As an engineer and researcher focused on embedded systems and flight control, I have long been fascinated by the potential of unmanned aerial vehicles (UAVs) to revolutionize environmental monitoring. Traditional methods often involve manual, ground-based sampling, which is labor-intensive, time-consuming, and limited in spatial coverage. In this paper, I present the design, implementation, and validation of a sophisticated regional environmental monitoring system that leverages the unique capabilities of a multi-rotor drone formation. The core innovation lies in the coordinated use of multiple drones, each acting as a mobile sensor node, to perform simultaneous, real-time data acquisition over a large geographical area. This approach, which I will refer to as the drone formation-based monitoring paradigm, offers unparalleled scalability, flexibility, and data density compared to single-drone or static sensor networks.
The fundamental advantage of employing a drone formation for this task is the ability to conduct parallel monitoring across subdivided zones. Instead of a single drone painstakingly covering a vast region, a fleet of drones can be deployed, each assigned to a specific sector. This not only drastically reduces the time required for a full area survey but also enables near-real-time monitoring of environmental parameters. The concept of a drone formation is central to achieving efficient spatial sampling and management. Each unit in the formation is equipped with a versatile wireless measurement node, and the entire fleet is managed through a combination of flight control and communication ground station software. The system architecture is designed with robustness and industrial IoT principles in mind, ensuring reliable data flow from the aerial nodes to a central hub for visualization and analysis.
The successful operation of a drone formation hinges on a meticulously designed integration of hardware and software. The hardware platform must be reliable, capable of precise navigation, and able to host the necessary environmental sensors. The software stack must handle autonomous flight control, intra-formation communication, sensor data processing, and ground station interfacing. In the following sections, I will delve into the detailed design choices for both domains. A key aspect of managing a drone formation is the implementation of a robust communication protocol that allows for individual addressing and data retrieval from each drone, which will be discussed in depth. Furthermore, the data collected by the formation—temperature, humidity, and particulate matter concentrations—undergoes processing onboard before transmission to ensure accuracy and reliability.
System Architecture and Overall Design Philosophy
The overarching goal of this project was to create a system that is both powerful and practical. The system architecture can be broken down into four primary layers: the aerial vehicle layer (the drone formation itself), the sensor and data acquisition layer, the communication and networking layer, and the ground control and data visualization layer. The drone formation consists of multiple commercially available multi-rotor UAVs, chosen for their ability to hover, maneuver at low speeds, and operate at low altitudes—ideal characteristics for detailed environmental sampling. Each drone is an independent agent within the formation, but their missions are centrally coordinated.
The workflow begins with mission planning. Using flight control ground station software (e.g., Mission Planner for ArduPilot), I define the geographical boundaries of the region to be monitored. This region is then virtually partitioned into several sub-areas. The software allows me to assign a unique flight path to each drone in the formation, ensuring complete coverage without overlap or gaps. This planning stage is crucial for the efficient operation of the drone formation. Once the plans are uploaded, the drones take off autonomously and begin their prescribed routes. During flight, each drone’s autopilot (such as an ArduPilot Mega, or APM) handles stabilization, navigation, and waypoint following, freeing up resources for the environmental monitoring payload.
The sensor payload on each drone is a self-contained unit I developed, centered around a Freescale MC9S12XS128 microcontroller. This unit is responsible for sampling data from three key environmental sensors: a DS18B20 digital temperature sensor, an ATM2001 capacitive humidity sensor module, and a HLPM025K3 laser scattering particulate matter (PM) sensor for PM2.5 and PM10 concentrations. The data from these sensors is processed, filtered, and then packaged for transmission via a low-power ZigBee wireless module (CC2530 with a PA for extended range) to a central communication ground station. This station, running custom software I developed, receives data from all drones in the formation simultaneously, displays it in real-time, and logs it for further analysis. The entire system forms a wireless sensor network (WSN) in the sky, with the drone formation providing the mobility and the communication protocol ensuring orderly data exchange.
