A new, low-cost ground-based NDVI sensor for manual and automated crop monitoring

Keywords

Proximal sensor; Sensor calibration; Spectral vegetation indices; Open-source agriculture technology

Abstract

Ground-based normalized difference vegetation index (NDVI) sensors are vital for accurate, localized crop condition and growth assessments, but their high cost and labor-intensive operation limit accessibility. To close this gap, this study presents a low-cost NDVI sensor priced under €250, offering an affordable yet high-accuracy crop monitoring tool. The device has dual functionality, operating in both manual (handheld) and automatic (standalone) modes, enabling continuous crop monitoring with higher temporal resolution and reduced labor costs. This study also identified and corrected the underestimation of measurements at higher NDVI values through sensor calibration. Subsequent field validation proved the accuracy of the low-cost sensor, showing a generally good overall agreement with results obtained with the reference sensor (r² = 0.99) after applying the derived calibration function. Extended field trials in Benin and the Philippines demonstrated the reliability of the device to adequately monitor treatment differences in various crop development and biomass accumulation. Further customization into automatic mode enabled continuous, high-frequency NDVI measurements, showing its ability to monitor crop phenological changes, such as senescence, in an additional field testing in Germany. Overall, this study demonstrates that the developed NDVI sensor device, made from affordable, off-the-shelf components, can be adapted into a scientifically usable NDVI sensor that is accurate, reliable, and cost-effective. It offers a viable alternative to expensive in-field monitoring systems and promotes accessibility to ground-based crop monitoring solutions, especially for research in the Global South.

1. Introduction

Spectral vegetation indices, like the Normalized Difference Vegetation Index (NDVI), play an important role in monitoring crop health and predicting yields [28]. The early detection of crop stress using indicators like the NDVI allows for timely interventions, preserving both yield and quality, thereby, contributing significantly to food security [54]. Moreover, the utilization of NDVI data facilitates informed decisions related to plant growth and nutrient status, particularly in terms of biomass nitrogen (N) estimation and yield prediction [6,18,32,52]. This valuable information from NDVI supports the optimization of irrigation and fertilizer management practices in the field [29].

NDVI data enables efficient monitoring of agricultural practices, delivering timely crop information essential for well-informed decision-making. NDVI measurements obtained from remote sensing, such as satellite or drone-based platforms, offer broad spatial coverage, enabling large-scale monitoring of vegetation health and land cover changes [45, 54]. However, these measurements may be limited due to their temporal resolution and atmospheric interferences, such as cloud cover [38,40]. Ground-based NDVI approaches, in particular, offer highly accurate and frequent data, addressing some of these limitations and thus can complement remote sensing data [23,47,51]. Several studies have emphasized the significance of ground-based vegetation spectral reflectance indices, including NDVI, in improving models for predicting crop traits, and have shown that combining satellite and ground-based NDVI data provides complementary phenological information, applicable to various ecosystems [13,15].

Ground-based NDVI measurements also offer cost advantages over remote sensing in localized applications. However, commercially-available ground-based sensors can still be expensive. In response to this, there has been a notable increase in the development of low-cost devices [3,10,14,30,41,48], aiming to democratize access to NDVI measurement technology.

Low-cost off-the-shelf sensor devices are increasingly explored due to their affordability, ease of deployment, and customizability. Several studies develop low-cost devices that are compatible with open-source software and modular hardware, allowing greater customization and wide applicability [12,19,31,34,58]. Notably, advances in the Internet of Things (IoT) and microcontrollers, such as the Node Microcontroller Unit (NODE MCU) and Raspberry Pi, have significantly enhanced their capabilities. These advancements enable real-time building monitoring and the tracking of various environmental parameters, such as temperature and energy efficiency [36]. This versatility enables low-cost off-the-shelf devices to be deployed across a wide range of applications. In particular, their use in vegetation monitoring offers an accessible yet accurate solution for enhancing crop monitoring and field management through indices such as NDVI, which is essential for assessing crop health and growth [17,24,25]. These devices offer cost-effective alternatives to high-cost devices, and allow field monitoring and automation technologies to be more accessible particularly for researchers in developing regions. Although some studies have demonstrated the accuracy of low-cost NDVI sensors [30,48], long-term field testing to ensure their reliability and robustness for frequent crop health and development monitoring remains limited. Additionally, low-cost sensors used for in-field, ground-based NDVI measurements are still labor-intensive because they are manually operated. A significant challenge is that most automatic ground-based NDVI sensor systems with high accuracy are primarily embedded within costly robotic platforms [53,56] or are based on commercial NDVI sensors [16]. Hence, an automatic, cost-effective NDVI ground-based sensor, especially one that is both affordable and portable, presents a compelling alternative. Such a device has the potential to democratize crop monitoring by providing affordability, scalability, and dependable long-term observations. Automated ground-based NDVI sensing provides a solution to not only increase the measurement frequency but also minimizing the need for human intervention. Despite multiple studies demonstrating accuracy of low-cost NDVI devices, an automatic, low-cost NDVI sensor with demonstrated accuracy and longer-term reliability remains scarce. Furthermore, there is a notable lack of low-cost NDVI sensors that have been validated through field applications over extended periods (e.g., > six months), particularly in regions such as the Global South, where these technologies could provide substantial benefits. This gap underscores the need for further studies to develop and validate affordable automated crop monitoring tools that can be deployed in the field for months. To fill that gap, this study introduces a low-cost NDVI sensor that functions as both a handheld manual and as an automatic stand-alone device.

We hypothesize that 1.) the developed low-cost sensor device can accurately measure NDVI and that it is 2.) applicable for longer-term measurements in the Global South. Additionally, we propose that 3.) it can function as an automatic sensor device for continuous and reliable NDVI measurements. To test these hypotheses, we first aim to assess its accuracy against a commercial reference sensor in a calibration and field validation experiment. We then aim to evaluate its applicability for longer-term NDVI measurements in two field trial experiments in the Philippines and Benin. Finally, we aim to determine its reliability as an automatic sensor device through a field application, performed in Germany.

2. Materials and methods

2.1. NDVI determination

NDVI can be obtained by measurements of the spectral reflectance in the red and near infrared (NIR) waveband region and calculated following Johansen and Tommervik [26] as follows:

NDVI = (ρ_NIR - ρ_RED) / (ρ_NIR + ρ_RED) (1)

where ρ_RED is the reflectance of a red band around 660 nm and ρ_NIR is the NIR band around 860 nm. NDVI values, usually range between –1 and +1 and provide an assessment of the crop health, cover and density as well as vegetation dynamics [8,37,50]. Negative values typically signify sparse or non-existent vegetation cover, whereas higher positive NDVI values indicate the presence of productive green vegetation [26]. NDVI values can thus be used as a proxy for vegetation productivity and biomass [42].

