1. Abstract
Planting the paddy crop generates large amounts of greenhouse gasses as bacteria thrive in the flooding water over the farmland that is required to grow this semi-aquatic plant. Today, the mass methane emission from rice paddy fields has become critical to resolve. This research project plans to use IoT technology to create systems of early detection and prevention of methane formation to solve the issue. The detection system identifies when methane is about to be generated which is indicated by 1) the paddy crop entering the reproductive growth phase and 2) the dissolved oxygen concentration level (DO level) being 6.5 mg/L or under. When these conditions are met, the prevention system will activate, which cycles the flooding water from the farmland through a turbine (in a flow meter) that is oxygenated by an air pump motor. However, the project doesn’t aim to make an actual product, but rather an experiment that verifies the feasibility and the credibility of this particular approach in resolving the problem of methane emission. In the end, the project was proven to be successful through experiments that showed the DO level increasing consistently when the prevention system is activated.
2. Introduction
In Taiwan, the most planted crop is the paddy [1], or Oryza sativa [6], which is used to make rice. Rice has always been one of the most essential ingredients in a Taiwanese meal. Its popularity has persisted since decades ago. Today, rice makes up around 20.42% of the total agricultural production value in Taiwan [1]. Furthermore, other countries – especially those in Asia such as Indonesia, India, Vietnam, etc. – also heavily export rice and rely on this crop for their economy [2]. Not only that but African countries like Benin [3] and Ethiopia [4] also depend on rice as a food source as they majorly import rice from other nations. From this, one can see the extent of influence rice has in different places all around the globe.
What makes growing rice different than just simply planting seeds in the soil, however, is how paddy is a semi-aquatic plant [5]. This means that the crop grows in both soil and water. [6] The surface on which the paddy is planted consists of two layers: soil which the plants are gripped in on the bottom and a few-inch-tall sheet of water on the top over the land [5]. This method of flooding their farmland keeps unwanted weeds and pests away as they cannot survive in the water that paddies can withstand [7]. In addition, this method also makes sure that there is enough water for the crops [7]. This project aims to detect and prevent methane from generating in the water of paddy farms.
Unfortunately, a few problems come from this specific way of agriculture. Firstly, soil salinization affects nitrogen uptake negatively and spreads toxic ions like chloride that can badly damage plants and influence their growth [8]. Furthermore, farmers tend to use a lot of fertilizers consisting of nitrogen to apply on the paddies; sometimes these crops may not absorb the fertilizer thoroughly, which can lead to the emission of nitrogen oxide gasses [9].
The most urgent issue is the generation of greenhouse gasses: ideal anaerobic (without oxygen) conditions are found in these fields under the water, which can lead to the emergence of bacteria in the residues of the crops [10]. As a result, the decay of organic matter can generate greenhouse gasses such as methane, carbon dioxide, nitrogen oxides, etc. According to the Asian Development Bank, the farming of rice plants contributes to around 8% of the total greenhouse gas emissions produced by agricultural practices in the world [11], which shows how problematic agricultural practices relating to this crop may be.
The emission of greenhouse gasses is thus an extremely urgent and serious concern and the very issue that the project aims to resolve. This issue is particularly chosen as it is the biggest problem that the method of flooding fields required for rice crops creates, significantly affecting the environment and the world itself.
As mentioned earlier, many different types of greenhouse gasses can be emitted from paddy fields — carbon dioxide, nitrous oxide, methane, etc. However, the most harmful gas out of all, and the gas this project targets to eliminate, is methane. Note that approximately 12% of total anthropogenic methane emissions in the world are caused by these rice fields, which account for 1.5% of all greenhouse gasses’ global warming effect [11] [26]. This itself shows how alarming the presence of methane gas in these farms can cause a huge global problem. Although other types of harmful gas are also present in the farmland, methane traps the most heat within the earth’s atmosphere [12] [13] [14]. According to the United Nations Environment Programme, 25% of global warming is due to the presence of methane [14], which only further reinforces the idea that it is a dangerous gas that must be targeted to be reduced first. According to NASA, there are indeed other major reasons for methane emissions like landfill waste decomposition and fossil fuels [15]. Even though that is the case, agriculture, specifically the farming of paddies, is one of the more controllable factors in reducing the emission of this particular greenhouse gas.
Methane gas can be found in various locations on the farmland. Due to the flooded land, bacteria emerge, which releases methane as they decay organic matter. Therefore, the greenhouse gases remain in the soil and water and eventually travel into the air and the atmosphere. In a paddy, the plant absorbs the methane gas from the soil water, carries it through its stem, and lastly emits it into the air from its micropores [16]. This is how methane is released into the air. To prevent that, one must reduce the amount of methane found in the source of the gas from which the plants absorb, which is the soil water itself. On a side note, what makes soil water different from the water used for flooding is that soil water contains a wide variety of different substances such as dissolved nutrients as well as organic and inorganic matters [17]. The factors explained above are why this project is focused on detecting and preventing methane formation in soil water in paddy farms.
