Kalman Filter-Enhanced Data Aggregation in LoRaWAN-Based IoT Framework for Aquaculture Monitoring in Sargassum sp. Cultivation

Abstract

This study presents a LoRaWAN-based IoT framework for robust data aggregation in Sargassum sp. cultivation, integrating multi-sensor monitoring and Kalman filter-based data enhancement. The system employs water quality sensors—including temperature, salinity, light intensity, dissolved oxygen, total dissolved solids, and pH—deployed in 6 out of 14 cultivation containers. Sensor data are transmitted via LoRaWAN to The Things Network (TTN) and processed through an MQTT-based pipeline in Node-RED before visualization in ThingSpeak. The Kalman filter is applied to improve data accuracy and detect faulty sensor readings, ensuring reliable aggregation of environmental parameters. Experimental results demonstrate that this approach effectively maintains optimal cultivation conditions, reducing ecological risks such as eutrophication and improving Sargassum sp. growth monitoring. Findings indicate that balanced light intensity plays a crucial role in photosynthesis, with optimally exposed containers exhibiting the highest survival rates and biomass. However, nutrient supplementation showed limited impact due to uneven distribution, highlighting the need for improved delivery systems. By combining real-time monitoring with advanced data processing, this framework enhances decision-making in sustainable aquaculture, demonstrating the potential of LoRaWAN and Kalman filter-based methodologies for environmental monitoring and resource management.

1. Introduction

The cultivation of sea algae, particularly Sargassum sp., has gained increasing attention in recent years due to its potential as a renewable resource with high economic value. Sargassum sp. can be utilized in various industrial sectors, including food allergy [1], food packaging [2], ruminant feed [3], biochar [4], as well as bioenergy sources like biogas [5,6,7] and bio-oil [8,9,10,11]. As a strategic marine commodity, Sargassum sp. presents significant economic opportunities while supporting sustainable practices, especially in the context of environmentally friendly alternative energy development.

To enhance the productivity and quality of Sargassum sp. as a source of bioenergy and other industrial products, it is essential to develop efficient, data-driven cultivation approaches. A deep understanding of the environmental factors affecting Sargassum sp. growth will yield significant benefits on both local and global scales. This not only supports sustainable marine resource management but also contributes to achieving sustainable development goals in the sectors of energy and food security.

Large-scale cultivation of Sargassum sp. is crucial to meet renewable energy needs in an environmentally friendly way. With high carbohydrate content, Sargassum sp. has great potential for conversion into bioenergy, such as bioethanol or biogas, which is more sustainable than fossil fuels. Its rapid growth and high biomass yield make it effective for large-scale production without impacting food security or requiring fertile land, and it supports marine health by absorbing excess nutrients.

However, the coastal cultivation of Sargassum sp. faces multiple challenges, primarily due to land-use conflicts with tourism and fisheries, as well as contamination from mining and agricultural activities, which can introduce harmful pathogens to the algae. Moreover, maintaining optimal environmental conditions is essential for ensuring healthy Sargassum sp. growth. This macroalgae is highly sensitive to water quality and nutrient availability, with factors such as light, temperature, salinity, and nutrients playing crucial roles in its development [12]. Fluctuations in these parameters, particularly due to global warming, declining water quality, and lack of supportive technology, can hinder productivity and reduce yield quality, affecting both its economic value and sustainability. Consequently, large-scale Sargassum sp. cultivation requires a sustainable approach, appropriate zoning, infrastructure support, and accurate environmental monitoring.

Research on macro-algae cultivation has been conducted in various countries with diverse approaches and objectives. For instance, in China, Gracilaria cultivation expanded significantly from 0.13 ha in 2000 to 1500 ha in 2011, contributing to eutrophication mitigation, algal bloom control, mariculture enhancement, and carbon dioxide (CO₂) absorption. Despite the positive impacts, challenges remain in large-scale management, such as technology efficiency, human resource management, and ecosystem impacts due to rapid expansion [12]. Similarly, in Southeast China, algae cultivation activities increased the total organic carbon (TOC) contribution from 23% (1963–1990) to 53% (1990–2022), highlighting its potential for climate change mitigation, although a long-term ecological impact analysis is still needed [13].

In Norway’s coastal areas, research identified key challenges in spatial planning for macroalgae cultivation, including potential conflicts with aquaculture, uncertainty regarding site suitability, and a lack of local knowledge. That study emphasized the importance of more inclusive policies for sustainability [14]. In Bangladesh, the cultivation methods of Hypnea musciformis were compared between off-bottom net and off-bottom long-line systems, with the former producing twice the biomass. Environmental factors such as temperature, turbidity, and total suspended solids (TSSs) negatively impacted results, while salinity and nutrient concentrations had a positive influence [15].

Although innovative cultivation systems have been explored globally, including in Maine Bay, USA, where an offshore system successfully supported the growth of Saccharina latissima under extreme conditions, there remains a need for technological advancements in attachment mechanisms and tension control for large-scale efficiency [16]. In Cornwall, UK, a seaweed–shellfish co-cultivation system was found to be ineffective in reducing dissolved nutrient concentrations due to its small scale and high seasonal variability [17].

On a global scale, seaweed cultivation not only absorbs (CO₂) but also produces oxygen, with carbon sequestration contributing economic benefits of up to $70.36 billion. However, a comprehensive life-cycle assessment is required to ensure long-term sustainability [18]. In the European Union, an integrative model suggests that the Atlantic region has the potential to produce up to 5 million tons of macroalgae per year. The Integrated Multi-Trophic Aquaculture (IMTA) approach has been proposed, although limitations in long-term environmental impact data and infrastructure challenges remain [19].

Despite these advances, challenges persist in large-scale management, technology, and ecological impacts. Land-based cultivation of Sargassum sp. offers a promising alternative, with vertical photobioreactors (V-PBR) achieving high biomass productivity. However, industrial-scale applications and long-term economic efficiency still require further development [20].