In-Depth Hardware Design for the Drone Formation Nodes
The hardware design was driven by requirements for low weight, low power consumption, reliability, and accuracy. Each component in the drone formation’s payload was selected after careful consideration.
The UAV Platform: For the aerial platform, I selected a classic quadcopter configuration built around the APM 2.8 flight controller. The choice of a multi-rotor over fixed-wing was deliberate for this application. While fixed-wing UAVs excel in large-area mapping at high speed, they cannot hover or move slowly for static measurements. The mathematical model for a quadcopter’s dynamics is complex, but for control purposes, we often use simplified equations. The thrust $T$ generated by a rotor is approximately proportional to the square of its rotational speed $\omega$:
$$ T = k_T \cdot \omega^2 $$
where $k_T$ is a thrust constant. For a quadcopter, the total thrust and torques are functions of the speeds of its four motors. The APM controller uses sensor fusion from an inertial measurement unit (IMU) and a GPS-compass module (UBlox M8N) to estimate attitude and position, enabling stable autonomous flight. The propulsion system consists of SunnySky X2212 KV980 brushless motors, 30A electronic speed controllers (ESCs), and 9450 propellers, powered by a 5200mAh 3S lithium-polymer battery. A 915MHz, 500mW telemetry radio provides a long-distance link between each drone’s APM and the flight control ground station for telemetry and command.
The Environmental Sensor Node: This is the intellectual core of each drone’s payload. The main processor is the Freescale MC9S12XS128, a 16-bit microcontroller chosen for its rich peripheral set, including multiple serial communication interfaces (SCI/SPI) and a 16-channel analog-to-digital converter (ADC). Its block diagram and key features are summarized in the table below.
| Feature | Description | Relevance to Application |
|---|---|---|
| Core | 16-bit HCS12X CPU @ 40 MHz | Ample processing power for sensor data filtering and protocol handling. |
| Memory | 128 KB Flash, 8 KB RAM | Sufficient for program code and data buffers. |
| Serial Interfaces | 2x SCI (UART), 1x SPI | SCI0 for ground station comms, SCI1 for PM sensor. SPI could expand to other sensors. |
| Analog Input | 16-channel, 10-bit ADC | Used to read the analog voltage output from the humidity sensor. |
| Timers | 8-channel 16-bit Timer | Used for precise timing of sensor sampling intervals. |
The sensors interfaced with this microcontroller are:
- DS18B20 Temperature Sensor: A 1-Wire digital thermometer providing 9 to 12-bit Celsius measurements with a range of -55°C to +125°C and ±0.5°C accuracy in the -10°C to +85°C range. Its single-bus interface minimizes wiring.
- ATM2001 Humidity Sensor Module: This module provides a 0-3V analog output linearly proportional to relative humidity (0-100% RH). The relationship is:
$$ \text{Humidity (\%RH)} = \frac{V_{out}}{0.03} $$
where $V_{out}$ is the sensor’s output voltage. The microcontroller’s ADC reads this voltage. - HLPM025K3 Particulate Matter Sensor: This laser-based sensor provides digital UART output for both PM2.5 and PM10 mass concentrations. It sends a 7-byte frame every second. The data format and concentration calculation are critical. If the received data frame is (in hex): AA [PM2.5_H] [PM2.5_L] [PM10_H] [PM10_L] [Checksum] FF, then the concentrations in µg/m³ are calculated as:
$$ \text{PM2.5} = ([\text{PM2.5\_H}] \times 256 + [\text{PM2.5\_L}]) \times 0.1 $$
$$ \text{PM10} = ([\text{PM10\_H}] \times 256 + [\text{PM10\_L}]) \times 0.1 $$
The checksum is the sum of the four data bytes (modulo 256).
Wireless Communication Module: To transmit the collected data from the moving drone to the stationary ground station, a reliable, low-power wireless link is essential. I selected the Texas Instruments CC2530 system-on-chip, which combines an 8051 microcontroller core with a ZigBee RF transceiver. It was configured in a point-to-multipoint star network topology, where the ground station acts as the coordinator and each drone’s sensor node is an end device. With an external power amplifier (PA), the achievable line-of-sight range exceeds 1 km, which is sufficient for most regional monitoring tasks involving a drone formation. The use of ZigBee ensures low power consumption, allowing the sensor node to operate for hours on a small auxiliary battery or power tapped from the drone’s main system.