2.2. Hard- and software implementation

The developed low-cost NDVI sensor consists of a SparkFun Red-Board microcontroller (SparkFun Electronics, Colorado, USA) connected to a micro SD card (Qwic OpenLog, SparkFun Electronics, Colorado, USA), and a real-time clock module (RTC RV-1805 I2C, SparkFun Electronics, Colorado, USA) for data storage and accurate time and date tracking. An I2C multiplexer (TCA9548A, SparkFun Electronics, Colorado, USA) is used to allow for connecting two identical spectral sensor modules (AS7263 NIR, SparkFun Electronics, Colorado, USA). The AS7263 is a digital 6-channel spectrometer that has 6 independent optical filters, with spectral responses defined from approximately 600 nm to 870 nm and a full-width half-maximum (FWHM) of 20 nm. One AS7263 sensor was placed facing upward and covered with a protective PTFE Teflon plate, acting as diffuser (thickness: 2 mm). The second AS7263 sensor was placed facing downward and not covered, to measure the reflectance from the ground and vegetation cover. The entire device was powered by 2AA rechargeable NiMH batteries connected to a voltage regulator (MT3608 DC-DC Adjustable step-up voltage regulator; CFStmbird, China) that ensured consistent 5 V energy supply to the device. For real-time monitoring of measurement data, the device is additionally equipped with a Bluetooth module (HC-06 Bluetooth Wireless RF, AZ-Delivery Vertriebs GmbH, Germany) allowing for connection to any Android device. To ensure consistent sensor height during every measurement, in its manual mode, the device was mounted on a 1.8 m pole constructed from metal tubes. Fig. 1A shows the assembled low-cost sensor in manual mode together with a schematic representation of its wiring. Based on results of the manual device, an improved automatic device (Fig. 1B) was developed. The top view of the manual sensor device is shown in Fig. 1C, while Fig. 1D illustrates the assembly of individual components to give a more detailed understanding of the overall assembly process. To minimize the power consumption of the automatic device, the manual device setup was changed as follows: 1.) the Arduino UNO like microcontroller was replaced by a Pro mini microcontroller, 2.) instead of two, four AA rechargeable NiMH batteries were used, thus avoiding the need for a power consuming voltage regulator, 3.) a low-power timer with an integrated MOSFET driver was incorporated (TPLS110; Adafruit Industries, USA). For protection, the system was placed in a weather-proof plastic housing. This was then set-up on a tripod, creating a standalone automatic sensor device suitable for field deployment (Fig. 1E). The hardware specifications and description of each component are summarized in Table 1 for manual handheld mode and in Table 2 for automatic standalone mode. Additionally, Tables 1 and 2 summarize the total cost of the manual and automatic sensor devices which were calculated by summing the individual cost of components used. The scripts used for the manual and automatic measurement mode, respectively, were developed and uploaded to the device using open-source Arduino Software (IDE) version 1.8.15.

Table 1. Specification of components and corresponding costs (in Euro) at time of writing, including sensor housing, and energy supply for the manual mode low-cost NDVI sensor device.

Table 2. Specification of components and corresponding costs (in Euro) at time of writing, including sensor housing, and energy supply for the automatic low-cost NDVI sensor device.

2.3. Sensor calibration

To test the accuracy of the low-cost sensor, a sensor calibration was performed by measuring the reflectance of various surfaces. In total, nine different colored cardboards (black, blue, brown, dark green, light green, orange, violet, red and white; 594 mm × 840 mm) were measured by the developed low-cost sensor and, as a standard, a commercial reference sensor in parallel. The reference sensor device consists of two 2-canal SKR1840 spectral sensors (Skye Instruments Ltd., UK) measuring reflectance in the red and NIR band at 656 ± 10 nm and 780 ± 10 nm, respectively. One of the two SKR1840 spectral sensors is upward directed and equipped with a cosine-correction diffuser for incident radiation, while the second sensor without diffusor is downward directed to measure the reflectance of the RED and NIR band of a 25° cone field of view. Both sensors were connected to a CR1000 data logger (Campbell Scientific Ltd., USA) for data storage. The reference sensor was inter-alia used in Schaller et al. [43]. Both, the low-cost and reference sensor, were fixed at a height of 50 cm above the colored cardboards to ensure that the measured area is fully covered by the colored cardboards. Since both, the developed low-cost and the reference sensor are passive NDVI devices, measurements were similarly taken by detecting sunlight reflected from the surface, without the need to emit their own radiation. NDVI measurements were taken outdoors between 11 am and 2 pm, when the sun was at its highest angle during a cloud free day on May 26, 2023 (ZALF, Müncheberg, Germany, 52° 30′ 57.2″ N 14° 06′ 46.4″ E). Measurements of both sensors for all nine colored cardboards were directly compared against each other to check for the low-cost sensor device measurement accuracy. Based on this, a calibration function was derived, used to adjust later on measurements in the field validation and field trial application for potential deviation of NDVI measurements by the low-cost sensor from the reference sensor.

2.4. Field validation

To further evaluate the accuracy of measurements from the low-cost device under real-life conditions, a field validation experiment was performed. This was done through parallel in situ measurements of various crops using the low-cost and reference SKR1840 spectral sensor device (Skye Instruments Ltd., UK). Measurements were conducted at the “PatchCrop” landscape laboratory of the Leibniz-Centre for Agricultural Landscape Research (ZALF). The “PatchCrop” landscape laboratory contains of a unique field site setup of multiple smaller patches (72 × 72 m), each under different crop rotation and management, with the aim to reduce e.g., the use of synthetic, chemical pesticides and fertilizers. It is located near the village of Tempelberg, Northeast Germany (52° 26,827′ N, 14° 8492′ E) and has a temperate climate characterized by a mean annual air temperature of 9.7°C and mean annual precipitation of 544 mm (ZALF weather station, 2010–2019). The medium loamy, sand textured soil can be classified as Luvisol (WRB). NDVI measurements were conducted on August 10, 2022 at four different patches along a transect with 10 measurement points per patch ranging from the edges to its center. The four patches were chosen to represent various surfaces (bare soil vs. crop cover) and crops (maize vs. sunflower vs. soybean) each at a different development stage. To ensure accurate measurements at each plot, both sensors were positioned in close proximity to one another, ensuring they measured from the same plot area. Each measurement was done in parallel with both the low-cost and reference sensors for 30 consecutive seconds.