In light of the limitations observed in existing Sargassum sp. cultivation systems—such as inefficiencies in environmental monitoring, limited responsiveness to fluctuating marine conditions, and the absence of scalable data-driven management frameworks—technological innovation becomes imperative. The application of a LoRaWAN-based IoT framework and real-time data processing has shown promise in various aquaculture contexts, offering improved precision, automation, and scalability. Leveraging such advancements may address the need for continuous environmental assessment and adaptive control mechanisms, which are critical for optimizing biomass yield and maintaining ecological balance.

To address these persistent challenges, this study focuses on the integration of advanced monitoring systems using a LoRaWAN-based IoT framework for large-scale data aggregation. By incorporating environmental parameters such as temperature, light, salinity, pH, dissolved oxygen, and nutrients, and simulating wave effects and air bubble treatments, we aim to optimize cultivation conditions. Additionally, a Kalman filter is employed to improve the accuracy and reliability of sensor data by detecting and correcting erroneous readings. This combination of advanced sensor networks and real-time data correction techniques is expected to significantly enhance Sargassum sp. cultivation practices.

This research contributes a novel approach to Sargassum sp. cultivation by integrating a LoRaWAN-based IoT framework and Kalman filter for robust environmental monitoring and data-driven management. The results of this study will not only advance macroalgae cultivation techniques but also offer a scalable solution for bio-energy production in sustainable marine resource management.

2. Materials and Methods

2.1. LoRaWAN-Based IoT Framework Architecture

The LoRaWAN-based IoT framework architecture proposed in this study was specifically designed for the Sargassum sp. cultivation system, aimed at enhancing the monitoring and management of environmental conditions critical for its growth. This framework focused on being easy to implement, highly reliable, and energy-efficient, making it ideal for coastal areas, particularly those with limited access to electricity and the Internet. The framework integrated a range of components that worked cohesively to enable seamless data collection, transmission, and analysis, supporting sustainable cultivation practices.

The framework architecture consisted of four layers, each with specific roles in ensuring the effective operation of the system, as illustrated in Figure 1. These were as follows:

• Perception layer: low-power microcontroller and sensors, responsible for collecting environmental data such as water quality and nutrient levels.
• Network layer: LoRaWAN, gateway, and GSM-based Internet, ensuring long-range, energy-efficient data transmission from the perception layer to the cloud.
• Service layer: network server, MQTT subscriber, and data aggregation, which manage data communication, secure storage, and real-time data analysis.
• Application layer: smartphone, computer, tablet, providing the user interface for monitoring and decision-making based on real-time data analysis.

In the perception layer, the system starts with end devices, which are equipped with sensors and low-power microcontrollers designed to monitor key environmental parameters, including temperature, light intensity, salinity, pH, dissolved oxygen, and nutrient levels (using a TDS sensor). In the network layer, these sensors continuously collect data, which are then transmitted using a LoRaWAN communication device. The LoRaWAN technology, optimized for long-range and low-energy operations, ensures the efficient transmission of data from remote coastal locations. A LoRaWAN gateway serves as a bridge between the end devices and cloud infrastructure, aggregating the collected data and forwarding them to the cloud, where they are securely stored and processed.

To handle the LoRaWAN protocol, in the service layer, the framework utilizes The Things Network (TTN), which provides backend services for device management and secure data transmission. Data are communicated between devices and cloud services using the MQTT protocol, which is lightweight and energy-efficient, making it well suited for use in remote areas with limited power resources. Real-time data aggregation and visualization are achieved through ThingSpeak, a platform that enables the monitoring of environmental variables, real-time data analysis, and the generation of actionable insights for optimal Sargassum sp. cultivation.

In addition, the framework uses a Kalman filter to enhance the accuracy and reliability of the monitored data by reducing sensor errors and ensuring precise readings. This is particularly important in remote areas where sensor readings can be influenced by various environmental factors. The Kalman filter is integrated into the data processing to detect and correct anomalies, providing more accurate data for decision-making.

At the application layer, the processed environmental data are presented to users through an intuitive and interactive interface accessible via smartphones, computers, or tablets. This layer plays a crucial role in decision-making by providing real-time visualizations, historical trends, and alerts based on predefined threshold values. Through a web-based dashboard and mobile applications, farmers can monitor critical water quality parameters, receive notifications when environmental conditions exceed optimal ranges, and take timely actions to maintain a healthy Sargassum sp. cultivation environment.

2.1.1. Perception Layer: Low-Power Microcontroller and Sensors

The main components of the end device in this system consisted of the STM32L072 microcontroller, LoRa communication module, and various sensors for monitoring water quality and nutrients. The STM32L072 microcontroller was selected due to its extremely low energy consumption, which is a critical requirement for IoT applications designed for long-term operation in field environments. This microcontroller is designed for low-power applications, utilizing ARM Cortex-M0+ technology that ensures sufficient performance while maintaining high energy efficiency.

For data communication, the LoRa (Long Range) module was used, offering extensive transmission coverage of up to several kilometers, even in areas with minimal signal interference, such as coastal regions. This technology is also energy-efficient, making it an ideal solution for an IoT system for Sargassum sp. that requires periodic data transmission without frequent power source replacements. The combination of STM32L072 and LoRa enabled the design of a reliable and efficient device capable of supporting real-time monitoring of critical parameters for Sargassum sp. growth, such as water quality and nutrient levels.

The sensors used in this system were designed to detect key environmental parameters crucial for Sargassum sp. growth. For water quality monitoring, sensors for temperature, light intensity, salinity, pH, and dissolved oxygen (DO) were utilized, collectively providing information about the aquatic environmental conditions. These sensors helped identify physical and chemical changes that could impact the health of Sargassum sp. Meanwhile, for monitoring nutrient levels in water, a total dissolved solids (TDS) sensor was employed. The TDS sensor measured the concentration of dissolved particles in the water, which was relevant for assessing the availability of essential nutrients for Sargassum sp. growth. The combination of these sensors ensured comprehensive and accurate data for managing the IoT-based cultivation system.

The end device and sensor components are illustrated in Figure 2, specifically in the section marked with a red circle labeled as point 1, which highlights the location and functions of these components within the system architecture.

