Positioning Accuracy of RTK-GNSS-Enabled Drones and Their Performance in Agricultural Crop Sensing

Pascal Izere^1,†, Biquan Zhao^1,2,†, James Schnable^3,4, Yufeng Ge^1,4, Yeyin Shi^1,*

Published in Journal of the ASABE 68(4): 559-571 (doi: 10.13031/ja.16306). Copyright 2025 American Society of Agricultural and Biological Engineers.

¹ Department of Biological Systems Engineering, University of Nebraska-Lincoln, Lincoln, Nebraska, USA.

² School of Natural Resources, University of Nebraska-Lincoln, Lincoln, Nebraska, USA.

³ Department of Agronomy and Horticulture, University of Nebraska-Lincoln, Lincoln, Nebraska, USA.

⁴ Center for Plant Science Innovation, University of Nebraska-Lincoln, Lincoln, Nebraska, USA.

^† Contributed equally to this work and share first authorship.

^* Correspondence: yshi18@unl.edu

The authors have paid for open access for this article. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License https://creative commons.org/licenses/by-nc-nd/4.0/

Submitted for review on 30 December 2024 as manuscript number ITSC 16306; approved for publication as a Research Article by Associate Editor Dr. Chenghai Yang and Community Editor Dr. Seung Chul Yoon of the Information Technology, Sensors, & Control Systems Community of ASABE on 2 June 2025.

Citation: Izere, P., Zhao, B., Schnable, J., Ge, Y., & Shi, Y. (2025). Positioning accuracy of RTK-GNSS-enabled drones and their performance in agricultural crop sensing. J. ASABE, 68(4), 559-571. https://doi.org/10.13031/ja.16306

Highlights

• The performance of RTK-GNSS-enabled drones without GCPs was assessed in agricultural applications.
• A centimeter-level positioning accuracy was achieved when drones were functioning with RTK-GNSS-enabled.
• RTK-GNSS-enabled drones without GCPs performed competitively with GCP-based methods in plant height estimation.
• Using RTK-enabled drones without GCPs provides a time- and labor-saving solution for agricultural practices.

ABSTRACT. Commercial drones equipped with Real-Time Kinematic GNSS (RTK-GNSS) technology have been available for several years and are now a standard feature in modern models. This technology has the potential to eliminate the need for ground control points (GCPs). However, despite its widespread adoption and the positioning accuracy claimed in product specifications, limited literature exists on quantifying and systematically evaluating its performance in agricultural field settings for specific applications, such as measuring canopy height profiles. This study aimed to evaluate the potential of RTK-GNSS-enabled drones to estimate plant height without GCPs by comparing it to a benchmark method that uses GCPs across three crops: triticale, maize, and soybean. Three methods were compared: (A) regular-GNSS with GCPs, (B) RTK-GNSS with GCPs, and (C) RTK-GNSS without GCPs. Prior to evaluating the performance in height estimation, the positioning accuracy of RTK-GNSS-enabled drones was assessed, and results confirmed the centimeter-level positioning accuracy achieved by drones with RTK-GNSS-enabled. For plant height estimation, Method-C (RTK-GNSS without GCPs) performed competitively with GCP-based methods, achieving R² values from 0.84 to 0.99 for drone-estimated versus manually measured plant heights for the three crops, with a mean RMSE of 12.15 cm (compared to Method-A R² of 0.86 to 0.99 with a mean RMSE of 11.86 cm, and Method-B R² of 0.84 to 0.97 with a mean RMSE of 11.49 cm). Results suggest that RTK-GNSS-enabled drones can reliably estimate plant height without GCPs, offering a time- and labor-saving solution for field measurements in breeding research and agricultural production.

Keywords.Global positioning system (GPS), Ground control point (GCP), High-throughput plant phenotyping, Plant height, Structure-from-motion, Unmanned aerial vehicle (UAV).

High-throughput phenotyping addresses many of the limitations of traditional field measurements and is changing the ways plant geneticists study the genes responsible for plant phenotypes and the ways plant breeders make selections (Araus and Cairns, 2014; Mir et al., 2019; Yang et al., 2020). Drones or unmanned aerial vehicles (UAVs) carrying sensors capable of capturing visible, spectral, hyperspectral, thermal, and/or fluorescence signals have emerged as a powerful platform for high-throughput plant phenotyping (Shi et al., 2016; Xie and Yang, 2020). This capability enables rapid and non-destructive evaluation of plant growth, biomass, nutrient levels, and other traits in the field (Ge et al., 2016; Pfeifer et al., 2012). For instance, estimating plant height is a common task in plant phenotyping, as it serves as a fundamental growth parameter linked to plant growth, light competition, health, biomass, and yield (Madec et al., 2017; White et al., 2012; Yang et al., 2020). Accurate and reliable measurements of plant height are essential for identifying genotypes with desirable traits, predicting biomass and yield, evaluating stress tolerance, and understanding underlying physiological processes that determine plant performance in different environments (Ji et al., 2022; Li et al., 2022; Xiao et al., 2022).

Drone platforms offer a non-destructive and rapid approach to field-based phenotyping for larger-scale field experiments in agricultural research, which can, in principle, be both more labor efficient and less subjective than manual measurements (Jiang et al., 2021; Volpato et al., 2021). Significant progress has been made in plant height estimation using two different imaging technologies: optical cameras or Light Detection and Ranging (LiDAR) sensors, both of which have demonstrated strong performance across various crops (Chang et al., 2017; Li et al., 2020; Malambo et al., 2018; Volpato et al., 2021). Although LiDAR sensors can provide a more accurate height estimation compared to optical sensors, LiDAR sensors are expensive. In addition to the cost-efficient advantage of optical sensors, using optical sensors requires less computation power for data processing compared to processing LiDAR cloud point data. Estimating plant height from optical RGB images captured will begin with using photogrammetry techniques to create one or more digital elevation models (DEMs) (Gao et al., 2022; Hassan et al., 2019; Iqbal et al., 2017). A DEM is a digital representation of the Earth’s elevation and includes two main types: the digital terrain model (DTM) and digital surface model (DSM). The DTM represents the bare earth surface, excluding objects like plants, trees, and buildings, while the DSM includes all surface features. In agricultural contexts, to obtain a DTM, pre-season drone flights can be conducted before seedling emergence. By subtracting the DTM from the DSM, a canopy plant height model can be generated for plant height estimation (Chang et al., 2017; Hassan et al., 2019; Holman et al., 2016).