3. Related Works
There have been many different ways to resolve the issue of paddy fields releasing massive amounts of methane and other greenhouse gasses of course. For example, a research called “Methane Emission from Paddy Fields and its Mitigation Options on a Field Scale” conducted by agricultural researchers Kazunori Minamikawa, Naoki Sakai, and Kazuyuki Yagi came up with certain practices that ameliorate the situation [18]. One of which is applying nitrogen fertilizers and organic matter into the farmland to reduce the amount of methane; another solution is to properly monitor and manage the water that is used to flood the fields. This research provides different solutions that can help with the brainstorming process of eradicating methane.
One work that implements embedded systems in helping to resolve the issue is “Autonomous Cyber Physical Systems for Monitoring of Methane Gas in Rice Field” by engineers Muhamad Komarudin, Hery Dian Septama, and Titin Yulianti [19]. This project monitors the amount of methane gas in the fields using a cyber-physical system. Additionally, there are findings that methane is found to have higher amounts at higher temperatures and lower humidity, typically found in the middle of the day. Also, the system functions autonomously and utilizes the Internet, solar power, and geo-location to identify certain factors. This work shows how one can monitor methane gas over a large piece of land.
There still needs to be a method to effectively separate the methane from the soil water, though. Engineering researchers Xiaoxiao Sun, Yanbin Yao, and Dameng Liu have proved the effectiveness of one particular method of displacement in their research “The behavior and efficiency of methane displaced by CO2 in different coals and experimental conditions” [20]. Their experiments use carbon dioxide to displace methane gas in coal samples. As a result, it is detected the methane would then be released due to the concept of partial pressure of gasses as the administration of CO2 into the coal samples is conducted. This shows one possible way that the methane can be separated from the water.
4. Methane-Prevention Methods
The goal is clear: the project will aim to create an autonomous system that somehow detects and prevents the generation of methane from the field. The first method is to detect when methane is in a large amount and use the technique of water displacement to extract methane from the water. The second method is to swap and oxygenate the water at the same time when detected that methane is about to form in the water.
Method 1: Methane Extraction
One method to approach this issue is to extract methane from soil water and trap it in a closed container using water displacement to prevent the gas from getting into the atmosphere [21].
This is what a simple diagram of a water displacement system looks like:
Figure 1. Water displacement diagram [22]
From the diagram, one can see that the gas residing in the water is being separated and released above the water due to its lower density. Methane is therefore applicable to the system here as it has a lower density than water.
Overall, the system this project aims to create will detect where methane is, measure the amount of methane that is present in an area, and remove the methane using an autonomous water displacement system. Because of the vast size of a typical paddy farmland, the full system would involve multiple smaller structures (a combination of methane-detecting and water displacement apparatus) dispersed as evenly as possible throughout the field to maximize the area that this system can function in. This entire system is necessary and important as it helps farmers swiftly identify sources of methane emissions autonomously and removes this greenhouse gas more efficiently than other methods without the use of embedded systems.
However, this system comes with its flaws. Firstly, it is very ineffectual to place multiple apparatuses all over a huge paddy field. The material may cost a lot and it is difficult to manage so many devices across such a large area. Furthermore, even if the methane is isolated, it is very difficult to find a way to collect and eradicate it without letting it out into the air. The methane is technically “separated” from the water through this system, but there is no way to eradicate the methane. Another system was needed as the solution.
Method 2: Water-Swapping
This new system needed to be more simple and direct and able to effectively resolve all the problems that were raised in the first system. First of all, something must be done with the methane. The problem is there is just no way to remove it after it is formed. Therefore, the solution is to swap the water used on the field before the methane is even formed. Specifically, the water needs to be transported to somewhere where it can be exposed to oxygen using a certain method [23] [24] [25], then brought back to the field. This prevents bacteria that thrive under anaerobic conditions from forming and creating methane gas. The process of transportation will involve tubes connected to different parts of the water-freshening system that intake and discharge the water. This process aims to prevent the generation of methane by renewing the water by maintaining its contained oxygen level.
Now, there are multiple techniques to allow the water to contain more oxygen. One technique is to increase the surface area of the water to increase exposure to oxygen [23]. Another technique is to produce great motion in the water [24] [25]. The third technique is to directly use an air pump to inject air. All three techniques are some ways to increase the water’s exposure to oxygen 24] [25].
Technique 1: Increasing Surface Area
The first technique is to increase the water’s surface area. The initial idea was to use hoses to transport the water from the field into the river or water banks, where it is exposed to oxygen, and bring in new water back into the field to do a swap. However, this method is infeasible as it requires a lot of work to bring the water, and certain rivers or water banks may be too far away. Instead, an alternate solution is to transport water through a water-freshening system to expose it to oxygen, taking the water from the paddy field and sort of filtering it through the cooler machine before sending it back into the field.
