Article Highlights

  • Drone-assisted water sampling can allow sample collection in difficult-to-reach places, but is limited by volume.

  • A multi-point automatic sampling system was developed.

  • A multimodal control and operation system with autonomous perception and dynamic path planning was constructed.

1 Introduction

Water resource protection plays an important role in the ecological environmental protection. Water resources are necessary for the survival of organisms, and also play a decisive role in agriculture, tourism, and aquaculture. The increasing risk of water degradation and human interactions with inland waters, such as through fishing and other activities, have increased the demand for periodic water sampling methods to protect public health [1]. According to the United Nations World Water Development Report [2], 80% of wastewater is discharged into rivers, lakes, and oceans without sufficient treatment. Over 3.4 million people die each year from water-related diseases [3]. The United States Centers for Disease Control and Prevention [4] reports that 780 million people worldwide lack access to clean water sources, indicating that protecting water sources is of urgent concern.

Scientific environmental sampling is an essential means of conducting environmental monitoring, determining the quality of subsequent inspection, analysis, and data processing. Existing water quality monitoring methods mainly involve manual sampling for precise laboratory analysis to evaluate changes in water quality. However, faced with complex and variable sampling environments, the difficulty faced by sampling personnel is high, and both the safety and standardization of operations are difficult to guarantee. With the development of technology, sampling equipment based on unmanned boats or drones has alleviated some of these problems. However, existing unmanned sampling systems only replace manual sampling, and still cannot meet the national requirements for standardized surface water sampling, especially for larger water bodies with greater surface areas. According to national standards from the Ministry of Ecology and Environment for surface water detection and sampling, two sampling points are required for waters with a width between 50 and 100  m, whereas three are required for waters with a width greater than 100  m (Table 1) [5]. For water with a depth within 5  m, the sampling depth should be at 40–50  cm. However, most autonomous sampling systems are relatively slow, spatially restricted, or difficult to use; no method captures water samples quickly at multiple locations while overcoming barriers or longer distances.

Table 1 Setting of sampling points for rivers and channels
Full size table

The traditional mechanical sampling method relies entirely on manual operation, resulting in low work efficiency, unreliable sampling data, and low accuracy. Some scholars [6, 7] have studied the use of electronic self-control sampling methods to collect water samples, replacing these primitive and rudimentary sampling devices. By setting parameters related to sampling (such as volume of sample and depth), the sampling accuracy can be improved to a certain extent.

Currently, research on intelligent devices for water collection has focused mainly on two directions: unmanned aerial vehicles (UAVs) and unmanned ships. Compared with unmanned ships, UAVs have better flexibility and a wider range of applications. UAVs offer several benefits, such as the capacity to swiftly reach and operate in hard-to-reach areas, as well as the ability to perform tasks efficiently while minimizing costs. Unmanned ships inevitably stir up the sampled water area, meaning the collected water samples may not reflect the water environment. Sampling devices based on unmanned ships are more suitable for large-scale and deep-sea water collection or occasional monitoring [8,9,10,11,12]. In 2022, Horricks et al. [13] compared sampling devices based on UAVs and ships as carriers for microbial water sampling in marine environments, concluding that a sampling device based on UAVs would replace traditional sampling methods that depend on ships.

There are two types of UAV-based sampling devices: one fixes the sampling device to the UAV to allow integrated flight and sampling, while the other has the sampling device hung on it for collection (Fig. 1). The direct flight UAV collection method has high efficiency and flexibility but requires more complex water environments and weather conditions. Placing samplers on the water surface and controlling them using UAVs for collection is relatively stable and reliable but requires more time and space for the collection process to be completed.