2.1.2. Network Layer: LoRaWAN, Gateway, and GSM-Based Internet

LoRaWAN (Long Range Wide Area Network) served as the primary communication protocol for connecting IoT devices in the monitoring of water quality and nutrient levels in Sargassum sp. cultivation. This protocol ensures efficient and secure data transmission through a multi-layered architecture. At the lowest level, the PHY layer manages wireless communication, while the LoRaWAN layer handles frequency management and device addressing. At the highest level, the customer application layer processes and prepares data for transmission to the gateway [21].

To accommodate different operational requirements, LoRaWAN supports three communication classes: Class A, Class B, and Class C. Class A, which was used in this study, offers the highest energy efficiency by limiting data reception to specific receive windows following uplink transmissions. This feature is particularly beneficial for remote deployments where low power consumption is crucial. In contrast, Class B allows scheduled data reception, and Class C enables continuous data reception at the cost of increased energy consumption. Additionally, LoRaWAN ensures data security through robust encryption mechanisms, safeguarding information integrity and confidentiality during transmission [22].

The Things Network (TTN) provides comprehensive support for LoRaWAN versions 1.0.0 through 1.1.0, offering improved security and interoperability [23]. TTN is powered by The Things Stack, which enhances scalability, enables firmware updates over-the-air (FOTA), and facilitates seamless network peering. LoRaWAN devices within TTN can be activated using either Over-the-Air Activation (OTAA) or Activation by Personalization (ABP). OTAA, which dynamically generates session keys for each connection, is preferred in this study due to its enhanced security. In contrast, ABP, while simpler to implement, is less secure as it uses pre-configured session keys [24].

To ensure effective data transmission, the LoRaWAN Gateway acts as an intermediary between end devices and the network server. It receives signals from LoRa devices, coordinates channel access, and forwards messages via the IP Stack. This architecture guarantees stable and efficient real-time communication for monitoring water quality and nutrients in Sargassum sp. cultivation [25].

Given the remote nature of Sargassum sp. cultivation sites, reliable connectivity was achieved by integrating LoRaWAN with GSM-based Internet. A GSM modem was employed to transmit data from the gateway to the network server via cellular networks, enabling real-time data accessibility even in areas with limited terrestrial network infrastructure. The combination of LoRaWAN’s long-range, low-power capabilities and GSM-based Internet connectivity ensured optimized IoT communication, balancing energy efficiency, data security, and broad network accessibility.

2.1.3. Service Layer: Network Server, MQTT Client, and Data Aggregation

In this study, The Things Network (TTN) served as the MQTT broker, while Node-RED, deployed on AWS, acted as an MQTT client (subscriber) that received data from TTN using the MQTT in node. As illustrated in Figure 2, marked with a red circle labeled 5, Node-RED facilitated real-time processing and dynamic control of data transmission intervals before forwarding data to ThingSpeak. This approach eliminated the need for microcontroller modifications when adjusting transmission rates, enhancing system flexibility in response to varying network conditions. Additionally, Node-RED enabled anomaly detection by identifying irregularities such as persistently constant values or sudden deviations, utilizing its flow-based programming interface.

The MQTT-based data transmission process can be mathematically represented as follows. Let $d_i\left(t\right)$ be the raw sensor data from sensor i at time t, let $w_i\left(t\right)$ represent the noise associated with the measurement. The transmitted data $z_i\left(t\right)$ received by ThingSpeak through the MQTT broker is given by:

$$z_i\left(t\right)=d_i\left(t\right)+w_i\left(t\right)$$

(1)

where $w_i\left(t\right)\sim N(0,{\sigma }^2)$ represents noise that follows a normal distribution with mean zero and variance ${\sigma }^2$. The frequency of data transmission is dynamically adjusted using a control function $f\left(T\right)$, which depends on network conditions and system requirements:

$$T_{new}=f(T_{prev},\Delta t)$$

(2)

where $T_{new}$ is the updated transmission interval, $T_{prev}$ is the previous interval, and $\Delta t$ is the adjustment factor based on network conditions.

For data aggregation and analysis, ThingSpeak functioned as an MQTT subscriber and received processed data from Node-RED for storage and visualization, as shown in Figure 2, marked with a red circle labeled 6. ThingSpeak provided real-time monitoring capabilities for water quality and nutrient levels, crucial for Sargassum sp. cultivation. Furthermore, ThingSpeak integrated built-in analytics tools such as MATLAB, enabling advanced data processing, trend analysis, anomaly detection, and predictive modeling.

The aggregation process in ThingSpeak involved combining multiple sensor readings using weighted values to enhance data accuracy and reliability. The aggregated data $X\left(t\right)$ at time t was computed as:

$$X\left(t\right)=\sum _{i=1}^N{w}_i{z}_i\left(t\right)$$

(3)

where $w_i$ represents the weight assigned to sensor i, ensuring that more reliable sensors contribute more significantly to the final aggregated value. The weighting factor $w_i$ is determined based on sensor reliability, historical accuracy, and environmental conditions.

To detect anomalies in the aggregated data, the system evaluated the residual error $r\left(t\right)$, defined as the deviation of the predicted value $\widehat{X}\left(t\right)$ from the actual aggregated measurement:

$$r\left(t\right)=X\left(t\right)-\widehat{X}\left(t\right)$$

(4)

If $\left|r\right(t\left)\right|$ exceeded a predefined threshold $\tau$, an anomaly was detected:

$$\left|r\right(t\left)\right|>\tau \Rightarrow \mathrm{anomaly}\mathrm{detected}$$

(5)

By leveraging the combination of Node-RED, AWS, and ThingSpeak, this IoT framework ensured efficient data flow management and robust data aggregation. Node-RED dynamically adjusted data transmission rates, while ThingSpeak provided a scalable platform for storing, analyzing, and visualizing sensor data. This architecture is particularly well suited for remote and resource-limited environments, such as Sargassum sp. cultivation, where adaptive data control and continuous monitoring are essential. Figure 3 illustrates the MQTT configuration in Node-RED.

