1 Introduction

In environmental protection, monitoring activities are crucial for determining preventive strategies or large-scale solutions; Earth observation from space is of particular interest. To date, such analyses are carried out via satellite [1], however, due to the low spatial and, above all, temporal resolution of this technology, the information on an area or site of interest is often insufficient and may have to be validated by other methods of measurement and analysis. For example, easily accessible satellite data concerns passages that can be up to 3 days apart and with a resolution of 30 m on the ground and therefore of little use for applications such as precision agriculture.

There are, of course, methods with high resolution, also satellite-based [2], but they could be considerably expensive while only partially addressing the issue. This paper, for example, will mention Planet [3], a constellation of multi-band satellites that are excellent for satellite comparisons but still confirm the need to use other technologies to complement current Earth observation data.

The ATEMO project [4], therefore, aims to increase knowledge on environmental issues with the development of a compact, low-cost and versatile device that can be used in different contexts and environments, from the analysis of the composition of the atmosphere, through the detection of substances and pollutants such as VOCs (Volatile Organic Compounds), NOx (Nitrogen Oxides), SOx (Sulphur Oxides), CO2 (Carbon Dioxide), O3 (Ozone), to the determination of light sources on the ground and their characterisation. Furthermore, in order to understand how these factors may influence ecosystems, ATEMO also aims to determine ground vegetation indices (e.g. NDVI—Normalized Difference Vegetation Index). Another important goal is to compare these data with those obtained from satellites and seek a correlation between the various parameters under investigation. Indeed, while the influence of atmospheric pollutants and night lighting on ecosystems [5], plants and humans is already known, dated [6] and more recent [7] studies have suggested that the combined effect of light and atmospheric pollution may lead to harmful and worsening phenomena.

The first field of application of ATEMO, during the summer of 2023, was the determination of vegetation indices for agriculture in the framework of the PRIN (Project of Significant National Interest) project called ‘Rewatering’ [8]. This activity exploits the collaboration between several departments of the University of Padova: DAFNAE (Department of Agronomy, Food, Natural Resources, Animals, and the Environment), DII (Department of Industrial Engineering) and CISAS “G. Colombo” (Centre for Space Studies), as well as other Italian research institutions. In this context, the device was used on board a tethered balloon to evaluate the results on crop performance following water stress on soybean crops [9].

Despite the existence of satellite and airborne technologies (such as microwaves [10]) in the case of water stress assessment, the type of analysis conducted by ATEMO focuses on vegetation indices processed from the visible and infrared spectrum, as is the case with the most common smart agriculture satellites [11]. Monitoring the growth status of crops and biomass using SAR (Synthetic Aperture Radar) is also still in the early stages of development [12] and the use of low-cost alternative technologies such as ATEMO will allow it to be implemented.

The remainder of this paper is organized as follows: the next sections will describe the design of the system and the ATEMO integration. After that, field tests are introduced, and a selection of the collected results is presented.

2 Design of ATEMO Experiment

2.1 Design Overview

ATEMO is designed to be a versatile, compact and easily scalable device; therefore, it is modular and its mass as well as the on-board devices change depending on the type of analysis conducted, area of interest and vehicle employed for the experiment. In fact, the platform can be integrated on board drones, stratospheric balloons and tethered balloons. The latter system proved to be a good trade-off for the first tests conducted in the framework of PRIN project: ATEMO was used for analysis on soybean fields, which required good resolution over long periods of time over the same area, both of which could not be satisfied by stratospheric balloons or drones.

Turning to how the experiment was developed, ATEMO represents the compact merge of similar devices designed and manufactured by the research group and which operated in the three fields of analysis of interest: MINLU [13], three cameras for characterization of the spectrum and the intensity of light pollution sources, ARIA [14], a system of sensors for air pollutants, O-ZONE [15] an air trapping system for stratospheric pollution analysis, and AREO (the results of which have yet to be published) for the estimation of vegetation indices.