Alternatively, there is another idea: to turn the water into droplets, increasing the total surface area of the liquid. For this system, water is exposed to air while it is in its droplet form. One system that can simulate this function is an air cooler. Below is a diagram of what the inside of an air cooler should look like:
Figure 2. Air cooling system [31]
The basic concept is that hot air will enter one side of the cooling medium, or the water curtain, which will create cool air when water is dripped from the water distributor to the water tank. The most important part of this system is how the water is exposed to air from being in the distributor to the tank. Therefore, this air cooler system can be used to transform the water. For this, a hose transport system with motors can be used to transport the water throughout the system. One hose can stretch from the field water to the water distributor, and the other hose can come from the water tank back into the field, completing a sort of filtering system that recycles the same water from the field. This method does not require multiple apparatuses to be set up across the field as one device is enough to perform water filtering.
However, although this system is effective in exposing air directly to droplets, it is too large and heavy, which makes it inconvenient to carry and place on farms. A smaller and more accessible apparatus is needed instead.
Technique 2: Creating Water Motion
The second technique is to make the water enter great motion. One device that can be used is a water turbine system. And in a flow meter, one can find precisely just that. Below is what the components of a flow meter look like:
Figure 3. Flow meter [32]
The flow meter itself serves to measure a substance’s flow rate by measuring the times when the magnet touches the coil at the top hole of the system. (The fluid would flow from one side of the apparatus to the other through the turbine.) The measurement will be shown on a small monitor when the data is sent from the coil through connections to the data logger as depicted in the diagram above.
The hose that sucks the water from the field can therefore enter one side of the flow meter and the other side will connect to the other hose that leads the water back to the field. Inside the system, the water will go through the turbine wheel, which is shown in the middle of the apparatus. This creates a fast circling motion in the water that creates bubbles, increasing the liquid’s surface area. However, one may notice how the system is useless regardless as there is no opening that allows the stirred water to be exposed to oxygen.
Therefore, the solution to that issue is to solely utilize the flow meter system for its structure and turbine system and not the measuring aspect. The top hole of the flow meter system would be removed (the connection to the data logger component along with all the measuring parts included).
Technique 3: Installing Air Pump
The third technique is to directly inject air into the water. Instead of looking at this technique as a separate solution, an air pump system can be combined with the flow meter system discussed earlier to create a more efficient device. Specifically, the components on the top of the flow meter (the connection to the data logger) will be replaced by an air pump that directly pumps air into the apparatus at the water and the turbine that is spinning it. This will ensure maximum exposure of oxygen for the water and thus confirm the success of this process. The project will utilize this very water-swapping technique: transporting the water into the flow meter where it will be oxygenated by the air pump (via tubes) then bring the water back into the field. The combination of these two techniques ensures the success of the project and would like the diagram below:
Figure 4. Flow meter and air pump motor system
5. Early Methane Detection Methods
The early detection of methane refers to checking for signs that indicate methane is about to form in the water soon, but before methane gas is produced. Ultimately, the system made for early detection will be used to activate the methane prevention system when these signs are detected.
Method 1: Growth Stages
However, there is still one problem. Doing this cycle of water filtering requires too much energy consumption for the machine to work. Therefore, the filtering action will be performed only when necessary. In a paddy crop’s life cycle, there are multiple stages, as shown in the diagram below.
Figure 5. Rice growth stages [27]
In a growth cycle, the water filtering action only needs to be performed at the start of the stage where methane levels are the highest.
However, how does one identify when a stage is entered? The life cycle of a paddy field is unpredictable, therefore one cannot use the time to measure how long to wait in between before activating the system. There are just too many outside factors to measure. As a result, the easiest way is just to set up a camera that looks at a paddy crop and use an AI machine learning model to check which stage the paddy crop is in. When the paddy is identified to be in the stage where methane is produced the most, it will send a signal to the water cooler system and activate it.
However, there were different answers to what exactly are the main stages of a paddy’s growth cycle. According to research called “Rice Growth Stage Classification via RF-Based Machine Learning and Image Processing” by Rodney Tai-Chu Sheng et al., the main stages of paddy are 1) sowing in the nursery bed, 2) transplanting, 3) tillering, 4) stem elongation, 5) panicle initiation, 6) flowering, 7) milk stage, 8) mature stage [27].
Figure 6. Rice growth stages [27]
According to the article “Rice Growth and Development” written by Richard Dunand and Johnny Saichuk from LSU AgCenter, the stages are 1) emergence, 2) seedling development, 3) tillering, 4) internode elongation, 5) prebooting, 6) booting, 7) heading, 8) grain filling, and 9) maturity [29]. According to EOS Data Analytics, the stages are 1) Germination, 2) Leaf Development, 3) Tillering, 4) Stem Elongation, 5) Booting, 6) Inflorescence Emergence/Heading, 7) Flowering/Anthersis, 8) Development of Fruits, 9) Ripening, 10) Senescence [33].
From just these three examples, one can observe a lot of things. Although there are several overlaps and similarities across the examples, their overall difference is still vast and apparent. This difference poses a huge problem as it makes it difficult to choose the categories of the database. Nevertheless, almost all sources agree that there are three main larger and longer phases that include these smaller stages as listed above. These three main phases are vegetative, reproductive, and ripening.
