Collecting and Processing Drone-based Remotely Sensed Data for Use in Forest Recovery Monitoring

Our research aims to facilitate the use of drone based LiDAR in forest recovery assessment through the development of a user-friendly protocol. Recent developments that make this work possible include advancement in LiDAR sensor technology and integration with deep loaning models. To begin, check the RPAS equipment and update the firmware if necessary.

Set up the RTK base station in a clear area that is away from obstacles and the tree line. Wait till the base station is completely powered up before starting the drone flight. Create a flight file for each mission and upload them to the remote controller.

Now, conduct the flight missions to collect LiDAR and multispectral data over the area of interest. Set up a GNSS base station over the RPAS's base station location and run the base station to obtain accurate coordinates for precise point positioning correction. Use a card reader to transfer the collected LiDAR and multispectral data from the sensors to a workstation.

Use photogrammetry software to pre-process the multispectral data and perform geometric and radiometric corrections. Generate a multispectral orthomosaic. Use appropriate software to reconstruct the raw LiDAR data into a point cloud file.

Use a geospatial tool to align the LiDAR point cloud with the multispectral orthomosaic. Run a Python script to merge the aligned LiDAR point cloud with the multispectral orthomosaic to generate a multispectral point cloud. Download CloudCompare, the three dimensional point cloud processing software, and install it.

Then, download the TreeAIBox plugin installer version 1 from GitHub, run the installer, and follow the onscreen prompts. Open CloudCompare from the desktop icon, or by selecting Start, followed by All Programs, and CloudCompare. Load the point cloud file by using Open one or several files, and click Apply.

If the point coordinates are large, accept the prompt to apply a global shift or scale. Select Input, which reads metadata from the file, and click Yes so the point cloud appears in the canvas. Open the Python plugin toolbar.

Expand the Script Register dropdown and click TreeAIBox to open the plugin graphical user interface. Ensure the Use GPU checkbox is selected if a Compute Unified Device Architecture supported graphics processing unit is available. From the top panel, select TreeFiltering and choose ALS if tree stems are not visible in the RPAS data.

Now, clear the tile size checkbox. From the predefined models dropdown, select treefiltering_als_esegformer. If using this model for the first time, click Download and confirm the popup showing the local path.

Select the point cloud in the canvas so it is highlighted with a bounding box. In the TreeFilter panel, click Apply. Confirm a new scale or field named TreeFilter is created with value 2 for tree points in red and value 1 for other points in blue before proceeding.

From the TreeAIBox top panel, select TreeisoNet. Enable Reclamation, ALS stem implicit, and Treeloc. From the dropdown, choose the required pre-trained model.

Ensure the point cloud in the canvas is selected, then click Apply. After processing, confirm that a new item named Treetops appears under the original point cloud in the DB tree window. Select this item and increase the point size, for example to 16, for improved visibility, and verify that treetop positions appear as white dots in the canvas.

For segmenting tree crowns, reselect the tree point cloud item. From the TreeAIBox top panel, select TreeisoNet and enable TreeOff. Download the required pre-trained model and then click Apply to run the model.

Next, confirm that a new scale or field named TreeOff is created. Verify that each tree is assigned a unique identifier with points from the same tree sharing the same identifier. Optionally, reset the treetop point size to Default to reduce visual clutter.

To improve visual contrast and randomize tree colors by identifier, clone the original point cloud to preserve the data by selecting Edit and Clone. Then, go to Edit, followed by Scalar Fields, and Convert to random RGB. Enter a large value to ensure discrete colors and click OK.View the point cloud displaying trees in random colors.

Finally, from the TreeAIBox top panel, select TreeisoNet and click Export stats to export segmentation results. Then, click Open output path to view the exported file in the Results folder. Confirm the output is a comma separated values file containing tree identifiers, coordinates, tree height, and crown area.

The TreeAIBox plugin successfully filtered tree points and delineated individual trees across all three sites and a total of 2, 755 individual trees were detected. The model identified 1, 706 trees on Site 1, 882 trees on Site 2, and 167 trees on Site 3. The detection rate for reference trees varied among the sites with 100%detected for site 2, 95%for Site 1, and 21%for Site 3.

The model achieved a 100%detection rate for all reference trees taller than one meter. Detection performance decreased for shorter trees with only 45%of trees between 0.5 to 1 meter detected and none of the trees below 0.5 meter identified. This protocol aims to address the research gap of individual tree detection and segmentation for young trees in complex forested environments.

This protocol offers a practical, effective, user-friendly, and versatile method for extracting individual tree metrics from LiDAR data. Our protocol will advance forest recovery and monitoring and is particularly useful for assessing rec climb well size, reducing time and cost of doing of plot surveys.