Optimization of Crop Yield in Precision Agriculture Using WSNs, Remote Sensing, and Atmospheric Simulation Models for Real-Time Environmental Monitoring

Abstract

Due to the increasing demand for agricultural production and the depletion of natural resources, the rational and efficient use of resources in agriculture becomes essential. Thus, Agriculture 4.0 or precision agriculture (PA) was born, which leverages advanced technologies such as Geographic Information Systems (GIS), Artificial Intelligence (AI), sensors and remote sensing techniques to optimize agricultural practices. This study focuses on an innovative approach integrating data from different sources, within a GIS platform, including data from an experimental atmospheric simulator and from a wireless sensor network, to identify the most suitable areas for future crops. In addition, we also calculate the optimal path of a drone for crop monitoring and for a farm machine for agricultural operations, improving efficiency and sustainability in relation to agricultural practices and applications. Expected and obtained results of the conducted study in a specific area of Reggio Calabria (Italy) include increased accuracy in agricultural planning, reduced resource and pesticide use, as well as increased yields and more sustainable management of natural resources.

1. Introduction

The necessity to interconnect wirelessly several MEMS or Micro-Electro-Mechanical Systems (MEMS), such as microsensors, microactuators and other microstructures, led to the birth of wireless sensor networks, or WSNs. In other words, WSNs are a group of MEMS, also known as sensing nodes, that are also able to collect, process and transmit the acquired information [1,2]. The sensing nodes being interconnected through the internet led to the well-known field of the Internet of Things (IoT) [3,4].

These networks are suitable in several sectors, industries and locations [5]. For example, some authors [6,7,8] explored the possibility and the main issues of process automation in factories using WSNs. In particular, the implementation of these networks has been discussed in detail in [6], while the issue of reliability of networks has been discussed more in detail in [8], where the need for a more robust design of these architectures has been mentioned. Jabeen et al. [9] proposed an Intelligent Healthcare System based on a WSN and IoT based on data encryption through a genetic algorithm and a subsequence user identification protocol. Premi et al. [10] presented an overview to protect the sensitive information related to health state in a WSN. Other authors also explored the use of WSNs for the development of smart cities; in detail, in [11], a method was proposed that was able to classify several kinds of urban solid waste thanks to IoT, WSNs and a machine learning; Asha et al. [12] proposed an IoT-based system for smart cities with network optimization using a Recurrent Neural Network and the honey badger algorithm; while Zhang et al. [13] developed a sensor-based pollution for analyzing environmental pollution in cities.

Despite their impact, WSNs also have some negative aspects—the foremost is low energy sustainability and coverage [14] but also congestion, loss connectivity and deteriorated Quality of Service [15,16,17]. There is also a need for fault-tolerant architecture and more interoperability of these platforms.

Scientists suggest some methods to face these limitations, for example Cui et al. [18] achieved energy saving by minimizing the transmission time while Costa and Ochiai [19] evaluated the performance of three modulation schemes. Another way to improve energy efficiency is the use of cooperative communication scheme [20,21]. Regarding the coverage problems, the authors in [22] proposed a method to minimize active nodes while ensuring connectivity and coverage in WSNs, extending the network lifetime. The necessary conditions, a queuing model, and a heuristic method (HCCVGA) were also developed, achieving near-optimal solutions with minimal impact on network longevity.

The use of WSNs has also affected the agricultural sector, leading to the emergence of so-called Agriculture 4.0 or precision agriculture (PA) [23]. The field of precision agriculture is certainly a constantly evolving sector that has attracted growing interest, especially in recent years, where, thanks to the advent of Artificial Intelligence, it is possible to create ‘intelligent’ systems capable of providing significant improvements compared to traditional agricultural practices. Precision agriculture stands out from traditional practices, which are characterized by the excessive exploitation of resources, through the use of cutting-edge technologies that enhance the efficiency of agricultural production while reducing the environmental impact [24]. Typically, this field of research is characterized by the use of advanced technological tools such as IoT, sensor networks, big data, and robotics to monitor and analyze in real time the variables that influence agricultural practices, with the main purpose of optimizing natural resources (such as water and energy) and limiting the use of chemical compounds in fertilizers and pesticides, promoting environmentally sustainable practices and encouraging biodiversity conservation. Productivity is optimized through the automation and efficient management of the vast amounts of data collected to understand and analyze the performance of crops and respond proactively to changes that may occur in a crop. These technologies stand in contrast to traditional agricultural practices, which rely mainly on human involvement, manual labor, and the experience of the farmer. So, this kind of agriculture is data driven and is based on new digital technologies such as IoT, big data analytics, cloud computing, AI (Artificial Intelligence), drones and robotics [25]. The main technologies of this kind of agriculture are presented in Figure 1.