Fig. 1
figure 1

Two types of UAV-based sampling devices

Full size image

Because drones do not need to consider human physiological conditions and endurance, they can perform tasks under various complex terrains and geographical conditions. Sampling methods based on drones have been favored by many scholars. Schwarzbach et al. [14] studied the use of helicopters to obtain water samples from distant wetlands. Their method used a small water pump attached to a 1.5-meter long inlet pipe through the helicopter to collect water samples into sampling bottles; in their study, they analyzed the stability of water collection under external environmental changes such as gusts and loads. In 2015, Ore et al. [15] developed a water collection device equipped with a multi-rotor drone, a water pump, sampling tubes, and sampling bottles, which could collect three 20-mL water samples per flight. Although this device allowed for the collection of multiple samples, the volume collected was relatively small, and it was difficult to accurately control the sampling depth; this limitation prevents it from meeting the requirements of situations that demand the analysis of multiple biochemical indicators. In 2017, Song et al. [16] proposed a device that uses drones to collect 25  mL water samples online and measure water temperature and conductivity. However, this sampling device also had a small sampling capacity and limited payload. Koparan et al. [17,18,19] designed a multi-rotor drone water sampler that integrated various sensors to collect environmental water samples and measure parameters such as pH, conductivity, dissolved oxygen (DO) and temperature. However, the endurance time and flight distance were comparatively short, and it could not achieve depth-controlled sampling. In 2020, Castendyk et al. [20] designed and verified a HydraSleeve sampling device loaded on a drone, which could collect large-volume (1.75 L) samples from pit lakes, replacing the previous method of sampling by ships and improving the safety for sampling personnel. However, this device required professional drone operators for operation. Zhang et al. [21] developed a water sampling and multi-parameter monitoring system fixed on a UAV. The process required no manual intervention and could meet daily environmental monitoring needs, but could only collect one sample at a time. In 2024, Lariosa et al. [22] developed a drone-based water sampling device with a tethered system and auto-lock mechanism to minimize contamination and to prevent leakage during water sample collection, however, it was also limited by the number of samples.

In a review article on unmanned aerial vehicle (UAV) water sampling methods, Lally et al. [23] concluded that UAVs can meet some aspects of the physical, chemical, and biological requirements for large-scale water body sampling in a safer, more effective, and cost-efficient manner. However, they also noted the main limitations of existing systems, including limited water collection volume (\(\le\) 500  mL) and difficult-to-guarantee sampling accuracy.

Current environmental water sampling devices are mostly used in the field of routine monitoring of water bodies. The number and capacity of samples in a single collection are very limited, resulting in low collection efficiency. In addition, both the collection device and the UAV are simple combinations, with sole control functions for both the sample collection device and UAV control. The UAV control mode is distinct and cannot achieve autonomous flight in complex environments. Additionally, professional pilots are needed to operate the UAVs, which cannot fully meet the requirements for daily monitoring and water sample collection related to health emergencies. Because of this, further research and development into water sampling devices are still needed.

Through discussions with water sampling workers from an environmental monitoring station, we derived four key high-level requirements for a UAV-assisted water sampler: it must be operated easily by one operator without drone operation skills; achieve automatic collection with multiple containers while accurately controlling the collection volume and depth; introduce no bias toward altering water properties compared with current practices, and be affordable to motivate a change in current practices.

The objective of this study was to develop and evaluate the functions of a low-cost UAV-based water sampling device to ensure the depth of sampling and meet the multi-point sampling requirements of different waterbody widths. This research is organized as follows: the achievement of the sampling device is explained below part two; and the field tests will be presented in part three; finally, the conclusions and future work will be addressed.

2 Materials and methods

A method for using the UAV device is illustrated in Fig. 2. An operator, equipped with a ground station computer, designates at least one pair of GNSS coordinates for water sampling. The UAV then travels to each location and collects water samples. The UAV-assisted sampling device lands on the water and captures samples while floating. Compared with the current hovering sampling method, this device floats on the water, allowing accurate, control of sampling depth; additionally, when large-capacity and multi-point sampling is conducted, it can effectively reduce the energy consumption of the system and extend the endurance time. While moving between sample locations, the vehicle can overcome various obstacles such as dams, bridges, or land. To prevent any mixture of samples, the system utilizes a water pump to flush out water from the current location and discharges it overboard. After the mission is finished, the system goes back to the ground station where scientists can exchange containers and commence another mission.

