The Impact of Cloud Versus Local Infrastructure on Automatic IoT-Driven Hydroponic Systems

Abstract

Technological advancements in the cloud field are becoming widely used on a large scale in increasing activity sectors. Agriculture is an important domain in everyday life, central to human existence. This research comparatively analyzes two proposed types of infrastructures that optimize the growth flow of plants in a hydroponic system for continuous monitoring, one full-cloud and one full-local. The study’s main objective is to determine which of the two infrastructures is more suitable for the hydroponic scenario by conducting seven types of tests. This research aims to fill a gap in the specialized literature through a detailed analysis of the configuration, implementation methods, and all implications of the two approaches from the perspective of the seven indicators. The seven indicators are response time, operational reliability, implementation costs, operational costs, configuration scalability, data accessibility, and data security. The cloud infrastructure uses Microsoft Azure technologies, while the local variant uses custom-made scripts and locally installed services. For both software infrastructures, the hardware components are identical, including an M5Stack module with sensors for monitoring temperature, humidity, electrical conductivity, and liquid level in the hydroponic container. The test results highlight that the local infrastructure offers a shorter response time (200 ms compared to 300 ms for the cloud infrastructure). The results also showed lower operational costs for the local infrastructure, making it more suitable for autonomous hydroponic systems. On the other hand, the results showed that cloud infrastructure has greater data accessibility than local infrastructure, and the security measures are advanced. These advantages of cloud infrastructure involve recurring costs of USD 82.57/month. The limitations of this research are associated with the exclusion of human errors and cybernetics simulations from the analysis. Another limitation concerns the real analysis of short-term costs. Future research will explore the real fluctuations of long-term costs. Additionally, infrastructure studies on different plant species and hydroponic farms will also be considered.

1. Introduction

Hydroponics is a method by which plants are not cultivated traditionally, that is, in soil, but through an artificial method. The plant roots are in a solution containing a mixture of water, minerals, and other substances necessary for its development. This growing technique is adapted to urban areas or those with limited arable land, and it is also very easy to grow, monitor, and harvest [1,2]. This technique minimizes resource consumption and maximizes the yield. To achieve this goal, the nutrient level must be provided according to specific standards for the cultivated plants to perform photosynthesis without rotting or becoming infected with various pests [3,4].

Hydroponic systems are classified based on their characteristics and operational methods, with the most commonly encountered being those implemented through the Nutrient Film Technique (NFT), Deep Water Culture (DWC), Ebb and Flow (Flood and Drain) Systems, and Drip Systems, as presented in

Table 1. Comparison of the leading hydroponic technologies.

Technology	Operating Principle	Advantages	Disadvantages
NFT	A thin layer of nutrient solution continuously flows over the roots	High water and nutrient efficiency Well-aerated roots	Susceptible to water flow interruptions Requires strict flow control
DWC	Roots are fully submerged in an oxygenated nutrient solution	Rapid plant growth; Simple system to implement	Requires constant oxygenation to prevent root suffocation
Ebb and Flow (Flood and Drain)	Alternates between flooding and draining the substrate	Provides good root oxygenation; Suitable for various plant types	High energy consumption Requires constant monitoring
Drip System	The nutrient solution is delivered directly to the roots via controlled dripping	Efficient for diverse crops; Low water consumption	Risk of clogged pipes; Requires regular maintenance

. NFT uses a thin layer of nutrient solution that flows over the plant roots. This method allows nutrient absorption while minimizing water use [5,6]. DWC is based on submerging the plant roots in a nutrient solution. Oxygen is supplied through air stones or other aeration methods [7,8,9]. Flood and Drain systems periodically flood the growing area with nutrient solution, then drain it, allowing the roots to absorb nutrients while also receiving oxygen during the drainage phase [5,7]. Drip systems deliver the nutrient solution directly to the base of each plant. In this way, it can be adjusted according to the specific needs of the crop [5,6,10].

Table 1 provides an overview of the four hydroponic technologies. This table highlights the advantages and limitations of each, as stated before in the literature.

Thus, by analyzing Table 1, the choice of methodology used in this study, which focuses on integrating hydroponic systems into an automatic IoT (Internet of Things) infrastructure, is justified. The study presented in this paper will utilize a hydroponic system of the Drip type, specifically the Recovery Drip System. This system recirculates the mixture of nutrient solution with water.

Research [3,11] demonstrates that different hydroponic technologies used for the same plant species will result in variations in nutrient uptake. This affects root development and plant health. Additionally, Mei et al. [12] introduce the role of microorganisms in hydroponic systems. The research demonstrates that microorganisms influence plant growth by increasing nutrient absorption.

Traditional hydroponic systems have evolved into improved variants that integrate aquaponics, a method that involves raising fish alongside plants [13,14]. In this symbiotic relationship, fish waste serves as a natural fertilizer for the plants. The plants help filter and purify the water for the fish, creating a closed-loop system that optimizes resource utilization [13,15].

Eliminating soil from the plant growing process reduces the risk of soil-borne diseases and pests. The absence of these leads to a decrease in the need to use chemical pesticides [1,2]. Unlike traditional agriculture, which is seasonal, hydroponics is not dependent on seasonal changes. Hydroponics is a form of controlled and imposed environment agriculture, which allows for a constant supply of products to retail chains [16,17]. Such agriculture is advantageous in regions where climate and agricultural seasons do not favor continuous delivery. Food production in controlled environments, such as greenhouses or indoor farms, is beneficial for urban areas, where demand is high and supply is limited [1,2].

Hydroponics replaces soil usage with water. Since water is a valuable resource, hydroponics recirculates nutrient solutions within a closed system, reducing water consumption [1,2]. Additionally, recycling water and nutrients leads to a reduction in waste [15]. Studies [1,6] have explored automation and remote monitoring technologies to align with modern technological standards. Monitoring environmental conditions, nutrient delivery, and system maintenance are elements studied by automatic systems dedicated to hydroponics.

1.1. Artificial Intelligence in Hydroponic Systems

Recent research has explored different computational methods to improve nutrient management in hydroponic systems. By optimizing the nutrient supply, the system will increase both the growth period and overall development of plants. In hydroponics, the absence of soil makes plants more vulnerable to environmental fluctuations and variations in nutrient levels [18]. As a result, accurately dosing nutrients faces challenges. Traditionally, nutrient mixing in these systems measures electrical conductivity (EC) and pH levels. This approach generates imbalances at the nutrient level, which compromises the hydroponic culture. For example, a study presented in work [19] highlighted that conventional systems cannot adequately address imbalances caused by the lack or excess of specific ions. This results in suboptimal crop growth. To solve this problem, the integration of machine learning (ML) algorithms is recommended [20,21]. The ML algorithms are trained using large datasets to determine nutrient amounts to make appropriate adjustments [22]. These infrastructures use software components and hardware elements, including modern sensors and actuation components [23].