However, obtaining an accurate estimate of plant height using drone RGB images requires maps with a high degree of accuracy in both horizontal and vertical dimensions (Chang et al., 2017; Gao et al., 2022). Drones used in plant phenotyping rely on onboard global navigation satellite system (GNSS) receivers, supplemented with inertial navigation systems (INS) to determine their location (Gyagenda et al., 2022; Rui et al., 2024). While these receivers are favored for their small size, low cost, wide availability, and ease of use, regular-GNSS receivers used on most consumer-grade drones often result in a positioning accuracy of 1 to 3 meters (Hassan et al., 2019; Küng et al., 2011; Pfeifer et al., 2012). This level of error in positioning can overwhelm the true biological differences in height of interest to plant geneticists and plant breeders (typically under 50 cm), which can also result in misidentification of individual research plots containing a specific plant genotype and/or agronomic treatment.

The inclusion of ground control points (GCPs) during DEM generation has proven to be an effective strategy to achieve the level of vertical and horizontal accuracy necessary to accurately estimate plant height using DTM and DSMs generated from drones equipped with regular-GNSS receivers (Holman et al., 2016; Xie et al., 2021). GCPs are visible markers whose locations are measured to centimeter-level positioning accuracy using survey-grade equipment. While the use of GCPs can significantly increase the precision and accuracy of maps derived from drones equipped with regular-GNSS receivers, setting GCPs requires additional time and labor in the field, and the placement of GCPs can require either entering areas which are hard to access and/or disrupt the design of field trials as a result of shading and/or plant removal (Volpato et al., 2021; Xie et al., 2021).

In the last several years an increasing number of drones equipped with Real-Time Kinematic (RTK) GNSS receivers, which can provide centimeter-level positioning accuracy during flights, have become commercially available (Štroner et al., 2021; Zhao et al., 2020; Zimmermann et al., 2017). For example, the Phantom 4 RTK from DJI, Shenzhen, China, is an industrial-grade drone equipped with an RTK-GNSS capable-receiver, claimed to achieve horizontal accuracy of up to 1 cm and vertical accuracy of up to 1.5 cm. Several recent studies have explored the positioning accuracy of RTK-GNSS-enabled drones for various applications, including mapping rockfalls and land cover classification (Sefercik et al., 2023), coastal areas (Taddia et al., 2019), and inaccessible forested areas (Tomaštík et al., 2019). Centimeter-level positioning accuracy was reported by these studies using RTK-GNSS-enabled drones, showing the capacity of RTK-GNSS-enabled drones for high-precision mapping tasks.

The widespread availability of RTK technology in drones may create the opportunity to eliminate the need for GCPs when conducting high-throughput phenotyping of plant height, but the effectiveness of RTK-GNSS-enabled drones for estimating plant height without the need for GCPs remains unexplored. This study aims to evaluate the potential of RTK-GNSS-enabled drones for estimating plant height without the need for GCPs by comparing this approach to both ground truth height measurements and current best practices for drone-based height measurement. A total of three methods are assessed and compared: (A) regular-GNSS with GCPs, (B) RTK-GNSS with GCPs, and (C) RTK-GNSS without GCPs. Specifically, this study (1) evaluates RTK-GNSS-enabled drones’ horizontal and vertical positioning accuracies from images collected while the drone was both stationary in static and in flight, and (2) compares the performance of three methods in estimating plant height for triticale, maize, and soybean. This research enhances our understanding of positioning performance for RTK-GNSS-enabled drones and provides valuable insights into the use of advanced RTK-GNSS-enabled drones for agricultural plant height estimation, potentially offering a more time-efficient, labor-saving, and cost-effective alternative to traditional drone-based methods that require GCPs.

Materials and Methods

Experiment Site Description

Field experiments were conducted at three research farms of the University of Nebraska-Lincoln (UNL) in 2022, including five total fields, two each of maize and triticale, and one of soybean (fig. 1). All these fields were in the shape of a rectangle with different areas. The field at Eastern Nebraska Research, Extension, and Education Center (ENREEC) was planted with triticale in an area of 0.74 hectares. The field at South-Central Agricultural Laboratory (SCAL) was 1.73 hectares, where soybeans were planted under different nitrogen level treatments and were irrigated with a parallel irrigation system. The other three fields were all located at Havelock Farm: the first field (Havelock-HIPS) was 2.7 hectares with maize hybrids under nitrogen treatments; the second field (Havelock-SAM) was 1.6 hectares with an inbred diversity panel of maize; and the third field (Havelock-triticale) was 1.23 hectares with triticale. In 2020, a pilot test assessing RTK-GNSS-enabled drone’s positioning performance was conducted on the East Campus at UNL (fig. 1).

Figure 1. Study sites in Nebraska, USA: A pilot test on the East Campus of University of Nebraska-Lincoln (UNL) in 2020 and three research farms, including five total fields for plant height estimation, in 2022. On the right side there was an example of the Havelock-SAM maize field, showing a raw orthomosaic RGB image, a raw digital surface model (DSM) image, and the digital terrain model (DTM) image estimated by using the inverse distance weighted (IDW) technique. Green dots in the DTM image were bare soil points used for the interpolation.

