Skip to content
Training Samples
Vector Buffer of 10 meters from all two-tracked roadsBuffer to Raster10-meter buffer from all trees
Final Project
Introduction
This project presents a comprehensive Unmanned Aerial Systems (UAS) mapping mission conducted to demonstrate the full workflow of geospatial data acquisition, processing, and analysis. The objective of this assignment was to apply concepts learned throughout the course to complete a start-to-finish mapping project that integrated aerial data into a Geographic Information System (GIS) environment for meaningful interpretation.
The mission was conducted over an assigned area at Purdue Wildlife Area (PWA), where aerial imagery was collected using a UAS platform under specified flight parameters, including consistent altitude and high image overlap to ensure data quality. Emphasis was placed on proper mission planning, metadata collection, and adherence to standardized coordinate systems to produce accurate and reliable spatial data.
Following data collection, imagery was processed into multiple geospatial products, including Orthomosaics, Digital Surface Models (DSMs), and classified land-cover maps. Additional analysis was performed to extract features such as road networks, demonstrating the practical applications of UAS data in spatial analysis. This project highlights the importance of transforming raw aerial imagery into structured geospatial information that supports decision-making and real-world applications.
Study Area
Figure 1. Mission Boundary
My partner, Diego Hernandez, and I were assigned boundary area 5 at Purdue Wildlife Area, as shown in Figure 1. The mission area was primarily of forested sections, agricultural land (bare earth), grass-covered terrain, and a small portion of the Purdue University Pond.
Upon arriving at the site, we conducted a thorough assessment of the terrain, access, and potential hazards. The area was accessible via a single gravel road and bordered by a nearby family-owned farm. Before flight operations, we communicated with the landowners and obtained permission to operate within the vicinity, ensuring safe and responsible flight practices.
Several potential hazards were identified during the assessment, including tall trees, overhead power lines, adjacent county roads, and wildlife. After identifying these risks, appropriate mitigation strategies and contingency plans were established to ensure safe flight operations and minimize potential impacts to both people and the surrounding environment. Below is an aerial view of our flight area.
Figure 2. Group 5 Mission Area
METADATA
General
Location: Purdue University Wildlife Area - Section 5
Date: 04/17/2026
Vehicle: DJI Matrice 300
Sensors: H20T Zenmuse Sensor
Battery: 2 LiPo Batteries
Flight Information
Flight Number: 2
Takeoff Time: 1200
Landing Time: 1240
Altitude (m): 121 meters (400 ft)
Sensor Angle: Nadir (-90 degrees)
Overlap: 80%
Sidelap: 80%
Total images captured: 524 images
Ground Control
Systems Used: None
Coordinate System: NAD83 (2011)/ UTM Zone 16N
Weather
Cloud Cover: Clear Skies 10 SM of visibility
Wind Direction: Variable
Wind Speed: Variable 6 knots
Temperature: 22 C
Crew
PIC: Diego Hernandez
VO: Diego Hernandez and Isabella Avedician
SO: Isabella Avedician
Flight Mission Debrief
On April 16, 2026, Diego and I went to our assigned mission area to operate the M300 in support of our final project. Weather conditions were partially cloudy with significant winds; the METAR indicated gusts up to 27 knots, approaching the M300’s operational limit of 30 knots. We initiated the flight to evaluate aircraft performance under these conditions; however, after approximately five minutes, we determined that continued operations would pose unnecessary risk. The mission was therefore terminated as a No-Go to ensure the safety of personnel, equipment, wildlife, and surrounding property.
On April 17, 2026, we returned to the site to reattempt the mission under more favorable conditions. The weather was calm and well within operational limits, with the METAR reporting: “KLAF 171554Z VRB06KT 10SM CLR 22/17 A2997 RMK A02 SLP147 T0220167.” Diego served as Pilot-in-Command and Visual Observer, while I performed duties as Sensor Operator and Visual Observer. The flight lasted approximately 30 minutes and resulted in the collection of 524 images. The mission required a brief interruption to land and replace the battery before completion.