The absence of soil directly results in plants’ heightened sensitivity to the amount of nutrients supplied. Therefore, the nutrient delivery system must be precise and operate continuously within optimal parameters. Based on these considerations, El-Ssawy et al. [24] emphasize the importance of advanced monitoring systems that detect real-time anomalies in environmental or nutrient conditions. In general, AI solutions can integrate predictive components that can anticipate potential causes of anomalies within the system. Furthermore, automatic systems have corrective action components to prevent situations that could compromise the hydroponic culture [25]. To develop such systems, the monitored environmental variables (such as temperature, humidity, and lighting conditions) that can influence nutrient uptake must be identified beforehand [26].

In addition to the monitoring component, automatic systems equipped with AI require specific calibration procedures tailored to the type of hydroculture. Depending on the crop type, the analyzed parameters vary, implying different configurations at nutrient and environmental parameter adjustment levels [4]. Kabir et al. [27] explore the possibility of using deep learning (DL) models to optimize nutrient mixtures based on crop type. The study highlights that DL models require large datasets, which is a disadvantage in practice due to the high variability of environmental conditions [28]. A direct consequence is reflected in the economic implications of adopting AI technologies for hydroponic solutions. Although AI tools can reduce labor costs and improve yields, the initial investment in technology and infrastructure can be prohibitive for many small-scale farmers [29,30]. Economic analyses indicate that the benefits of AI integration must be evaluated in terms of the costs of developing hardware and software [31,32]. In addition to the initial hardware and software development costs, the economic feasibility analysis of using AI in hydroponics must also include long-term operational costs. Cloud-based AI solutions involve monthly subscriptions. For example, for the Microsoft Azure cloud platform, the calculator used to estimate long-term costs can be consulted at the address [33]. Local systems may have lower costs, but they require constant maintenance.

1.2. IoT Sensors and Actuators in Hydroponic Systems

Studies [34,35] have shown the importance of the calibration process in the hydroponic context. These studies have highlighted the need to recalibrate pH and EC sensors periodically. Otherwise, the measurements of these sensors may drift due to temperature fluctuations and deposits on the sensor surface. The calibration process compares the sensor readings with a standard. The process itself involves adjusting the sensor output to this standard. For pH sensors, the process uses buffer solutions with specific pH values [36,37]. Similarly, EC sensors are calibrated using solutions with known conductivities [35]. Failure to follow the calibration protocol generates erroneous readings. The consequence of this attitude affects nutrient management strategies [38].

Integrating IoT technology with sensors has greatly improved nutrient monitoring in hydroponic systems [39]. IoT-based solutions enable remote supervision and control of nutrient mixtures, allowing for real-time adjustments in response to changes in nutrient levels or environmental conditions [40]. These setups typically combine sensors that monitor pH, EC, temperature, and other key parameters, providing a comprehensive view of the nutrient solution’s status [41].

Automation in hydroponics is further enhanced by using actuators that respond to sensor feedback, enabling precise control over nutrient dosing and the recirculation of the solution. Pumps, valves, and dosing systems are included in the category of actuators operating in agriculture. The primary function of actuators in hydroponic systems is the controlled release of the correct amounts of nutrients. Studies [42,43] have shown that hydraulic ram pumps are the most suitable for hydroponic systems due to the way their valves are designed [44].

In the context of automation, the synchronization between sensors and actuators influences the reaction speed of the hydroponic system as a whole [45]. For example, microfluidic systems improve the performance of nutrient dosing through precision, and they are released based on data provided by sensors [46]. These systems can mimic the functions of vascularized tissues, allowing the transport of nutrients to specific locations within a substrate or biological system [47].

Biohybrid actuators combine biological components with synthetic materials, allowing them to respond to environmental stimuli dynamically [48,49,50,51]. Smart hydrogels also represent another research direction regarding the adaptability of nutrient delivery systems in hydroponic systems [52,53].

Additionally, intelligent polymer-based actuators have been researched to automate hydroponic systems. These actuators respond to triggering events, such as the change in pH or temperature [54,55]. Automating nutrient addition based on sensor feedback in hydroponic systems can significantly impact crop yields. For example, integrating nutrient solution sensors with delivery systems enables the precise adjustment of fertilizers to meet the specific requirements of the crops [56].

Several studies have investigated IoT-driven hydroponic systems. Despite all this, there is a notable lack of comparative analysis in the literature between practical cloud-based solutions and their local counterparts. Additionally, there are no studies in the literature that focus exclusively on the performance indicators of each infrastructure, which would enable a relevant comparative analysis. To prevent this gap in the literature, the present research aims to evaluate the two concrete infrastructure proposals through seven performance indicators. This research provides practical insights for agricultural digitalization.

To clarify the key elements of the research, the following Research Questions (RQs) are proposed:

• RQ1: What are the configuration differences specific to each type of hydroponic infrastructure, both at the hardware and software levels, in a cloud versus local approach?
• RQ2: Which infrastructure is superior in a comparison based on performance indicators such as response time, operational reliability, implementation costs, operational costs, scalability, data accessibility, and security?
• RQ3: What implications do the performance differences have for the practical implementation of automatic hydroponic systems?

The evaluation of these criteria provides insights into the trade-offs between cloud-based and local hydroponic infrastructures. The research will guide researchers, engineers, and agricultural practitioners in selecting the most effective solution.

1.3. Paper Contributions and Structure

The main contributions of the paper are as mentioned below:

• Identification of a hardware infrastructure configuration that enables the automation of a hydroponic solution;
• Identification of possible software infrastructures compatible with the required hardware infrastructure;
• Analysis of the possibility of integrating an IoT infrastructure;
• Detailed comparative analysis between cloud infrastructure and local infrastructure for the two reference proposals of the research;
• The research analyzes the impact of using IoT equipment (M5Stack, hydroponic monitoring sensors) in both cloud and local environments;
• Detailed presentation of a method for implementing the two infrastructures;
• Assessment of the impact of response time (RT), operational reliability (OR), implementation costs (IC), operational costs (OC), configuration scalability (CS), data accessibility (DA), and data security (DS) through the two infrastructures;
• The research provides applicable recommendations for farmers who wish to adopt modern and intelligent solutions to improve decision-making processes and achieve agricultural digitalization.

The paper is structured into five sections. Section 2 presents the proposed hardware infrastructure, followed by the two software infrastructures analyzed, the automation scheme for liquid feeding of the hydroponic container, and the presentation of the seven indicators associated with the evaluation metrics of the two software infrastructures. Section 3 includes the results highlighting the comparative analysis of the two infrastructures. The discussion of the results, research limitations, and future works are depicted in Section 4. Finally, Section 5 describes the conclusions of the paper.

2. Materials and Methods

The primary objective of this project is to investigate two infrastructures that optimize the growth flow of plants in a hydroponic system integrated into an IoT configuration for continuous monitoring. A hydroponic system of the Drip System type was used for testing, which ensures the recirculation of the nutrient solution at the level of the plant roots. This type of system is defined as an irrigation system in which the nutrient solution is delivered to the plant roots through drip emitters or drip heads. These are connected to a network of pipes or hoses that distribute the nutrient solution directly to the plants’ roots. This type of system is designed for crops in small spaces. This plant was selected due to its rapid growth, which occurs in 14 days after sowing [57].