In more details, ATEMO has a weight of 2.5 kg, a base of 17 cm × 17 cm and a height of 25 cm as shown in Fig. 1, the instrument consists of an external aluminium frame to which 3D-printed components are attached, guaranteeing both the secure integration of the various sensors and cameras, and their easy attachment and removal in the event of a change of setup.

Fig. 1
figure 1

On the left, ATEMO experiment front view; on the right, ATEMO experiment bottom view (camera compartment)

The heart of ATEMO is the sensitive part consisting of several sensors, mainly consisting in cameras. In more details, the scientific payload consisted in the following instruments:

  • Two Basler ace 2 cameras, one colour and one monochrome, both with Sony’s IMX546, an 8MP square CMOS sensor with a wide relative response spectrum (from approximately 400 nm to 1000 nm).

  • A FLIR Vue Pro R, a thermal imaging camera (7.5–13.5 μm) adding useful information in both day and night contexts.

  • A GPS tracking system.

  • One or two SQM-L (Sky Quality Meter—type L), measuring the brightness of the night sky in magnitudes per square arc second.

  • Environmental sensors for temperature, humidity, and pressure.

As far as the air pollution aspect is concerned, on the other hand, the use of sensors that can provide a qualitative picture of the composition of the air is still being considered. It is expected to detect the presence of harmful compounds or pollution indicators and, after extensive testing, turn this qualitative results in a quantitative one to measure the concentration of the most harmful pollutants.

The choice of the cameras depended on several factors: such as the possibility of acquiring images from each camera at the same time with ease, the size of the sensor (square) which was therefore easier to manage for reprocessing and comparison with satellite acquisitions, the possibility of easily removing any IR filters and the friendly customisation in general, and finally the low cost for advanced and medium–high-resolution sensors.

The systems and sensors described above are managed and acquired by a Raspberry Pi3 with the Raspbian operating system; however, the acquisition is carried out autonomously by means of a C +  + programme that guarantees the acquisition of the cameras and saves raw scientific data on an SD card on board. In the case of non-stratospheric experiments, real-time connection with ATEMO is still possible for remote control, data exchange or manual setting of acquisition parameters.

Except for the Basler cameras, which were chosen for the reasons listed above and because the company also procures space grade instrumentation; all on-board instrumentation is the cheapest alternative that meets the functional requirements. Starting with the power bank, which is the only one on the market above 16,000 mAh that can select voltage up to 24 Volt and simultaneously supply 5V to the USB sockets. But also, the FLIR which is one of the cheapest models at low weight, the raspberry which is in fact a miniaturised computer that is easy to use and low cost, and the SQMs which are basically among the few sensors of this type at such low prices. The easy availability of the instruments was also considered.

Both the cameras and the SQM are connected via Ethernet by means of a switch that allows the Raspberry port to be split. Everything is powered by a 21,000 mAh power bank that provides a 5V output for all the systems and a 15V output for the Basler cameras.

The power consumption of the experiment and, therefore, its duration is strictly dependent on the acquisition frequency. Table 1 shows a comparison between different acquisition rates and the relative autonomy of ATEMO.

Table 1 Power budget with difference acquisition rates

Obviously, it is not only the frequency of images captured that is characteristic of the type of analysis underway but also the actual setup of ATEMO is closely linked to the application sector and the host aircraft. In fact, the optics combined with the two Basler-type cameras are not the only possible configuration; two possible alternative solutions used according to the flight in progress are the Kowa LM35HC (mainly for high altitude flights) and the Tuss LYM0814 (for low altitude flights). The comparison of these solutions can be seen in Table 2.

Table 2 Optical systems comparison

As will be discussed in more detail later, in the case of both light pollution assessment and calculation of vegetation indices, it is essential to receive specific information from the two on-board cameras; therefore, readings from the two cameras are filtered with band-pass or triple-band filters. For this reason, the next chapter focuses on describing the acquisition strategies and the preventive actions taken to maximise the scientific output.