According to research called “Mitigating Methane Emission from the Rice Ecosystem through Organic Amendments” by Kandasamy Senthilraja et al., the reproductive phase is the phase where the paddy generates the most methane gas [26]. Therefore, a program that activates the water-pumping system at the first instance in which the paddy is detected to be in the reproductive phase needs to be created.
Method 2: Dissolved Oxygen Level
However, our renewed design aims to prevent the production of methane gas in the first place, so using a methane sensor would be no use. Instead, a dissolved oxygen sensor is implemented to collect data on the dissolved oxygen level (DO level) present in the water. If the amount increases from before to after the system does its job, then the project works. If the sensor reads a concentration DO level that indicates uncontaminated water, then the system should deactivate as the job is done. According to the Government of Northwest Territories, the DO level for healthy water is above 6.5-8 mg/L, so that means the level for water to be considered as dangerous is below 6.5 mg/L [30].
***
To further solidify the accuracy of the project, the process can be improved so that the water pumping system is activated in two conditions instead of just one – in the conditions that the camera detects the plant to be in either of the two stages in which methane is produced and DO level is detected to be lower than 6.5 mg/L.
Although the water pumping system now knows when to activate based on machine learning and dissolved oxygen systems, it doesn’t know when to stop. Similar to the process of detecting early methane signs, there can be a method to detect when the environment is safe and doesn’t require the system to function. In this case, when the DO level is equal to or above 8 mg/L, in which the water condition is healthy, the pumping system will stop until the DO level of the water reaches 6.5 mg/L again. Now the water pumping system knows when to start and stop. However, this cycle of starting and stopping the system based on DO level will end when the crops are no longer in the reproductive growth stage phase.
6. Process
This project will just be a sort of prototype for a potential product that can be on the market in the future, so the approach will be more experimental – to verify whether the device will work rather than creating a finalized product. The experiment would be done in a home lab setting: at room temperature with healthy water conditions. As a result, many variables and factors will be different than those for a real farmland setting, where dissolved methane levels are higher and temperature varies over time.
As a result, the process will be split into three parts: 1) the water pumping system, 2) the machine learning system, and 3) the dissolved oxygen system. Specifically, these three systems only work if: 1) the first system successfully swaps water through a flow meter while oxygenating it with an air pump in and of itself, 2) the second system can successfully identify the growth stage of a paddy crop and signal the first system (of water pumping) to activate when a plant in the reproductive stage is recognized, 3) the dissolved oxygen sensor can work to tell the first system to activate whenever a dissolved oxygen level of below 6.5 mg/L is detected.
Below is a diagram that shows what the final product will look like:
Figure 7. Final product diagram
*The lines that connect the devices simply mean there are wires (however many) between them, it does not necessarily mean there is only one wire between the devices connected.
As one can see from above, the embedded system collects data from the sensor to send to the local server and drives the motors when a trigger from the local server is detected. Since the distance between the embedded system board and the local server is rather massive in a vast farmland setting, the connection between these two devices would be wireless for the actual product. However, the experiment, which is conducted in a home lab setting, simply uses a USB cord for the connection.
For the making of the project, here are all the materials, tools, and applications needed:
General & Tools | Water Pumping System | Machine Learning System: Applications | Dissolved Oxygen System |
Arduino IDE Program | L298N motor driver | p5.js | SEN 0237 Dissolved oxygen sensor [38] |
Arduino IDE board | 2 water pump motors | Teachable Machine | Temperature sensor or thermometer |
Arduino USB cable | 2 tubes for water transport | p5 serial control | |
Laptop | Flowmeter | ||
Wires | Air pump motor | ||
Battery holder | Tube for air transport | ||
Two batteries | Pot (big container) | ||
Breadboard | Water | ||
Screwdriver | Jar (smaller container) | ||
Water-proof tape | |||
Drill |
Part 1: Water Pumping System
The first part was to make the main water pumping system. As seen from the diagram above, there are two containers. The first one is a pot (used to simulate the flooding water in the field). The second one is a jar, a separate container that acts as a temporary storage for the recycled water. Water motor 1 will then take the water from the pot to the jar through the flow meter. Additionally, the air pump motor will pump oxygen into the flow meter as the water flows through the turbine. Once that process is done, water motor 2 will take the refreshed water back into the pot.
The flow meter (left) and air pump motor (right) are needed for the first step.
Figure 8. Flow meter (left) and air pump motor (right)
The data logger (the wire-and-sensor component) is taken out of the flow meter since this project only uses the flow meter for its structure, which allows water to flow in and out through a turbine inside. A small hole was drilled into the flow meter and an air-transport tube was connected from the hole to the air pump opening. Now the air pump can successfully supply oxygen to the flow meter.