2.1. Prototype Infrastructure

The system, as illustrated in Figure 1, comprises a pipe serving as a reservoir for the nutrient solution, a PVC channel for supplying the reservoir with the nutrient solution, a hydraulic ram pump, and a monitoring system equipped with sensors. The plant chosen for conducting the tests was Lactuca sativa L. (green lettuce). For these tests, this type of plant was used due to its rapid growth cycle. Practically, this plant generates experimental results in the shortest time. This work will be followed by further research, including a comparative analysis of the degree of development of various species in a hydroponic system, as well as a study on the growth of a specific class of species in a mixed hydroponic system. The authors declare that at the time of writing this material, they are monitoring and acquiring data for several species.

The element responsible for monitoring the solution is the M5Stack module. It has the following components connected:

• A liquid temperature sensor (DS18B20) for measuring the temperature of the nutrient solution;
• An electrical conductivity sensor (EC-5 Atlas Scientific) for determining the concentration of nutrients;
• A liquid level sensor (SEN18) for monitoring the amount of available nutrient solution;
• An environmental temperature and humidity monitoring sensor (DHT22).

The data are collected at one-hour intervals, as the monitoring process exhibits minimal variability. The data is transmitted to a cloud database stored in Azure, which is made possible by the Wi-Fi element integrated into the M5Stack module.

The novel element introduced in this study focuses on integrating a hydroponic system into an Internet of Things (IoT) infrastructure. This is necessary for Industry 5.0 for the following reasons:

• The growth of the population and the standard of living leads to an increase in food consumption, especially in urban agglomerated areas. In this context, traditional agriculture no longer represents a solution. For this reason, the integration of plant growing systems under non-traditional conditions becomes an acute necessity;
• Hydroponic systems require careful monitoring of nutrients, as they lack soil and are grown in environments that are generally not conducive to plant development;
• This type of plant cultivation requires reduced costs for plant maintenance, human resources, and the raw materials used, such as water, nutrients, and energy consumption. This is necessary because hydroponic agriculture must have lower production costs than traditional agriculture to be viable.

Therefore, Industry 5.0 explicitly directs the integration of hydroponic systems into an IoT infrastructure that allows for the monitoring and automation of the plant-growing process in urban environments [39].

In contrast to conventional agriculture, hydroponic agriculture is based on growing plants without soil. The advantages of this system include water savings, better control over nutrients, the ability to produce in atypical environments, without influence of bad weather conditions, and higher yields per unit of surface area. However, the major disadvantage of this approach lies in the need for constant monitoring of the environment in which these plants are grown. Therefore, the prototype proposed in Figure 1 aims to continuously monitor parameters such as water temperature, EC, solution level in the tank, and ambient temperature and humidity and adjust the nutrient concentration of the solution for the plants. Automating the entire process implies minimal human intervention, increasing production profitability.

2.2. Proposed Software Infrastructure

The automatic IoT hydroponic system proposed in this study has a modular software architecture. The study aims to analyze in detail the type of infrastructure suitable for the specific problem to be solved. Technological advancements sometimes allow for a new approach, but it is not always a viable or optimal solution. In the following, two monitoring methodologies for the hydroponic system will be proposed for analysis: full-cloud or full-local.

2.2.1. AFCI

The first proposed infrastructure studied the Azure platform services to ensure that all the required hydroponic tasks are fully accomplished in the cloud. The IoT-Based Azure Full Cloud Infrastructure (AFCI) solution includes the following services:

• Azure IoT Hub for managing the M5Stack devices and collecting data from sensors;
• Azure Stream Analytics for processing events in real-time and triggering the activation of the pump based on decisions made using the ML model;
• Azure SQL Database stores data that allows the creation of a history. Analyzing the variability of the data enables the development of trends for ML models;
• Azure Machine Learning is used to develop predictive models that optimize the parameters of the nutrient solution.

The software infrastructure, schematically presented in Figure 2, is entirely cloud-based. The study aims to highlight the possibility of building such an infrastructure in the context of technological evolution, which, from a software perspective, seeks to completely change the programming paradigm by software modularity at the cloud level. In Figure 2, the hardware elements are marked with blue, the software elements for analysis and decision-making are green, and the software elements for data are red.

The proposed architectural workflow includes sending data from the sensors to Azure IoT Hub via the M5Stack. Azure IoT Hub directs the data to Azure Stream Analytics, which preprocesses it and then redirects it to Azure Machine Learning. This Azure component makes predictions and identifies whether activating the nutrient supply pump is necessary. This action is reported back to the Azure SQL Database through a flag. If the flag value is set to 1, the pump is activated. A cron job running on the process computer verifies the state of the flag each hour and activates the pump when necessary. After the pump starts, the flag is reset to the zero value.

2.2.2. LSI

The authors present an entirely IoT-based Local Software Infrastructure (LSI) configuration to perform a relevant comparative analysis of a complete cloud software infrastructure and a local one. This includes:

• A local SQL Server database;
• A local Web project that can be accessed locally, allowing the M5Stack module to make requests to a local address and use it to add sensor values to the database;
• A script that retrieves data from the database via a cron job and uses a local ML component. Depending on the result returned by the ML model, it sets a flag in the database;
• Another script that runs via a cron job associated with the pump checks if the indicator is active, starts the pump, and deactivates the indicator in the database.

This local software infrastructure is presented in Figure 3.

Section 3 will compare the two proposed infrastructures, where multiple tests will be analyzed and presented.

In the workflow associated with this scheme presented in Figure 3, the data is collected by the M5Stack module. The module sends a request to a Web address accessible within the local network. This request triggers a script that adds the data to an SQL Server database. Next, a cron job periodically runs a script that retrieves the most recent entry from the database and sends it to the local ML model, which decides whether or not the pump should be started. If the pump needs to be started, the script sets an indicator in the database to the active state. At the same time, another cron job runs on a process computer connected to the pump, executing a script that checks the indicator in the database. If the indicator is active, the pump starts, and then the indicator is deactivated in the database.

By analyzing the hardware structures of AFCI and LSI, it can be observed that LSI requires an additional central computer that acts as a central server. This central server exists in AFCI, except that Microsoft Azure owns it. The central computer is integrated into the processing computer for the LSI, providing multiple tasks. The processing computer is necessary for both infrastructures.

Section 3 also details the implementation method for each software module from the two schemes presented in Figure 2 and Figure 3. Several ways exist to effectively implement the conceptual modules presented for the two software infrastructures at the code level. The authors chose integration through .NET technologies using the C# programming language.