2.5. Field trial application

To test sensor handling and longer-term reliability, the low-cost, handheld sensor device (in manual mode) was subsequently used for an entire crop season at an experimental field trial in Calauan, Laguna, Philippines (14° 7′ 51.3444″ N, 121° 18′ 37.425″ E) and at the Koussin Lele irrigated scheme (7° 14′ 0.848303″ N, 2° 16′ 42.625102″ E), in the Cove district of southern Benin. The field trial site in Calauan, Laguna, Philippines, experiences a tropical monsoon climate with two distinct seasons: dry season from November to April and wet season from May to October [7]. The field trial, established as an on-farm experiment under pineapple cultivation (Ananas comosus (L.) Merr.), adopted a randomized block design featuring four different treatments. Treatments included pineapple residue incorporation into the soil without fertilizer application (PR only), pineapple residue incorporation with mineral fertilizer application (MIN+PR), pineapple residue incorporation with vermicompost organic fertilizer (Arnold and Paz Organic Farm, Laguna, Philippines) application (ORG+PR), and mineral fertilizer application only (MIN only). Monthly NDVI measurements were typically taken between 3 PM and 5 PM (GMT+8) from March 10, 2022, to June 8, 2023, with the device placed at the edge of pre-inserted frames, each containing one pineapple plant (Fig. 2A). In order to determine the utility of the NDVI measurements, the relationship between NDVI and dry biomass weight was determined. For this purpose, repeated biomass samples were conducted over the crop growth period. Harvested biomass samples were oven dried at 70°C for at least 48 h to obtain dry biomass weight.

The field trial site at the Cove district in southern Benin features a transitional climate [44] with two rainy and two dry seasons [1]. The soil in the Koussin Lele irrigated scheme is classified as Gleysois [57]. The field trial adopted a split-plot design with water management as the main plot treatment and high (120 kg N ha^-1) and low (90 kg N ha^-1) mineral N fertilizer levels as the subplot treatment. Water management methods included continuous flooding (CF) and alternate wetting and drying irrigations at a 25 cm threshold (AWD). Each plot was 20 m² in size with specific distances between main plots and subplots. To prevent N movement and lateral flows, a 200-micron plastic layer was placed around the levees. Biweekly NDVI measurements for rice (Oryza sativa) were taken from November 18, 2022, to February 16, 2023. The device was positioned at the center of pre-installed frames covering four rice hills each. Measurements were typically carried out between 9 and 11 AM (CET) (Fig. 2B). Identical to the field trial application in the Philippines, NDVI measurements were aligned with measurements of dry biomass weight, sampled during the crop growth period. To obtain dry biomass weight, harvested biomass samples were oven dried at 70 °C for 72 h.

Similar to the manual handheld low-cost sensor device, also its automatic standalone version was field tested at the aforementioned “PatchCrop” landscape laboratory. To test its capability for continuous NDVI measurements, the sensor was placed from July 6 to July 26, 2023 on a tripod at a fixed location in an oat field patch. Subsequently, the sensor automatically measured NDVI for 30 seconds at a 20 min interval during the entire 20 days measurement period.

2.6. Data analyses

NDVI was calculated from the spectral measurements obtained from the reference and low-cost sensor following equation 2:

NDVI = (ρ_{NIR_down} / ρ_{NIR_up} - ρ_{RED_down} / ρ_{RED_up}) / (ρ_{NIR_down} / ρ_{NIR_up} + ρ_{RED_down} / ρ_{RED_up}) (2)

where, ρ_NIR is the measurement at 780 ± 10 nm and ρ_RED is the measurement at 656 ± 10 nm from up- and downward sensor of reference sensor, respectively. For the low-cost sensor device, the average of measurements between 760 ± 33 nm and 810 ± 33 nm for ρ_NIR, and at 660 ± 33 nm for ρ_RED were used. The accuracy of measurements from the low-cost sensor was assessed by determining measurement trueness and precision [35]. The trueness of the low-cost sensor measurements was evaluated by comparing NDVI values obtained by the low-cost sensor and the reference sensor during sensor calibration and field validation experiments. To test for significant differences (p < 0.05) between NDVI values measured by both sensors, the Wilcoxon Signed-Rank Test (One-Sample Wilcoxon Test) was used. A non-parametric test was performed after applying Shapiro-Wilk test (p < 0.05) to check for normal distribution. Precision was evaluated by assessing the spread of the 30 s consecutive NDVI measurements (standard deviation; SD) from the low-cost sensor and comparing it against the SD of the reference sensor. All statistical analyses were conducted using R version 4.2.2 within the RStudio integrated development environment (RStudio 1.4.1106).

3. Results and discussion

3.1. NDVI sensor calibration

During sensor calibration, similar spectral reflectance ratios, and thereon calculated NDVI, were recorded for the different colored cardboard surfaces by the low-cost and reference sensor, respectively (Fig. 3). Several studies emphasize the crucial role of calibrating spectroscopy data to known surface reflectance for ensuring accuracy [4,27, 46]. Calculated NDVI values of the low-cost sensor were able to show differences between darker green (NDVI = 0.46 ± 0.01) and lighter green (NDVI = 0.32 ± 0.01) colored cardboard surfaces. This was found to be consistent with the results of the laboratory validation performed by Kitić et al. [30]. NDVI calculated from spectral measurements using their developed low-cost multispectral optical device also showed higher NDVI for dark green (between 0.40 and 0.65) and lower NDVI for lighter green (between 0.20 and 0.30) target surfaces. Fig. 3 also shows a comparably low SD within each NDVI measurement for the low-cost (SD = 0.008) and reference sensor (SD = 0.006), demonstrating their similar good precision. Furthermore, no significant difference between calculated NDVI values of the low-cost and reference sensor was obtained (Wilcoxon Signed-Rank Test, p = 0.11). Hence, in general, the low-cost sensor demonstrated acceptable accuracy in measuring spectral reflectance from different colored cardboard surfaces, indicating its ability to capture variations in NDVI, which is crucial for effectively monitoring crop health. However, noticeable slight underestimation was observed, with the low-cost sensor NDVI values deviating from the reference sensor at higher NDVI values (Fig. 3). This observation was also noted by Stamford et al. [48], who found an underestimation at higher NDVI values in their low-cost NDVI imaging system in a validation experiment. Their study emphasized that, despite generally lower NDVI values obtained from their low-cost imaging system, its higher sensitivity at lower NDVI values is beneficial for discriminating chlorophyll content in leaves and plant greenness. To correct for the observed slight deviation at higher NDVI, we hence derived a calibration function through fitting a linear regression on the NDVI values of the 1:1-agreement plot (Fig. 3), thus ensuring consistency and comparability of measured NDVI values. As demonstrated in studies by Crain et al. [10] and Stamford et al. [48], sensor calibration ensures accuracy also during extended periods of use. The NDVI calibration function derived during sensor calibration was subsequently applied to all further NDVI measurements using the presented low-cost device.