For building the water motors system, a cooking pot that simulates a farmland was prepared and Water Motor 1 was taped near the bottom of the container. A water-transport tube was connected from the opening of Water Motor 1 to one of the openings of the flow meter. A plastic jar was prepared and Water Motor 2 was taped at the bottom of it. Two holes were drilled in the lid – one of which allowed the other opening of the flow meter to be inserted in (this allows the water that Water Motor 1 sucks to be transported into the jar while being oxygenated), and the other enabled the other water-transport pipe to be connected from Water Motor 2 back to the pot.
Then, the wires of both the water motors were attached to their respective output pins of the L298N Motor Driver. This allows the motors to be able to activate when their respective inputs are signaled for. This requires programming in the Arduino IDE Application, so the L298N Motor Driver was connected to the Arduino IDE board, and the board was connected to a local server (laptop) via a USB cord. (The code will be created in later parts of the process.) To have an additional power source that does not just come from the laptop, batteries are connected to the Arduino Board through a breadboard to provide electricity for the system to work. Below is what the completed water pumping system looks like:
Figure 9. Completed pumping system
Part 2: Machine Learning System
The first system involves using AI machine learning (ML). An ML model must be created that can identify if a paddy crop is in the vegetative, reproductive, or ripening growth stage phase. The water pumping system needs the model to tell when a paddy crop enters the reproductive phase when the most amount of methane is about to be produced. To create an accurate model, a dataset of images of all the paddy’s growth stage phases needs to be created for the model to be trained successfully with a training image data subset and tested out by a testing image data subset.
Therefore, to prepare for the testing and training data subsets, we need to find multiple images for the three growth stage phases. For this experiment, a total of 90 images were gathered, all of which can either be in the testing or the training subset depending on how the experimenter chooses to split the dataset.
These images were sorted into an online machine learning software called Teachable Machine in these three separate categories, which runs a program that allows my dataset to be trained successfully. Teachable Machine is a free platform made by Google that contains a feature to automatically sort images into categories.
The user simply needs to add new classes (the categories) and then upload several images for each class as datasets to tell the program what each category looks like visually. The more images, or the larger the dataset, the better and more accurate the model will become. Lastly, press the respective buttons on the page to train and export the model.
According to its official website [35], the platform uses transfer learning to train its models. Transfer learning refers to when knowledge from a previous task is used to improve the model for future tasks. The platform uses pre-trained mobile models built with the TensorFlow.js algorithm as well as the technique of transfer learning to accurately identify the class the input belongs to.
The Teachable Machine program was set up and the images were uploaded into their respective categories.
Figure 10. Images in Teachable Machine [35]
After that, the model was tested out. To provide an image input, one can either upload a photo from files or open the webcam, in which the program will automatically update how likely the image shown on the camera is to be in each class (in percentage). The program was shown an image of a paddy in the transplanting stage, which was one of the stages in the vegetative phase. If the model works, the program should indicate a much higher percentage that the image is in the vegetative class than the other two classes. The model had successfully identified the image as being in the vegetative class. The same process was repeated for the reproductive and ripening phases.
Figure 11. Teachable Machine class identification
As seen above, the algorithm does work for these three specific images. However, there needs to be an accurate test to check if the machine learning does work effectively. To do this, the dataset is to be split into training and testing subsets at different numbers. For example, the 90 images can be split into 45 images for the training subset and 45 for testing or 75 for training and 15 for testing. Each way of splitting (based on the number of images in the subsets) is one trial, and the experiment consists of five trials. In each trial, the training images will be trained into a model that will be tested with the testing images which produces a model accuracy percentage – the percentage of how many images from the testing subset are identified correctly by the model out of all images; in each trial, this process will repeat for three times with different images sorted in the specific number of images in training and testing subsets for the trial each time to ensure accuracy; the average of the model accuracy percentage for the three times will be the data observed for each trial.
Here are the results:
Figure 12. Graphs of average model accuracy based on the number of images in subsets
The two graphs show how the average model accuracy (in percentage) will be affected by the number of images in both the training and testing subsets over these five trials. As one can see, the two graphs are nearly the same (other than the order of the number on the x-axis) as both graphs show the same experiment. The separation of the graphs is just to show the respective numbers of images in training and testing subsets (in which the number of images in training and testing always adds up to 90 in each trial).
From this data, it’s clear that the lack of images in either the training or testing subsets can result in a rather low accuracy (56% and 63%). On the other hand, having the same number of images in both subsets led to a higher accuracy (92%), which makes it the best option to use for the project.
Now there is the model from Teachable Machine, the model needs to connect to an Arduino IDE code that tells the water pump motors to activate when a reproductive phase is produced. The method was to type a set of code on another program that produces three separate outputs based on which of the three classes from our model is identified through a camera. This output will then be sent to the Arduino IDE code, which controls the motors. Therefore, this program needs an area to code, can include functions that open a camera, has an add-on that can transport the Teachable Machine model to the code, and has a serial control that can connect the code of the program to the Arduino IDE app through a port.
p5.js, an online open-source JavaScript library that allows coding for a sort of web, was a perfect program for the tasks required. The interface is split into two: the left side is where the user’s code goes and the right is where it shows what the web looks like based on the code. Using the concept of web editing, the function of opening a space for the camera can be implemented. There is indeed an add-on to import a machine learning model from Teachable Machine. Additionally, there is a serial control for p5.js, which can connect the program to the Arduino IDE code using ports on the laptop itself.