2.2 Acquisition System Description

The two on-board cameras are used for two main purposes: the estimation of vegetation indices during the day and the analysis of light sources during the night. For these two cases, the configuration of the cameras changes slightly depending on the required light filtering. In the case of vegetation analysis, the colour camera is equipped with a triple-band filter (475, 550, and 850 nm) and the monochrome camera with a band-pass filter (735 nm), both designed to reproduce indices already employed in literature and calculated by Planet satellites. The rationale behind the choice of these specific bands is related to vegetation indices (which will be described in more detail later in Chapter 3), but which essentially express the health and vigour of plants based on the fact that chlorophyll, a descriptive variable of health, emits at certain frequencies [16]. In fact, the first real comparison that is made is between what is calculated through ATEMO’s acquisition and ordinary GIS (Geographic Information System) results, which for these application sectors can be obtained mainly from satellite data.

In the night-time application, however, the two cameras are used with a filtering solution similar to that of the MINLU experiment, another project conducted by our research group [13]. To date, the setup of the experiment is decided and fixed according to the flight scientific goals; in the future, the implementation of a filter wheel driven by a stepper motor will allow changing the filters during the flight. The filter wheel attachment points can be seen in Fig. 1, right, at the centre of the camera compartment.

As mentioned, the camera’s field of view (FOV), is another key point to consider. In fact, the choice of optics is made according to the host vehicle; in the case of integration on drones or balloons tethered at an altitude of 50–100 m, optics with wide apertures (Tuss) are used, while in the case of stratospheric balloons, optics with much narrower apertures (Kowa) are favoured.

For proper interpretation of the data provided by ATEMO, calibration with a commercial spectrometer (Black Comet C-SR-14) was necessary; the on-board cameras were characterized both individually and with some filtering conditions, since the input information is affected by the transmission of the optics, the filter, the response of the CMOS sensor, and, in our case, reflections from the soybean leaves. First, the chamber was characterized with only the optics minus the filters, including the IR filter (integrated into the chamber and physically removed), in fact the 3 channels sum around the NIR spectrum above 700 nm wavelength (Fig. 2 left side) showing that the filter is no longer present. Next, using a selection of known sources and the spectrometer, the response of the instrument was thus defined: the sources were first acquired with the spectrometer and then with the chambers obtaining the RGB channel response for the light sources (these, on the right in Fig. 2, are at the filter passbands in the triple-band plus single-band filter configuration mentioned at the beginning of Section “Acquisition system description”). In this way, by normalizing the obtained signals, the right relationship between camera input and output is obtained. This is important in order to have consistent results in fact, a significant example is the one that highlights the response for the 850 nm source, which is redundant for all 3 channels and therefore, without calibration, the light information would erroneously appear to be three times as large as it actually is.

Fig. 2
figure 2

On the left is a graph representing the camera response over the entire visible and NIR spectrum; on the right, also on the same spectrum, is the cameras response acquiring three light sources (475, 550, and 850 nm) and considering the response of the spectrometer, the CMOS sensor, optical path, filtering, and soybean leaf reflections to normalize the output signal. For both graphs, the lines correspond to the 3 RGB (Red Blue Green) channels, respectively

The calibration process and the setup described in this section were chosen for the application of the PRIN project called “rewatering”; the aim was to investigate the reduced irrigation effects on Soya crops with the estimation of vegetation indices. The next section will describe the activities, procedures, and results of the field tests carried out during the summer of 2023.

3 Field Tests and Results

3.1 Tests Framework

The goal of ATEMO is certainly to provide a global image of a given area of interest; therefore, in order to ensure consistent results, it will be necessary in the future to carry out tests on the other areas of analysis (air and light pollutants) as well. As mentioned, the configuration described will mainly deal with the analysis of aspects related to the health of vegetation that is also the main goal of the PRIN project previously highlighted: differentially irrigated soybean crops were monitored (one part at 70% of requirements and the other at 100%) investigating on the consequences of this action (deficit irrigation can be a strategy to reduce water consumption) [8].