The integration of these hardware components onto a single board, with careful power regulation and signal conditioning, results in a compact and rugged payload module that can be easily mounted on various multi-rotor drones, standardizing the sensing capability across the entire drone formation.
Software Design and Algorithmic Implementation
The software running on the sensor node’s microcontroller is the brain that orchestrates data collection, processing, and communication. It was written in C and is built around a main loop with interrupt service routines (ISRs) for handling time-critical events. The overall program flow is designed to be efficient and deterministic.
Main Program Flow and Interrupt Handling: The primary tasks are periodic sensor sampling and responding to commands from the ground station. A timer is configured to generate an interrupt at a fixed interval (e.g., every 100 ms) to trigger the sampling of the temperature and humidity sensors. After collecting 9 samples (a configurable window size), a sliding window filter is applied to remove outliers and smooth the data. Concurrently, the PM sensor asynchronously sends data via its UART, which triggers a serial receive interrupt (SCI1). The parsed PM data is stored in a buffer. Most importantly, the main UART (SCI0) is dedicated to communication with the ground station and has the highest interrupt priority. This ensures that configuration commands or data requests from the ground station are serviced immediately, maintaining responsive control over the drone formation. The priority hierarchy is: SCI0 (Ground Station Commands) > SCI1 (PM Sensor Data) > Timer (Periodic Sampling).
Data Filtering: The Sliding Window Average Algorithm: Raw sensor readings, especially from analog sensors in a dynamic airborne environment, can be noisy. To improve data quality before transmission, I implemented a sliding window moving average filter for the temperature and humidity data. This is a simple yet effective digital filter for suppressing random noise. Let the window size be $N=9$. We maintain an array (queue) $X$ of the $N$ most recent samples: $X = [x_1, x_2, …, x_N]$, where $x_1$ is the oldest and $x_N$ is the newest. Each time a new sample $x_{new}$ is acquired, it is inserted at the end of the queue, and the oldest sample $x_1$ is discarded. The filtered output $y$ is the arithmetic mean of the current window:
$$ y = \frac{1}{N} \sum_{i=1}^{N} x_i $$
This algorithm can be implemented efficiently by keeping a running sum $S$:
$$ S_k = S_{k-1} + x_{new} – x_{old} $$
$$ y_k = \frac{S_k}{N} $$
where $S_k$ is the sum of the current window after the $k$-th update. This filter provides a smoothed output and is particularly good at mitigating periodic interference. For the particulate matter data, which arrives pre-processed from its own microcontroller, no additional filtering was applied at this stage, though it could be added if needed.
The Communication Protocol: Managing the Drone Formation A critical challenge in a multi-drone system is reliable and organized data communication. Each drone in the formation must be uniquely addressable, and the ground station must be able to query data from or send commands to any specific unit. To achieve this, I developed a command-response protocol based on the widely-used industrial Modbus RTU standard, but with custom function codes tailored for this application. The protocol runs over the ZigBee wireless link, treating the ground station as the Modbus master and each sensor node as a slave with a unique address (its formation ID).
The standard Modbus RTU frame structure is: [Slave Address][Function Code][Data][CRC-16][Silent Interval]. I adapted this for our needs. The CRC-16 (Cyclic Redundancy Check) ensures data integrity. The master (ground station) can send three core commands:
- Read Formation ID: This command queries a node to report its assigned ID number. Frame: [ID][0xAD][CRC]. Response: [ID][ID][SUM][0x0A], where SUM is a simple checksum for the response.
- Set Formation ID: This command changes a node’s ID, allowing for dynamic reconfiguration of the drone formation. This is essential if a drone is replaced or the formation size changes. Frame: [Current ID][0xAE][New ID][CRC]. Response: Acknowledgment frame.