3.2. Field validation

A good overall agreement was found between NDVI values measured during field validation by the low-cost and reference sensor, respectively (Fig. 4). The highest average NDVI (± SD) was found for soybean (0.81 ± 0.08), followed by sunflowers (0.45 ± 0.13), maize (0.31 ± 0.11), and then bare soil (0.03 ± 0.01). These results are in alignment with the condition of the plants during the time of measurement, as the soybean crops had the most dense and verdant foliage, while the maize crops were already wilting and showing partly brownish hues. Nevertheless, similarly to results obtained during the sensor calibration experiment an underestimation at higher NDVI values was observed when NDVI values of the low cost sensor where compared to those of the reference sensor. Thus, during field validation, NDVI values measured using the low-cost sensor were found to be significantly less than the NDVI values measured with the reference sensor (Wilcoxon Signed-Rank Test, p<0.05). However, after applying the derived NDVI calibration function, no significant difference from measurements obtained from both sensors was found (Wilcoxon Signed-Rank Test, p = 0.19) and the regression line using the adjusted NDVI measurements, falls rather close to the 1:1-agreement line. Both indicate the trueness of measured NDVI using the low-cost sensor (Fig. 4). The precision of the low-cost sensor measurements was demonstrated by the low SD of the repetitive NDVI records (n = 30) per measurement point (ranging from 0.001 to 0.090) falling close to those of the reference sensor (ranging from 0.001 to 0.081). Both together, the good 1:1 agreement and low SD of repetitive NDVI measurement records indicates the overall good accuracy of the presented low-cost sensor. This is further substantiated by measured, crop specific NDVI values being comparable to values presented in literature using satellite, drone or commercial handheld sensor devices. For example, in the case of soybean and sunflower, NDVI values ranging from 0.80 to 0.90 are reported during the highest crop vegetative growth period using satellite images by Andrade et al. [2] and Stepanov et al. [49]. Yu and Shang [59] also utilized satellite images, showing NDVI values for sunflower crops ranging from 0.35 to 0.45. Using optical sensor measurements, the study by Martin et al. [33] also reported lower NDVI values for maize during the end of the growing stage, ranging from 0.2 to 0.4, which align with the conditions observed for maize crops during our field validation experiment. Lastly, in line with our findings, Choudhary et al. [9] noted that the bare soil surface exhibited the lowest NDVI values, attributed to the lack of vegetation cover on the surface. Overall, the field validation experiment showed that the low-cost NDVI sensor device can accurately measure and distinguish NDVI of various crop and bare soil surfaces, pivotal for reliable crop monitoring. Accordingly, the NDVI calibration function applied in the field validation was also used for the NDVI measurements obtained in the field trial application.

3.3. Field trial application in the Philippines and Benin

NDVI values obtained for the Philippine field trial were able to reflect the incremental growth of the pineapple crops under various treatments over the measurement period (Fig. 5A). This is substantiated by a slight positive correlation (r² = 0.48) obtained between dry biomass weight of pineapple plants and NDVI measured at the same date (Fig. 5C). This is in line with a study by Putra et al. [39], who also reported a rather weak positive correlation between NDVI and pineapple biomass, especially when compared to other vegetation indices. A gradual increase in NDVI values was observed in our study from the beginning of the measurement period in March 2022 until flowering in February 2023. This aligns well with the observed development of pineapple crops in the field, which increased substantially in size and vigor during this period. After flowering stage, pineapple crops of all treatments showed a slow decline in NDVI values. The decline observed in measured NDVI following the flowering stage can be attributed to decreasing chlorophyll levels, a connection supported by the findings of Gomez Herrera et al. [20], who also documented a reduction in NDVI after flowering in pineapple leaves. The correlation between reduced chlorophyll content and declining NDVI was also reported by Cui et al. [11]. Measured NDVI values during flowering stage in our study, ranging from 0.29 to 0.50, are furthermore in a good agreement with NDVI values (0.33–0.43) given for pineapple crops during flowering by Putra et al. [39] using UAV.

Similarly to the Philippine field trial, NDVI values obtained at the Benin field trial were also able to clearly reflect the crop growth of the cultivated rice under various treatments over the entire crop growth period (Fig. 5B). This is again substantiated by a high positive correlation (r² = 0.82) between dry biomass weight of rice and same date NDVI measurements (Fig. 5D). In general, both investigated factors, water level and fertilizer amount, showed a clear difference in NDVI, with the water level having a greater effect (Fig. 5B). While continuous flooding resulted in a higher NDVI, reduced fertilizer application resulted in a lower NDVI. Thus, it was found that the treatment with the high MIN fertilizer applied (120 kg N ha^-1) and continuous flooding for water treatment (CF) reflected the highest NDVI throughout the measurement period. Our results align with the study by Kimaro et al. [29] who investigated the effects of different fertilizer soil amendments and irrigation water management on rice production in Tanzania. They reported a close relationship (r² = 0.82) between rice NDVI and dry matter accumulation and found that the NDVI was higher under continuous flooding irrigation than under the system of rice intensification. In addition to this, Guan et al. [22] also found that NDVI values increases for rice with higher nitrogen application, as a result of increased biomass production. In our study, we observed an increase in NDVI during the rice growing and tillering stage (November to December 2022; Fig. 5B). The maximum NDVI was measured on January 03, 2023 during the heading stage with 0.68 ± 0.02 for MIN high (CF) treatment. In contrast, the lowest NDVI with 0.42 ± 0.01 was observed during heading stage for alternative wetting and drying (AWD) treatment with MIN low (90 kg N ha^-1). During rice maturity, a decline in NDVI for all treatments was found. The decrease in NDVI observed in our study was consistent with the reported decline in NDVI for rice during maturity (from 0.6 to 0.1) presented in a study by Gonzalez-Betancourt and Mayorga-Ruiz [21]. These results show that the low-cost sensor not only differentiated well between the corresponding effects of different water and fertilizer treatments, but also reflected temporal dynamics in crop growth. Overall, the field trial experiments in the Philippines and Benin demonstrated that the developed low-cost NDVI sensor is a reliable tool for longer-term NDVI measurements and crop health monitoring.