Nebraska is located in the Great Plains, and the elevation of Nebraska rises gradually from southeast to northwest. All our study fields in southeast Nebraska were relatively flat with little terrain variation, as shown by our DTM example of the Havelock-SAM maize field, the difference in IDW interpolated DTM was only 0.6 m (fig. 1). The little variation in terrains was consistent across all our study fields, as those fields have been well managed for crop planting for years.

Pilot Test of Positioning Accuracy of RTK-GNSS-Enabled Drone in 2020

The DJI Phantom 4 RTK (DJI, Shenzhen, China) was tested in 2020, as it was the earliest commercially available small drone (released in October 2018) equipped with the RTK-GNSS receiver (fig. 2). The Phantom 4 RTK captured images with a built-in RGB camera at a resolution of 5472×3648 pixels, and the location information (latitude and longitude) for each image taken was recorded by the drones’ antennas and stored into metadata of each image. To enable the RTK-GNSS for the Phantom 4 RTK in this study, a DJI D-RTK 2 GNSS mobile station, acting as the base station, was set near the area (the star sign in fig. 1).

In an open area of 105 m × 35 m at the East Campus of UNL (fig. 1), four black-and-white wooden boards, being used as GCPs, were put at the edge of the field, and six testing objects were placed inside the area to evaluate the positioning accuracy for the DJI Phantom 4 RTK. The six objects were initially put on the ground level and then elevated to various heights in this test, denoting two testing scenarios: ground level and elevating level. The ground-truth locations of the base station, the four GCPs, and the six objects at the two height scenarios were measured by a survey-grade receiver (Topcon Positioning Systems, Inc., Tokyo, Japan) at a centimeter-level positioning accuracy.

In this pilot test, positioning accuracy of the DJI Phantom 4 RTK was assessed using a combination design involving two object height scenarios (ground level and elevating level) × two drone statuses (in flight and sitting stationary on the testing objects with the propellers off) × three GNSS settings (regular-GNSS, RTK-uncorrected-base, and RTK-corrected-base). When the drone was placed on the testing objects with the propellers off, it remained completely stationary and captured ten pictures in this static condition. GPS coordinates information from images taken at the six testing objects was extracted for analysis. For the in-flight scenario, orthomosaic maps and DEMs were generated using images taken in the air after implementing drone flights. Images were collected with three different GNSS settings. In the first setting, images were collected with RTK-GNSS disabled (regular-GNSS). Both the second and third settings collected images with RTK-GNSS-enabled (RTK-GNSS), but in the second setting, the location of the base station was measured by the base station itself (RTK-uncorrected-base) while in the third setting, the base station’s location was corrected using the ground-truth coordinates measured by the survey-grade receiver (RTK-corrected-base).

Figure 2. Two drones from DJI (DJI, Shenzhen, China) equipped with RTK-GNSS receivers (antennas), including the Phantom 4 RTK and the Matrices 300 RTK. The Phantom 4 RTK was used in both the pilot test in 2020 and plant height estimation in 2022. The Matrices 300 RTK was used in plant height estimation in 2022.

Absolute difference was calculated for the base station using its uncorrected coordinates (x₁, y₁, z₁) and the ground-truth coordinates (x₂, y₂, z₂) measured by the survey-grade receiver, both for horizontal (?_XY) and vertical (?_Z) directions as shown in equations 1 and 2:

(1)

(2)

where ?_XY and ?_Z represents the absolute differences in the horizontal and vertical directions, respectively, between two coordinates: P₁ (x₁, y₁, z₁) and P₂ (x₂, y₂, z₂).

Positioning Accuracy Assessment when the Drone was Stationary in Static

Ten images were taken on each testing object for the assessment of positioning accuracy in static when the drone was stationary in static on the testing objects. The mean coordinates (X, Y, Z) of each testing object were calculated by averaging the coordinates of the ten raw images (eq. 3). For each testing object, using the mean coordinates (X, Y, Z) and its ground-truth coordinates measured by the survey-grade receiver, absolute differences for horizontal and vertical directions were calculated afterward (see eq. 1 and eq. 2). Eventually, based on the absolute differences of six testing objects, either horizontally or vertically, root mean squared error (RMSE) was calculated, i.e., RMSE_XY and RMSE_Z.

(3)

(4)

(5)

where

(X, Y, Z) represents Mean coordinate of ten images taken at a testing object.

RMSE_XY represents Root mean square error in the horizontal direction (X and Y), based on absolute differences across six testing objects.

RMSE_Z represents Root mean square error in the vertical direction (Z), based on absolute differences across the same six objects.

Positioning Accuracy Assessment Based on Orthomosaics and DEM Maps

After generating orthomosaics and DEM maps by using the images taken in flight, we evaluated horizontal positioning accuracy by using orthomosaic maps and vertical positioning accuracy by using DEMs, respectively, for the six testing objects. Pix4D Mapper (PIX4D, Lausanne, Switzerland; http://pix4d.com) was used to process the in-flight RGB images for each flight. Three methods were used for processing: (A) regular-GNSS with GCPs, (B) RTK-GNSS with GCPs, and (C) RTK-GNSS without GCPs. Method-A and Method-B served as benchmarks, using GCPs, while in Method-C, GCPs were excluded. Eventually, RMSE_XY and RMSE_Z were calculated for the assessment.

Agricultural Crop Height Estimation in 2022

Manual Plant Height Measurements

During the 2022 growing season, plant height was manually measured multiple times, with each time randomly selecting 30 plots across the five fields (table 1). In each plot, four plants that were uniform and representative of the plot were measured using a stick. A marked plastic cylindrical tube, a marked wooden ruler, and a metallic marked ruler were used to measure plant height for maize, triticale, and soybean, respectively (fig. 3). To minimize sources of variance in ground truth measurements, all the plant height measurements were collected by the same surveyor. The plant height of each plot at each time point was calculated by averaging the measurements from the four plants within the plot.

Figure 3. Plant height measurements for maize, triticale, and soybean in the fields.