To cite some works: Mekala and Viswanathan [26] assessed the possible use of IoT in different agricultural sectors such as water and energy savings; Van Klompenburg et al. [27] achieved, using different features, yield prediction using machine learning techniques; Boursinais et al. [28] highlighted the relevance of UAVs in smart farming. The spread of robotics in agriculture has brought several advantages such as speeding up operation and productivity increases enhancing security [29].

Existing studies often fail to provide fully integrated frameworks that combine multisource data and enable actionable insights in real time. Another issue that needs to be studied, although not covered in this paper, is the processing of these data in systems with limited computational power, when Deep Learning models are being employed. It is also important to consider that despite the growing use of machine learning in many applications, they may slow down some particular processes, which hinders their potential in some particular contexts. This need could become more evident in sectors like precision agriculture, wildfire conservation, and smart grids. Despite their already discussed increased computational cost, machine learning methods can not only provide increased accuracy, but also help in faster processing of data, by learning precise patterns present in data, and avoid unnecessary calculations which would otherwise be performed by the use of other methods. In summary, the increased complexity of machine learning systems can be a hurdle but also an opportunity in the design of systems using wireless sensor networks.

In this context, our research proposes a comprehensive system that integrates sensor data, atmospheric simulations, and UAV imagery using advanced data fusion methods. In particular, sensor data are provided by different kinds of sensors (linked through WSNs) that communicate with each other with an advanced communication protocol proposed by the authors. The resulting data are visualized within an open-source GIS platform, facilitating real-time planning for irrigation and fertilization, as well as optimizing autonomous drone and tractor operations. The originality of this research mainly concerns the use of an experimental atmospheric simulator that, by discretizing the atmosphere through three-dimensional cubes, accurately simulates its functioning through a particular physical–mathematical model, Smoothed Particle Hydrodynamics (SPH). This simulator also acts as a pattern detector to estimate atmospheric variables over time, which are useful for defining the areas most affected by such atmospheric phenomena. Furthermore, the novelty of this research lies precisely in the integration of different data—coming from simulations, ground truth, and satellites—within a single evaluation system, which allows for obtaining significant results in the field of precision agriculture. Additionally, this study includes the use of a sensor network (WSN), which has been appropriately studied and optimized with a high-performance communication protocol, enabling the achievement of promising results in identifying the most suitable areas for growing plants and trees typical of the region. This approach aims to bridge the gap in current solutions by enhancing resource efficiency and sustainability, aligning with the goals of Agriculture 4.0 and advancing in WSN technologies.

2. Materials and Methods

2.1. Study Area

A portion of the northern territory of the municipality of Reggio Calabria was chosen as the study area. The testing area for the simulator is highlighted with the red polygon in Figure 2. The entire methodology was applied in a hilly area always within the municipality of Reggio Calabria, with an extension of approximately 50 hectares.

2.2. Sensors

As previously said, WSNs are based on nodes, called sensors, that are able to collect information about an event or a physical object. WSNs can be grouped by several characteristics, and one of these is the type of sensor [14]. We can distinguish these as:

• Seismic,
• Thermal,
• Acoustic,
• Visual,
• Magnetic,
• Radar, and
• Infrared.

Another possible classification is based on the environment in this classification we can identify:

• Terrestrial WSNs,
• Underwater WSNs,
• Multimedia WSNs, and
• Underground WSNs.

Regarding the agricultural field, sensors are able to save resources during the process without reducing production, and thus reducing the environmental footprint [30,31,32].

For this study, five types of sensors were employed:

• A soil moisture sensor;
• A Leaf wetness sensor;
• A pH sensor;
• A Temperature and humidity sensor, and
• A barometric sensor.

Soil moisture sensors are widely used by the farmers because they enable identifying the appropriate time to irrigate, avoiding over or under watering.

Soil moisture sensors can be classified into two categories based on the technology they use: those that measure Volumetric Water Content (VWC) and those that measure soil tension when placed in the soil profile. Volumetric water content refers to the volume of liquid water in the soil, typically expressed as a percentage. On the other hand, soil water tension measures the energy required by plant roots to extract water from soil particles. As water is removed from the soil, soil tension increases [33]. For this study, we used a TEROS 12 soil moisture sensor.

A leaf wetness sensor is used to evaluate the amount of moisture present on the surface of leaves, helping to identify the right time to irrigate and avoid the spread of fungal diseases. The sensor functions by measuring the electrical resistance of two conductive plates, collocated on the leaf’s surface. So, the sensor is able, by checking the changes in electrical resistance, to determine when the leaves are wet or dry. Often these kinds of sensors are able to give other relevant information such as temperature and humidity [34]. In this experiment, we used LP-80 Leaf Wetness Sensor ACCUPAR.