Gained experience during its application in the field furthermore demonstrated the device's ease of use due to its lightweight design and straightforward operation. Additionally, real-time calculation of NDVI values and their convenient display on any Android device through the attached Bluetooth module contributed greatly to the overall user-friendly functionality of the device. Based on field trials performed in the Philippines and Benin, measurement of an experimental site with 12 measurement plots require at most an hour, including walks in between plots within the field site. During this time, the fully-charged batteries used did not require change or charging of batteries to complete the measurement in the field. Since the developed low-cost device can be powered for up to 8 h efficiently by only 2 AA NiMH batteries, the amount of times needed for battery change and recharging was low. This is extremely beneficial to areas without access to charging stations near the field and especially suitable for use in areas with limited availability of electricity and charging options. Thus, the device open manifold possibilities in terms of a more accessible research tool especially in the Global South. Irrespective of that, the manual measurements were still labor intensive which restricted their frequency and thus temporal measurement resolution.

3.4. Field trial application of the automatic sensor device

Measurements from the low-cost NDVI sensor as an automatic device reflected well the temporal dynamics of the oat crops during maturity. Fig. 6 shows the gradual decrease in NDVI of oat as it goes through senescence. The decline in NDVI for oats during senescence can be attributed to the discoloration of oat leaves, which is primarily due to a decrease in chlorophyll content. The highest average daily NDVI (±SD) measured with 0.51 ± 0.01 was obtained at the beginning of the measurement period. This was consistent with the much greener state of the oat crops on July 06, 2023. Over time, a decline in crop vigor and a shift towards a more yellow hue was observed, resulting in a 62 % decrease in average daily NDVI values to 0.20 ± 0.01 by July 26, 2023. This decrease is consistent with reported decline in measured NDVI (0.60 to 0.10) for oat during ripening and yellowing stage reported in the study by Wittich and Kraft [55] performed at a field site in Braunschweig, Germany. Furthermore, the study by Bytyqi and Kutasy [5] observed lower NDVI (0.34–0.38) for various oat varieties measured during senescence. In view of these results, the automatized low-cost NDVI sensor accurately reflected oat crop maturity over the measurement period, highlighting its suitability for continuous crop monitoring in the field. In terms of practical use, the automatic mode was found to be more efficient than manual operation, since it can provide higher frequency of measurements with less human labor. It was also energy efficient, with rechargeable batteries that can last up to 3 months before needing replacement. Furthermore, the device demonstrated its robustness as it was able to withstand velocity wind (8.71 m s^-1) and a reported thunderstorm event on the 24th of July 2023 (ZALF weather station, 2023). The measurement remained stable during the extreme weather event, with the SD (SD = 0.03) falling within the overall range (SD = 0.004–0.04) of calculated SD across all measurements. Hence, the developed automatic low-cost sensor device proved not only its reliability in providing accurate continuous NDVI values, but also its field readiness. However, a limitation encountered when using the automatic sensor device was that the data extraction has to be done manually. This means that the device must be turned off for some time while the data is being extracted from the device and transferred to the computer or separate storage disk. Additionally, the automatic low-cost sensor device was only tested to monitor the senescence of the oat crop, limiting its evaluation to a single phase of crop development. The automatic device's performance in detecting other critical growth stages, such as germination, vegetative growth, and flowering, was not assessed. While the manual mode of the device has already demonstrated its capabilities in this regard, future studies can assess the automatic sensor's long-term performance and versatility across different crop stages, including its ability to provide consistent and reliable data throughout the entire crop cycle. Overall, the developed automatic sensor device can still be further improved to enhance remote monitoring and access to measurement data for an extended period of time. Here, the modular design of the device is beneficial. It can be easily further customized by adding a wireless communication modules, such as Wi-Fi or LoRaWAN wireless network. These modules might then enable the device to transmit data to cloud platforms or remote servers, where it can be accessed and analyzed in real-time through web or mobile applications. This connectivity will allow users to monitor NDVI remotely, while in remote areas, the version presented here with local data storage might still be used. Building on this study, it is recommended to evaluate various low-cost spectral sensors and their accuracy in deriving other vegetation indices to further enhance the understanding of their performance and applicability. In addition to this, integration of additional low-cost sensors (e.g., distance sensor for plant height measurement) to enhance data collection and provide a more comprehensive assessment of crop development can be explored. Future studies could also expand the customizability of the device by utilizing 3D printing to enhance the structural adaptability and scalability of the device. Lastly, it is recommended to investigate whether deploying multiple low-cost, automatic standalone sensors over an extended period can enable real-time mapping of vegetation health across a field. Accordingly, a cost-benefit analysis of the mass production of these low-cost devices should be conducted to evaluate their viability for large-scale deployment in various agricultural fields.

4. Conclusion

Our study proved the reliability of the developed low-cost NDVI sensor (priced < 250 Euros) for cost-effective and accurate ground-based measurements of NDVI values across various crops. The importance of performing sensor calibration when using low-cost devices to improve measurement accuracy was also demonstrated. The field trials conducted in the Philippines and Benin underscored the device's robustness for extended monitoring across diverse agricultural landscapes, making it an accessible and valuable device that can support agricultural research especially in the Global South. Given its open-source nature, the device is in addition, easily customizable for various applications. This is demonstrated, for example, by its further development towards automation aiming at continuous NDVI measurements, thus offering advantages of a higher temporal measurement resolution and reduced labor-intensity. This makes the presented low-cost sensor an ideal practical solution for the long-term monitoring of crops as part of inter-alia experimental field trials with numerous treatment comparisons. Overall, our study hence enriches the understanding of the accuracy, applicability, and reliability of low-cost NDVI sensor devices. With its affordability, open-source design, and demonstrated efficacy in diverse agricultural settings, this device emerges as a valuable tool for advancing ground-based crop monitoring solutions, contributing to the democratization of access to crucial agricultural technologies.

Ethics statement

Not applicable: This manuscript does not include human or animal research.

CRediT authorship contribution statement

Reena Macagga: Writing - review & editing, Writing - original draft, Visualization, Methodology, Investigation, Conceptualization. Geoffroy Sossa: Writing - review & editing, Visualization, Formal analysis, Data curation. Yvonne Ayaribli: Writing - review & editing, Methodology, Investigation. Rinan Bayot: Writing - review & editing, Visualization, Methodology, Formal analysis, Data curation. Pearl Sanchez: Writing - review & editing, Supervision, Project administration, Methodology, Conceptualization. Jürgen Augustin: Writing - review & editing, Validation, Supervision, Conceptualization. Sonoko Dorothea Bellingrath-Kimura: Writing - review & editing, Validation, Supervision. Mathias Hoffmann: Writing - review & editing, Visualization, Supervision, Project administration, Methodology, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was funded by a grant (2819DOKA06) from the German Federal Ministry of Food and Agriculture (BMEL). Geoffroy Sossa was supported by the West African Science Service Center on Climate Change and Adapted Land Use (WASCAL), the Prince Albert II of Monaco Foundation, and the International Foundation for Science (grant no. I-1-C-6642-1, IFS). The maintenance of the “PatchCrop” infrastructure is supported by the Leibniz Centre for Agricultural Landscape Research (ZALF). We would also like to extend our deepest gratitude to Natalija Obradović for helping us in the sensor calibration experiment. Special thanks also to Shukrona Giyasidinova, Mario Belío Miranda, and Isabel Zentgraf for assisting us during the field validation experiment. Lastly, we would like to thank our partner farmers Mr. Rodrigo Cachuela and Mr. Michael Alcantara for supporting us in the field trial experiment in the Philippines.