Drone Campaigns and GCP Settings

Throughout the growing season, drone flights were conducted on the same dates as the plant height measurement. In addition to the Phantom 4 RTK used in the 2020 pilot test, another DJI drone with RTK-GNSS receivers, Matrices 300 RTK, was used for RGB image collection (fig. 2). The Matrices 300 RTK, equipped with a DJI Zenmuse P1 camera, collected images at a resolution of 8192×5460 pixels. Differing from the previous test using a base station (DJI D-RTK 2 GNSS mobile station) in a close range (nearest Trimble base stations were less than 30 km away from the testing area), the rover in this study connected to a subscribed RTK correction network (commonly known as a network RTK), Trimble VRS Now. The subscription allows the RTK-GNSS to be enabled with the known location of the Trimble base station (equivalent to the RTK-corrected-base setting in the pilot test), achieving a horizontal accuracy of 2 cm and a vertical accuracy of 3 cm.

During data collection, the two drones captured RGB images of each field with RTK-GNSS-enabled (i.e., RTK-GNSS) and disabled (i.e., regular-GNSS). For example, four flights were conducted at a field: the first flight used the Matrices 300 RTK with RTK-GNSS, the second flight used the Matrices 300 RTK with regular-GNSS, the third flight used the Phantom 4 RTK with RTK-GNSS, and the fourth used the Phantom 4 RTK with regular-GNSS. In between the two flights using the same drone, we completely turned off and turned on the drone again to interrupt the function of RTK-GNSS of the previous flight. The two RTK-GNSS-enabled drones were set to capture RGB images at 1-second intervals during each flight. Due to maintenance and repairs, data from the Matrice 300 RTK was unavailable after August 3^rd (table 1). All flights were set with a 90% side overlap and an 85% front overlap. For triticale and soybean fields, flights were conducted at 25 m above ground, while flights over maize fields were at 35 m above ground. The flights were scheduled around solar noon to reduce shadow effects. Eight or nine GCPs were placed evenly across each field before the drone flights (fig. A1). The same survey-grade receiver was used to measure the GPS coordinates of these GCPs. The mean pixel size of the output for the Phantom 4 RTK was 0.8 cm and it was 0.33 cm for the Matrice 300 RTK.

Positioning Accuracy Assessment Based on Orthomosaics and DEM Maps

The same three processing methods, (A) regular-GNSS with GCPs, (B) RTK-GNSS with GCPs, and (C) RTK-GNSS without GCPs, were used to process the raw images taken during flights to generate orthomosaic and DEM maps. Similar to the 2020 pilot test of evaluating positioning accuracy based on orthomosaic maps for horizontal and using DEMs for vertical, RMSE_XY and RMSE_Z were calculated for the three methods across the five fields and two drones. ANOVA was used to test differences among the three methods of the two drones.

Performance in Plant Height Estimation

To calculate plant height, a DTM representing the bare earth surface is needed. Since pre-season drone flights to produce DTM were not conducted, we used an inverse distance weighted (IDW) technique to generate bare earth surface for each field from bare soil sampling points. IDW is one of the standard methods for generating maps from sampled data (Polidori and El Hage, 2020). We first created a set of points representing soil surfaces in each field using the high-spatial-resolution orthomosaic maps and extracted their elevation values from the DEMs. These points were manually identified when the soil was visible on orthomosaic maps. We then used the IDW interpolation tool in ArcMap (ArcGIS, ESRI, Redlands, CA, USA) to create the DTM. In this study, we had around 300 bare soil points for the ENREEC field, 440 points for the SCAL field, 570 points for the Havelock-HIPS field, 300 points for the Havelock-SAM field, and 260 points for the Havelock-triticale field, with a mean of nearly 230 points per hectare across the five fields. By using the interpolated DTMs and derived DSMs, the canopy plant height maps were created as follows:

(6)

where DSM stands for Digital Surface Model; output directly from the processing in Pix4D software using drone images, and DTM stands for Digital Terrain Model; represents the bare earth surface interpolated using a set of bare soil sampling points and the inverse distance weighted (IDW) interpolation technique in ArcMap software.

Canopy plant height maps were generated throughout the growing season for the five fields across three crop types, two drones, and three methods. To extract plot-level plant height, the value corresponding to the 99^th percentile of the pixel readings within a plot boundary was used as the plant height of the plot. In this study, manual plant height was measured at the top of the canopy, while pixels in the plant height maps (eq. 6) represent not only the top of canopy but any surface areas of the field, including bare soil ground and lower layers of canopy as long as they are visible in the drone images. So other metrics, such as lower percentiles or mean value, would not be appropriate to reflect the height of the canopy top. On the other hand, instead of using the maximum value (100^th percentile) of the pixel readings within the plot, we selected the 99^th percentile to leave out potential noise readings generated during data collection or pre-processing. It was demonstrated to have a slightly better performance compared to the 90^th, 95^th, and maximum values on our datasets, though the differences were not obvious. Additionally, as described previously, all our study fields in southeast Nebraska were pretty flat with little terrain variation. The plot-level plant height values were then correlated to the field-measured plot-level plant heights using a linear regression. The performance in plant height estimation was assessed by the coefficient of determination (R²), RMSE, relative RMSE (rRMSE), and standard error (SE).

Results

Positioning Accuracy of RTK-GNSS-Enabled Drones

Positioning Accuracy in the Pilot Test for Phantom 4 RTK in 2020

In the pilot test, the RTK-GNSS mode of the Phantom 4 RTK was enabled using a DJI D-RTK 2 GNSS mobile station (served as the base station). The uncorrected coordinate of the base station and its ground-truth coordinate resulted in a horizontal absolute difference of 11.57 cm and a vertical absolute difference of 3.63 cm. In a static setting when the drone was stationary, results showed that enabling RTK-GNSS can improve both horizontal (RMSE_XY) and vertical (RMSE_Z) accuracies compared to the regular-GNSS setting (table 2). For example, the RMSE_XY of using regular-GNSS was 1.76 cm, which was improved to 1.15 cm when using RTK-GNSS with an uncorrected base station’s location, and it was greatly improved to 0.68 cm when the RTK-GNSS was enabled using the corrected base coordinates. The RMSE_Z of using regular-GNSS setting was also improved from 3.24 cm to 2.38 and 2.45 cm when RTK-GNSS-enabled with uncorrected or corrected coordinates, respectively.