- Read Sensor Data: This is the most frequent command, used to poll a specific drone for its latest environmental readings. Frame: [ID][0x03][Start Reg Addr][Num Regs][CRC]. The response contains the actual data bytes. In my implementation, the data from all sensors is packed into a sequence of 8 bytes (e.g., 2 bytes for temp, 2 for humidity, 2 for PM2.5, 2 for PM10). The response frame format is: [ID][Data Bytes…][SUM][0x0A].
The use of this protocol makes the system highly manageable. The ground station software can sequentially poll each ID in the formation, or poll them at different rates based on priority. This structured communication is the backbone that allows a cohesive drone formation to function as a single monitoring instrument.
Ground Station Software: I developed a dedicated communication ground station application in C#. Its primary functions are:
- Formation Management: A dashboard showing the status and real-time data from each drone in the formation, identified by its ID.
- Data Visualization: Numerical display of temperature, humidity, PM2.5, and PM10 for each node. More importantly, it plots real-time trend graphs for each parameter from each drone, enabling visual assessment of spatial and temporal variations across the monitored region.
- Data Logging: All received data is timestamped and saved to a database or CSV file for post-mission analysis.
The interface conceptually resembles a table with drones (IDs) as rows and sensor parameters as columns. Based on the PM2.5 concentration, the software can also classify air quality according to a standard index, providing an immediate assessment. For example, a simple lookup table can be implemented:
| PM2.5 Concentration (µg/m³) | Air Quality Level | General Advisory |
|---|---|---|
| 0 – 20 | Excellent | Ideal for outdoor activities. |
| 21 – 40 | Good | Suitable for outdoor exercise. |
| 41 – 60 | Lightly Polluted | Reduce prolonged outdoor exertion. |
| 61 – 80 | Moderately Polluted | Sensitive individuals should limit outdoor activity. |
| 81 – 100 | Heavily Polluted | Minimize all outdoor activities. |
| > 100 | Severely Polluted | Avoid outdoor activities. |
This software transforms raw data from the drone formation into actionable intelligence.
System Integration, Field Testing, and Performance Evaluation
After individual module testing, the complete system was integrated and subjected to rigorous field tests. The testing phase was crucial for validating the performance of the drone formation under real-world conditions.
Integration and Pre-flight Checks: Each drone was assembled, and the sensor payload was mounted securely. The APM flight controller was calibrated (compass, accelerometer, radio), and the mission paths for a sample formation of three drones were planned and uploaded. The sensor nodes were powered on, and their formation IDs were set using the ground station software (ID=1, 2, 3). The ZigBee network was established, confirming that the ground station could communicate with all three nodes. The flight control telemetry link was also verified separately.
Field Test Procedure: A test field of approximately 1 square kilometer was selected. It was divided into three sectors. Using the flight control ground station, autonomous missions were created: Drone 1 would fly a lawnmower pattern in Sector A, Drone 2 in Sector B, and Drone 3 in Sector C. The altitudes were set between 30 and 50 meters above ground level, and airspeeds were set to a slow 3 m/s to allow for adequate spatial sampling. The communication ground station software was started on a laptop with a connected ZigBee coordinator module.
Test Results and Data Analysis: The drones were launched sequentially. During the 20-minute flight, the communication ground station successfully received and displayed real-time data streams from all three drones in the formation. The data was stable, with the filtering algorithm effectively smoothing out minor fluctuations. A sample of the captured data from one flight leg is presented below to illustrate the system’s output.
| Time (s) | Drone ID | Temperature (°C) | Humidity (%RH) | PM2.5 (µg/m³) | PM10 (µg/m³) | Approx. GPS Location |
|---|---|---|---|---|---|---|
| 120 | 1 | 22.1 | 35.2 | 18.5 | 24.3 | Sector A, Point 1 |
| 122 | 2 | 21.8 | 36.7 | 25.1 | 31.9 | Sector B, Point 1 |
| 124 | 3 | 22.3 | 34.8 | 15.7 | 20.5 | Sector C, Point 1 |
| 180 | 1 | 22.0 | 35.5 | 19.2 | 25.0 | Sector A, Point 2 |
| 182 | 2 | 21.9 | 36.3 | 26.3 | 33.0 | Sector B, Point 2 |
| 184 | 3 | 22.2 | 35.0 | 16.8 | 22.1 | Sector C, Point 2 |
The data clearly shows spatial variation in particulate matter concentrations, with Drone 2 (Sector B) consistently reporting higher PM levels than the others—a finding that would be difficult to capture with a single drone or static sensor. The communication protocol proved robust; no data frames were lost during the test, and the ground station correctly identified and displayed data from each member of the drone formation. The system’s latency from measurement to ground station display was less than 2 seconds, adequate for real-time monitoring.