A pH sensor is a tool able to measure the pH value of a liquid or a soil, and it functions thanks to an electrode able to detect variations in H⁺ concentration. These sensors are widely used in agriculture due to the ability of pH to influence nutrient availability for plants [35]. A HALO soil pH-meter was used for this experiment.

A temperature and humidity sensor measures air temperature and humidity levels. It is commonly used in environmental monitoring. In agriculture, these sensors are crucial for monitoring conditions that affect plant health and growth. They are also used to manage greenhouse climates. Bi-Sensor by Mahier Smart Agrocontroller was used (MAHER ELECTRÓNICA, Almería, Spain).

A barometric pressure sensor measures atmospheric pressure and is widely used in weather monitoring, aviation, and altimeters. The sensor typically consists of a flexible membrane, often made of silicon, which responds to changes in atmospheric pressure by expanding or contracting. This movement causes a change in the sensor’s electrical resistance. To account for temperature-related variations in atmospheric pressure, modern sensors may also include temperature sensors. Since temperature influences air density, which affects pressure, this compensation is crucial. Barometric sensors play an important role in agriculture, such as in weather forecasting and managing livestock, and are key in understanding weather patterns and their impact on crop growth and yield. For this study, we used a Digital Barometer PS-0060-AD (Netsens s.r.l., Toscana, Italy).

The use of these sensors was crucial for this study because they were employed to verify the data obtained from the atmospheric simulator, which will be described further in this paper (Section 2.5).

2.3. Communication Protocol

Wireless sensor nodes are small, autonomous units that collect and transmit data such as temperature, humidity, light, motion, pressure, or chemical constituents in real time wirelessly within a network. These nodes are essential to wireless sensor networks (WSNs) used in applications like environmental monitoring, agriculture, health care, and smart cities.

A basic wireless sensor node consists of four main components: one with sensing capability, thus the sensor to monitor a specific parameter; the process unit, usually a microcontroller that processes the data derived from sensors; the communication module, which incorporates a wireless communication protocol calibrated for the required range (long range or short range); lastly, the power source. Furthermore, a basic wireless communication system is comprised of three principal component:

• The transmitter encodes the source of information to a readable signal, encrypts it and sends it to the encoder to reduce noise and create a modulated signal.
• The channel serves as the medium that transfers signals from the transmitter to the receiver.
• The receiver is responsible for reconstructing the original information signal through a series of processes, including demultiplexing, demodulation, channel decoding, decryption, and source decoding.

Communication protocols are vital for ensuring that these sensors inside the network can transmit and receive data effectively. When selecting communication protocols [36,37] for agricultural systems, ensuring uninterrupted data transmission is essential, but it is also useful to consider coverage, energy usage, data transfer speed, and costs. Each protocol has its advantages and disadvantages, making the selection process complex. Low-Power Wide-Area Network (LPWAN) protocols like LoRa and Sigfox are gaining popularity for their energy efficiency over large areas, while traditional protocols like Wi-Fi and ZigBee remain important in smart farming. It is essential to evaluate wireless communication protocols based on specific application needs, so focusing on low power consumption and reliable data transmission LoRa (long-range protocol) represents the best candidate for remote agricultural areas.

This protocol uses Chirp Spread Spectrum (CSS) modulation for signal resilience and reduced collision probability. It operates on various frequency bands: 868 MHz in Europe, 915 MHz in the U.S., and 433 MHz in Asia. Data rates range from 0.3 kbps to 50 kbps, and communication distances are 2–5 km in urban areas and up to 15 km in rural locations. Its star topology connects end devices to a central gateway for internet access, enhancing efficiency and management. LoRa features AES-128 encryption for secure communication.

Other uses of the LoRa protocol can be found for monitoring wetlands [38], for monitoring sensors in greenhouses [39] and for monitoring water quality in coastal aeras [40].

2.4. Geographic Information Systems and Atmospheric Simulators

Geographic Information System or GIS software are able to store, manage and analyze several kinds of georeferenced data (i.e., soil characteristics or corps yields). The use of GIS helps farmers to analyze data, make more informed decisions and apply more competitive strategies. In our research, Geographic Information Systems (GIS) play an important role in managing data collected from various sources and providing valuable insights related to Agriculture 4.0 [41,42]. The selection of optimal areas for cultivation is influenced by factors such as climate, soil type, irrigation, topography, and market demands, all of which impact crop yield and quality. For this study, we used QGIS, an open-source GIS software (version 3.34), which enables the processing, visualization, and analysis of geographic data, including maps, satellite imagery, and sensor data, through Python scripting.