On p5.js, one can add a camera to the web and transfer the model from Teachable Machine into the program, producing different outputs based on the class identified from the camera. If the input is detected to be in the vegetative phase, the output would be 1. If it is in the reproductive phase, the output is 2. If it is in the ripening phase, the output is 3. The output will then be sent to the p5.js serial control through one of the ports on the laptop. All there is left to do is connect the Arduino IDE board to that same port on the laptop using a USB cable. The output then can be used in the Arduino IDE application to control the water motors.
Specifically, the output from the p5.js program will be used as input into the Arduino IDE program. Whenever the input reads in 2 (being the indicator that the plant is in the reproductive phase), the motors will activate, but if the input reads in 1 and 3 (indicators for the vegetative and ripening phases, respectively), the motors will remain dormant. With that, the machine learning system is complete.
Part 3: Dissolved Oxygen System
The last stage involves making sure that the dissolved oxygen sensor is capable of detecting certain levels of dissolved oxygen (DO) and checking if the DO level increases before and after the water pumping system activates. This system will then work with the AI machine learning system to decide when to activate the water pumping system.
As mentioned earlier in the paper, a healthy water condition should have a DO level above 6.5-8 mg/L. Therefore, the water pumping system will activate when the sensor reads a level of 6.5 mg/L or below and stop when it reads 8 mg/L or above.
This is what the dissolved oxygen sensor looks like:
Figure 13. DO sensor
The black stick on the right is the probe, which is the part inserted inside the water to receive the input. The other device on the left of the probe is the signal converter board, which transfers the input received from the probe to the Arduino board. On the signal converter boards, the black and red wires are connected to the battery, and the blue wire is inserted in the digital pin A1 on the Arduino IDE board, which means that the input from the probe will be sent to A1. The Arduino IDE programming application will tell the value of the input from A1 via the Serial Monitor (a part of the interface of the Arduino IDE app that shows certain information or data depending on the function of the program).
The code used came from a page from DFRobot that provided instructions on how to use the SEN0237 DO sensor [37]. The sensor had to be first calibrated before being ready for use. To calibrate, the voltage and temperature of the water needed to be measured on two different points.
Once the calibration was done, the program was told what temperature the water was and it would receive the input from the sensor (which is constantly updated). The input is then put into the formula to find the DO level in mg/L which will be updated via the serial monitor.
The DO sensor system is then combined with the water pumping system. Below is the setup of the combined system:
Figure 14. Full setup
The probe is inserted into the water. That data will be transported through the signal converter board into the Arduino IDE Board, which will printed in the serial monitor.
A test is conducted to verify if the water pumping system works by checking if the DO level increases before and after one pump. For this, multiple trials were tested. For each trial, the DO level is measured once before the pumping. The water motor pumping system is then activated before a second reading is measured at the end. Each trial is done every three hours apart from 6:00 a.m. to 6:00 p.m., and the pumping system runs for ten minutes each trial. Between each trial, the DO level decreases as the water is left there and is slowly contaminated by the air.
Figure 15. Bar graph
As seen from the graph above, the data is consistent in how all trials show the DO level increases from before to after the pumping is activated. This indicates that the pumping system is adequate and successful in providing oxygen to the water.
Another experiment was conducted in which the pumping system was activated five times, each time for 20 seconds more than the last time, starting from 20 seconds to 100 seconds. The final and initial DO levels were measured within each period during each trial to get the results below:
Figure 16. Line graph
This further solidifies that the experiment works as the DO level changes between each trial either remains the same or increases, which proves that the pumping system does work (given a certain amount of time).
However, in a home lab setting, the DO level typically should not go anywhere below 8 mg/L. Therefore, the DO sensor is not able detect when the DO level is below or equal to 6.5 mg/L (where water conditions become dangerous) in this condition. However, one can still check if the system works by changing the threshold value (6.5 mg/L) to something higher that is possible to reach in a standard household setting. In this case, 10 mg/L was used as the threshold. Thus, if the motor is activated when a DO value of 10 mg/L or below is detected, then the code works. If this code works, then it would also work in an actual farmland setting.
The probe was once again inserted into the water. The moment the sensor read 10 mg/L, the pumping system activated. Therefore, it is confirmed that the dissolved oxygen sensor does successfully work for starting the water pumping system under or equal to the threshold of 6.5 mg/L and stopping it over or equal to 8 mg/L. Additionally, the fact that the DO level increased from before to after the water pumping shows that the whole process does indeed work.