2.3. Automation Scheme for the Monitoring and Control Process of the Hydroponic System

The automatic system from this research uses a Recovery Drip System type. It allows the recirculation of a solution consisting of water mixed with nutrients. The nutrient solution is redirected to the reservoir upon meeting certain conditions. The Recovery Drip System belongs to the class of Drip Irrigation Hydroponics systems. Figure 4 presents the constituent elements of this system, where the seed is placed. The seed is deposited in the Dripper, which absorbs the substance, providing a favorable environment for plant development. Essentially, the Dripper replaces the soil, which supplies the nutrients necessary for plant growth. In the absence of soil, the Dripper is soaked with the water-soluble nutrient mixture. The drippers and the seed are placed in a support called a Holder. The Holder is periodically supplied with the liquid mixture in a PVC pipe.

Figure 4 presents the components of the Recovery Drip System hydroponic system. This illustrates how the nutrient solution is distributed to the plants. Figure 5 shows the nutrient solution tank and the feed pump. This figure highlights the automatic nutrient feeding system. The automatic system presented in Figure 5 is an integral part of the overall system, analyzed from the perspective of the two viewpoints. Although this element is common to both infrastructures, it is considered a constitutive part of the entire system.

Figure 5 shows the nutrient solution tank and the pump that delivers this solution to the dripping system and, therefore, to the plants.

The proposed automatic system aims to maintain the concentration of nutrients within a range which depends on the plant type. The block diagram of this system is presented in Figure 6.

The Process block mainly refers to the areas where the plants and the solution tank are located. The Controller block comprises the following hardware and software components: M5Stack, Azure IoT Hub 2025, Azure Stream Analytics 2025, Azure SQL Database 2022-16.0.1000.6, Azure Machine Learning 2025, and a Process Computer for the AFCI, as well as M5Stack, a Process Computer, a Local SQL Database, and a Local Machine Learning Model for the LSI. This block receives and processes information from the sensors, as explained in Section 2.2, and sends a control signal to the Nutrients Pump to adjust the nutrient concentration of the solution that feeds the plants’ roots.

2.4. Comparison Metrics Between AFCI and LSI

The comparative analysis between AFCI and LSI will be carried out based on seven tests that focus on the infrastructure’s performance in terms of RT, OR, IC, OC, CS, DA, and DS.

RT aims to measure the time required for the process when an event is identified. In other words, RT quantifies the time elapsed between collecting sensor data and activating the pump. The authors start with the following two hypotheses:

• RT-H1: The time for AFCI varies depending on the Internet connection. Sending data in the cloud may increase the RT, compared to LSI;
• RT-H2: The Internet connection does not affect the response time for LSI, as all processes run locally. However, the execution time per iteration within the process must be quantified for a relevant comparison with AFCI.

The main objective for OR is the total error-free operating time over a predetermined period. This performance indicator reflects the level of trust in the quality of the software infrastructure used. The authors start from the following two hypotheses in the OR-associated test:

• OR-H1: AFCI may record downtime when Azure performs maintenance;
• OR-H2: LSI has no risk of downtime.

The IC indicator refers to the programming team’s costs for implementing the solution. The more complex the implementation and the more advanced programming knowledge and modern technologies it requires, the higher the costs directly impact the overall business profit. In the case of IC, the authors start with the following two hypotheses:

• IC-H1: AFCI has significantly higher IC than LSI because Azure programmers are experienced professionals trained explicitly in using these technologies, and the implementation time is much longer;
• IC-H2: LSI does not require senior-level programming skills. A mid-level programmer can implement such an infrastructure, lowering IC.

The OC indicator refers to the actual costs of the software infrastructure, meaning how much is paid for each auxiliary service integrated into the infrastructure. These services are typically billed monthly based on their usage frequency. For this indicator, the authors propose the following hypotheses:

• OC-H1: The costs for Azure services (IoT Hub, server, SQL Database, cron) are higher than those of a local infrastructure;
• OC-H2: The costs for LSI can be zero.

The CS indicator is associated with the flexibility of the proposed infrastructure when adding new hydroponic crops. Extending the monitoring needs by introducing a new crop involves adding more monitoring elements. A professional solution should allow this with minimal changes to the software infrastructure. Therefore, the authors start with the following hypotheses:

• CS-H1: AFCI requires no fundamental modifications from the CS perspective;
• CS-H2: LSI requires a series of minimal modifications.

The DA indicator refers to the ability to view data and the status of the hydroponic system remotely. The authors will analyze the following hypotheses:

• DA-H1: Through AFCI, data can be accessed from anywhere in the world;
• DA-H2: LSI does not allow data visualization outside the intranet.

The DS indicator tests the data’s vulnerability to attacks by competitors who might attempt to sabotage hydroponic crops. The authors begin their tests with the following hypotheses:

• DS-H1: AFCI benefits from all security layers through Azure’s pre-implemented mechanisms, even though the data is stored externally;
• DS-H2: LSI may have multiple security vulnerabilities if not explicitly addressed during the software infrastructure implementation.

Section 3’s analysis of results concerning these hypotheses will provide a clear image regarding the choice of the best solution in the context of hydroponic systems through an IoT infrastructure associated with the two scenarios. The results are disseminated in detail in Section 4.

3. Results

This section compares the AFCI and the LSI. This analysis covers both hardware aspects and software implications. This approach enables a comparative study based on the seven tests defined earlier: RT, OR, IC, OC, CS, DA, and DS. In addition to presenting the raw results, this analysis will include correlations between the tested parameters.

3.1. Hardware Comparison Infrastructure Between AFCI and LSI

The hardware infrastructure includes the same equipment for both AFCI and LSI. The authors proposed using M5Stack equipment and sensors to monitor hydroponic cultures. The module was programmed for both types of networks using Arduino IDE, utilizing the logical schema from Figure 7. This logical schema highlights a comparative analysis of the distinctive elements between AFCI and LSI. In Figure 7a, it can be observed that the request enabling data transmission to the cloud is made using Message Queuing Telemetry Transport (MQTT) technologies, whereas in the case of LSI, Figure 7b, the request is made via Hypertext Transfer Protocol (HTTP).

Figure 7a,b show that the hardware structure is not fundamentally altered between the two approaches. The controlling element is the nutrient pump activated by the processing computer. The process computer used in these tests is a Hewlett Packard personal computer (PC), as it has small dimensions, integrated Wi-Fi capability, and connectivity features with other devices. The functions performed by the PC differ between the two configurations. In the case of AFCI, the process computer runs a cron job that checks if a flag is set in the database stored in the cloud to trigger the activation of the pump. In the case of LSI, the process computer performs many functions, hosting the database, the ML model, and the cron job that runs the database’s flag check to control the pump’s activation.

Figure 8 presents the hardware structure of the monitoring equipment, which includes the DS18B20 temperature sensor, the DHT22 temperature and ambient humidity sensor, the FC-28 electrical conductivity sensor, the SEN18 liquid level sensor, and the M5Stack monitoring unit. The unit is responsible for data acquisition and transmission to the data server for both methods.