Fig. 2
figure 2

Overview of the proposed method

Full size image

2.1 Design of a multirotor UAV

Multirotor UAVs have significant advantages over fixed-wing aircraft for sampling. Small multirotor UAVs can achieve vertical takeoff and landing [24]. Multirotor UAVs have many beneficial characteristics, such as a simple structure, easy and flexible operation, and strong maneuverability. During flight, attitude can be controlled by the motor speed. Multirotor UAVs can satisfactorily meet the requirements of low altitude, low speed, and large range needed in the process of collecting water samples. Combined with the requirements of multiple containers and large capacity collection, the developed system uses an octo-rotor electric UAV as the mounting platform for the collection system to achieve a large payload of more than 8  kg while reducing the size of the UAV body for easy transport. The design process of the UAV mounting platform is illustrated in Fig. 3.

Fig. 3
figure 3

Design process of the multirotor UAV

Full size image

The main components of the octo-rotor UAV include the frame, propellers, flight controller, motors and ESCs, power supply, landing gear and global positioning system (GPS), etc. To design the body, analysis and simulation were conducted using the RflySim unmanned aerial vehicle design and simulation platform [25], released by the Reliable Flight Control Group at Beijing University of Aeronautics and Astronautics. From this, a multirotor unmanned aerial vehicle with a payload of 8  kg (excluding the weight of the body) should use a frame measuring 800–1200  mm. After analysis and simulation, an octo-rotor UAV frame with a center frame diameter of 1050  mm was ultimately selected. The motors used were 4120 brushless direct current motors, paired with Hobbywing 40A ESCs. The propeller blades were made of 15-inch all-carbon-fiber paddles. A 6 S 16000 mAH lithium battery with sufficient endurance capacity was used. In terms of drone control system design, the commonly used open-source PX4 architecture was adopted, and the CUAV V5+ flight controller that supports this firmware was selected. The UAV components utilized in this work are illustrated in Table 2, while its mechanical design is indicated in Fig. 4.

Table 2 Specifications of multirotor UAV components
Full size table
Fig. 4
figure 4

Mechanical design and manufacture of the UAV

Full size image

2.2 Design and fabrication of automatic water capture mechanism

A multicontainer water capture mechanism was developed with SolidWorks(Dassault Systémes Americas Corp., Waltham, MA, USA) and fabricated. The water sampling device is shown in a three-dimensional (3D) rendering in Fig. 4a; it had a water collection unit, a multi-position rotating unit, a sampler bottle fixing rack, a depth-adjustment mechanism, and an embedded controller. The water collection unit was responsible for collecting water samples; the multi-position rotating unit worked through a servo to direct the water to different sampling bottles to allow samples to be collected from different sampling locations and to flush the residual water in the pump from the previous collection point. The bottle fixing rack held five 500  mL sampling bottles. The embedded controller automatically controlled the servo and pump in the device. The mechanical design of the water sampling device is illustrated in Fig. 5.

Fig. 5
figure 5

Mechanical design of the water sampling mechanism

Full size image

Almost all the components of the sampling mechanism were made of acrylonitrile butadiene styrene (ABS) plastic and were fixed to the airframe using four mounting points(Fig. 5b). The containers were filled through a plastic tube named the "nozzle", controlled by a servo. The servo restricted the nozzle’s rotation to a single plane, allowing a total rotation of \(360^\circ\) with \(36^\circ\) in each direction from the center (Fig. 5b). The nozzle was rotated by the servo to one of ten preset positions, consisting of five positions for filling containers and five positions for expelling water. These water-jettisoning positions were used to flush water through the system and clean it after each sample. To minimize the risk of cross-contamination, five gaps were set up between the five containers to discharge water from the flushing water pump and delivery pipeline before collecting water samples. The flushing phase could be adjusted and was typically set to 10  s, approximately two times the duration required to fill a container with 100  mL.