2.2. Enhanced Water Quality Monitoring Through Kalman Filter-Based Fault Detection

In this study, the Kalman filter was applied to process and refine sensor data collected from various environmental sensors, such as TDS (total dissolved solids), DO (dissolved oxygen), temperature, salinity, pH, and light sensors. These sensor systems often generate data that are contaminated with noise, either from external disturbances or from uncertainties in the measurement process. By applying a Kalman filter, the uncertainty in the sensor readings can be minimized, and more accurate results are obtained, which can then be used for further analysis [26].

The more accurate estimates of environmental parameters produced by the Kalman filter provide more reliable results compared to the raw data, which are often affected by noise. This allows for more precise and dependable measurements in applications like water quality monitoring or IoT systems.

Thus, the Kalman filter not only improved the accuracy of sensor measurements but also enhanced the overall reliability of the system, enabling the implementation of more effective and efficient environmental monitoring systems.

2.2.1. Water Quality Monitoring System Model

The system model used in water quality monitoring can be represented as a state vector that includes various environmental parameters being monitored, such as temperature, pH, dissolved oxygen (DO), salinity, total dissolved solids (TDS), and light intensity. The state vector at time t is denoted by:

$$D\left(t\right)={\left[\begin{array}{c}T\left(t\right),pH\left(t\right),DO\left(t\right),S\left(t\right),TDS\left(t\right),L\left(t\right)\end{array}\right]}^T$$

(6)

where each element in this vector represents the value of the respective sensor parameter used to measure environmental conditions. These values at time t include the following: $T\left(t\right)$: temperature; $pH\left(t\right)$: pH of water; $DO\left(t\right)$: dissolved oxygen concentration; $S\left(t\right)$: salinity; $TDS\left(t\right)$: total dissolved solids; and $L\left(t\right)$: light intensity. This model assumes that the true value of each measured parameter is part of the state vector, which is continuously updated over time.

In an IoT-based water quality monitoring system, the sensors used to measure various environmental parameters often generate data contaminated by noise and sensor bias. Noise typically arises from external disturbances or uncertainties in the measurement process, while sensor bias results from systematic errors in the sensor device. The sensor bias, denoted as $b_i\left(t\right)$, may occur due to sensor drift or environmental factors, such as improper calibration or persistent environmental disturbances. Therefore, Equation (1) incorporates a sensor bias factor, allowing the sensor reading $z_i\left(t\right)$ of sensor i at time k to be modeled by:

$$z_i\left(k\right)=d_i\left(k\right)+w_i\left(k\right)+b_i\left(k\right)$$

(7)

where $w_i\left(k\right)\sim N(0,Q)$ is measurement noise assumed to be normally distributed with zero mean and covariance Q.

2.2.2. Kalman Filter Model for Data Processing

The Kalman filter was used to estimate the true state of water parameters based on sensor measurements. The process consisted of two main stages [27]:

(a)

Prediction: the prediction state is expressed in the form of a linear state equation as follows [28]:

$${\widehat{x}}_{k|k-1}=A{\widehat{x}}_{k-1|k-1}+B{u}_k$$

(8)

$$P_{k|k-1}=A{P}_{k-1|k-1}{A}^T+Q$$

(9)

where ${\widehat{x}}_{k|k-1}$ is the predicted state of the system at time k based on the previous state, $P_{k|k-1}$ is the predicted error covariance matrix, A is the state transition matrix, B is the control matrix, with control input $u_k$ (if available), and Q is the process noise covariance matrix.

(b)

Correction (updating estimates with sensor data):

after sensor data are received, the estimate is updated using the following equations [28]:

$$K_k=P_{k|k-1}{H}^T{(H{P}_{k|k-1}{H}^T+R)}^{-1}$$

(10)

$${\widehat{x}}_{k|k}={\widehat{x}}_{k|k-1}+K_k(z_k-H{\widehat{x}}_{k|k-1})$$

(11)

$$P_{k|k}=(I-K_{k}H)P_{k|k-1}$$

(12)

where $K_k$ is the Kalman gain, which determines how much the new measurement influences the estimate, H is a measurement matrix that maps the system state to sensor measurements, R is the measurement noise covariance matrix, and I is the identity matrix of appropriate dimensions.

2.2.3. Sensor Fault Detection

The fault detection mechanism employed in this study was based on the residual analysis from the Kalman filter, where the residual was defined as the difference between the actual sensor measurement and the predicted value. A fault was indicated when the residual exceeded a statistically defined threshold. The threshold $\tau$ was set to $2\sqrt{R}$, where R is the measurement noise covariance estimated from historical sensor data collected under controlled, fault-free conditions. This threshold corresponded to a 95% confidence interval, assuming normally distributed residuals. By using this statistical approach, the system could effectively distinguish between normal fluctuations and significant deviations that may indicate sensor faults.

To control the false alarm rate, a temporal validation strategy was applied: a fault was only confirmed if the residual exceeded the threshold over three consecutive time steps. This persistence check helped prevent transient noise or short-lived anomalies from being misclassified as faults. The combination of data-driven thresholding and multi-step validation provided a simple yet robust solution for real-time fault detection in multi-sensor environments, ensuring both sensitivity and reliability in the presence of environmental variability.

To detect faults in the sensor, a residual analysis between the prediction and the received measurements was performed. This residual was calculated as follows [29]:

$$r_k=z_k-H{\widehat{x}}_{k|k-1}$$

(13)

If the residual $r_k$ exceeded a certain threshold $\tau$, then the sensor reading was considered invalid, indicating a fault in the sensor:

$$\parallel r_k\parallel >\tau \Rightarrow \mathrm{sensor}\mathrm{fault}\mathrm{detected}$$

(14)

The threshold $\tau$ can be determined based on the residual statistics from historical data.

2.2.4. Parameter Range Calculation (Min–Max Range)

After the data were corrected using the Kalman Filter, the system calculated the range of parameters for each sensor. This range was calculated as:

$$\mathrm{Range}\left(x\right)=\max \left(x\right)-\min \left(x\right)$$

(15)

If the range of a parameter exceeded the defined limits based on water quality standards of the Indonesian National Standard (SNI, Standar Nasional Indonesia), the system issued a warning to notify the operator that the parameter was outside the normal range.