In our test, using GCPs to generate orthomosaic and DEM maps from in-flight images, it provided centimeter-level positioning accuracy in both horizontal (a mean of 1.39 cm) and vertical (a mean of 4.23 cm) directions for Method-A and Method-B. Without using GCPs, centimeter-level positioning accuracy in both horizontal (a mean of 6.91 cm) and vertical (a mean of 5.27 cm) directions was observed (table 3). Results confirmed the effectiveness of using GCPs to improve the positioning accuracy of orthomosaic and DEM maps derived from drone images while indicating a competitive accuracy of orthomosaic and DEM maps without using GCPs.

Positioning Accuracy of Two Drones in 2022

Centimeter-level positioning accuracy was consistently observed in 2022 with an additional RTK-GNSS-enabled drone used in this study (table 4). Horizontally, Method-C exhibited a mean RMSE_XY of 2.36 cm, which is comparable to the mean RMSE_XY of 2.21 cm in Method-A and 2.20 cm in Method-B. Similarly, for vertical accuracy, Method-C achieved a mean RMSE_Z of 4.66 cm, while the mean RMSE_Z for Method-A was 4.34 cm and it was 4.47 cm for Method-B. ANOVA results indicated no significant differences (p-value ranged from 0.22 to 0.79) among the three methods, the two drone platforms, or their interaction, in terms of either horizontal or vertical accuracy (table 5). The competitive performance of Method-C, using RTK-GNSS without GCPs, suggests that it may be feasible to eliminate the use of GCPs when RTK functionality is enabled on drones.

Plant Height Estimation of Three Methods by Dates, Crops, and Drones

A linear regression analysis was performed among the three methods for each crop type by the two drones throughout the season (fig. 4). No significant differences (p > 0.05) were observed among the three methods regarding the R², indicating consistent performance across all three methods for the three crops. For example, Method-C achieved R² values between 0.84 and 0.99 for drone-estimated versus manually measured plant heights for the three crops. Method-A achieved R² values of 0.86 to 0.99, and Method-B had R² values of 0.84 to 0.97. Performance of estimating plant height of three crops was detailed by dates and by using the two drones (tables A1-A3). The mean RMSE for estimating plant height of three crops was 11.86 cm for Method-A, 11.49 cm for Method-B, and 12.15 cm for Method-C (table 6). Overall, performance was consistent across the three methods, the three crops, and both drone platforms, with no apparent trends over the dates. Similar to R² and RMSE, the other evaluation metrics of rRMSE and SE showed Method-C’s performance to be comparable to Method-A and Method-B.

Discussion

Positioning Accuracy Assessment

The use of GCPs is a well-established method for providing ground-truth coordinates as reference in photogrammetric processing of drone images (Colomina and Molina, 2014; Sanz-Ablanedo et al., 2018; Yeh et al., 2016). GCPs help minimize image distortion and enhance positioning accuracy in drone applications such as plant height estimation and surveying. In this study, results verified that the inclusion of GCPs consistently resulted in centimeter-level horizontal and vertical positioning accuracy in orthomosaic and DEM maps derived from drone images (tables 2-4). The benchmark methods using GCPs produced orthomosaic maps with a mean 2D positioning accuracy of 2.21 cm RMSE and a vertical positioning accuracy of 4.34 cm RMSE (table 4). The alternative method tested in this study, which utilized drones with RTK-GNSS-enabled and without GCPs for processing, demonstrated comparable performance, achieving a mean horizontal accuracy of 2.36 cm RMSE and a vertical accuracy of 4.66 cm RMSE (table 4). Our results are consistent with previous studies, reporting centimeter-level positioning accuracy using RTK-GNSS-enabled drones for various applications (Sefercik et al., 2023; Taddia et al., 2019; Tomaštík et al., 2019; Ladino and Sama, 2024). Results suggest that it will be feasible to eliminate GCPs when using RTK-GNSS-enabled drones while still producing high-quality orthomosaic and DEM maps at high positioning accuracy for various applications.

Figure 4. Correlation between drone-estimated and manual-measured plant height throughout the growing season in triticale (A, B, C), soybean (D, E, F), and maize (G, H, I) across the three methods: Method-A: regular-GNSS with GCPs, Method-B: RTK-GNSS with GCPs, and Method-C: RTK-GNSS without GCPs. Datapoints in black circles: Phantom 4 RTK; Datapoints in orange triangles: Matrice 300 RTK.

Though, there would be variations/errors introduced during the observations over time (multiple observations during the growing season) and space (across different fields), as well as different drone platforms. For example, we observed some instances where the method of RTK-GNSS with GCPs performed worse than the method with only RTK-GNSS, which were (1) using Phantom 4 RTK for the Havelock-SAM field, (2) using Phantom 4 RTK for the SCAL field, and (3) using Matrice 300 RTK for the ENREEC field. In addition, in our study, although we used the GCPs in the same locations for all observations during the growing season. However, their locations were not permanently fixed on the ground. Errors can be introduced as we may not be able to put the GCPs exactly in the same postures each time in exactly the same posture and position as when they were during the measurement of their locations using the survey-grade GPS equipment. For instance, if the GCPs were inclined a little bit, it may result in some errors when we processed the images in the Pix4D software, particularly with such high spatial resolution of 0.3–0.8 cm in our drone images. This result confirms the previous recommendation of having permanently fixed GCPs as a standard drone protocol for plant phenotyping for over-time observations, which in turn highlights the importance of our study that explores the use of RTK-GNSS-enabled drones under the elimination of GCPs. The use of RTK-GNSS-enabled drones without GCPs provides a cost- and efficiency-friendly option.