The nozzle was connected to the outlet of a micro water pump attached to a flange plate; the other end of the flange plate was connected to the servo motor. A flexible tube connected the inlet of the pump and a plastic bellow mounted with a linear actuator used to change the lengths of the bellow. The length of the bellow extending into the water could be controlled through the linear actuator.

The five sampling bottles were made of plastic to be lightweight and clear, allowing for quick visual inspection of their contents. The 500  mL plastic bottles are available in several formats. These bottles were placed on a circular tray primarily for easy disassembly and compactness, ensuring that the center of mass of the water aligns with the center of mass of the UAV. The ABS plastic components used in the sampling mechanism are suitable for 3D printing, reduce production costs, and offer good corrosion resistance and strength.

2.3 Design of floating attachment

To ensure the stability and accurate control of the sampling depth, the drone was designed to land and float on the water surface during sampling collection. The balance of the drone while floating and stopping was ensured by fixing four floating plastic balls on its tripod, which also provided sufficient buoyancy. The selection of floating balls was based on the overall weight of the sampling device and the available buoyancy of individual floating balls. The floating attachment is shown in Fig. 5b.

2.4 Control system design

The high-level control architecture is shown in Fig. 6. The control system had three parts: a ground station handled low-level communication with the mini-pc placed on the UAV, a mini-pc attached to the UAV that uses the ROS (Robot Operating System) to read sensor data and run SLAM(Simultaneous Localization and Mapping), and the water-sampling subsystem controlled by an embedded controller that is attached to the UAV that receives instructions from the ground station.

Fig. 6
figure 6

Control system architecture

Full size image

2.4.1 UAV control system

The UAV used the Pixhawk flight controller as it is a fully open-source software and hardware flight control system that is very powerful, allows for the development of autonomous mapping and navigation functions for UAVs, and can perfectly support ROS [26]. In terms of flight control hardware, a CUAV V5+ that supports PX4 firmware was selected, which integrates sensors such as accelerometers, gyroscopes, electronic compasses, and barometers; it also has various types of interfaces such as I2C and SBUS, as well as support for multiple communication protocols. The Pixhawk flight control system has a processor with fast computing capabilities and can be used in conjunction with ground workstation software such as Mission Planner and QGroundControl. Communication between the ground station and the flight control board can be established through data transmission modules to construct an automatic control system.

2.4.2 Multi-modal control system design for a UAV

Considering the needs of operators and to expand the scope of use, a multi-mode control system was developed for the UAV, including manual radio control, remote navigation control, and autonomous control. The mode selection is set by the user at the ground station. In manual radio control, when the remote controller sends a command, the receiver on the UAV receives the remote command and then sends it to the flight control system to achieve control. This mode is mainly used for the device debugging and temporary use by professional operators to capture a small amount of samples. Remote navigation control mode uses a GPS satellite navigation system as the positioning system for the UAV. A GPS signal receiver is installed at the top of the device (Fig. 4). The GPS module can accurately locate during takeoff, hovering, and flight. Figure 7 displays a schematic diagram of a water sampling operation facilitated by a UAV in autonomous control mode.

Fig. 7
figure 7

Schematic diagram showing the operation of the UAV-assisted water sampling system

Full size image

The above two control modes have no significant difference from the current control methods of most drones. However, these methods have the disadvantages of difficult operation, difficulty in automatic obstacle avoidance during low-altitude flight, and inaccurate positioning when the GPS is obstructed. The sampling system developed here was to mainly be used in rivers, lakes and other areas, the banks of rivers and lakes are generally planted with landscape trees, which inevitably leads to UAV positioning failure when GPS positioning is used for sampling. Therefore, an autonomous control mode was also added to compensate for the shortcomings of current UAV sampling devices. In autonomous control mode, the positioning and navigation of the UAV rely on visual sensors and inertial measurement unit (IMU) sensors, and the Ego-Planner algorithm is used to quickly plan the UAV’s flight path [27]; in this way, even if the drone flies in an environment with obstacles, it can avoid them through visual sensors.