2.2.5. Applying Kalman Filter for Data Processing in ThingSpeak

After applying the Kalman Filter and analyzing the data, the corrected results were sent to the ThingSpeak platform for further analysis and visualization, as described in Algorithm 1. The system operated as follows: the sensor transmitted data to ThingSpeak, where MATLAB retrieved and processed it using the Kalman filter. Fault detection was then conducted based on the residual analysis and parameter range checks. Finally, the corrected and analyzed results were uploaded back to ThingSpeak for continuous monitoring and visualization.

Algorithm 1 Sensor data analysis with Kalman filter-based fault detection.

Initialization: API keys and channels Set readChannelID, readAPIKey, writeChannelID, writeAPIKey Time range for data retrieval Set startDate, endDate Kalman filter parameters Set $A\leftarrow 1$, $H\leftarrow 1$, $Q\leftarrow 1e-5$, $R\leftarrow 0.01$ Input: Retrieve data, timeStamps from ThingSpeak Output: Display detected faults per sensor Plot raw vs. filtered data for each sensor    1: Define sensorNames    2: Set actualNumSensors based on data size    3: for each sensor i in actualNumSensors do    4:     Extract sensorData    5:     $x_{est}$$\leftarrow firstvalidvalue$    6:     $P\leftarrow 1$    7:     for each data point k do    8:         if $sensorData\left(k\right)$ is valid then    9:            $x_{pred}\leftarrow A·x_{est}$   10:            $P_{pred}\leftarrow A·P·A^T+Q$   11:            $K\leftarrow P_{pred}·H^T/(H·P_{pred}·H^T+R)$   12:            $x_{est}\leftarrow x_{pred}+K·(sensorData\left(k\right)-H·x_{pred})$   13:            $P\leftarrow (1-K·H)·P_{pred}$   14:         end if   15:         Store $x_{est}$ in kalmanData   16:     end for   17:     Save kalmanData to estimatedValues   18: end for   19: Set threshold $\leftarrow 2\times \sqrt{R}$   20: Detect faults where $|\mathit{data}-\mathit{estimatedValues}|>\mathit{threshold}$   21: Display detected faults per sensor   22: Plot raw vs. filtered data for each sensor   23: for each row in estimatedValues do   24:     Upload to ThingSpeak   25: end for

2.3. Air-Bubble-Based Sargassum sp. Cultivation

The cultivation of Sargassum sp. was carried out in 14 containers with the assistance of air bubbles to promote the natural growth of this algae without additional nutrient supplementation, as shown in Figure 4 and Figure 5, which illustrate the experimental setup. Each container was filled with seawater as a medium that replicates the natural habitat of Sargassum sp. and was equipped with an air bubble system using an aerator. The air bubbles played a vital role in providing oxygen and buoyancy, helping the algae stay at the water surface to receive optimal sunlight for photosynthesis. The containers were placed in areas with sufficient sunlight. Although no additional nutrients were provided, regular monitoring was conducted to ensure that water conditions remained supportive and the air bubble system functioned properly, allowing the algae to grow healthily. This method was simpler as it did not require added nutrients and closely mimicked the natural growth conditions of Sargassum sp., making it an efficient and environmentally friendly approach for sustainable Sargassum sp. research and production.

The Sargassum sp. used in this cultivation was a brown macroalga commonly found in tropical waters. Sargassum sp. is classified as follows: Empire Eukaryota, Kingdom Chromista, Phylum Heterokontophyta, Subphylum Ochrophytina, Class Phaeophyceae, Subclass Fucophycidae, Order Fucales, Family Sargassaceae, and Genus Sargassum sp. Morphologically, Sargassum sp. has flattened stems with regular branching on both sides, where the main branches are closely spaced. Its leaves are oval-shaped, serrated, and lack a distinct central vein. Additionally, Sargassum sp. has spherical bladders with rounded or pointed, flattened, and serrated tips. This alga grows in waters with depths ranging from 0.5 to 10 m, in areas with strong currents and waves, and typically attaches to the substrate of the seafloor.

The air-bubble-based cultivation method described earlier aligned with the natural conditions required for Sargassum sp. growth, as detailed in its classification and morphological characteristics. Using seawater as the growth medium and placing the containers in well-lit areas mimicked the natural habitat of Sargassum sp., which thrives in shallow tropical waters with strong currents. The aerator system not only ensured adequate oxygen supply but also helped replicate the buoyancy effect from water movement, which is essential for keeping the algae near the surface for optimal photosynthesis. This approach enhances the ecological suitability and efficiency of the cultivation system, providing a practical method for studying and producing Sargassum sp. in a controlled environment.

2.4. Nutrient-Supplemented Sargassum sp. Cultivation

An experimental setup for cultivating Sargassum sp. is shown in Figure 6, designed to evaluate the impact of nutrient supplementation on algae growth within three vertically arranged containers. This setup ensured the effective distribution of nutrient-enriched seawater among the containers while continuously monitoring water quality parameters. In the bottom container, seawater was enriched with nutrients at specific concentrations and then pumped to the top container. From the top container, the water flowed downward by gravity to the middle container through a pipe equipped with a stop valve, allowing precise control over the water flow rate. A similar process occurred between the middle and bottom containers, creating a structured water circulation system. This water flow setup not only facilitated even nutrient distribution but also simulated wave-like effects that mimicked natural coastal conditions, where macroalgae such as Sargassum sp. are typically cultivated. These wave-like effects are crucial as they enhance gas exchange, nutrient distribution, and mechanical stimulation, all of which contribute to optimal algae growth.