Thanks to the combination and analysis of remote sensing and in loco data with QGIS (Quantum GIS), it was possible to establish the most suitable area for cultivation and to establish the path for a drone dedicated to monitor crops and the route for an automatic tractor used for the main agricultural activities. In detail, the plugins used to achieve this advanced tracking analysis in QGIS were “NAI (Network Analysis Library)” and “PgRouting”. These instruments enable identifying, after calculating many variables, and evaluating the optimal paths. In this work, PgRouting was used to calculate the optimal paths, and different ‘weights’ were defined to take into account the type of terrain, the speed of the tractor, the slope, and the characteristics of the soil.

The atmospheric simulator [43], which is the main innovation proposed from this work, generates punctual climate data related to the topography of a given area by processing and combining DEM, radar, and sensor inputs. The simulator produces outputs for various atmospheric variables, such as wind, temperature, and humidity, for a cubic cell measuring 50 m, even with limited input data. The atmosphere in the simulator is discretized, and its microphysics are governed by equations that describe the interactions between numerous particles. This atmospheric system consists of a group of particles that are centered within a cube (cell). Each particle’s mass exchanges and interactions with other particles, as well as the environment, are determined by the specific quantities of gas, water, and humidity that it contains. These particles are influenced by forces arising from their energy state, their interactions with other particles, or their contact with the surface. In essence, the simulator is capable of replicating the behavior of the Earth’s atmosphere concerning climate trends, while also detailing the discretization and interactions within the atmosphere. The simulator consists of two solvers, L1 and L2, that run simultaneously and handle tasks at a higher level. These solvers can transform atmospheric processes related to the troposphere and vertical exchange into a particle model. The first solver can perform a variety of tasks, including:

• Assigning both physical and chemical attributes to the particles,
• Applying force to the particles based on their energy state.
• Defining particles and objects inside the environment original state, and
• Dividing the 3D space into macro-clusters and offering a separately identifiable forecast field for each one.

This solver also accepts as input real- or hypothetical-time data values.

Solver L2 is a Newtonian Simulator which works with the Smoothed Particle Hydrodynamics (SPH) software library (Developed by NVIDIA) and is able to simulate, in a realistic way, fluid or gas behavior. This solver also combines the forces of the particles, obtained from the previous solver, and starts dynamic development using the SPH fluid rules. For example, based on specific parameters (such as mass, partial compressibility, viscosity, and friction), repulsive forces are activated when cells locally intersect at time t. This mechanism simulates wind in a three-dimensional environment, accounting for minor pressure variations and intricate turbulence behavior. In regions with complex orography, like the Apennines, where it is difficult to manage rapid altitude changes using traditional NWP (Numerical Weather Prediction) methods, this approach proves especially effective. An ASCII DEM representing the simulation’s orographic domain is examined as part of Solver L2’s initialization phase. The environment is filled with uniformly distributed particles that grow until they reach a stable point. Following the processing of the initial condition from sensor or radar data, Solver L1 instructs Solver L2 on how to allocate forces, velocities, and positions in clusters (a particular spatial volume). L2 progresses after a set period during which these instructions are executed. Once the evolution is completed, Solver L1 is updated with the particles’ new state, including their position, velocity, density, and any changes from the initial state. Following this, Solver L1 links particle characteristics and atmospheric conditions, such as temperature and humidity, using the data supplied by L2. Based on the data provided by Solver L2, the pattern detector, a part of L1, creates groups with recurring atmospheric characteristics, such as cloud cover, precipitation, humidity, mean temperature, and pressure. The output produced by the simulator is represented by the findings that have been classified as clusters.

2.5. Vehicles Automation

Self-driving tractors are autonomous machines equipped with advanced technologies like sensors, cameras, GPS, and machine learning algorithms. These tractors are becoming increasingly popular in the agricultural sector due to the numerous advantages they offer to farmers. One major benefit is their ability to operate continuously, improving the efficiency of farming activities. Unlike human operators, self-driving tractors can work without breaks or rest, making them especially valuable during critical periods such as planting and harvest seasons. Additionally, these tractors provide greater precision and accuracy in farming tasks. They can operate with high consistency, reducing the use of fertilizers, pesticides, and other inputs, which leads to better crop yields and lower operational costs [44]. Moreover, self-driving tractors are capable of adapting to various weather conditions and terrains, helping to reduce soil erosion and enhance soil health. They can also operate in hazardous environments where human presence might be unsafe.