The autonomous control system of the UAV relied mainly on a mini-pc running with the ROS system, which obtains map-related data through sensors for localization and mapping. A motion planner was used to optimize the flight trajectory of the UAV. Finally, the attitude control information was calculated and sent to the flight controller. As a low-level controller of the UAV, the flight controller implemented position and velocity closed-loop control along with attitude control to achieve stable operation of the UAV. The autonomous SLAM localization and navigation control scheme of the UAV is shown in Fig. 8. In this design, the positioning of the UAV was achieved mainly by Visual-Inertial Odometry through Intel’s RealSense stereo camera and IMU sensor. The mapping and local planning for the UAV were achieved by an Euclidean signed distance field algorithm. The autonomous SLAM of the UAV was mainly based on the ROS framework, and communication was established between each module to collect sensor data and achieve autonomous position control.

Fig. 8
figure 8

Design scheme of autonomous SLAM localization and navigation for a UAV

Full size image

During the process of autonomous mapping and navigation, an open-source algorithm with strong real-time performance, high positioning accuracy, and smooth trajectory planning known as Ego-Planner was used. By installing the ROS function package of this algorithm, configuring relevant parameters of the UAV, and calibrating sensor information, the algorithm could be executed to quickly construct a point cloud map and plan the trajectory of the UAV. Depth images obtained from the stereo camera are transmitted to the onboard computer to construct a corresponding point cloud map using the Ego-Planner algorithm, achieving real-time planning of the UAV’s trajectory. The three control modes are usually set before the drone flies, and users can set them through a ground workstation (local PC) or remote controller. Users can interact with the drone through a remote controller and manually switch the control model based on flight conditions, with the manual radio control mode having the highest priority.

2.4.3 Control system design for automatic water capture mechanism

The control unit of the water collection device used the OpenRB-150 microcontroller from ROBOTIS as the control core. This microcontroller supported programming using the Arduino integrated development environment and can control the servo motor used in the capturing mechanism. It could also communicate with the mini-PC through a serial port, making it easier to use the ROS platform to control the capturing mechanism indirectly. The control system framework is presented in Fig. 9, mainly consisting of an OpenRB-150 controller, relays, servo motors, linear actuator, and host computer devices. The hardware connection and communication between components are also presented in Fig. 9. The action of the sampling mechanism can be achieved through programming and debugging. After debugging, the sampling mechanism can be installed on the drone for overall sampling testing.

Fig. 9
figure 9

Control system framework of the sampling device

Full size image

The working process of the water capture mechanism was as follows: after the water sampler was launched to the target point and floated on the water surface, the mini-PC sent the number to be collected and the sampling capacity of sampling containers to the OpenRB-150 controller through the Rosserial protocol. OpenRB-150 first controls the relay to drive the linear actuator such that the inlet of the pipe has a depth of approximately 0.4 m; following this, it then controls the servo motor to drive the nozzle to rotate to the previous gap position of the target sampling container, before driving another relay to start the pump; after 10 s, the pump stops and completes the cleaning requirements. Subsequently, the servo motor drives the nozzle to rotate to the target sampling container and then starts the pump. After the set working time of the pump, it automatically stops, completing sample collection; the above process is repeated for the collection of other samples as the UAV ascends and returns to the next launch location.

3 Experimental assessment and discussion

The evaluation of the device focused on four distinct aspects. The payload and endurance capabilities of the drone were first evaluated (Sect. 3.1), followed by the system’s effectiveness in descending to float at the water surface and takeoff (Sect. 3.2). Next, the system’s effectiveness in collecting samples was evaluaed (Sect. 3.3), followed by a comparison of the properties of water samplers collected manually with those collected using the aerial water sampler (Sect. 3.4).

3.1 Payload capacity and endurance test

Good stability and maneuverability are key to ensuring the safe and normal flight of the octo-rotor drone. Before carrying out tasks with the sampling device, it was necessary to test the flight performance of the drone, including stability under no load and performance with different loads. Testing the changes in flight speed, flight stability, and endurance time under different loads is crucial, as these are important indicators for evaluating whether the designed drone platform can meet sampling requirements.