To monitor the environmental parameters affecting the growth of Sargassum sp., five types of water quality sensors were installed in the top container: a light sensor to measure light intensity, a temperature sensor to monitor water temperature, a salinity sensor to evaluate salt concentration, a dissolved oxygen (DO) sensor to assess oxygen levels, and a pH sensor to measure the water’s acidity. Additionally, a total dissolved solids (TDS) sensor was used to measure the concentration of dissolved solids, serving as an indicator of nutrient levels in the water. In the middle and bottom containers, water quality monitoring was limited to TDS sensors, which ensured consistent nutrient distribution throughout the water circulation system. This approach provided a comprehensive assessment of the nutrient distribution system’s effectiveness in maintaining environmental conditions that were optimal for the growth of Sargassum sp.

An ANOVA was performed in the data analysis section of ThingSpeak using MATLAB to evaluate significant differences in the growth parameters of Sargassum sp. between the three containers with different nutrient flows. The ANOVA algorithm presented in Algorithm 2 calculated the F-value and p-value from the received data to determine whether the observed differences were statistically significant. The results of this analysis provided insights into the impact of nutrient distribution on the growth of Sargassum sp.

Algorithm 2 ANOVA for final weight of nutrient-supplemented Sargassum sp. cultivation

1: Input: Final weight data of Sargassum sp. in three containers with circulation system    2: Output: ANOVA results and significance test    3: container1 ← [112], container2 ← [84], container3 ← [47]    4: data ← [container1, container2, container3], group ← [1, 2, 3]    5: grandMean ← mean(data), meanPerGroup ← [mean(container1), mean(container2), mean(container3)]    6: n1, n2, n3 ← length of container1, container2, container3    7: tss ← sum((data - grandMean)2)    8: bss ← n1 * (meanPerGroup[1] - grandMean)2 + n2 * (meanPerGroup[2] - grandMean)2 + n3 * (meanPerGroup[3] - grandMean)2    9: wss ← sum((container1 - meanPerGroup[1])2) + sum((container2 - meanPerGroup[2])2) + sum((container3 - meanPerGroup[3])2)   10: dfBetween ← 2, dfWithin ← n1 + n2 + n3 - 3   11: F ← (bss / dfBetween) / (wss / dfWithin)   12: pValue ← 1 - betainc(dfBetween / (dfBetween + dfWithin), dfBetween / 2, dfWithin / 2)   13: if pValue < 0.05 then   14:     Display “Significant difference between groups”   15: else   16:     Display “No significant difference”   17: end if

3. Results and Discussion

In this study, we developed a Kalman filter-enhanced data fusion system for multi-sensor monitoring within a LoRaWAN-based IoT framework tailored to aquaculture environments. The architecture followed a layered IoT design: the perception layer comprised environmental sensors integrated with the ultra-low-power STM32L072 microcontroller; the network layer utilized LoRa communication via The Things Network (TTN) for reliable, long-range, and low-power data transmission; the service layer performed real-time data aggregation and filtering using a Kalman filter; and the application layer supported data visualization and decision-making through an interactive monitoring dashboard. While communication performance metrics such as packet loss rate and latency are essential in assessing LoRaWAN reliability, these have been extensively studied in the literature under various deployment scenarios [30,31,32] and were not the focus of our work. Instead, our contribution emphasizes the implementation of robust data aggregation and error correction techniques in dynamic marine conditions, which remains under-explored in existing research.

To enhance transparency and reproducibility in system design, we explicitly describe the communication parameters applied in our deployment. We configured the system with a Spreading Factor (SF) of seven, a Bandwidth (BW) of 125 kHz, and a Transmission Interval (T) of 10 min—settings chosen to balance energy efficiency with communication reliability in coastal monitoring contexts. These values are consistent with prior studies: SF7 is preferred for its shorter time-on-air and reduced energy usage [33], a 125 kHz bandwidth provides a compromise between data rate and sensitivity [34,35], and transmission intervals between 5 and 15 min are widely used in environmental IoT systems to preserve battery life while ensuring adequate temporal resolution [36,37]. Given this established context, our study advanced the state of the art by integrating these configurations into a unified aquaculture-specific IoT framework with enhanced data fidelity through Kalman-based processing.

To assess the real-world performance of the proposed framework, we conducted two experimental deployments in different aquaculture scenarios involving the cultivation of Sargassum sp. These practical evaluations were designed to test the system’s robustness in monitoring water quality under varying physical and environmental conditions. The first scenario involved the cultivation of Sargassum sp. in 6 out of 14 cultivation containers, where water quality was monitored in real time throughout a 20-day growth period using the developed IoT system. During that period, key water quality parameters were continuously measured and maintained within the thresholds specified by Indonesia’s National Standard (SNI). This scenario aimed to assess the influence of sunlight exposure on biomass accumulation and survival rate.

Building upon the insights from the first scenario, the second scenario explored a different cultivation method by utilizing three vertically stacked containers. Water was pumped from a nutrient-rich reservoir into the top container, then flowed successively into the middle and bottom containers, simulating natural wave-induced nutrient delivery. This design allowed us to investigate the vertical distribution of nutrients and its effect on plant growth. In this setting, real-time monitoring focused on total dissolved solids (TDS) as a proxy for nutrient levels, while ensuring all water quality metrics remained within national standards. The goal of this scenario was to evaluate how nutrient availability influenced the growth performance of Sargassum sp.

Finally, while our experimental setup involved a limited number of nodes (six containers), we acknowledge the importance of evaluating the system’s scalability in broader deployments. The impact of increasing network size on LoRaWAN performance has been extensively examined in prior research. Bor et al. [38] demonstrated that the packet success rate (PSR) significantly declined as the number of nodes increased, primarily due to packet collisions arising from the ALOHA-based MAC protocol used in LoRaWAN. Similar findings were reported by Adelantado et al. [33] and Georgiou and Raza [39], highlighting the inherent scalability limitations when spreading factors and channel assignments are not optimally managed. Therefore, for future deployments involving 60 to 600 nodes, adopting strategies such as adaptive data rate (ADR), channel planning, and multi-gateway architectures is critical to maintaining network reliability and performance [30,36,40].

3.1. Evaluation of Kalman Filter

The numerical parameters used in the Kalman filter, as presented in

Table 1. Kalman filter parameters.