Ground data included the measurements of leaf saturation, humidity, and soil temperature, and finally the results were compared with each other and with what was seen from satellites and with ATEMO. Space images were obtained using Sentinel [2] or Planet [3] data; however, only in the second case (Planet) images could be compared, as the ground resolution is higher and the passes more frequent.

Image data analysis for agriculture and biomasses is usually carried out by means of vegetation indices, i.e. dimensionless numbers that give information on plant vigour and can be of various formulations depending on the analyte and the type of information sought. In practical terms, these are calculated by spectral analysis and as an algebraic recalculation between results on the various levels of a spectrum. The index most employed in literature is the NDVI (Normalized Difference Vegetation Index) a recombination of the information of the response of instrument on the bands of NIR (Near InfraRed) and Red, which formula can be found below.

$$NDVI=\frac{NIR-Red}{NIR+Red}$$
(1)

The specific wavelength it depends on the acquisition system and on the procedure, to be consistent with the presented research we will take the values defined in Table 3.

Table 3 Band selection comparison between ATEMO and Planet [3]satellites

In the case of ATEMO, we focused on 3 additional indices: GNDVI (Green-NDVI), ENDVI (Enhanced-NDVI) and NDRE (Normalized Difference Red-Edge), defined according to the following formulas.

$$GNDVI=\frac{NIR-Green}{NIR+Green}$$
(2)
$$ENDVI=\frac{\left(NIR+Green\right)-2Blue}{\left(NIR+Green\right)+2Blue}$$
(3)
$$NDRE=\frac{NIR-RE}{NIR+RE}$$
(4)

where Green, Blue and RE (Red Edge) values are, another time, the one expressed in Table 3.

In order to determine these indices, the specific spectral bands required were then acquired using bandpass filters on the monochrome camera and triple bands on the colour camera. Below, in Table 3 you can see the bands that can be acquired with ATEMO in this specific setup and in the same table, the 8 bands of the Planet satellite case [3]. Figure 3 shows the optical filtering system on board ATEMO.

Fig. 3
figure 3

Basler cameras with internally mounted filters (C-mount)

Obviously, it is not easy to have the same band average values for ATEMO and Sentinel, mainly due to the type of technology being different. However, especially in the present case, ATEMO’s bands intersect in a good way with the ones of Planet, and it is, therefore, easy to correlate the two results. In general, there are many strategies to determine vegetation indices, and the easy access to them makes them useful tools for consistent simplified information from lot of instruments [17].

3.2 Tests Setup

As explained above, the payload was placed on board of a tethered balloon. This is the most congenial solution, with respect to drones or free balloons, as it allows ATEMO to be stationed on the same field and be able to observe significant changes within the same day. The setup includes three safety rods arranged at 120° to each other (Fig. 4) and a central emergency rod which is not tensioned and is only employed for emergency or in case of strong wind. With this configuration, the balloon can remain at an altitude of about 50 m for several hours (6–7); both the heat and the balloon yield and porosity can influence the duration of the floating phase; in any case, the balloon, which contains about 2 m3 of helium to lift the 2.5 kg of ATEMO, can be quickly lowered and refilled to ensure a few extra hours in the event of prolonged testing. In the event of an accidental breakage of the balloon, a parachute is inserted halfway between the balloon and the payload. In addition, this parachute is connected through two joints (one upper and one lower), making the flight chain similar to a double pendulum and helping to dampen the various oscillations caused by the wind.