The drone platform was tested with three different working states of 2  kg (only carrying water sampling device), 5.5 kg (water sampling device and floating attachment), and 7.5 kg (sampling device, floating attachment, and 2  kg water sample), measuring flight speed, stability, and voltage. Each experiment was repeated three times, and the average values are given in Table 3. As shown in Table 3. With increasing load weight, the performance of the drone was not significantly affected, and both the working voltage and the flight speed remained stable. These results also indicate that if the flight conditions change due to increased payload, endurance will be reduced, because the UAV consumes more battery power to increase the thrust to overcome the payload. The flight tests with different loads indicate that the design of the octo-rotor UAV platform meets the on-site operational requirements and that it could be applied to onsite water sampling experiments.

Table 3 Performance of UAV platform with different loads
Full size table

3.2 Maintain floating on water and takeoffing from water surface while sampling

The outdoor assessment of the floating control was carried out at Hongyuan Lake at Jiaxing University. The water at this location was less than 1  m depth and a lightly breeze day with a wind speed below 3  m/s (as measured by a hand-held anemometer) was chosen. As the working state of the sampling system requires the UAV to float on the water surface, it was necessary to verify the device’s floatability and whether the UAV could take off normally from the water surface after sampling. The UAV was first remotely controlled to land on the water’s surface; after it stabilized, it was controlled to vertically ascend into the air. The state of the UAV taking off from the water is shown in Fig. 10 while Fig. 11 illustrates the flight performance chart, developed based on the calculated averages of thrust and payload at each level.

Fig. 10
figure 10

State of the UAV taking off from the water surface

Full size image

In this experiment, the main goal was to determine how much throttle the drone needed to take off from the water under different loads. These tests confirm that the floating sampling method worked well under windy conditions.

Fig. 11
figure 11

Changes in the thrust for taking off on the water surface with different payloads

Full size image

3.3 Water sampler performance

The water sampling performance and autonomous flight quality of the system were tested at Jiaxing University campus over Hongyuan Lake. To conduct outdoor sampling experiments, five locations in the pond were chosen. The center of the pond specifically facilitated manual water sampling by boat to verify the results obtained from the water sampling system assisted by the UAV. The letter “o" represented the launch position for the autonomous missions, while the sampling locations are represented by numbers, as depicted in Fig. 12. Water samples were gathered within a 2-meter radius of the sampling points.

Fig. 12
figure 12

Water sampling points on Hongyuan Lake at the Jiaxing University campus

Full size image

The effectiveness of the water sampling system was tested for both manual and autonomous flights. The UAV was operated in manual mode during the manual flight tests, controlling the drone with a radio controller. Each test included five samples, and the water sample containers were subsequently inspected. Any amount less than the set sampling values was recorded as less than full. In total, ten trials requiring 4–5  min of flight time each were conducted. Additionally, we allocated an extra 5–8  min for the system setup, emptying containers, and regular battery replacements.

3.3.1 Performance of the UAV-assisted sampling device assessed by manual flights

The automatic activation function of the water sampling mechanism was tested to ensure its reliability during a water sampling mission in an outdoor environment. The success of each attempt to collect water samples with UAV assistance was recorded. Each sampling attempt was carried out manually with individual flights. A total of ten attempts were made to collect water samples using a radio controller, with a total of 50 samples. Each trial took 6–8 min to fly and sample, with an additional 3–5 min to set up the system and empty the sampling containers. Overall,the 50 water samples were collected at the same lake and on the same day within 2  h.

To assess the effectiveness of the sampling mechanism, the UAV was operated remotely in “Loiter mode" to facilitate easy control of the flight path toward the setting target points. The UAV was launched from a location with good GPS reception and was controlled using the radio controller until the water sampler was positioned on the water’s surface. After the water sampler was launched to the target point and floated on the water’s surface, the sampling system started to collect water. The UAV ascended and returned to the next launch location after it finished sampling for a container. The success of the mission was determined by whether the sampler captured the water, and the quantity of water collected was measured; If the collected water sample was measured to be less than 95% of the set volume (100  mL), the mission was recorded as less than full and the causes of the failure were recorded as well. The results are summarized in Table 4.