Parameter	Symbol	Value
Process noise covariance	Q	$1\times {10}^{-5}$
Measurement noise covariance	R	$0.01$
Fault detection threshold	$2\sqrt{R}$	$0.2$

, were defined based on both theoretical considerations and empirical tuning to optimize filter performance. The process noise covariance (Q) was set to a small value to reflect minimal process noise, indicating a high confidence level in the system model. The measurement noise covariance (R) was determined from the statistical variance observed in sensor calibration data, representing the expected noise level of the measurement system. The Kalman gain (K) was derived iteratively during filter implementation, dynamically adjusting the balance between model predictions and sensor observations. The fault detection threshold was defined as $2\sqrt{R}$, which corresponds to a 95% confidence interval for zero-mean, normally distributed residuals. This threshold ensured that deviations beyond this bound were statistically significant, thus indicating potential faults in sensor readings. These parameter definitions were selected to enhance the robustness and reliability of the Kalman filter in the proposed system.

Table 2. Fault detection by the Kalman filter.

Parameters	Number of Faults	Percentage (%)
Temperature °C	148	1.85
pH	127	1.59
Dissolved oxygen (mg/L)	289	3.61
Salinity (ppt)	195	2.43
Light intensity (lux)	119	1.49
Total dissolved solids mg/L	189	2.36

summarizes the fault detection results for each sensor during the 20-day monitoring period. The Kalman filter successfully identified anomalies in the sensor data, such as abrupt spikes, sustained drifts, or missing values [27]. These faults were flagged for further validation, ensuring the reliability of the water quality monitoring system.

Figure 7 illustrates the visualization of sensor data after applying the Kalman filter, displayed on the ThingSpeak platform. The plots highlight the smoothed data trends over the monitoring period, with noisy fluctuations significantly reduced. Each sensor’s data are represented with markers indicating flagged anomalies, facilitating quick identification and interpretation of faults.

To improve the quality of sensor readings and minimize the effect of environmental noise, a classical Kalman Filter (KF) was employed. The KF is known for its effectiveness in real-time estimation when the system dynamics are linear and the noise is Gaussian [40]. Although some sensors—such as TDS—can exhibit nonlinear behavior over wide operational ranges, in our application context (i.e., controlled aquaculture conditions), the TDS, temperature, and pH sensor outputs were observed to follow near-linear patterns, and the noise closely approximated a Gaussian distribution. Similar assumptions have been validated in previous studies applying a KF to water quality and environmental monitoring [26,41].

While parameters such as temperature and dissolved oxygen can exhibit coupling in natural aquatic ecosystems, their variations in the controlled aquaculture environment used in this study were largely independent. This justified the use of a decoupled state-space model, which simplified computation without significantly compromising accuracy [42].

Therefore, the use of a KF was considered justified, offering a balance between computational efficiency and estimation accuracy. Compared to the Extended Kalman Filter (EKF) or Unscented Kalman Filter (UKF), the classical KF is computationally simpler and more suitable for real-time embedded systems with limited processing capacity, such as the STM32L0-series microcontrollers used in our framework. Moreover, the relatively stable and constrained environmental conditions did not warrant the complexity of nonlinear filters [43]. Preliminary simulations using the EKF and UKF on a subset of the dataset did not demonstrate substantial improvement in estimation performance compared to the KF, further supporting our selection of a linear filtering approach.

The application of the Kalman Filter not only improved the clarity and accuracy of real-time data visualization but also enhanced the reliability of the monitoring system by promptly detecting and flagging faulty data points. This ensured that the cultivation environment for Sargassum sp. could be maintained within optimal conditions, ultimately promoting sustainable growth and reducing the risk of adverse ecological impacts such as eutrophication.

3.2. Impact of Air-Bubble Systems on Sargassum sp. Growth

The integration of Kalman filter-based fault detection with range analysis provided a robust framework for monitoring water quality in the cultivation of Sargassum sp. This method ensured not only accurate data acquisition but also the detection and correction of sensor faults, thereby addressing potential reliability concerns.

The analysis results, particularly the computed ranges for critical parameters, are detailed in the “Range” column of

Table 3. Water quality values of the cultivation medium for Sargassum sp.

Parameters	Range	Reference
Temperature °C	26.25–29.97	24–30 (SNI 7904:2013)
pH	6.63–7.85	7–8.5 (SNI 7904:2013)
Dissolved oxygen (mg/L)	4.01–12.06	>5 (SNI 7904:2013)
Salinity (ppt)	30.14–38.72	30–35 (SNI 7904:2013)
Light intensity (lux)	6134–24,968	6000-24,000 (SNI 7904:2013)
Total dissolved solids mg/L	997.89–1037.22	28,000–35,000 (SNI 7904:2013)

. These ranges offer valuable insights into the dynamics of water quality over the monitoring period, highlighting the stability and variability of key factors such as temperature, dissolved oxygen, pH, salinity, TDS, and light intensity. This information is crucial for assessing the effectiveness of the air-bubble systems used to enhance Sargassum sp. growth by improving aeration and nutrient distribution within the cultivation medium.

The consistent performance of the air-bubble systems was reflected in stable water quality parameters, which contributed to optimal growth conditions for Sargassum sp. This highlights the importance of integrating reliable monitoring techniques like the Kalman filter with innovative cultivation methods to promote sustainable aquaculture practices.

3.2.1. Growth Assessment of Sargassum sp.

The study showed that the average final weight of Sargassum sp. cultivated in controlled environments ranged from 12 g to 37 g, as illustrated in Figure 8. The highest final weight was recorded in container 7 (37 g), while the lowest was observed in containers 2, 8, and 10 (12 g). All monitored water quality parameters, namely, sunlight intensity, temperature, salinity, pH, dissolved oxygen, and total dissolved solids, met the SNI 7904:2013 standards for optimal Sargassum sp. growth. This confirmed that variations in final weight were not due to sub-optimal water quality but may have been influenced by other factors, such as environmental conditions or container placement.

Container 7, which achieved the highest final weight, was positioned to receive maximum sunlight exposure, suggesting that light intensity played a significant role in supporting the growth of Sargassum sp. In contrast, containers with lower growth rates likely experienced less sunlight, despite having uniform water quality conditions.