Estimation of Plant Height Using Drone Imagery

While several studies have explored using plant height estimated from RGB drone images as a key indicator of crop growth, the reliance on GCPs raises concerns about costs and the time-consuming nature of pre-flight setup (Sanz-Ablanedo et al., 2018; Villanueva and Blanco, 2019; Xie et al., 2021). Despite the fact that many drones now come equipped with the advanced GPS antennas supporting RTK-GNSS mode, limited research has examined the ability of RTK-GNSS-enabled drones to estimate plant height without GCPs. This study compared the performance of plant height estimation across three crop types, over an entire growing season, using three methods and two commercial drones. Our findings confirmed the feasibility of drones for crop height estimation (Chang et al., 2017; Holman et al., 2016; Iqbal et al., 2017). Additionally, Method-C (RTK-GNSS without GCPs) achieved a comparable performance in plant height estimation for triticale, maize, and soybean compared to the benchmark methods using GCPs (tables 6-7, fig. 4). Our findings suggest the potential of excluding GCPs when using RTK-GNSS-enabled drones for plant height estimation, facilitating a time-efficient, labor-saving, and cost-effective alternative to traditional methods that rely on GCPs while maintaining high accuracy.

This study did not conduct pre-season flights to generate the DTMs under the bare ground condition. Instead, we applied the IDW method to create a bare soil map representing the DTMs. Indeed, using DTMs from a pre-season flight could potentially improve further plant height estimates. Nevertheless, previous research has demonstrated that plant height estimation using interpolated soil points from DSMs to create DTMs is comparable to results from imaging bare soil surfaces (Holman et al., 2016; Schirrmann et al., 2016; Yuan et al., 2018). Thus, the DTMs generated through the IDW approach in this study were a reasonable alternative to overcome the lack of pre-season flights to create DTMs.

Previous research has noted the potential impact of wind on the accuracy of drone-based plant height estimation (Han et al., 2018; Panday et al., 2020). Wind disturbances can negatively affect plant height estimates derived from RGB drone images by destabilizing drones and reducing image quality (Han et al., 2018). This effect may become more pronounced depending on crop type and actual plant height. For example, among the three crops studied, a better estimation was observed for soybean (the shortest), followed by triticale (intermediate in height) and maize (the tallest) (fig. 4). Higher crop plants may be more significantly sensitive to wind speed and thus influenced by the wind distribution, resulting in greater errors in plant height estimation. In this study, the average wind speed during data collection was 4.24 m/s, and we found few substantial correlations between wind speed and RMSE throughout the growing season (fig. A2). Incorporating wind speed data or maps in the future may enhance the accuracy of plant height estimation.

Potential Saving on Costs

To respond to concerns about potential labor and equipment costs associated with GCP use, this study simulated the time required to set up GCPs in the field, obtain coordinates using surveying equipment, and add GCP information into image processing based on our current setup (table 7). Method-C (15 mins) demonstrated a significant reduction in working time compared to Method-A and Method-B (60 mins) from our study (i.e., 8 GCPs across a field averaging 1.6 hectares), suggesting 75% of time saved. This also implies a substantial reduction in costs, such as labor expenses and equipment costs. These results highlight the efficiency and cost-effectiveness of the RTK-GNSS-enabled drone method in plant height estimation by eliminating the need for GCPs.

Conclusions

This study evaluated the positioning accuracy of RTK-GNSS-enabled drones and their effectiveness in plant height estimation when eliminating the use of GCPs. Three methods were compared: Method-A (regular-GNSS with GCPs), Method-B (RTK-GNSS with GCPs), and the proposed Method-C (RTK-GNSS without GCPs). All three methods achieved centimeter-level horizontal and vertical positioning accuracy in orthomosaic and DEM maps. Additionally, the three methods performed similarly in estimating plant height across triticale, maize, and soybean throughout the growing season. Method-C showed an overall R² range of 0.84 to 0.99, comparable to Method-A (0.86 to 0.99) and Method-B (0.84 to 0.97). These findings suggest that Method-C has the potential to eliminate the need for GCPs in plant height estimation, providing an efficient, cost-effective way to estimate plant height with reduced labor and equipment costs. Future studies could test RTK-GNSS-enabled drones on varied terrain, such as uneven or hilly areas, to further validate performance without GCPs. Overall, this research provides valuable insights into a practical and time-saving solution for agricultural monitoring of plant height with the advanced RTK-GNSS-enabled drones.

Acknowledgments

Authors would like to acknowledge the financial support for this study provided by the USDA-NIFA (USDA-NIFA High Intensity Phenotyping Sites Program (Award # 2020-68013-32371) and NIFA-funded project An Intelligent Unmanned Aerial Application System for Site-Specific Weed Management (Award # 2021-67021-34412)), which has been crucial in funding this research.

References

Araus, J. L., & Cairns, J. E. (2014). Field high-throughput phenotyping: The new crop breeding frontier. Trends Plant Sci., 19(1), 52-61. https://doi.org/10.1016/j.tplants.2013.09.008

Chang, A., Jung, J., Maeda, M. M., & Landivar, J. (2017). Crop height monitoring with digital imagery from Unmanned Aerial System (UAS). Comput. Electron. Agric., 141, 232-237. https://doi.org/10.1016/j.compag.2017.07.008

Colomina, I., & Molina, P. (2014). Unmanned aerial systems for photogrammetry and remote sensing: A review. ISPRS J. Photogramm. Remote Sens., 92, 79-97. https://doi.org/10.1016/j.isprsjprs.2014.02.013

Gao, M., Yang, F., Wei, H., & Liu, X. (2022). Individual maize location and height estimation in field from UAV-borne LiDAR and RGB images. Remote Sens., 14(10), 2292. https://doi.org/10.3390/rs14102292