Out of the 50 consecutive samples collected, 46 samples passed the qualification criteria, resulting in a success rate of 92% (ten trials with five samples each). The reasons for the unsuccessful water collection were errors made by the operator and the servo motor not being fastened securely, caused by the rigidity of the flange plate as a 3D-printed plastic component. The operator errors could be minimized with more practice, and the fastening problem on the motor was minimized by remanufacturing the flange plate as a metal part.

Table 4 Sampling success rates for manual flights
Full size table

3.3.2 Water sampling autonomous flight missions

To assess the efficiency of the sampling system under autonomous control, individual autonomous flight missions were executed for a total of 50 samples. Each trial took 6–8 min to fly and sample, with an additional 3–5 min to set up the system and empty the sampling containers. Overall,the 50 water samples were collected within 2  h. Before each flight, the sampling mechanism was configured, and the flight controller and power unit were connected. To initiate the mission, the throttle level was adjusted to 65% to provide sufficient thrust. Once the motors started rotating at this throttle level, the UAV ascended to the preset altitude and proceeded toward the predetermined location. When the UAV arrived at the specified sampling site, it descended to the water surface, and the sampling system began to collect a 100  mL water sample in one container. After the UAV completed all five sampling locations, it ascended to navigation altitude and returned to the launch siteto complete the mission.

If the collected water sample volume was less than 95  mL (the difference was set to 5  mL), the mission was recorded as less than full along with the reason for the failure. The results are summarized in Table 5. Out of the 50 consecutive samples collected, 48 samples passed the qualification criteria with a success rate of 96% (ten trials with five samples each); two were categorized as unsuccessful (4%). The cause of the two failed water sampling missions could be attributed to a malfunctioning servo motor; to decrease the chances of servo malfunction, the plastic flange plate was replaced with a metal plate.

Table 5 Sampling success rates for autonomous flights
Full size table

3.4 Comparison of UAV-assisted and manual water sampling

An experiment was conducted to verify whether the water samples collected using the UAV mechanism have similar chemical properties compared with those collected manually. There were potential variations to consider, such as those caused by pumping, the journey through the tube, agitation during flight, and changes in water properties during the time delay between sample collection and measurement on land. Manual water samples were obtained immediately after the UAV ascended from the sampling location by a person in a kayak at the same sampling location. Both sets of water samples were immediately transported to the shore for measurement. The measurements for both the manual and UAV-assisted water samples were conducted at the same time within 2  h on the sampling day.

Sampling at five locations on Hongyuan Lake at the Jiaxing University campus was used to confirm the consistency between manual and UAV-based sampling methods. Three samples were collected near the shore and two were collected closer to the middle of the pond (Fig. 12). At each location, two samples were collected manually and two were collected with the UAV-assisted sampling mechanism for a total of ten samples with each method. Each sample was 100  mL, as only four water quality indices were measured: dissolved oxygen (DO), temperature, chemical oxygen demand (COD) and pH. It took approximately 1  h to obtain these data because of the time to kayak, collect samples manually, and conduct initial analysis and filtering on site. Capturing the samples with the UAV took around 15  min.

The manual water sampling method was verified to be comparable to the UAV-assisted water sampling method. It is necessary to take temperature, pH, and DO measurements either in situ or on-site because it is not feasible to properly analyze these components once the sample has been transported to a laboratory [28]; because of this, the measurements of these parameters were conducted onsite using portable measurement sensors. After the measurements were taken at the site, water samples were transported to the laboratory to measure the COD. The lab determines the COD using non-portable equipment, as ti does not change rapidly after sampling and filtering.