The constituent elements of the AFCI and LSI proposals are identical at the hardware infrastructure level. The difference between the two is exclusively at the software infrastructure level.

3.2. Software Comparison Infrastructure Between AFCI and LSI

The programming of the M5Stack module is carried out similarly for both AFCI and LSI, as presented in Figure 7. In practice, the only difference at the monitoring and data acquisition module level is how the request is made to an external service. The fundamental differences are identified in the techniques used to manage the data after the acquisition process.

The study of the ML model is excluded from the comparative analysis because it is the subject of another article that will be published following this material, which is dedicated to choosing the best solution from an infrastructure perspective. Consequently, the authors emphasize that the absence of the study of acquired data, ML algorithms, and their performance results is dispersed into another work. In this way, the authors do not confuse readers with multiple development technologies.

The AFCI approach requires minimal coding and focuses more on settings performed at the Azure cloud level. This integration challenges the Azure programmer by requiring knowledge of available services, their compatibilities, specific settings for each service, etc. The LSI approach extends the code volume, as the local cronjob runs a custom C# script. Additionally, the data reaches the database through an API request made by a C# script using the ASP.NET MVC 5 framework.

3.2.1. Software Infrastructure of the AFCI

From a software perspective, AFCI constructs its M5Stack request in the form of JSON, as shown in the following example:

{

 “temperature_DS18B20”: 28.19,

 “temperature_DHT22”: 30.9,

 “humidity_DHT22”: 43.3,

 “soil_moisture”: 1,

 “water_level”: 1.87,

 “EventProcessedUtcTime”: “2025-02-28T16:39:43.2929208Z”,

 “PartitionId”: 0,

 “EventEnqueuedUtcTime”: “2025-02-28T15:41:54.6700000Z”,

 “IoTHub”: {

   “MessageId”: null,

   “CorrelationId”: null,

   “ConnectionDeviceId”: “1”,

   “ConnectionDeviceGenerationId”: “638763404011884782”,

   “EnqueuedTime”: “2025-02-28T15:41:54.5810000Z”

 }

}

In this example, the DS18B20 sensor recorded a temperature of approximately 28.19 °C in the liquid. The DHT22 sensor reported an ambient temperature of 30.9 °C and measured the relative humidity of the air at 43.3%.

The liquid conductivity value is 1%, which suggests that the liquid is not detected at all. This value indicates the need to trigger the event of starting the irrigation pump. The water level has been measured at 1.87%. This parameter indicates how much liquid is available in the reservoir, suggesting the necessity of starting the pump.

The EventProcessedUtcTime indicates the moment when the cloud service processed the event. It is expressed in UTC (Coordinated Universal Time). The identifier of the partition from which the event originates is 0. Partitions are used in distributed systems to identify the origin of the data. Extending the monitored plants needs to consider multi-partitioning. The EventEnqueuedUtcTime indicates when the event was added to the queue for processing, which is also expressed in UTC format.

The IoT Hub section contains metadata about the connection of the IoT device with the hub:

• MessageId is null, meaning the message does not have a specific ID;
• ConnectionDeviceId is the unique identifier of the device, explicitly set in the Azure graphical interface for a specific device. The ID of the connected device is 1, uniquely identifying the device that sent the data;
• ConnectionDeviceGenerationId is the identifier for device generation;
• EnqueuedTime is the exact moment when the message was queued in IoT Hub.

An Azure Stream Analytics Job connects m5stackInstance (IoT Hub) and the SQL Server Database. This service identifies the data event in the Hub, retrieves it, and deposits it into the database. To achieve this, the software pipeline needs to be studied, analyzed, and built, as illustrated in Figure 9. Here, m5stackInstance is the job’s input, while the database table is the output. A server for SQL Database hosting is also set up in the Azure portal.

The most important software component is the Azure Stream Analytics service, as presented in Figure 9. The image highlights how the data from a device (M5Stack), used for managing the hydroponic system, is processed and monitored. In the central part of the image, the Query section shows the SQL script that selects all data from the stream. The stream service acquires the data from m5stackInstance and inserts it into the hydroponicDB database. This infrastructure is configured to continuously capture and store data from the hydroponic system sensors, which are monitored at one-hour intervals.

In the lower part of Figure 9, the Job simulation (preview) section displays a chart of the data processing workflow. In this chart, an input node can be observed fetching data from the m5stackInstance service, followed by an intermediate processing step labeled Computing hydroponicdb, then a merging step (Merger), and finally, the output into the hydroponicDB database.

3.2.2. Software Infrastructure of the LSI

The data acquired from the M5Stack device is transmitted to the database through an API implemented in the C# language using the ASP.NET MVC 5 framework. This API is a Web application that runs on a local server, which is represented by a PC equipped with the Windows 10 operating system. Internet Information Services (IIS) supports web applications. IIS is configured to handle HTTPS requests coming from the M5Stack device.

In the case of LSI, the M5Stack makes the API request via localhost, which is designed as follows: https://localhost:44330/Home/ReceiveData?tempDS18B20=22.5&tempDHT22=23.1&humDHT22=60&soilMoist=99&waterLevel=82 (accessed on 10 February 2025). This web address can be accessed only by users from the same network as the server which hosts the applications.

This API request is a URL that transmits the data received by the M5Stack to a database using the parameters in the URL. Consequently, the URL is structured to transmit key-value parameters containing information about the state of the hydroponic system. The temperature recorded by the DS18B20 sensor is 22.5 °C, the ambient temperature detected by the DHT22 sensor is 23.1 °C, and the relative humidity of the air measured by the DHT22 sensor is 60%. The electrical conductivity value is 99%, and the water level in the reservoir is 82%.

This approach can be disadvantageous from the perspective of parameter visualization through the URL. Anyone connected to the internal network who knows this URL construction method could add fake data to sabotage the hardware infrastructure.

The script running on the process computer is identical for AFCI and LSI. This script checks in a table whether the value in the column associated with the pump is 1 or 0. A value of 1 indicates the need to start the pump. For this purpose, the script will send a command to the pump’s corresponding installation, and subsequently, it will set the value to 0. The value will be modified later when the plant’s optimal development conditions are no longer met. This value is set by the output of the ML service, which is not the subject of study of the present article. A detailed analysis of this issue will be presented in a subsequent article, which is expected to be published within a maximum of one month.

Beyond the hardware functionality, the cronjobs ensure the pipeline’s operation. The difference between a local cronjob and an Azure one through Stream Analytics lies in the financial resources. A locally set cronjob via Windows Task Scheduler is a free resource, whereas Azure Stream Analytics charges hourly based on the resource group to which it belongs. The configuration for each service also changes the monthly costs.

The cost analysis for the Microsoft Azure service used for processing and storing data from the hydroponic system outlines that the most considerable cost comes from the hydroponic import job, the Stream Analytics job, which generated a cost of EUR 8.40 per day. Other services, such as IoT Hub (m5stackInstance) and SQL Server (hydro server/hydroponicdb), did not incur significant costs on the same day.