Ge, Y., Bai, G., Stoerger, V., & Schnable, J. C. (2016). Temporal dynamics of maize plant growth, water use, and leaf water content using automated high throughput RGB and hyperspectral imaging. Comput. Electron. Agric., 127, 625-632. https://doi.org/10.1016/j.compag.2016.07.028

Gyagenda, N., Hatilima, J. V., Roth, H., & Zhmud, V. (2022). A review of GNSS-independent UAV navigation techniques. Rob. Auton. Syst., 152, 104069. https://doi.org/10.1016/j.robot.2022.104069

Han, X., Thomasson, J. A., Bagnall, G. C., Pugh, N. A., Horne, D. W., Rooney, W. L.,... Cope, D. A. (2018). Measurement and calibration of plant-height from fixed-wing UAV images. Sensors, 18(12), 4092. https://doi.org/10.3390/s18124092

Hassan, M. A., Yang, M., Fu, L., Rasheed, A., Zheng, B., Xia, X.,... He, Z. (2019). Accuracy assessment of plant height using an unmanned aerial vehicle for quantitative genomic analysis in bread wheat. Plant Methods, 15(1), 37. https://doi.org/10.1186/s13007-019-0419-7

Holman, F. H., Riche, A. B., Michalski, A., Castle, M., Wooster, M. J., & Hawkesford, M. J. (2016). High throughput field phenotyping of wheat plant height and growth rate in field plot trials using UAV based remote sensing. Remote Sens., 8(12), 1031. https://doi.org/10.3390/rs8121031

Iqbal, F., Lucieer, A., Barry, K., & Wells, R. (2017). Poppy crop height and capsule volume estimation from a single UAS flight. Remote Sens., 9(7), 647. https://doi.org/10.3390/rs9070647

Ji, Y., Chen, Z., Cheng, Q., Liu, R., Li, M., Yan, X.,... Yang, T. (2022). Estimation of plant height and yield based on UAV imagery in faba bean (Vicia faba L.). Plant Methods, 18(1), 26. https://doi.org/10.1186/s13007-022-00861-7

Jiang, Z., Tu, H., Bai, B., Yang, C., Zhao, B., Guo, Z.,... Zhang, J. (2021). Combining UAV-RGB high-throughput field phenotyping and genome-wide association study to reveal genetic variation of rice germplasms in dynamic response to drought stress. New Phytol., 232(1), 440-455. https://doi.org/10.1111/nph.17580

Küng, O., Strecha, C., Beyeler, A., Zufferey, J. C., Floreano, D., Fua, P., & Gervaix, F. (2011). The accuracy of automatic photogrammetric techniques on ultra-light UAV imagery. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XXXVIII-1/C22, 125-130. https://doi.org/10.5194/isprsarchives-XXXVIII-1-C22-125-2011

Ladino, K. S., & Sama, M. P. (2024). Short-term performance evaluation of real-time kinematic global navigation satellite system receivers in unmanned aircraft systems. J. ASABE, 67(6), 1433-1446. https://doi.org/10.13031/ja.16030

Li, B., Xu, X., Zhang, L., Han, J., Bian, C., Li, G.,... Jin, L. (2020). Above-ground biomass estimation and yield prediction in potato by using UAV-based RGB and hyperspectral imaging. ISPRS J. Photogramm. Remote Sens., 162, 161-172. https://doi.org/10.1016/j.isprsjprs.2020.02.013

Li, J., Schachtman, D. P., Creech, C. F., Wang, L., Ge, Y., & Shi, Y. (2022). Evaluation of UAV-derived multimodal remote sensing data for biomass prediction and drought tolerance assessment in bioenergy sorghum. Crop J., 10(5), 1363-1375. https://doi.org/10.1016/j.cj.2022.04.005

Madec, S., Baret, F., de Solan, B., Thomas, S., Dutartre, D., Jezequel, S.,... Comar, A. (2017). High-throughput phenotyping of plant height: Comparing unmanned aerial vehicles and ground LiDAR estimates. Front. Plant Sci., 8. https://doi.org/10.3389/fpls.2017.02002

Malambo, L., Popescu, S. C., Murray, S. C., Putman, E., Pugh, N. A., Horne, D. W.,... Bishop, M. (2018). Multitemporal field-based plant height estimation using 3D point clouds generated from small unmanned aerial systems high-resolution imagery. Int. J. Appl. Earth Obs. Geoinf., 64, 31-42. https://doi.org/10.1016/j.jag.2017.08.014

Mir, R. R., Reynolds, M., Pinto, F., Khan, M. A., & Bhat, M. A. (2019). High-throughput phenotyping for crop improvement in the genomics era. Plant Sci., 282, 60-72. https://doi.org/10.1016/j.plantsci.2019.01.007

Panday, U. S., Shrestha, N., Maharjan, S., Pratihast, A. K., Shahnawaz, Shrestha, K. L., & Aryal, J. (2020). Correlating the plant height of wheat with above-ground biomass and crop yield using drone imagery and crop surface model, a case study from Nepal. Drones, 4(3), 28. https://doi.org/10.3390/drones4030028

Pfeifer, N., Glira, P., & Briese, C. (2012). Direct georeferencing with on board navigation components of light weight UAV platforms. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XXXIX-B7, 487-492. https://doi.org/10.5194/isprsarchives-XXXIX-B7-487-2012

Polidori, L., & El Hage, M. (2020). Digital elevation model quality assessment methods: A critical review. Remote Sens., 12(21), 3522. https://doi.org/10.3390/rs12213522

Rui, Z., Zhang, Z., & Chen, T. (2024). A review of field plant phenotyping platform, sensing and navigation technology. In Z. Zhang, Y. Jiang, C. Wen, S. Men, & Y. Zhang (Eds.), Transforming technologies in smart agriculture: New applications during the past decade (pp. 1-46). Singapore: Springer. https://doi.org/10.1007/978-981-97-4141-0_1