The main focus of testing was to ensure that the UAV-assisted sampling mechanism did not introduce any bias in the measurements. The DO levels measured manually at the site were compared with the measurements obtained using the UAV mechanism (Fig. 13a). The DO values at the five sample locations were similar and displayed the same overall trend. That findings suggest that the UAV mechanism and the delay in measurements (which was longer when done by kayak than when flying), had a minimal influence on the DO levels. Location 2 appeared to deviate slightly from the general trend, potentially owing to its proximity to the shore, where there may be more surface plants. The use of UAV technology to collect samples swiftly could help clarify these factors. Measurements of pH revealed no significant difference between manual and the UAV sampling, though the COD values revealed some differences between the two methods (Fig. 13b and c). However, these variations can likely be explained by the normal fluctuation in the samples, and the differences do not suggest a significant bias caused by the UAV sampling mechanism. The typical range for COD in lakes is less than 40  mg/L [29], indicating that the observed variation is minimal. To ensure the accuracy of measurements, further field and lab tests will be conducted. The temperature recorded manually at the sample location remained relatively constant (Fig. 13d). Conversely, the temperatures measured in the samples obtained through the UAV mechanism varied during transportation, particularly at locations 4 and 5. To address this issue, future iterations of the system should include a temperature probe mounted at the end of the pumping tube to accurately measure water temperature at the sample location.

The experiments demonstrate that the UAV-assisted sampling mechanism is capable of capturing samples that are comparable to those collected manually while significantly decreasing the amount of effort and time required. This will allow hydrologists to obtain a greater number of samples from a single lake or river, enabling the development of a high-resolution map. This is particularly useful in scenarios such as after a rainstorm or a factory leakage, where it is important to identify the source of chemical or biological contaminants. Additionally, reducing collection time is crucial as various water properties, such as DO, can change rapidly within a matter of hours. By utilizing this UAV system, the collection time can be reduced by nearly ten-fold.

Fig. 13
figure 13

Measurements of water chemistry are collected both through manual sampling and using an UAV device. Points represent the average of two replicate measurements

Full size image

There are several limitations to this UAV-assisted autonomous water sampling device. The device cannot be used on very windy or rainy days because of issues with flight stability. Additionally, the 25-minute flight time limits the use of a sampling system to collect water samples for a lake with a width greater than 10  km. Although contamination was avoided in this study by washing the pump and tubes before sampling, the sample still may end up being a mixture of sites if the user collects polluted water samples and flushing protocols are sufficient to prevent cross-contamination; this will be need to be assessed in future studies. Finally, users will need to obtain permission from the aviation administration to perform UAV flights in some areas.

4 Conclusions and future work

To meet the needs of water monitoring and sampling for environmental sampling, an unmanned intelligent sampling system was developed that includes an octo-rotor drone-carrying platform, a multi-container automatic collection device, and a multi-mode control system. The designed platform has a maximum payload of 8  kg, a flight time of 20–25  min, and a flight speed of 1  km/min; it also has multi-mode control system functions to meet the needs of monitoring and sampling while being simple to operate. The sampling system can land on water surfaces and collect water samples while remaining afloat; its five 500  mL water sample collection containers can be used to accurately collect samples at 40–50  cm depth and allow for multiple point sampling for larger water bodies with greater surface areas. The characteristics of the water in the samples collected by the UAV were confirmed to be consistent with those obtained through conventional manual sampling methods demonstrating that this system can be used to increase geographic coverage and temporal resolution of water sampling efforts.

The utilization of this UAVs for water sampling could benefit areas where access to water sampling sites is challenging. It also compensates for the current inability of UAV sampling devices to perform multi-point sampling with large volumes per flight. The multi-modal control system reduces the skill requirements for operators. Overall this system provides a feasible solution for monitoring environmental areas where current sampling methods may pose challenges. Future work will focus on expanding and improving the system’s outdoor capabilities, particularly in exploring its potential for sampling at greater depths and integration with other sensing and sampling mechanisms. Additionally, the integration of more sensors will be pursued to allow for the measurement of on-site indicator parameters and the upload of data to the cloud. The performance of the system will also be tested on a more diverse set of water bodies, such as in coastal waters.