Processing
After completing the flight mission, we transitioned to data collection and processing. Because the imagery was stored on Diego’s SD card, the data was first transferred to a temporary folder on the local network so it could be accessed and copied to my workstation. Due to a battery change during the mission, the dataset was split into two separate image collections. To streamline processing, I organized and merged both datasets into a single project folder.
Once the data was consolidated, it was ready for processing. Using Drone2Map, I imported the imagery and created a dedicated project folder within my local temp drive. I configured the processing settings to generate key 2D products, including a True Orthomosaic, Digital Surface Model (DSM), and Digital Terrain Model (DTM). After confirming all parameters, processing was initiated at 09:18 on April 22, 2026. The data finished processing on April 22, 2026, around 1600. After the data had been processed and the 2D products were created, I imported them into ArcGIS Pro to begin creating a series of maps.
Building Maps and Deliverables
Locator Map
A locator map is important for providing spatial context, allowing for a quick understanding of where the main map is located relative to a larger, more familiar region. For this specific map, I imported the Orthomosaic created in Drone2Map based on an ESRI Basemap to show a highly accurate aerial image and included a zoomed-out map to show exactly where the mission area is located relative to the state of Indiana. I’ve also included the flight path to show where exactly the M300 flew.
Mission Orthomosaic
Next, I took the 2D Orthomosaic and made a layout to show the raw imagery. I added two insets with extent indicators to show the detail of this image.
Shaded DSM
Shaded DSMs are created to make elevation data easier to interpret visually and analytically. DSMs show the elevation at the top of all surfaces, including trees, cars, and buildings. For this dataset, I used the Hillshade function in the Spatial Analyst Toolbox. I used the 2D DSM created earlier as my input. I put the Hillshade on top of the DSM and changed the transparency to 33% to show a more detailed image of the elevation in the area.
Side-by-Side Comparison
For easier interpretation and viewing, I have created a side-by-side comparison map of the Orthomosaic and the shaded DSM for the viewers.
Digitized Roads
I began by creating a new feature class labeled “Roads” and defined it as a line feature class. Prior to creation, I ensured the spatial reference matched that of the project data. I then opened the attribute table and added a new field titled “Road Type.”
Next, within the geodatabase, I created a domain named “RoadTypeDomain” and defined the classification categories: gravel, trail, dirt, paved, and two-track. With the schema established, I proceeded to digitize each road within the mission area.
After completing the digitization, I returned to the attribute table to assign the appropriate road type to each feature. Finally, I developed a layout displaying the classified road network overlaid on the orthomosaic.
Vector-Based Land Cover
For this map, I developed a vector-based land cover classification of the mission area. To ensure spatial accuracy and logical consistency, I implemented topology rules so that features connect and interact in a way that reflects real-world conditions.
I began by creating a new geodatabase and, within it, established five polygon feature classes: water, grass, farm area, gravel, and trees. Using the editing tools, I digitized each land cover type across the dataset. Throughout this process, I enabled snapping and utilized the autocomplete, tracing, and merge tools to produce clean, continuous features.
After completing the digitization, I applied topology rules to validate the dataset. Figure 3 illustrates the specific rules used to ensure data accuracy and integrity. The final land cover map is shown below.
Figure 3. Topology Rules
Raster-Based Classification of Land Covers
Unclassified
After several iterations addressing challenges associated with the small pixel size and large dataset, I identified parameters that produced optimal results. I began by resampling the orthomosaic to a pixel size of 20 centimeters (0.2 meters) to improve processing efficiency.
For the unclassified, raster-based land cover classification, I utilized the Image Classification Wizard tool. During the segmentation process, I applied the following parameters: Spectral Detail = 12, Spatial Detail = 15, Minimum Segmentation Size = 50 pixels, and Number of Classes = 4.
The classification process was initiated at approximately 1100 and completed at around 1300.