Data availability

https://doi.org/10.4228/zalf-db3a-6q09 (The data and code referred to in this study are publicly accessible at)

References

1. S. Aliou, G.K. Denis, B. Ibouraïman, H. Romain, K. Valentin, Effet de l’urée et du NPK 15-15-15 perlés et super granulés sur la productivité des variétés de riz IR841 et NERICA-L14 en zone de bas-fond au Sud-Bénin (In French), J. Appl. Biosci. 77 (2014) 6575–6589.
2. T.G. Andrade, A.S.D. Andrade Jr, M.O. Souza, J.W.B. Lopes, P.F.D.M.J. Vieira, Soybean yield prediction using remote sensing in Southwestern Piauí State, Brazil, Revista Caatinga 35 (2022) 105–116, https://doi.org/10.1590/1983-21252022v35n111rc.
3. C. Barrows, D. Bulanon, Development of a low-cost multispectral camera for aerial crop monitoring, J. Unmanned Veh. Syst. Sys (2017), https://doi.org/10.1139/juvs-2017-0008 juvs-2017-0008.
4. W.M. Baugh, D.P. Groeneveld, Empirical proof of the empirical line, Int. J. Remote Sens. 29 (2008) 665–672, https://doi.org/10.1080/01431160701352162.
5. B. Bytyqi, E. Kutasy, Leaf reflectance characteristics and yield of spring oat varieties as influenced by varietal divergences and nutritional supply, Acta Agraria Debreceniensis (2023) 29–34, https://doi.org/10.34101/actaagrar/1/12144.
6. L. Cabrera-Bosquet, G. Molero, A. Stellacci, J. Bort, S. Nogués, J. Araus, NDVI as a potential tool for predicting biomass, plant nitrogen content and growth in wheat genotypes subjected to different water and nitrogen conditions, Cereal Res. Commun. 39 (2011) 147–159, https://doi.org/10.1556/CRC.39.2011.1.15.
7. J.G. Cadondon, E.A. Vallar, A.B. Beltran, A.H. Orbecido, M.C.D. Galvez, Variation in dissolved organic matter using absorbance and fluorescence measurements during dry season in sta. Rosa and Cabuyao Rivers, Philippines, Water 14 (2022) 1444, https://doi.org/10.3390/w14091444.
8. B. Chen, G. Xu, N.C. Coops, P. Ciais, J.L. Innes, G. Wang, R.B. Myneni, T. Wang, J. Krzyzanowski, Q. Li, L. Cao, Y. Liu, Changes in vegetation photosynthetic activity trends across the Asia–Pacific region over the last three decades, Remote Sens. Environ. 144 (2014) 28–41, https://doi.org/10.1016/j.rse.2013.12.018.
9. K. Choudhary, W. Shi, M.S. Boori, S. Corgne, Agriculture phenology monitoring using NDVI time series based on remote sensing satellites: a case study of Guangdong, China, Opt. Mem. Neural Netw. 28 (2019) 204–214, https://doi.org/10.3103/S1060992X19030093.
10. J. Crain, I. Ortiz-Monasterio, B. Raun, Evaluation of a reduced cost active NDVI sensor for crop nutrient management, J. Sensor. 2012 (2012) 1–10, https://doi.org/10.1155/2012/582028.
11. D. Cui, M. Li, Q. Zhang, Development of an optical sensor for crop leaf chlorophyll content detection, Comput. Electron. Agricult. 69 (2009) 171–176, https://doi.org/10.1016/j.compag.2009.08.001.
12. D.D. Drajić, N.R. Gligorić, Reliable low-cost air quality monitoring using off-the-shelf sensors and statistical calibration, Elektronika Ir Elektrotechnika 26 (2) (2020) 32–41, https://doi.org/10.5755/j01.eie.26.2.25734.
13. A.H. Elmetwalli, A.N. Tyler, F.S. Moghanm, S.A.M. Alamri, E.M. Eid, S. Elsayed, Integration of radiometric ground-based data and high-resolution quickbird imagery with multivariate modeling to estimate maize traits in the Nile Delta of Egypt, Sensors 21 (2021), https://doi.org/10.3390/s21113915.
14. F.J. Fernández-Alonso, Z. Hernández, V. Torres-Costa, A cost-effective portable multiband spectrophotometer for precision agriculture, Agriculture 13 (2023), https://doi.org/10.3390/agriculture13081467.
15. G. Filippa, E. Cremonese, M. Migliavacca, M. Galvagno, O. Sonnentag, E. Humphreys, K. Hufkens, Y. Ryu, J. Verfaillie, U. Morra di Cella, A.D. Richardson, NDVI derived from near-infrared-enabled digital cameras: applicability across different plant functional types, Agricult. For. Meteorol. 249 (2018) 275–285, https://doi.org/10.1016/j.agrformet.2017.11.003.
16. J.A. Gamon, O. Kovalchuck, C.Y.S. Wong, A. Harris, S.R. Garrity, Monitoring seasonal and diurnal changes in photosynthetic pigments with automated PRI and NDVI sensors, Biogeosciences 12 (2015) 4149–4159, https://doi.org/10.5194/bg-12-4149-2015.
17. S.R. Garrity, L.A. Vierling, K. Bickford, A simple filtered photodiode instrument for continuous measurement of narrowband NDVI and PRI over vegetated canopies, Agricult. For. Meteorol. 150 (3) (2010) 489–496, https://doi.org/10.1016/j.agrformet.2010.01.004.
18. J. Geipel, J. Link, J. Wirwahn, W. Claupein, A programmable aerial multispectral camera system for in-season crop biomass and nitrogen content estimation, Agriculture 6 (2016) 4, https://doi.org/10.3390/agriculture6010004.
19. M. Ghamari, H. Kamangir, K. Arezoo, K. Alipour, Evaluation and calibration of low-cost off-the-shelf particulate matter sensors using machine learning techniques, IET Wirel. Sens. Syst. 12 (5-6) (2022) 134–148, https://doi.org/10.1049/wss2.12043.
20. M.D. Gómez Herrera, P. Alayón Luaces, M.V. Avanza, Organic compounds determined at different levels of ripening of the pineapple (Ananas comosus L. Merr.) Cv cayenne in two cultivation systems under subtropical conditions, Int. J. Fruit Sci. 20 (2020) 371–384, https://doi.org/10.1080/15538362.2019.1650409.
