high costs associated with high-resolution imagery. UAVs (Unmanned Aerial Vehicles), or drones, offer an effective alternative by capturing high-resolution spatial data without the restrictions of cloud interference. They can be equipped with various sensors, such as LiDAR, multispectral, hyperspectral, and infrared cameras, to collect detailed plant data. UAV-based photogrammetry, which processes aerial images to create maps and 3D models, has become a popular method for forest monitoring. Among the commonly used photogrammetry software, Agisoft Metashape and PIX4D are widely adopted for forest science. However, there is limited research comparing the efficiency of these two software programs in estimating critical structural features of forests, such as tree height, canopy area, and density. The primary aim of this study is to compare the accuracy and efficiency of Agisoft Metashape and PIX4D in estimating forest structural characteristics, specifically tree density, canopy area, and tree height, in the Zagros forests of western Iran. These forests are under pressure from over-exploitation, and effective management requires accurate, up-to-date, and cost-effective monitoring methods in these regions.
Materials and Methods
The study was conducted in the protected forests of Qalaje, located between the Kermanshah and Ilam provinces in Iran. The study area covers 15 hectares, with an elevation of 1,700 meters above sea level, and consists of cold, semi-arid climate condition. The dominant tree species in the study area are Quercus brantii, Pronus microcarpa and Crataegus pontica. A UAV (Phantom 4-RTK) equipped with a real-time kinematic (RTK) system was used to capture high-resolution aerial images over the 15-hectare area. The study area was divided into three groups of sample plots based on canopy coverage density: low (less than 25%), medium (25-50%), and high (more than 50%) canopy coverage. Ten replicates were randomly selected in each category for image processing and field measurement. The flight plan was designed using GS RTK software with a flight altitude of 100 meters. The longitudinal and transversal overlaps were 75%. The images were processed in both Agisoft and PIX4D to create dense point clouds. In Agisoft, a Structure from Motion (SfM) algorithm was used to create dense point clouds, followed by 3D model generation, orthophoto mosaic creation, and extraction of Digital Surface Models (DSM) and Digital Terrain Models (DTM). Similarly, PIX4D followed a comparable process with slight variations in parameter settings. After point cloud generation, the canopy area and tree height were estimated using ArcGIS, where the tree crown area was calculated from the DSM and DTM layers. Field measurements were taken for each plot, including tree density, crown diameter, and tree height. These measurements were used to validate the UAV-based estimates from the two software programs. Paired T-tests were used to compare the estimated tree density, canopy area, and tree height from the UAV images with the field measurements. Linear regression models were also developed to assess the correlation between the UAV-based estimates and field data, with the coefficient of determination (R²) calculated. Additionally, the Root Mean Square Error (RMSE) was used to quantify the estimation error.
Results and discussion
Regarding the comparison of the performance of the two software programs for generating point clouds, the results showed that the point cloud density produced by Agisoft (1708 points per square meter) was higher than that produced by PIX4D (1498.5 points per square meter). Additionally, the image processing time in Agisoft (242 minutes) was less than that in PIX4D (273 minutes). The results indicated that only in sample plots with low canopy cover density, the estimated number of trees in Agisoft (3.3) and PIX4D (3) did not show a significant difference from the measured number (3.9). In sample plots with medium and high canopy cover density, the error in the estimated number of trees by Agisoft (with values of 55.15% and 70.29%, respectively) was lower than the estimated error by PIX4D (with values of 61.13% and 78.10%, respectively). Regarding canopy cover estimation, the results showed that in Agisoft and PIX4D, the highest RMSE% error in canopy cover estimation was related to sample plots with low canopy cover density, with values of 30.77% and 22.90%, respectively. The results also indicated that with an increase in canopy cover density in the sample plots, the percentage of error in canopy cover estimation decreased relatively similarly in both software programs. Specifically, the canopy cover estimation error for Agisoft at medium and high canopy densities was 27.63% and 8.63%, respectively, while for PIX4D, it was 12.67% and 11.15%, respectively. Regarding tree height estimation, the results showed that the estimated height error at low canopy density was the lowest in both Agisoft and PIX4D, with values of 14.77% and 5.84%, respectively, compared to other canopy density classes. Furthermore, the results indicated that as the canopy density in the sample plots increased from low to medium, the estimated RMSE% error in tree height for the outputs of both Agisoft and PIX4D increased (with values of 23.12% and 12.59%, respectively). However, as the sample plot density increased from medium to high, this error decreased for both Agisoft and PIX4D (with values of 21.34% and 11.69%, respectively).
Conclusion
In this study, the performance of Agisoft and PIX4D software was compared in estimating the structural features of Zagros forests. Considering the processing time and quality of dense point clouds, it was concluded that Agisoft's performance is superior to that of PIX4D. However, the performance of both software programs for estimating the number of trees, height, and canopy cover is entirely dependent on the level of canopy density.