Fig. 4
figure 4

Tie ropes arrangement

It is precisely this latter which is the most problematic part of the experimental phase; in fact, in the ideal case, in the absence of wind, the balloon sits independently for several hours, while in the case of wind, the adjustments to be made by means of the tie-rods are various and the altitude varies several times during the day. Figure 5 compares the altitudes on two different test days, in the case of wind (left) and in the case of calm (right), both on the same field in Cortona, for the half hour following the balloon release and stabilization. This comparison of the two ascent phases shows that the practical difficulty of the tests is to manoeuvre the ball to keep it in the desired position. The tests are not affected by this as the acquisition takes place at the desired height but keeping the balloon in place is complicated in windy conditions and must be considered.

Fig. 5
figure 5

Comparison of the altitude graph in the early stages (ascent phase) of flight on 19 July and 24 August both at the Cortona field

Before commissioning the device, several function checks are carried out, but above all, the exposure and gain values of the cameras are set. After a few on-site tests, the balloon is inflated and slowly raised to the desired height. For the evaluation of vegetation indices, the acquisition rate is about 1 per min, which is done in favour of safety, but this ensures a good amount of data that is also useful for fine georeferencing and more accurate attitude estimation.

4 Results

The test campaign took place in a total of five sessions from May to August 2023, three of which were carried out at a soybean field in Castelfranco Veneto (Treviso, Italy) and the other in Cortona (Arezzo, Italy), both composed of 4 closely spaced plots, 2 irrigated at 70% of water requirements and 2 at 100%.

Being a test methodology never carried out before, summer 2023 was mainly a way to test the functionality of the systems, understand the goodness of the acquisitions, define which indices were most descriptive of water stress and then compare the various sessions with each other and with satellite data.

Figures 6 and 7 show an acquisition carried out in Castelfranco Veneto by the two cameras and the calculation of the GNDVI, ENDVI and NDRE indices.

Fig. 6
figure 6

Images acquired by the two cameras in Castelfranco Veneto on 9 August 2023, on the left the image of the monochrome camera with the Red-Edge filter and on the right the colour camera with the triple-band filter

Fig. 7
figure 7

Starting from left to right ENDVI index, GNDVI index and NDRE index made starting from the acquisition in Fig. 6

Another information that can be obtained via ATEMO is the ground temperature; this is not redundant but rather very useful for understanding the temperature of the cultivated field under water stress; in Figs. 8, 9 and 10, it is possible to see an acquisition made in Cortona, the relative vegetation indices, and the thermal camera images showing the irrigation that has just taken place and, therefore, the different temperature of the two portions of the field.

Fig. 8
figure 8

Images acquired by the two cameras in Cortona on 24 August 2023, on the left the image of the monochrome camera with the Red-Edge filter and on the right the colour camera with the triple-band filter

Fig. 9
figure 9

Starting from left to right ENDVI index, GNDVI index and NDRE index made starting from the acquisition in Fig. 8

Fig. 10
figure 10

two different orientation images of the thermal camera on the Cortona field following a partial irrigation of the field

Furthermore, a comparison of the same indices calculated by satellite can be seen in Fig. 11. This comparison has only been made for the Cortona field; the reason is related on the fact that the data for the day worked on Castelfranco are absent, or rather the images present clouds that therefore prevent visibility in the field. In fact, this reconfirms what the limitations of low-cost satellite-only observation can be.

Fig. 11
figure 11

calculation of the indices in Fig. 13 (ENDVI, GNDVI, NDRE) using Planet [3] satellite data

Considering the index evaluation images (Figs. 7, 9 and 11), visually, it can be seen quite immediately how the differences in vegetation are, especially in the Castelfranco Veneto area, almost imperceptible. As a matter of fact, the images considered take a portion of the field irrigated at 70% and a portion of the field irrigated at 100% in a single shot, therefore, they should present a different and clear behaviour through graphic representation. To investigate the reasons for this discrepancy between the PRIN objectives and the ATEMO results, rainfall data were retrieved during the season in which the tests were conducted.

Table 4 shows these data, which are available and can be obtained from ARPAV (Veneto Regional Agency for Environmental Protection) [18] or ISPRA (Institute for Environmental Protection and Research) [19].