A remote-controlled system allows an operator to manage a tractor from a distant location using data from the tractor’s cameras and sensors. Successful remote operation requires meeting specific latency and bandwidth criteria to ensure smooth transmission of video and control data. This system typically includes cameras, sensors, collision avoidance systems, and control interfaces. The ISO 11783 standard [45], also known as ISOBUS, facilitates communication between tractors and agricultural equipment. Based on the SAE J1939 protocol, ISOBUS uses the Controller Area Network (CAN), enabling electronic control units (ECUs) to exchange data through the CAN bus. Autonomous control systems for agricultural tractors rely on communication methods that comply with the ISOBUS standard. The Cannelloni software, which connects area network controllers within a local network, supports this communication. However, ensuring the security of data transmission over the internet or mobile networks remains a concern. Tractor control itself is governed by the CAN bus protocol, which supports a bit rate of up to 1 Mbps. The main components required for remotely operating a tractor include a remote-controlled cabin, a remote-controlled tractor, and a computer system for CAN tunneling over a mobile network [46]. The remote-controlled cabin allows the operator to monitor and control the tractor from a distance, while the remote-controlled tractor is equipped with sensors and actuators to perform tasks autonomously. The computer system enables the transmission of data through CAN tunneling, allowing communication between the tractor’s control units over a mobile network. This setup ensures that the operator can manage the tractor’s functions in real time, no matter the location.

Network traffic for remote tractor operation is routed through a Virtual Private Network (VPN), with communication occurring over a shared local area network (LAN). A router connects the tractor to the tactical network using multiple mobile networks, automatically switching between them to maintain optimal performance. The system aims to assess the feasibility of remote control by transmitting CAN messages over the network. The SocketCAN [47] interface allows CAN data to be handled like a standard network interface, while the Cannelloni tool facilitates CAN data transmission via an Ethernet tunnel using either UDP or SCTP protocols. UDP is fast but unreliable, whereas SCTP offers reliable delivery at the cost of slower performance. To securely transfer CAN traffic, Secure Shell (SSH) is used; however, since SSH is incompatible with SCTP, conversion to TCP is performed using the Socat utility.

A Raspberry Pi 4 with a Kvaser USB adapter is used onboard the tractor to connect to the local CAN system. The Raspberry Pi on the tractor acts as a client, linking to a server Raspberry Pi in the remote-controlled cabin. The cabin server has a fixed internet address, while the tractor may be behind a Network Address Translation (NAT) or firewall. If the connection drops, the tractor continuously attempts to reconnect. To minimize irrelevant CAN traffic, both the tractor and remote-controlled cabin implement filters that enable only the necessary messages to pass through. These filters are managed by adjusting the Cannelloni software to better control the data flow [48].

2.6. Visual Representation

The proposed methodology involved different technologies and tools: Geographic Information Systems (GIS) enable the collection and integration of geospatial data within a single computer system to monitor different aspects related to agriculture, such as crop distribution or the identification of areas with different levels of fertility in the field, and to plan targeted interventions for each type of crop. It is therefore possible to create zoning and prescription maps. Within these systems, which act as collectors of organized information, it is possible to integrate data from different sources, such as those from sensors and remote sensing, which are essential for obtaining ground truth information, as well as data from satellites (a field that is constantly advancing thanks to the ability to obtain increasingly detailed information), with acquisitions that can also be periodic and specific for certain variables that are fundamental for the correct management of agricultural crops. As for remote sensing, there are numerous satellites and even drones equipped with sensors useful in this field, which can provide essential information, such as multispectral or thermal sensors. Furthermore, the possibility of using Artificial Intelligence guarantees the opportunity to make predictions through the analysis of historical data, predicting crop yields, detecting diseases and pests quickly, and, especially, optimizing irrigation and avoiding unnecessary waste. The previously mentioned methodology is presented in Figure 3.

3. Results

All the data collected were imported into QGIS, in order to show the results obtained.

The punctual values, extracted from the clustering activity of the simulator’s Solver L1, are the simulator’s output and represent the climatic variables. These punctual values are then imported into QGIS, as shown in Figure 4. In order to achieve a more accurate analysis, local meteorological data were initially inserted inside the simulator.

The data were subjected to an interpolation technique in order to assess regions that are more vulnerable to extreme and adverse meteorological events. This process helps in identifying geographic areas that are most at risk, based on various environmental and climatic factors. By interpolating the data, a more detailed and continuous spatial representation of risk can be created, allowing for a comprehensive understanding of the areas most affected by these events.

To obtain a continuous spatial representation of the study domain of the geospatial and atmospheric variables taken into consideration, it was therefore necessary to interpolate the values of the variables on the basis of a linear estimator, through which the values measured at points near a given value are obtained, according to this relationship (1):

$$Z^{\ast }\left(x_0\right)={\displaystyle \sum }_{\alpha =1}^n{\lambda }_{\alpha }\times z\left(x_{\alpha }\right)$$

(1)

where $x_{\alpha }$ represents the so-called estimate neighborhood with respect to $x_0$, and the coefficient ${\lambda }_{\alpha }$ is the weight of the linear combination. For the following study, a statistical estimation model called Ordinary Kriging was used, which belongs to the models of univariate geostatistics and allows obtaining a forecast map as a linear combination of the values observed at nearby points, optimally weighted, where the weight depends on the distance and spatial correlation between the points. This statistical model has the advantage of being one of the most precise interpolation methods and is able to provide an estimate of the error associated with each forecast. The variogram parameters used for Ordinary Kriging are the follow:

Spherical model parameters:

-

c₀ = 1.6428070365928094,

-

c₁ = 2.1590749251648447, and

-

a = 56,645.970600413806.