The higher growth in container 7 highlights the importance of light intensity, as it directly influences photosynthesis, a key factor in the growth of Sargassum sp. [44,45,46]. To improve growth rates and achieve uniformity across all containers, it is essential to consider other influential factors, such as nutrient distribution and water movement.

This consideration aligned with the second research scenario, which focused on the cultivation of Sargassum sp. with nutrient supplementation. In that scenario, the use of water flow treatment was explored to enhance nutrient distribution across all containers. Water movement significantly aids in distributing nutrients, resulting in more uniform growth [15]. Integrating controlled water flow can ensure that supplemented nutrients are evenly distributed, reducing the dependency on container positioning or light exposure.

Moreover, in controlled environments, the availability of nutrients naturally decreases over time due to absorption by the plants. The addition of water flow mechanisms can maintain nutrient availability by improving its distribution [46]. Utilizing the IoT framework in this study further enhances this process, as it enables real-time monitoring and management of critical water quality parameters, ensuring optimal growth conditions throughout the cultivation period.

3.2.2. Survival Rate of Sargassum sp.

The survival rate of Sargassum sp. was measured by observing the physical condition of the seaweed, including leaf health, size, growth, and signs of damage or degradation. Signs such as color fading or wounds indicate environmental stress that could affect survival.

The observation results showed that the survival rate of Sargassum sp. cultivated in the laboratory ranged from 24% to 74%, as shown in Figure 9. The highest survival rate was found in container 7 at 74%, while the lowest survival rate was found in containers 2, 8, and 10, each at 24%. Container 7 had the highest survival rate compared to the other containers, while the lowest survival rates were observed in containers 2, 8 and 10.

The survival rate of Sargassum sp. is a crucial indicator of the environmental conditions in each container. In this study, container 7 exhibited the highest survival rate at 74%, suggesting more favorable conditions such as better sunlight exposure, temperature, and possibly water circulation. These factors are essential for the health of Sargassum sp., as they directly influence photosynthesis, nutrient uptake, and stress reduction. On the other hand, the lowest survival rates were observed in containers 2, 8, and 10, with rates of 24%. This was likely due to insufficient sunlight penetration, which hindered the seaweed’s ability to perform photosynthesis. Sunlight is essential for Sargassum sp. to carry out photosynthesis effectively, as macroalgae convert solar energy into usable energy for growth [47].

In addition to light, the lower survival rates in these containers could also be attributed to environmental conditions that were not suitable for the seaweed’s natural habitat. Adjusting to a new environment makes Sargassum sp. more vulnerable to growth challenges, as it requires more energy for adaptation, competing with the energy needed for food reserve production from photosynthesis [48]. Furthermore, both internal and external factors influence growth. External factors, such as light, temperature, water, organic matter, and nutrient availability, are shaped by the environment, while internal factors, such as photosynthesis rate, respiration, and genetic influence, depend on the plant itself. If both factors are optimized, they enhance photosynthesis, leading to the production of photosynthates that support plant growth [49].

3.3. Effects of Nutrient Supplementation on Sargassum sp. Growth

The study revealed that the average final weight of Sargassum sp. in the circulation system varied across container levels, with 112 g in level 1, 84 g in level 2, and 47 g in level 3 as shown in Figure 10. The Analysis of Variance (ANOVA), calculated using MATLAB Analysis Thingspeak with Algorithm 2, indicated that the application of fertilizer through the circulation system did not result in statistically significant differences in the survival rate of Sargassum sp. (p > 0.05). The highest final weight was observed in level 1 containers, likely because the fertilized water from the main container was initially directed into these containers, providing a nutrient concentration higher than the other two containers. In contrast, nutrient concentrations in level 2 and level 3 containers decreased due to distribution effects.

The use of NPK (nitrogen, phosphorus, and potassium) fertilizer at a dose of 0.13 g/L was chosen due to its essential nutrients—nitrogen, phosphorus, and potassium—that are critical for seaweed growth. Nitrogen supports photosynthesis, phosphorus contributes to ATP (adenosine triphosphate) formation, and potassium aids root development while enhancing resistance to stress and disease [50,51]. The fertilizer dose applied in this study resulted in higher final weights compared to previous research, which reported a final weight of 10.13 g with a dose of 2.5 g/L [44]. This finding indicates that the circulation system used may enhance fertilizer efficiency.

However, the results suggest that higher nutrient concentrations or optimized distribution across containers could further improve growth outcomes. Studies on alternative fertilizers, such as urea and liquid fertilizers, have shown their potential to promote higher growth rates compared to TSP (triple superphosphate) fertilizer [52]. Furthermore, inappropriate nitrate and phosphate concentrations could negatively affect the formation of carbohydrates, proteins, and other metabolites crucial for the growth of Sargassum sp. [53,54].

4. Conclusions

This study demonstrated the success of the proposed IoT framework in monitoring water quality and nutrient levels in real time, supporting more efficient management of the cultivation environment for Sargassum sp. The application of the Kalman filter improved data accuracy, detected sensor faults, and maintained optimal conditions for Sargassum sp. growth, thereby supporting sustainable aquaculture practices and reducing ecological risks, such as eutrophication. The air-bubble systems demonstrated stable performance, maintaining water quality parameters at optimal levels.

The growth of Sargassum sp. is significantly influenced by water quality, particularly light intensity and nutrient availability. Balanced light intensity is crucial for successful photosynthesis and growth, although the nutrient distribution system in this study is not yet optimal, affecting the overall growth performance. This study underscores the importance of integrating modern technologies, such as the IoT framework and Kalman filter, with innovative cultivation methods to support sustainable aquaculture practices.

As future work, this study will develop an automated system for controlling water quality and nutrient distribution in Sargassum sp. cultivation. This system will be designed to adjust parameters such as pH, salinity, dissolved oxygen, and nutrient concentrations in real time based on collected data, creating a more controlled and optimal environment. This advancement is expected to improve nutrient distribution efficiency and enhance the overall growth performance of Sargassum sp.