Table 4 Seasonal rainfall data for the sowing and cultivation period in the Castelfranco area [18]

The irrigation provisions of the 2023 summer season were defined according to the previous season, which, as shown in Table 4, were significantly lower. While this represented a significant limitation for the scientific results of the experiment, it only partially affected the validation of the ATEMO acquisition system: the comparison with satellite data still provides valuable information.

Precisely regarding this topic, before moving on to the conclusions, it is interesting to note that for some indices what is obtained by satellite has a slight difference in terms of the relative scale of the index. One of the reasons for this could be the different software elaboration of the data between the acquisitions of the balloon device compared to the satellite. Not only, it could be connected to the consideration of certain variables like the reflections (such as those of the soya leaves) and atmospheric filtering that could have different models used for the two technologies.

To effectively compare the results obtained with ATEMO and the Planet data, we rescaled the GNDVI values to have the same average and variance as the index calculated from the satellite. The ATEMO-based GNDVI is then down sampled to match the spatial resolution provided by Planet (3 m) by averaging the values of neighbouring pixels (Fig. 12).

Fig. 12
figure 12

ATEMO-based GNDVI down sampled on Planet [3] (Fig. 11) spatial resolution

The difference between the two is computed and the distribution of the pixel GNDVI values is represented in Fig. 13 and is plotted in Fig. 14: the distribution presents an average value of − 0.0137 and a standard deviation of 0.0424.

Fig. 13
figure 13

Graphic representation in pixels of the modulus of difference between the two technologies

Fig. 14
figure 14

Pixel Differences representation between ATEMO and Satellite

This operation accounts for the differences in the acquisition methods, e.g. cloud coverage, atmospheric absorption and different camera sensitivities.

5 Conclusions

Looking at the results, ATEMO confirms all functional requirements, thus managing to acquire shots of the fields below with a high resolution. The tethered balloon solution proved to be an excellent solution in several aspects, above all the stationary time and the autonomy even in the altitude; it shall be mentioned that the latter needs some correction in case of wind gusts but is very stable in calm conditions.

The device is also functional when it comes to calculating vegetation indices; and for this reason it is noticeable that the effect of differential irrigation is very low. This is not due to the ability of the soybean to withstand water stress, but rather, by appropriately cross-referencing the seasonal rainfall data, it is due to the large quantities of water that, especially in the Castelfranco Veneto field, resulted in an almost total cancellation of irrigation at 70% in the half of the field under stress.

The ground temperature can also provide important information on the analyses conducted and also anticipate future uses in the night-time field, which, also for the other analytes for which ATEMO was built, will arise in the coming years of research.

In addition to what has been observed via balloon, confirming the usefulness and consistency of the data collected, it can be seen that via satellite the results are compatible with what has been done. Another important note is the slight discrepancy that exists between the average value of the satellite and balloon observations: this is due to the automatic adjustments that are made from satellite (removal of atmospheric disturbances and reflections) which do not follow the same set-up as the calibrations made for Basler and described in Section “Design of ATEMO experiment”. However, although there is this shift, the correlation between satellite and balloon data can be found. In the future, it could prove useful to use the same correction factor used by balloon for satellite data as well, which considers more variables such as, for example, reflections from soybean leaves.

Last, it should be emphasized that in the cases of the more economically accessible satellite services, the quality and resolution of satellite images is lower (4 m for services such as Planet and 0.0042 m for ATEMO mounted on a wire balloon). In the case of the more expensive satellites, on the other hand, again looking for the evaluation of vegetation indices in the visible spectrum, less frequent passes and, in some cases, poor visibility due to cloudy conditions, result in obstacles to clear vision on specific dates of interest.

It can be concluded that, with a view to future analysis, the use of ATEMO can be of extreme interest for an in-depth analysis during a satellite observation campaign, reducing costs by using a compact, inexpensive device that adapts to the user’s needs.