Values obtained:

-

Nugget (c₀) = 1.6428070365928094,

-

Sill (c₀ + c₁) = 3.801881961757654, and

-

Range (a) = 56,645.970600413806 m.

The process by which these parameters were obtained starts from a dataset containing rainfall data from the atmospheric simulator simulation. With the aim of modeling the spatial correlation between precipitation values in different locations, the empirical semivariogram was calculated, verifying that it fit well with the spherical model. The distance between all pairs of points was then calculated using the Haversine formula. For each pair, the squared difference between the precipitation values was calculated, in order to obtain the semivariance. The distances between the points were divided into bins to summarize how the difference in values changes with distance.

Subsequently, the spherical semivariogram was estimated, as an analytical function used to describe the spatial correlation, in which the following parameters appear:

-

c₀ (nugget): represents the variance at very small distances;

-

c₁ (sill): represents the total variance at large distances;

-

a (range): represents the distance beyond which the spatial correlation becomes negligible.

The estimate of the parameters c₀, c₁ and a for the representation by Ordinary Kriging was obtained through a least squares fitting process, with the aim of identifying the optimal values so that the spherical model best fits the empirical semivariogram. This was achieved by identifying the combination of c₀, c₁ and a that minimized the error between the predicted semivariance values and those observed for each distance interval.

The value of 1.64 for the nugget is reasonable and indicates some spatial variability at small distances. This is consistent with the fact that precipitation is not uniformly distributed across the territory. The value of 3.80 for the sill indicates spatial variability over larger distances, while the spatial correlation is significant up to approximately 56.6 km, beyond which it becomes negligible.

As depicted in Figure 5, the analysis reveals that the regions most susceptible to extreme and adverse meteorological events are primarily found in two key areas: the coastal regions and mountainous terrains.

Integrating spatial and temporal data fusion techniques with data collected from satellites and UAVs (Unmanned Aerial Vehicles) has proven to be highly effective in calculating vegetation indices and in other fields of research [49,50]. This method enables the production of Very High-Resolution (VHR) images, which offer significantly more detail than traditional satellite imagery. By combining data from multiple sources, a more thorough and accurate representation of the terrain can be achieved.

The VHR images generated through this fusion process enable the extraction of vegetation indices, which are crucial indicators of plant health and density. These indices, computed from the spectral data of the images, provide valuable insights into the condition of vegetation. One of the key indices used is the Normalized Difference Vegetation Index (NDVI). The combination of satellite data, which provides broad spatial coverage, with UAV data, which allows for more localized and precise measurements, enables a more comprehensive view of the vegetation dynamics over time.

The NDVI is an index commonly used in geospatial analysis to define the most vigorous vegetative zones from the dry ones and the bare rocks [51,52]. It is a value that varies from −1 (typically attributed to bodies of water), passing through zero (representing bare soil and rocks), up to 1—a value attributed to vigorous vegetation. The estimation of this vegetation index is obtained with a simple operation between the bands (2):

$$NDVI={\displaystyle \frac{NIR-RED}{NIR+RED}}$$

(2)

This index is mainly useful for determining lush areas that grow without problems or diseases. Yellowing of the leaves, in fact, leads to a decrease in the NDVI value, which could indicate stress in the plants, suggesting the need for an intervention with pesticides, fertilizers, or other corrective measures, depending on the cause of the problem.

As shown in Figure 6a,b, the NDVI calculation derived from the VHR image demonstrates the increased resolution and accuracy made possible by this data fusion technique. The enhanced clarity of the vegetation indices leads to a better understanding of plant health, which in turn allows for more effective monitoring and management of vegetation in the study area. This advanced approach not only improves the precision of vegetation analysis but also enriches the overall information available for ecological research and environmental management.