When comparing this solution with the local implementation, referred to as LSI, which uses Task Scheduler in Windows, there are advantages and disadvantages for each approach:

1.

Costs:

• Azure Stream Analytics is relatively expensive because it processes real-time data and uses paid cloud resources;
• Task Scheduler is free and runs locally on the PC, eliminating cloud infrastructure-related costs.

2.

Complexity:

• Configuring a Stream Analytics job requires integration with IoT Hub, SQL Server, and other Azure services, which can be highly complex for users without experience;
• Task Scheduler runs the script locally for data processing and storage, making it much simpler to configure and not requiring an experienced programmer. This directly impacts the financial resource.

3.

Performance:

• Azure offers pre-implemented, tested, and fully functional long-term services that relieve the owner of hardware infrastructure concerns, provide long-term support, and enable continuous processing, making it ideal for large volumes of data [58];
• Task Scheduler depends on the availability of the local PC, and if the device is turned off, the job will no longer run. This means the entire burden of continuous functionality falls on the owner.

These considerations led to the idea that Task Scheduler is a cheaper and simpler option for a small project. On the other hand, for a large-scale project, Azure remains superior but has the disadvantage of costs. The proposed tests analyze these considerations in detail and present them in the next section.

3.3. Metrics Results for Performance Comparison Between AFCI and LSI

To measure RT, 240 test cycles were performed for each infrastructure, using the same network to ensure the Internet connection was identical. The time monitored during each test was obtained as the difference between the moment the data was sent and the moment it was recorded in the database. The results are as follows:

• For AFCI, the average RT was approximately 300 ms;
• For LSI, the average response time was 200 ms.

Figure 10 presents the RT evolution for AFCI and LSI. The graph shows the superiority of the LSI infrastructure.

AFCI performance depends on the quality of the Internet connection. An Internet connection outage would lead to the hydroponic system’s inability to function. LSI demonstrates superiority in terms of RT, making it more suitable for the automation typology of the hydroponic system. As a result, both RT-H1 and RT-H2 are confirmed. The superiority of the LSI configuration is marked by the value 1 in

Table 2. Comparison of AFCI and LSI infrastructures based on key metrics.

Metric Indicator	AFCI	LSI
RT	0	1
OR	0	1
IC	0	1
OC	0	1
CS	1	1
DA	1	0
DS	1	0
Total	3	5

, while the value 0 is assigned to AFCI. In this way, after analyzing all the proposed metrics, it will be easy to visually determine the superiority of one of the configurations.

For OR, the following observations were recorded over 10 days:

• AFCI missed one measurement due to Azure unavailability, which could have been caused by maintenance or other service issues;
• LSI did not record any interruptions, confirming hypotheses OR-H1 and OR-H2.

However, it should also be noted that a power outage would have the same consequences as an internet outage, affecting both types of infrastructure. Nevertheless, this time, LSI receives a point in Table 2, as it demonstrates superiority in the OR metric.

The estimation of IC was based on the skill level of one of the authors, who was assessed as an intermediate programmer. The number of hours required to implement the AFCI infrastructure was 31 h. In contrast, the number of hours needed for the LSI implementation was approximately 20. The authors acknowledge that this analysis is subjective, given that a programmer may be more experienced with cloud infrastructures or vice versa. Additionally, counting hours may be influenced by factors such as fatigue or others; however, the difference in the number of hours is significant enough to confirm both hypotheses, IC-H1 and IC-H2. This metric demonstrates the superiority of LSI over AFCI, adding one point to Table 2 for LSI.

An experimental Azure account was used to evaluate the OC metric. Within this account, for the 10 days of monitoring, an 82.57 USD/month cost was obtained for Azure services.

Table 3. Cost analysis of Azure Services for the hydroponic system in production.

Service Category	Service Type	Description	Estimated Monthly Cost (USD)
Internet of Things	Azure IoT Hub	Standard Tier, Free: 500 devices, 8000 msgs/day, 0.00 USD/mo, 0 IoT Hub Units;	0
Databases	Azure SQL Database	Single Database, vCore, General Purpose, Serverless, Standard-series (Gen 5), Locally Redundant, 1 Billed vCores, RA-GRS Backup Storage Redundancy, 0 GB Point-In-Time Restore, 0 × 5 GB Long Term Retention	1.27
Analytics	Azure Stream Analytics	Standard Type, 1 Streaming Unit(s) × 730 h; Stream Analytics on 1 Device(s) with IoT Edge	81.30
Total	-	-	82.57

presents the detailed cost analysis of the Azure Infrastructure. This account exclusively includes server services, database, IoT Hub, and cron through Azure Stream Analytics. The OC for LSI is 0 USD/month, confirming hypotheses OC-H1 and OC-H2. LSI’s superiority in terms of OC leads to adding one point in Table 2.

Table 3 depicts the cheapest scenario that can be used in a production phase. Unlike the development phase used in this research, the production phase may include auxiliary costs. Microsoft provides an online calculator that allows the configuration of all services needed for the production phase. In Table 3, the production configuration is similar in cost to the development one. The configuration set for the production phase includes the “Standard Tier, Free” plan for the IoT service, which consists of 500 devices and 8000 messages per day, with no additional costs. This is a free option for implementations of the size of the hydroponic system proposed in this work. The Azure SQL Database service uses serverless resources, a setting with minimal costs. Therefore, this service generates an auxiliary monthly cost of 1.27 USD. The Azure Stream Analytics service is the most expensive, as it captures the real-time data stream. This service generates a monthly cost of 81.30 USD. Finally, the total estimated cost is 82.57 USD per month.

A fundamental note is that although AFCI involves recurring costs, these also cover advanced data security. This is a significant advantage; however, hydroponic farming should not generate considerable data security issues. On the other hand, while LSI appears more economical, it may incur hidden costs through unexpected technical interventions.

In the case of adding a new type of crop or a new monitoring area, the following OC observations can be made:

• AFCI requires configuring a new device in the Azure platform and setting up the unique identification code in the firmware loaded on the device, with no other modifications needed;
• LSI requires similar changes, such as associating a unique device code that must be transmitted as an additional parameter through the URL. In this case, the changes are similar to those required for AFCI.

The fluctuations in Azure service costs and the long-term projections are based on extensive monitoring of cloud resource consumption. In this regard, the following fluctuations have been identified:

• The estimated minimum cost is approximately 75 USD/month for low resource usage;
• The estimated maximum cost is approximately 120 USD/month in the case of intensive use, including the increase in the volume of data stored and processed.

For a long-term projection, the scenarios are as follows:

• Cost for 1 year (average scenario, 82.57 USD/month) results in approximately 990 USD/year;
• The cost over three years (without volume discounts) is approximately 2970 USD.