7. Conclusion
The project aims to resolve the issue of methane generation from rice fields. Ultimately, the idea was to add oxygen to the water before the methane was generated. To achieve that, a water pumping system transported water through a flow meter and an air pump motor pumped air into the water. To conserve energy, the pumping system only activates under two conditions – 1) when a plant in the paddy field is identified to be in the reproductive phase (where it produces the most methane) by an AI machine learning model and 2) when the dissolved oxygen level is detected to be lower than or equal to 6.5 mg/L. The water pumping system, AI model, and dissolved oxygen sensor have all proven to work effectively. Therefore, it is confirmed that this project is effective in resolving the issue of methane generation in paddy fields. For future consideration, the project aims to be made into a product in the market and integrate with renewable energy such as solar energy to become even more environmentally friendly.
Work Cited
[1] E. Yuan, “Ministry of Agriculture, Executive Yuan, R.O.C.(Taiwan),” Moa.gov.tw, 2024.
[2] “Rice consumption by country 2023/24 | Statista,” Statista, 2023.
[3] L. Estelle et al., “Characterization of rice farming systems, production constraints and determinants of adoption of improved varieties by smallholder farmers of the Republic of Benin,” Scientific reports, vol. 12, no. 1, Mar. 2022.
[4] T. Molla, K. Tesfaye, Firew Mekbib, T. Tana, and Tilahun Taddesse, “Farmers’ perspectives on drivers of rice yield in the Fogera Plain of Ethiopia,” Heliyon, vol. 8, no. 12, pp. e12021–e12021, Dec. 2022.
[5] “How Rice Grows,” Think Rice, 2020.
[6] N. ALBAKRI, “QuickCheck: Do rice plants have to be grown in water?,” The Star, Sep. 12, 2023.
[7] “Flooded rice systems: Emissions & mitigation strategies» AgLED,” AgLEDx Resource Platform, May 05, 2023.
[8] “Soil Salinization - ESDAC - European Commission,” Europa.eu, 2018.
[9] J.-R. Park, Y.-H. Jang, E.-G. Kim, G.-S. Lee, and K.-M. Kim, “Nitrogen Fertilization Causes Changes in Agricultural Characteristics and Gas Emissions in Rice Field,” Sustainability, vol. 15, no. 4, pp. 3336–3336, Feb. 2023.
[10] “Greening the rice we eat,” World Bank Blogs, 2022.
[11] Asian Development Bank, “How Can We Incentivize Reducing Methane Emission in Rice Farming in Asia?,” Asian Development Bank, 2023.
[12] UCAR, “Some Greenhouse Gases Are Stronger than Others | Center for Science Education,” Ucar.edu, 2014.
[13] Silva, Marina Leite Fonseca, A. Lima, Faustino, and Flávio Aparecido Rodrigues, “Environmental Impacts of Rice Cultivation,” American journal of plant sciences, vol. 06, no. 12, pp. 2009–2018, Jan. 2015.
[14] UN Environmental Programme, “What’s the deal with methane?,” UNEP, 2022.
[15] “Methane | Vital Signs – Climate Change: Vital Signs of the Planet,” Climate Change: Vital Signs of the Planet, Apr. 11, 2024.
[16] “Eliminating Methane Emissions in Paddy Farms: A Sustainable | Course Hero,” Coursehero.com, 2024.
[17] “Soil Management,” Hawaii.edu, 2024.
[18] Kazunori Minamikawa, N. Sakai, and K. Yagi, “Methane Emission from Paddy Fields and its Mitigation Options on a Field Scale,” Microbes and environments, vol. 21, no. 3, pp. 135–147, Jan. 2006.
[19] Muhamad Komarudin, Hery Dian Septama, and Titin Yulianti, “Autonomous Cyber Physical Systems for Monitoring of Methane Gas in Rice Field,” Advances in biological sciences research/Advances in Biological Sciences Research, Jan. 2022.
[20] X. Sun, Y. Yao, and D. Liu, “The behavior and efficiency of methane displaced by CO2 in different coals and experimental conditions,” Journal of natural gas science and engineering, vol. 93, pp. 104032–104032, Sep. 2021.
[21] “Collecting Gases Over Water | Introduction to Chemistry,” Collegesidekick.com, 2024.
[22] “Collection of Gases and Measurement of their Volumes - Mini Chemistry - Free O Level Chemistry Notes & Learn Chemistry From The Chemistry Authority,” Mini Chemistry, Apr. 27, 2016.
[23] P. C. Wilson, “Water Quality Notes: Dissolved Oxygen,” EDIS, vol. 2010, no. 2, Apr. 2010.
[24] “3 Quick Ways to Add Oxygen to Your Pond - GardenSite.co.uk,” GardenSite, 2024.
[25] Apure, “How to add oxygen to a fish tank?” Apure, Jul. 04, 2022.
[26] Kandasamy Senthilraja et al., “Mitigating Methane Emission from the Rice Ecosystem through Organic Amendments,” Agriculture, vol. 13, no. 5, pp. 1037–1037, May 2023.
[27] Government of Northwest Territories, “Dissolved Oxygen (DO).”
[28] Abhishek Singh, “Fig. 3. Life cycle of rice paddy from germination to ripening phase....,” ResearchGate, 2023.