Thanks to the powerful capabilities of Geographic Information Systems (GIS) in storing, analyzing, and integrating a wide variety of data types, we were able to effectively assess and categorize the area into three distinct classes based on multiple parameters. These parameters included vegetation indices derived from both satellite imagery and UAV (Unmanned Aerial Vehicle) data, sensor data, and climatic variables. Each of these data sources provided critical insights into the environmental and agricultural conditions of the area, allowing for a comprehensive and data-driven classification. The integration of different data sources, including those from the atmospheric simulator, is particularly complex and articulated for several reasons. First of all, there is the difference in data formats. The geospatial data that can be integrated into the GIS platform can be in the raster format (such as images from satellites) or the vector format (such as field boundaries or sensor positions). Satellite data, in particular, are often present in scientific geospatial formats such as GeoTIFF and NetCDF, while sensor data are expressed in CSV and JSON formats. Consequently, for the visualization and management of the data, it was necessary to ensure coherence and correct georeferencing and spatial alignment of the data, in order to correct distortions (considering that the images can have different angles and resolutions).

The main difficulty lies in the different spatial resolutions and temporal frequencies. Satellite remote sensing data often have, for freely distributed data, a lower spatial resolution compared to data collected on the ground by sensors (up to the centimeter). At the same time, temporal frequencies can vary considerably, and in such cases, a methodology to adopt is certainly temporal analysis to manage the data at regular intervals. Obviously, it is necessary to ensure interoperability between different systems and platforms, thus adopting APIs (Application Programming Interfaces) and interoperability standards such as OGC (Open Geospatial Consortium).

In this specific case, the various formats of the sensors (CSV), satellite images (pre-processed), and drones (GeoTIFF) were all compatible with the GIS system used and were all georeferenced and aligned with the appropriate geoprocessing techniques within the GIS system. Regarding spatial resolution, satellite data used (10 m) were integrated, where missing, with drone data, operating a data fusion process. In this application case, the simulator was essential for predicting the variables of rainfall and wind in the study area, provided in the CSV format. The simulations were conducted at specific time intervals (with weekly intervals), and the neural network incorporated in the simulator predicted the behavior of these variables for the following season. These data were then used to create the zonation maps and identify the areas that require irrigation. The same process was performed with the simulator data.

In relation to the method of data-driven classification, there are many methodologies for grouping areas with similar characteristics to proceed with an accurate geospatial analysis: the common K-means, Hierarchical Clustering (agglomerative and divisive methods), Fuzzy C-means, Spectral Clustering, Gaussian Mixture Models, and Self-Organizing Maps, to name a few, each with its advantages and disadvantages and implementation possibilities in a QGIS environment or via a Python algorithm. Among the various possible methods, DBSCAN (Density-Based Spatial Clustering of Applications with Noise) was chosen for this work. DBSCAN is a clustering algorithm that is based on the spatial density of the data and not on a predetermined number of groups, such as K-means. Conceptually, clusters are defined as areas of high point density separated by areas of low density, and enable labeling and discarding points that do not belong to any cluster as outliers. The main parameters for using DBSCAN are Eps (the maximum distance between two neighboring points) and MinPts (the minimum number of neighboring points required to form a cluster). To identify areas with a similar density of features, variables were selected that were used to group areas with a similar spatial distribution. In this experimental and initial phase, the data-driven classification was performed by selecting the meteorological variables, the information obtained from satellite data, and all the data collected by the different sensors on site.

The first class, referred to as “Class A”, represents areas that are suitable for crops that do not require constant irrigation. These crops are more resilient to varying water availability, making them ideal for regions where irrigation needs are less frequent or intensive.

The second class, “Class B”, identifies areas that are suited for crops requiring near-constant irrigation. These crops are highly dependent on regular and consistent water supply to thrive, often found in areas where water management systems are vital for sustaining crop health and maximizing yields.

The final class, “Class C”, is designated for areas suitable for arboreal (tree-based) culture. These areas are typically characterized by conditions conducive to growing trees, such as specific soil types, water availability, and climate factors that support long-term tree growth.

Figure 7 illustrates this classification, clearly showing the distribution of the three classes across the study area, providing a visual representation of how different environmental and climatic factors influence agricultural practices in the region.

Lastly, using the methodology previously mentioned, we were able, thanks to the GIS software, to identify the best path for a drone, used to check the state of already growing crops and the route for an automatic tractor used to carry out irrigation and fertilization operations (Figure 8).

4. Discussion

This study aimed to explore and test an automated integration system designed to optimize and manage crop cultivation. The methodology combines an experimental atmospheric simulator, satellite and UAV imagery, local sensor data, and data transmission protocols, all integrated within an open-source GIS platform. The primary goal was to identify various crop types and determine the efficient routes for an automated tractor and drone to monitor crop growth and health. The research was conducted in a specific area of Reggio Calabria, selected for its suitability to implement the described technologies.