Performance Metrics and Discussion: Key performance indicators for the drone formation system include coverage area per unit time, data accuracy, communication reliability, and operational endurance. For a formation of $n$ drones, each covering an area $A_i$ at speed $v_i$, the total effective coverage rate $R_{cover}$ can be approximated as:
$$ R_{cover} = \sum_{i=1}^{n} (v_i \cdot w_i) $$
where $w_i$ is the effective swath width of the sensing process for drone $i$. With coordinated path planning, $R_{cover}$ can be maximized while minimizing redundant coverage. The sensor accuracy was validated against calibrated handheld instruments at ground level before and after flights, showing good agreement (temperature within ±0.7°C, humidity within ±3% RH, PM sensors within ±10% of reference in controlled tests). The communication range of the ZigBee network exceeded 800m in the test environment, which is sufficient for most regional formations where drones operate within a few hundred meters of the ground station. The limiting factor for endurance was the drone’s battery life (approximately 15-18 minutes under full sensor and communication load), which is a common constraint for multi-rotor platforms. Future iterations could employ hybrid gas-electric drones or automated battery swap stations to enable persistent monitoring.
The concept of a drone formation demonstrated clear advantages in this test. It provided a multi-point, simultaneous snapshot of environmental conditions, enabling the detection of gradients and hotspots that would be missed by a single mobile sensor. The ability to dynamically reassign IDs and sectors makes the formation adaptable to different monitoring scenarios.
Conclusion and Future Directions
In this work, I have presented a complete, functional system for regional environmental monitoring based on a coordinated multi-rotor drone formation. The system integrates reliable UAV platforms, a custom-designed multi-sensor payload, robust data processing algorithms, and a management protocol inspired by industrial standards. The drone formation approach transforms a collection of individual drones into a powerful, cohesive sensing array capable of providing high-resolution spatial and temporal data on air quality parameters.
The successful field tests confirm the viability of the architecture. The use of the Modbus RTU-based protocol for formation management proved to be a simple and effective solution for ensuring reliable communication and control within the drone network. The sliding window filter enhanced data quality, and the ground station software provided an intuitive interface for real-time situational awareness.
Looking forward, there are several exciting avenues for extending this work. First, enhancing the autonomy of the drone formation itself is a major goal. This could involve implementing more advanced swarm algorithms where drones communicate with each other (not just the ground station) to dynamically adjust their paths based on real-time sensor readings—for example, having the formation concentrate around a detected pollution plume. This would move from a pre-planned, static formation to an intelligent, adaptive one. Second, integrating more diverse sensors is straightforward given the modular design. Sensors for gases like $NO_2$, $O_3$, $CO$, and $SO_2$ could be added to create a comprehensive air quality monitoring formation. Third, data fusion techniques and machine learning algorithms could be applied to the rich datasets generated by the formation to predict pollution dispersion patterns or identify emission sources. Finally, work on improving operational logistics, such as autonomous charging and longer endurance platforms, will be crucial for large-scale, continuous deployment.
The potential applications of such a drone formation system are vast, extending beyond environmental monitoring to agriculture (crop health assessment), disaster response (mapping gas leaks after an earthquake), and industrial site inspections. The core principles demonstrated here—modular sensor design, robust formation communication, and centralized data visualization—provide a strong foundation for building increasingly sophisticated and autonomous aerial sensor networks. The era of drone formations as standard tools for earth observation and environmental science has undoubtedly begun.