[29] Rodney Tai-Chu Sheng, Y.-H. Huang, P.-C. Chan, Showkat Ahmad Bhat, Y.-C. Wu, and N.-F. Huang, “Rice Growth Stage Classification via RF-Based Machine Learning and Image Processing,” Agriculture, vol. 12, no. 12, pp. 2137–2137, Dec. 2022.
[30] “Louisiana Rice Production Handbook | 41.”
[31] Orient Electric, “How to choose the right Air Cooler - Air Cooler Buying Guide | Orient Electric Blog,” Orientelectric.com, Feb. 15, 2021.
[32] Fuel Dump, “Guide To Choosing The Correct Fuel Flow Meter,” Fuel Dump, Sep. 06, 2016.
[33] “Rice Growth Stages: Crop Development From Seed To Grain,” EOS Data Analytics, Feb. 27, 2024.
[34] “Fig. 2 Paddy growth stages (a transplanting b tillering c grain filling...,” ResearchGate, 2023.
[35] “FAQ,” Teachable Machine, 2024.
[36] “Teachable Machine,” Withgoogle.com, 2024. https://teachablemachine.withgoogle.com/train/image (accessed Jul. 22, 2024).
[37] “Gravity__Analog_Dissolved_Oxygen_Sensor_SKU_SEN0237-DFRobot,” Dfrobot.com, 2024.
[38] “Gravity: 類比式酸鹼值感測模組 Analog pH Sensor / Meter Kit For Arduino V2 – 機器人王國商城,” Robotkingdom.com.tw, 2018.
Database Images
[39] “Fig.8 Phenology stages of rice growth cycle.,” ResearchGate, 2021.
[40] “Fig. 2 Paddy growth stages (a transplanting b tillering c grain filling...,” ResearchGate, 2023.
[41] Rodney Tai-Chu Sheng, Y.-H. Huang, P.-C. Chan, Showkat Ahmad Bhat, Y.-C. Wu, and N.-F. Huang, “Rice Growth Stage Classification via RF-Based Machine Learning and Image Processing,” Agriculture, vol. 12, no. 12, pp. 2137–2137, Dec. 2022.
[42] A. Nazir et al., “Estimation and Forecasting of Rice Yield Using Phenology-Based Algorithm and Linear Regression Model on Sentinel-II Satellite Data,” Agriculture, vol. 11, no. 10, p. 1026, Oct. 2021.
[43] H. Yang, B. Pan, N. Li, W. Wang, J. Zhang, and X. Zhang, “A systematic method for spatio-temporal phenology estimation of paddy rice using time series Sentinel-1 images,” Remote Sensing of Environment, vol. 259, pp. 112394–112394, Jun. 2021.
[44] Space4Good, “Uncovering the grain of truth,” Space4Good, Jun. 27, 2022.
[45] Fadhlullah Ramadhani, Reddy Pullanagari, Gabor Kereszturi, and J. Procter, “Automatic Mapping of Rice Growth Stages Using the Integration of SENTINEL-2, MOD13Q1, and SENTINEL-1,” Remote Sensing, vol. 12, no. 21, pp. 3613–3613, Nov. 2020.
[46] R. Zhao, Y. Li, and M. Ma, “Mapping Paddy Rice with Satellite Remote Sensing: A Review,” Sustainability, vol. 13, no. 2, pp. 503–503, Jan. 2021.
[47] “Figure 2.3: Rice phenological stages.,” ResearchGate, 2019.
[48] X. Guo, J. Yin, K. Li, and J. Yang, “Fine Classification of Rice Paddy Based on RHSI-DT Method Using Multi-Temporal Compact Polarimetric SAR Data,” Remote Sensing, vol. 13, no. 24, pp. 5060–5060, Dec. 2021.
[49] B. Wang, Y. Liu, Q. Sheng, J. Li, J. Tao, and Z. Yan, “Rice Phenology Retrieval Based on Growth Curve Simulation and Multi-Temporal Sentinel-1 Data,” Sustainability, vol. 14, no. 13, pp. 8009–8009, Jun. 2022.
[50] “Photos depicting the general appearance of rice crop at different rice...,” ResearchGate, 2019.
[51] “Fig. 2. Rice growth stages list.,” ResearchGate, 2022.
[52] dhz2014, “Mature rice paddy,” 123RF, 2014.
[53] “The examples of rice conditions in the field campaign.,” ResearchGate, 2021.
[54] “Illustration of the rice fields under study at some phenological...,” ResearchGate, 2021.
[55] Growth Stages of Paddy Plant, National Council of Education Research & Training.
Programs
[56] “Paddy growth stage detection” P5js.org, 2024. https://editor.p5js.org/andersonchung1031/sketches/7PGJLyxSe
[57] “Teachable Machine,” Withgoogle.com, 2024. https://teachablemachine.withgoogle.com/models/9duh3FLam/
[58] Arduino IDE code for motor activation based on crop growth stage identification. https://justpaste.it/3aylm
[59] Arduino IDE code for motor activation based on dissolved oxygen level detection. https://justpaste.it/g0b6v
Comments