This study highlights both the advantages and challenges of this approach when compared to other methods. One major strength is the comprehensive integration of various technologies, which offers a holistic approach to Agriculture 4.0. The experiment provides valuable insights into crop conditions, aiding in the precise optimization of irrigation and fertilization. However, Agriculture 4.0, including the proposed methodology, still faces challenges, particularly regarding the availability of accurate and timely data, as well as the expertise required to interpret and utilize this information effectively. Moreover, the high initial setup costs and the need for skilled personnel to operate and maintain the system may limit its widespread adoption.

Despite these challenges, which are common in Agriculture 4.0, this research contributes significantly by demonstrating how integrated technologies can optimize irrigation and monitor crop health. The authors emphasize the methodology itself rather than immediate outcomes, leaving room for further analysis by experts to identify optimal cultivation areas. Promising technologies for integration in drones, such as advanced sensors for applying fertilizers or natural pesticides, are also discussed. These developments are especially relevant as fertilizer production increasingly depends on evaporation and crystallization processes, and as the demand for fertilizers grows alongside rising raw material costs. New production techniques based on these methods are being developed to address this challenge.

In conclusion, this study advances the field of Agriculture 4.0 by demonstrating the integration of diverse technologies and approaches. While challenges remain, the potential benefits of adopting such integrated systems for precision agriculture are significant. The findings highlight the value of advanced technologies in improving agricultural practices and provide a pathway toward more efficient and sustainable farming in the future

5. Conclusions

The proposed innovative system offers distinct advantages in terms of time efficiency, human resource management, and overall productivity. By automating key agricultural tasks traditionally performed by humans, it helps reduce both time and costs. Furthermore, the system’s ability to analyze data on soil conditions and crop health can improve productivity and quality while also conserving the valuable resource of water. Water usage is optimized through weather data provided by the simulator. A key innovation in this system is the use of software that, with minimal data input, enables accurate weather condition assessments. Another significant development is the increasing influence of self-driving tractors on the agriculture sector. While still relatively new, technological advancements suggest that these machines will become more prevalent and continue transforming farming practices.

In any case, the main results obtained from this work, from the point of view of efficiency and sustainability, mainly concern the reduction in water consumption and the efficient use of fertilizers and pesticides. In fact, it must be considered that the chosen study area is located in southern Italy, a region of the peninsula that is suffering from drought and climate change, with consequent disruptions in productivity due to continuous phenomena of scorching heat and a lack of water resources. Furthermore, it is an area where the use of natural practices is still favored, and from an efficiency perspective, the use of this system is particularly beneficial in this region. Additionally, the area is affected by the cultivation of a particular tree, the bergamot (which grows only in this area due to its specific climatic conditions), used throughout Italy and the world for perfumes because of its unique essence. This method would enable the identification of areas most suitable for these crops, as well as for vineyards. For this reason, through this study, the efficiency and management of this area would be significantly increased, bringing important advantages both environmentally and economically.

So, from an environmental perspective, it is worth noting that the area lacks specific irrigation systems, and waste is high. Such a system would enable the saving of substantial amounts of water annually. From an economic perspective, however, the increased productivity of specific and unique crops in the region would lead to economic growth for the entire area, since this particular product is exported worldwide, being unique.

Therefore, significant advantages have been identified for the application of this method: optimization of natural resources already in decline, improvement in the agricultural productivity of specific native species, and a reduction in environmental impact. However, the method proves to be effective not only for the specific characteristics of the study area but also in other regions with similar conditions. Indeed, thanks to its ability to optimize the use of natural resources and improve agricultural efficiency, it can be successfully applied in contexts facing similar challenges, such as water scarcity and the need for sustainable agricultural practices. The flexibility and adaptability of the system also make it useful in other regions characterized by difficult climatic conditions or specialized crops.

It should be highlighted that the research was conducted in a specific region of Reggio Calabria, selected for its favorable morphological and orographic features, making it an ideal location for implementing various technologies. This shows that, when properly planned and executed, such approaches can be tailored to suit specific agricultural settings, maximizing the benefits of technology based on local conditions. This research work, in fact, fully addresses the challenges related to the consumption of natural resources, adopting automated and optimized techniques to improve productivity while minimizing environmental impact; however, this study is not limited only to the current situation of the area but also provides long-term forecasts on the conditions of the areas affected by this technology. Furthermore, it can be applied to any other area, with equally efficient results.

Looking ahead, future improvements should focus on enhancing data processing and transmission protocols. Experimenting with more efficient algorithms could further optimize productivity and data management. Additionally, validating and scaling up these integrated systems will be crucial. Collaboration among research institutions, agricultural organizations, and farmers will be key to promoting the widespread adoption of these innovative solutions. The ultimate goal is to develop a sustainable agricultural model that efficiently uses resources, minimizes environmental impact, and meets farmers’ needs. Achieving this will require the continuous advancement of technologies and methods, with a focus on adapting them to the specific demands of various crops, environmental conditions, and farmer requirements.