Expanding infrastructure for more hydroponic systems will result in a proportional increase in the costs of Azure services. Regarding the hidden costs of local infrastructure (LSI), the initial operational costs are zero. In the category of hidden fees, the following situations can be considered:

• Hardware maintenance is associated with the costs of replacing and maintaining local servers;
• Energy consumption of the local server that runs continuously. It can consume an average of 30–50 kWh per month, equivalent to approximately 3.65–6.07 USD/month. The authors note that these costs vary depending on the specific kWh price in each country and that the previous report is comparable to Romania.

The OC results contradict the initial hypotheses, OC-H1 and OC-H2, where the authors expected LSI to require more modifications. In reality, the changes were the same at the infrastructure level. Based on these considerations, both infrastructures receive one point in Table 2 for the proposed infrastructure’s flexibility and scalability.

Regarding the DA indicator, the tests confirmed both hypotheses DA-H1 and DA-H2:

• For AFCI, global data accessibility is advantageous. Figure 11 lists data from the database stored in the cloud. This database can be accessed from anywhere;
• For LSI, access is limited to the local network only.

Therefore, AFCI allows the system to be monitored from any location, an advantage for distributed farms. At the same time, LSI can operate autonomously, but the lack of remote access is a limitation. Based on these considerations, in Table 2, AFCI is awarded one point, while LSI receives 0, as it is inferior to AFCI from the perspective of DA.

For AFCI, DS is advanced, with no risk of unauthorized access. The way data is sent by the M5Stack is primarily secured through the MQTT protocol, which uses SAS 2023 (Shared Access Signature) data encryption mechanisms. In addition, communication between the M5Stack device and the cloud infrastructure is carried out via an encrypted channel using SAS encryption mechanisms, which ensure connection security. SAS encryption generates a temporary shared access signature with a session-specific authentication token. This approach ensures that only authorized entities can communicate with Azure services, thus restricting cloud resource access to specific periods and restricting permissions. SAS tokens are generated based on a shared secret known only to Azure and the client device, eliminating any possibility of interception or identity spoofing. Figure 12 presents a screenshot that enabled the generation of the SAS for 24 h from the corresponding Azure Bash interface. Figure 12 presents the Shared Access Policies section for m5stackInstance. The access policies can be observed in the upper part of the image, each having specific permissions such as Registry Read, Registry Write, Service Connect, and Device Connect. These permissions control how devices and services can interact with the IoT Hub. The command in the lower part of the image generates a SAS for the access policy named cosminaSAS. This custom access policy provides a security token to control access to Azure resources without exposing the primary or secondary access keys.

Furthermore, Azure IoT Hub, which serves as the entry point for data in the cloud infrastructure, uses SAS to manage device-level access. Each device, including M5Stack, receives a unique SAS key to generate temporary authentication tokens. The M5Stack also uses the primary access key of the IoT Hub service, which is securely stored. As a result, data transfer is triple-secured through the MQTT protocol, SAS, and the primary key of the service.

Regarding LSI, no security measures are in place unless they are explicitly implemented. Furthermore, anyone who knows the URL can add data through URL parameters. These parameter values are directly inserted into the database. Therefore, assuming someone sets the liquid level to 0 via the URL, the system will trigger the pump activation, even if it is unnecessary.

In summary, AFCI excludes the risks of cyberattacks, while LSI requires protection measures to be implemented from scratch. In correlation with OC, it follows that securing LSI may introduce additional costs. Based on these considerations, one point will be added for AFCI and zero points for LSI in Table 2. The results confirm the two initial hypotheses, DS-H1 and DS-H2.

By analyzing Table 2, it can be observed that LSI is superior in terms of RT, OR, OC, and IC. AFCI is superior in terms of DA and DS. Both infrastructures are equal regarding CS. Based on these considerations, an LSI-type infrastructure is sufficient when the scales are specific to a single building for modeling the challenges of hydroponic systems. Extending to an infrastructure that can centrally manage multiple hydroponic systems in different places should consider the AFCI proposal.

4. Discussion

The comparative analysis of the AFCI and the LSI has shown that, from a hardware perspective, the two systems are identical, featuring the same monitoring and control components. The two approaches’ hardware infrastructure uses the same equipment for monitoring and controlling the hydroponic system. The main hardware components include the M5Stack unit for data acquisition and transmission, the DS18B20 sensor (for measuring liquid temperature), DHT22 (for ambient temperature and humidity), FC-28 (for liquid electrical conductivity), and SEN18 (for water level).

The hardware element ensuring the automation of the liquid feeding process for the container is controlled by the process computer, represented by a compact HP PC. Both AFCI and LSI use the same hardware infrastructure. The differences between AFCI and LSI arise at the software functionality level. In AFCI, this computer runs a cron job that checks a flag in the cloud database to activate the irrigation pump. In the case of LSI, the PC equipment has a more complex role, managing both the database and the Web application. At the same time, it executes the cron job that controls the pump activation.

Regarding communication, AFCI uses the MQTT protocol to send data to the cloud, while LSI uses HTTP to send requests to the local server. Thus, although the equipment used is identical, the differences between AFCI and LSI appear at the software and data management levels.

Significant differences emerge at the level of software infrastructure and data management. In the case of AFCI, the M5Stack unit sends data to Azure IoT Hub as a JSON object. This message is picked up by the Azure Stream Analytics service, which processes and stores it using the SQL database managed on Azure. A major challenge of this method is understanding Azure services, their compatibility, optimal configuration to avoid high costs, and programming at the code level when necessary. Programming is performed in Arduino IDE, and interaction with Azure requires knowledge of SQL, JSON, and Azure development.

In contrast, LSI uses a different software approach. The data is sent from the M5Stack to a local API implemented in C# and ASP.NET MVC 5, which runs on a local web server based on IIS. This method allows greater autonomy over the software infrastructure but presents security risks since the information is transmitted via a URL accessible within the internal network. A vulnerability is the possibility of unauthorized users who know the URL structure inserting false data. For data flow management, in the case of LSI, Windows Task Scheduler runs a local cron job that checks the database and activates the irrigation pump when necessary. This is a cost-effective, free solution, but the local PC must always be turned on to ensure continuous operation.

The tests have shown that the AFCI infrastructure offers a longer response time than LSI. Additionally, AFCI comes with higher operational costs. On the other hand, the LSI solution presents the advantage of total control over data, with reduced long-term costs, but involves greater complexity in managing the local infrastructure and dependency on proprietary hardware resources. Furthermore, the tests highlighted that data security is more straightforward to ensure in the case of AFCI. At the same time, LSI requires the development of custom security layers, which cannot match the Microsoft security services developed and tested by large teams of programmers and testers.

4.1. Interpretation of the Results

Beyond the technical aspects highlighted by the tests, this study’s results clearly show the necessary trade-offs between performance and costs. AFCI, thanks to its scalability capabilities, uses the Azure platform to process and analyze data efficiently. However, the significant costs associated with the entire infrastructure offset this advantage.