Classified
Similar to the unclassified approach, I created a new map and resampled the original orthomosaic to a 20-centimeter (0.2-meter) pixel size. I then used the Image Classification Wizard tool, this time selecting a supervised classification method and applying the following segmentation parameters: Spectral Detail = 12, Spatial Detail = 15, and Minimum Segmentation Size = 50 pixels.
For classification, I defined four classes—water, grass, bare ground, and trees—by creating corresponding training feature classes. For each class, I generated approximately 45 to 55 training samples (see Figure 4).
The classification process required multiple attempts to complete successfully. The final run began at approximately 0830 and finished around 1000.
Figure 4. Training Samples
DSM 6 meters
The objective of this map was to identify and display all features exceeding 6 meters in height. To accomplish this, I first used the Raster Calculator to subtract the Digital Terrain Model (DTM) from the Digital Surface Model (DSM), resulting in a normalized surface representing object heights above ground level.
Next, I applied the Reclassify tool and manually defined two class breaks. I set the minimum threshold at 6 meters and retained the existing maximum value, effectively isolating all features taller than 6 meters. This process produced the final output shown below.
Data Analysis
The objective of this map was to generate a 10-meter buffer around all two-track roads. To accomplish this, I first used the Select By Attributes tool to isolate the two-track roads and created a separate layer from the selection.
Next, I applied the Buffer Analysis tool, using the two-track road layer as the input and specifying a buffer distance of 10 meters. I enabled the dissolve option to merge all output features into a single polygon, preventing overlap between buffered segments.
Finally, I used the resulting buffer layer to create the map shown below.
To support subsequent map algebra operations, I converted the buffer layer into a raster format. Using the Feature to Raster conversion tool, I selected the buffer layer as the input and designated the OBJECTID field for rasterization.
Within the Environments settings, I specified the cell size to match the orthomosaic resolution (0.015209) and enabled the Snap Raster option to ensure proper pixel alignment with existing datasets. After processing, the resulting raster layer was used to create the layout shown below.
Finally, I generated a 10-meter proximity analysis of trees located near two-track roads using map algebra. Utilizing the Raster Calculator tool, I applied the following expression:
(“SupervisedClassificationLayer” == 1) & (“BufferLayer” > 0)
This operation isolated pixels that met both conditions: those classified as trees in the supervised classification layer and those falling within the 10-meter buffer of two-track roads. The resulting output highlights only the areas where trees exceed 6 meters in height and are located within the specified proximity to the road network.
Summary
This project demonstrates a complete Unmanned Aerial Systems (UAS) mapping workflow, integrating data collection, processing, and geospatial analysis to produce meaningful GIS outputs. Conducted at Purdue Wildlife Area (PWA), the mission involved acquiring high-resolution aerial imagery using a DJI Matrice 300 under carefully planned flight parameters to ensure data accuracy and consistency. Despite an initial No-Go decision due to high winds, the mission was successfully completed the following day under favorable conditions, resulting in 524 images.
Post-flight processing was completed using Drone2Map, where imagery was transformed into key geospatial products, including an orthomosaic, Digital Surface Model (DSM), and Digital Terrain Model (DTM). These datasets were then imported into ArcGIS Pro for further analysis and map production. A series of deliverables were created, including a locator map, mission orthomosaic, shaded DSM, and side-by-side comparison for enhanced visualization.
Advanced spatial analysis techniques were applied to extract and classify features within the study area. This included digitizing and classifying road networks, developing a vector-based land cover map with topology validation, and performing both unsupervised and supervised raster-based classifications. Additional analysis identified features exceeding 6 meters in height and examined spatial relationships between trees and nearby two-track roads through buffering and raster-based map algebra.
Overall, this project highlights the effectiveness of UAS technology in capturing high-quality spatial data and demonstrates the importance of structured workflows in transforming raw imagery into actionable geospatial information. The results showcase practical applications of GIS in environmental analysis, land cover classification, and infrastructure assessment.
Want to print your doc?
This is not the way.
This is not the way.
Try clicking the ··· in the right corner or using a keyboard shortcut (
CtrlP
) instead.