21. M. González-Betancourt, Z.L. Mayorga-Ruíz, Normalized difference vegetation index for rice management in El Espinal, Colombia, DYNA 85 (2018) 47–56, https://doi.org/10.15446/dyna.v85n205.69516.
22. S. Guan, K. Fukami, H. Matsunaka, M. Okami, R. Tanaka, H. Nakano, T. Sakai, K. Nakano, H. Ohdan, K. Takahashi, Assessing correlation of high-resolution NDVI with fertilizer application level and yield of rice and wheat crops using small UAVs, Remote Sens. 11 (2019) 112, https://doi.org/10.3390/rs11020112.
23. G. Hmimina, E. Dufrˆene, J.-Y. Pontailler, N. Delpierre, M. Aubinet, B. Caquet, A. De Grandcourt, B. Burban, C. Flechard, A. Granier, P. Gross, B. Heinesch, B. Longdoz, C. Moureaux, J.-M. Ourcival, S. Rambal, L. Saint Andr´e, K. Soudani, Evaluation of the potential of MODIS satellite data to predict vegetation phenology in different biomes: an investigation using ground-based NDVI measurements, Remote Sens. Environ. 132 (2013) 145–158, https://doi.org/10.1016/j.rse.2013.01.010.
24. F.H. Holman, A.B. Riche, M. Castle, M.J. Wooster, M.J. Hawkesford, Radiometric calibration of ‘commercial off-the-shelf’ cameras for UAV-based high-resolution temporal crop phenotyping of reflectance and NDVI, Remote Sens. 11 (14) (2019) 1657, https://doi.org/10.3390/rs11141657.
25. S.Y.Y. Jewan, V. Pagay, L. Billa, S.D. Tyerman, D. Gautam, D. Sparkes, H.H. Chai, A. Singh, The feasibility of using a low-cost near-infrared, sensitive, consumer-grade digital camera mounted on a commercial UAV to assess Bambara groundnut yield, Int. J. Remote Sens. 43 (2) (2022) 393–423, https://doi.org/10.1080/01431161.2021.1974116.
26. B. Johansen, H. Tømmervik, The relationship between phytomass, NDVI and vegetation communities on Svalbard, Int. J. Appl. Earth Observ. Geoinform. 27 (2014) 20–30, https://doi.org/10.1016/j.jag.2013.07.001.
27. E. Karpouzli, T. Malthus, The empirical line method for the atmospheric correction of IKONOS imagery, Int. J. Remote Sens. 24 (2003) 1143–1150, https://doi.org/10.1080/0143116021000026779.
28. L. Karthikeyan, I. Chawla, A.K. Mishra, A review of remote sensing applications in agriculture for food security: Crop growth and yield, irrigation, and crop losses, J. Hydrol. 586 (2020) 124905, https://doi.org/10.1016/j.jhydrol.2020.124905.
29. O.D. Kimaro, S.L. Gebre, P. Hieronimo, N. Kihupi, K.-H. Feger, D.N. Kimaro, Handheld NDVI sensor-based rice productivity assessment under combinations of fertilizer soil amendment and irrigation water management in lower Moshi irrigation scheme, North Tanzania, Environ. Earth Sci. 82 (2023) 78, https://doi.org/10.1007/s12665-022-10730-0.
30. G. Kitić, A. Tagarakis, N. Cselyuszka, M. Panić, S. Birgermajer, D. Sakulski, J. Matović, A new low-cost portable multispectral optical device for precise plant status assessment, Comput. Electron. Agricult. 162 (2019) 300–308, https://doi.org/10.1016/j.compag.2019.04.021.
31. T. Kramer, V. Garcia-Hansen, S. Omrani, climateBOX: A low-cost and open-source monitoring device for personal thermal comfort evaluation, Energy Build. 283 (2023) 112830, https://doi.org/10.1016/j.enbuild.2023.112830.
32. D. Lapidus, M.E. Salem, R.H. Beach, S. Zayed, I. Ortiz-Monasterio, Greenhouse gas mitigation benefits and profitability of the GreenSeeker Handheld NDVI sensor: evidence from Mexico, Precisi. Agricult. 23 (2022) 2388–2406, https://doi.org/10.1007/s11119-022-09925-z.
33. K.L. Martin, K. Girma, K.W. Freeman, R.K. Teal, B. Tuba´na, D.B. Arnall, B. Chung, O. Walsh, J.B. Solie, M.L. Stone, W.R. Raun, Expression of variability in corn as influenced by growth stage using optical sensor measurements, Agron. J. 99 (2007) 384–389, https://doi.org/10.2134/agronj2005.0268.
34. C.R. Martin, N. Zeng, A. Karion, R.R. Dickerson, X. Ren, B.N. Turpic, K.J. Weber, Evaluation and environmental correction of ambient CO₂ measurements from a low-cost NDIR sensor, Atmosph. Measure. Techniq. 10 (7) (2017) 2383–2395, https://doi.org/10.5194/amt-10-2383-2017.
35. A. Menditto, M. Patriarca, B. Magnusson, Understanding the meaning of accuracy, trueness and precision, Accredit. Qual. Assur. 12 (2007) 45–47, https://doi.org/10.1007/s00769-006-0191-z.
36. B. Mobaraki, F.J. Castilla Pascual, F. Lozano-Galant, J.A. Lozano-Galant, R. Porras Soriano, In situ U-value measurement of building envelopes through continuous low-cost monitoring, Case Stud. Therm. Eng. 43 (2023) 102778, https://doi.org/10.1016/j.csite.2023.102778.
37. S.S. Nati, D.A. Mahmood, M.H. Al-Jiboori, Seasonal normalized difference vegetation index response to air temperature and precipitation in Baghdad, Open Agricult. 5 (2020) 631–637, https://doi.org/10.1515/opag-2020-0058.
38. V.H.R. Prudente, V.S. Martins, D.C. Vieira, N.R.d.F.e Silva, M. Adami, I.D. Sanches, Limitations of cloud cover for optical remote sensing of agricultural areas across South America, Remote Sens. Applic. Soc. Environ. 20 (2020) 100414, https://doi.org/10.1016/j.rsase.2020.100414.
39. P.M. Putra, W. Kristiavani, D.C. Mumtayzabi, T. Auggarwati, R. Amitsa, D. H. Shollash, D. Okiyanto, Sudarto, Pineapple biomass estimation using unmanned aerial vehicle wave-rotates forcing stage vegetation index approach for ultra-high-resolution image, Smart Agricult. Technol. 1 (2021) 100025, https://doi.org/10.1016/j.atech.2021.100025.