On the other hand, LSI, through a local server and a customized software solution, offers a higher degree of independence but raises issues related to maintenance and scalability. The data show that the local infrastructure can quickly become a limitation as the volume of processed data increases. Consequently, it becomes necessary to make additional investments in hardware equipment and qualified human resources.

An important point to note is the different ways in which the two infrastructures manage data security. AFCI uses a highly developed authentication system based on Azure infrastructure, reducing the risk of unauthorized access. However, dependency on an external service raises privacy concerns that are not problematic in hydroponic systems. These systems do not handle sensitive data involving temperature, humidity, electrical conductivity, and liquid levels in containers. Being completely autonomous, LSI eliminates this risk but is vulnerable to attacks and potential hardware failures that could lead to data loss.

Summarizing the research results, the answers to the RQs are:

• RQ1: AFCI uses Microsoft Azure for all processes involving data from the IoT infrastructure. This underscores the need for a permanent internet connection. LSI manages data locally, using its own servers and dedicated scripts. Therefore, in the case of LSI, there is no dependency on the internet;
• RQ2: The performance indicators studied in the results show that LSI has a response time of 200 ms compared to 300 ms for AFCI. Additionally, LSI is superior to AFCI in terms of reliability and costs. AFCI offers scalability, global accessibility, and advanced security, but with recurring costs of USD 82.57/month;
• RQ3: LSI is ideal for small, autonomous farms where cost is a priority. AFCI is more suitable for large, distributed farms where accessibility and interoperability are central elements.

The results obtained have implications for how infrastructure solutions are chosen based on the typology of each application. This study demonstrates that there is no universally optimal solution; the choice must be made according to each project’s priorities. LSI is a viable solution for a small business with a low volume of data and limited financial resources. On the other hand, for a project with high scalability and global accessibility requirements, AFCI is a more suitable choice despite the higher costs.

4.2. Limitations of the Study

This paper examines the infrastructures related to the challenges of automatic hydroponic systems. However, it was conducted under the constraints of a controlled environment defined by a limited number of variables. The study did not account for many monitoring elements or distributed automation. In practice, the approach was simplified to emphasize the unit tasks of the problem, but it does not consider the monitoring of multiple crop types, which require different conditions. Furthermore, the analysis did not include numerous monitoring elements.

Unlike traditional methods, AI-based hydroponic systems require operators with technical skills for setup, monitoring, and maintenance. This transition poses a barrier for small farms. Qualified personnel involve additional costs compared to traditional methods. Therefore, the adoption of AI must also be evaluated from the perspective of the availability and cost of the corresponding labor factor.

Furthermore, the authors acknowledge that the study did not consider potential human errors in implementing LSI or real cyberattacks. The authors consider such vulnerabilities to be of no significant concern in the context of hydroponic systems, where the data are not sensitive. Human errors or those resulting from cyberattacks are mentioned in the category of errors, which can lead to the incorrect introduction of nutrient concentrations. This behavior leads to nutritional imbalances, affecting plant growth and crop yield. For example, an error in recording electrical conductivity generates nutritional deficiencies or toxicity in the hydroponic solution. Similarly, cyberattacks on the IoT infrastructure used for hydroponic monitoring and control lead to disruptions in automatic processes. Although data such as temperature, humidity, and nutrient solution levels are not considered sensitive from an information security perspective, their compromise can affect system decisions, leading to over-irrigation, excessive or insufficient nutrient administration, or even complete system shutdown. In future research, the impact of human errors and cyberattacks on the operation of hydroponic systems will be explored, focusing on developing solutions to prevent or mitigate these risks.

Moreover, the cost analysis was performed over a short term. Over the long term, certain costs may fluctuate. For example, the initially lower costs of LSI could be offset by maintenance expenses.

4.3. Future Research Directions

In light of the study’s conclusions, future research directions include the development of hybrid solutions that combine the advantages of cloud infrastructure with those of local infrastructure. Such a solution would analyze the integration of benefits from both approaches to create an optimized system.

Additionally, the authors propose several studies in the field of hydroponics, including:

• Analysis of multiple ML algorithms for modeling the hydroponic system from the perspective of pump activation. This research will examine both local ML.NET-specific algorithms and pre-implemented Azure ML algorithms. The study will also present the specific costs associated with the Azure ML Studio service;
• Optimization of chemical processes using pH predictions and anomaly detection in hydroponic measurements. This direction aims to improve the efficiency and reliability of hydroponic systems through advanced predictive techniques;
• Automatic water flow control systems in hydroponic systems with IoT infrastructure. Developing and testing systems capable of dynamically adjusting water flow based on real-time data;
• Study of environmental variability’s impact on plant growth in hydroponic systems. Investigating how fluctuations in environmental conditions affect crop development and yield;
• Prediction of water requirements for hydroponic systems dedicated to specific plant cultures for operational cost analysis using intelligent prediction tools. This research will focus on optimizing resource allocation and reducing costs through data-driven insights.

By exploring these future research directions, the authors aim to delve deeper into modern technologies within the context of hydroponic systems. They believe such systems will be widely developed at the urban area level as cities continuously expand.

5. Conclusions

This study addressed the issue of configuring an infrastructure for monitoring and controlling a hydroponic system. The authors analyzed two structures: AFCI and entirely LSI. Based on the presented study, it was concluded that each proposal is suitable for a hydroponic system depending on its specific requirements and characteristics. The central research question of this study focused on optimizing the infrastructure for automatic hydroponic systems. These infrastructures depend on classic factors: cost, security, scalability, and implementation complexity. Based on the conducted study, it was demonstrated that cloud-based solutions are more suitable when a distributed monitoring infrastructure is desired. In practice, this solution can monitor multiple facilities located in different places. From a security and reliability perspective, AFCI is superior to LSI but has the disadvantage of higher operational costs compared to local solutions.

Each infrastructure was detailed in terms of hardware and software. The AFCI solution was based on the specific services of the Microsoft Azure platform. The LSI solution used local custom-made software services. The two proposed infrastructures were implemented and analyzed from the perspective of seven indicators, which were summarized in a table after a thorough analysis. The seven parameters (RT, OR, IC, OC, CS, DA, and DS) were assessed based on tests and showed the superiority of LSI for monitoring a small-scale hydroponic system. The comparative analysis based on scoring showed that for the proposed type of hydroponic system, AFCI has a score of 3 benefits, compared to LSI, which totals 5 advantages. The analysis highlighted the superiority of LSI for the proposed automatic hydroponic system. The comparative study also highlighted the advantages of AFCI if a hydroponic system is desired that can be accessed remotely or can centrally monitor multiple hydroponic systems located in different places.

Most current research promotes the integration of IoT infrastructures. This research highlighted that such infrastructures are helpful when the application’s specific requirements demand this approach. In practice, the research demonstrated that not every problem intended for optimization can be modeled through IoT technologies.

Through this research, the authors believe they have contributed by introducing and analyzing modern technologies into the agricultural field through a future-oriented sector represented by hydroponic systems. The authors consider this research a foundation for further farm developments considering IoT architecture elements.