40. M. Reyniers, E. Vrindts, Measuring wheat nitrogen status from sparse and ground-based platforms, Int. J. Remote Sens. 27 (3) (2006) 549–567, https://doi.org/10.1080/01431160500170792.
41. Y. Ryu, G. Lee, S. Jeon, Y. Song, H. Kimm, Monitoring multi-layer canopy spring phenology of temperate deciduous and evergreen forests using low-cost spectral sensors, Remote Sens. Environ. 149 (2014) 227–238, https://doi.org/10.1016/j.rse.2014.04.015.
42. H. Sanin-Janin, M. Garel, J.-L. Chapuis, D. Pontier, Assessing the performance of NDVI as a proxy for plant biomass using non-linear models: a case study on the Kerguelen archipelago, Polar Biol. 32 (2009) 961–871, https://doi.org/10.1007/s00300-009-0586-8.
43. J. Schaller, E. Scherwietes, L. Gerber, S. Vaidya, D. Kaczorek, J. Pausch, D. Batkasky, M. Sommer, M. Hoffmann, Silica fertilization improved wheat performance and increased phosphorus concentrations during drought at the field scale, Scientif. Rep. 11 (2021) 20852, https://doi.org/10.1038/s41598-021-00464-7.
44. O.I. Sintonijn, S. Atdeigha, A. Ayedelini, M. Ahbanasou, A. Agbossou, Characterization of irrigation techniques and topology of vegetable gardens in the districts of Cove and Zangnanodo in center of Benin, Int. J. Eng. Res. Applic. IJERA 7 (10) (2017) 40–51.
45. R.P. Shhodia, R.L. Ray, S.K. Singh, Applications of remote sensing in precision agriculture: a review, Remote Sens. 12 (2020) 3136, https://doi.org/10.3390/rs12193136.
46. G.M. Smith, E.J. Milton, The use of the empirical line method to calibrate remotely sensed data to reflectance, Int. J. Remote Sens. 20 (1999) 2653–2662, https://doi.org/10.1080/014311699211994.
47. K. Soudani, G. Hmimina, N. Delpierre, J.-Y. Pontailler, M. Aubinet, D. Bonai, B. Caquet, A. De Grandcourt, B. Burban, C. Flechard, D. Guyon, A. Granier, P. Gross, B. Heinesh, B. Longdoz, D. Loustan, C. Moureaux, J.-M. Ourcival, S. Rambal, L. Saint Andre, E. Dufrene, Ground-based Network of NDVI measurements for tracking temporal dynamics of canopy structure and vegetation phenology in different biomes, Remote Sens. Environ. 123 (2012) 234–245, https://doi.org/10.1016/j.rse.2012.03.012.
48. J.D. Stamford, S. Vialet-Chabrand, I. Cameron, T. Lawson, Development of an accurate low cost NDVI imaging system for assessing plant health, Plant Method. 19 (9) (2023), https://doi.org/10.1186/s13007-023-01081-3.
49. A. Stepanov, K. Dubrovin, A. Sorokin, T. Aseeva, Predicting soybean yield at the regional scale using remote sensing and climatic data, Remote Sens. 12 (2020) 1936, https://doi.org/10.3390/rs12121936.
50. T.R. Tenreiro, M. Garcia-Vila, J.A. Gomez, J.A. Jimenez-Berni, E. Ferrers, Using NDVI for the assessment of canopy cover in agricultural crops within modelling research, Comput. Electron. Agricult. 182 (2021) 106038, https://doi.org/10.1016/j.compag.2021.106038.
51. A. Tittebrand, U. Spank, Ch. Bernhofer, Comparison of satellite- and ground-based NDVI above different land-use types, Theor. Appl. Climatol. 98 (2009) 171–186, https://doi.org/10.1007/s00704-009-0103-3.
52. O. Vergassi-Diaz, M.A. Zaman-Allah, B. Masuka, A. Hontero, P. Zarco-Tejada, B. M. Prasanna, J.E. Cairns, J.L. Araus, A novel remote sensing approach for prediction of maize yield under different conditions of nitrogen fertilization, Front. Plant Sci. 7 (2016), https://doi.org/10.3389/fpls.2016.00666.
53. N. Virlet, K. Sabermansell, P. Sadeghi-Tehran, M.J. Hawkesford, Field Scanalyzer: An automated robotic field phenotype platform for detailed crop monitoring, Front. Plant Biol. 44 (2017) 143, https://doi.org/10.1071/FP16163.
54. M. Weiss, F. Jacob, G. Duveiller, Remote sensing for agricultural applications: a meta-review, Remote Sens. Environ. 236 (2020) 111402, https://doi.org/10.1016/j.rse.2019.111402.
55. F.-P. Wittich, M. Kraft, The normalized difference vegetation index obtained from agrometeorological standard radiation sensors: a comparison with ground-based multiband spectral measurements during the phenological development of an oat canopy, Int. J. Biometeorol. 52 (2008) 167–177, https://doi.org/10.1007/s00484-007-0108-5.
56. Y. Yamasaki, M. Moroi, N. Roguchi, Development of a high-accuracy autonomous sensing system for a field scouting robot, Comput. Electron. Agricult. 193 (2022) 106630, https://doi.org/10.1016/j.compag.2021.106630.
57. I. Youssouf, M. Lawani, Benin Soils: Classification of the World Reference Database, in: Report on the Soil Resources of the World, Fourteenth Meeting of the West and Central African Subcommittee on Soil Correlation for Land Reclamation, 2009, pp. 29–50.
58. Y. Xu, Q. Li, Z. Tang, J. Liu, B. Xiang, Towards accurate, cost-effective, ultra-low-power and non-invasive respiration monitoring: A reusable wireless wearable sensor for an off-the-shelf KN95 mask, Sensors 21 (2021) 6698, https://doi.org/10.3390/s21206698.
59. B. Yu, S. Shang, Multi-year mapping of maize and sunflower in Hetao Irrigation District of China with high spatial and temporal resolution vegetation index series, Remote Sens. 9 (2017) 855, https://doi.org/10.3390/rs9080855.