Sanz-Ablanedo, E., Chandler, J. H., Rodríguez-Pérez, J. R., & Ordóńez, C. (2018). Accuracy of Unmanned Aerial Vehicle (UAV) and SfM photogrammetry survey as a function of the number and location of ground control points used. Remote Sens, 10(10), 1606. https://doi.org/10.3390/rs10101606

Schirrmann, M., Giebel, A., Gleiniger, F., Pflanz, M., Lentschke, J., & Dammer, K.-H. (2016). Monitoring agronomic parameters of winter wheat crops with low-cost UAV imagery. Remote Sens., 8(9), 706. https://doi.org/10.3390/rs8090706

Sefercik, U. G., Kavzoglu, T., Çölkesen, I., Nazar, M., Öztürk, M. Y., Adali, S., & Dinç, S. (2023). 3D positioning accuracy and land cover classification performance of multispectral RTK UAVs. Int. J. Eng. Geosci., 8(2), 119-128. https://doi.org/10.26833/ijeg.1074791

Shi, Y., Thomasson, J. A., Murray, S. C., Pugh, N. A., Rooney, W. L., Shafian, S.,... Yang, C. (2016). Unmanned Aerial Vehicles for high-throughput phenotyping and agronomic research. PLoS One, 11(7), e0159781. 10.1371/journal.pone.0159781

Štroner, M., Urban, R., Seidl, J., Reindl, T., & Broucek, J. (2021). Photogrammetry using UAV-mounted GNSS RTK: Georeferencing strategies without GCPs. Remote Sens., 13(7), 1336. https://doi.org/10.3390/rs13071336

Taddia, Y., Stecchi, F., & Pellegrinelli, A. (2019). Using DJI Phantom 4 RTK Drone For Topographic Mapping Of Coastal Areas. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W13, 625-630. https://doi.org/10.5194/isprs-archives-XLII-2-W13-625-2019

Tomaštík, J., Mokroš, M., Surový, P., Grznárová, A., & Merganic, J. (2019). UAV RTK/PPK method — An optimal solution for mapping inaccessible forested areas? Remote Sens., 11(6), 721. https://doi.org/10.3390/rs11060721

Villanueva, J. K., & Blanco, A. C. (2019). Optimization of Ground Control Point (GCP) configuration for Unmanned Aerial Vehicle (UAV) survey using structure from motion (SfM). Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-4/W12, 167-174. https://doi.org/10.5194/isprs-archives-XLII-4-W12-167-2019

Volpato, L., Pinto, F., González-Pérez, L., Thompson, L. G., Borém, A., Reynolds, M.,... Rodrigues, F. A. (2021). High throughput field phenotyping for plant height using UAV-Based RGB imagery in wheat breeding lines: feasibility and validation. Front. Plant Sci., 12. https://doi.org/10.3389/fpls.2021.591587

White, J. W., Andrade-Sanchez, P., Gore, M. A., Bronson, K. F., Coffelt, T. A., Conley, M. M.,... Wang, G. (2012). Field-based phenomics for plant genetics research. Field Crops Res., 133, 101-112. https://doi.org/10.1016/j.fcr.2012.04.003

Xiao, Q., Bai, X., Zhang, C., & He, Y. (2022). Advanced high-throughput plant phenotyping techniques for genome-wide association studies: A review. J. Adv. Res., 35, 215-230. https://doi.org/10.1016/j.jare.2021.05.002

Xie, C., & Yang, C. (2020). A review on plant high-throughput phenotyping traits using UAV-based sensors. Comput. Electron. Agric., 178, 105731. https://doi.org/10.1016/j.compag.2020.105731

Xie, T., Li, J., Yang, C., Jiang, Z., Chen, Y., Guo, L., & Zhang, J. (2021). Crop height estimation based on UAV images: Methods, errors, and strategies. Comput. Electron. Agric., 185, 106155. https://doi.org/10.1016/j.compag.2021.106155

Yang, W., Feng, H., Zhang, X., Zhang, J., Doonan, J. H., Batchelor, W. D.,... Yan, J. (2020). Crop phenomics and high-throughput phenotyping: Past decades, current challenges, and future perspectives. Molecular Plant, 13(2), 187-214. https://doi.org/10.1016/j.molp.2020.01.008

Yeh, M. L., Chou, Y. T., & Yang, L. S. (2016). The evaluation of GPS techniques for UAV-based photogrammetry in urban area. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLI-B1, 1079-1084. https://doi.org/10.5194/isprs-archives-XLI-B1-1079-2016

Yuan, W., Li, J., Bhatta, M., Shi, Y., Baenziger, P. S., & Ge, Y. (2018). Wheat height estimation using LiDAR in comparison to ultrasonic sensor and UAS. Sensors, 18(11), 3731. https://doi.org/10.3390/s18113731

Zhao, B., Li, J., Wang, L., & Shi, Y. (2020). Positioning accuracy assessment of a commercial RTK UAS. Proc. SPIE Defense + Commercial Sensing - Autonomous Air and Ground Sensing Systems for Agricultural Optimization and Phenotyping V.11414. SPIE. https://doi.org/10.1117/12.2557899

Zimmermann, F., Eling, C., Klingbeil, L., & Kuhlmann, H. (2017). Precise positioning of UAVS – dealing with challenging RTK-GPS measurement conditions during automated uav flights. ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., IV-2/W3, 95-102. https://doi.org/10.5194/isprs-annals-IV-2-W3-95-2017

Appendix

Figure A1. GCP locations of the five study fields: (A) ENREEC field planted with triticale (8 GCPs), (B) Havelock field planted with triticale (8 GCPs), (C) SCAL field planted with soybean (9 GCPs), (D) Havelock-HIPS field planted with maize (9 GCPs), and (E) Havelock-SAM field planted with maize (9 GCPs).

Figure A2. Correlation coefficient between wind speed and RMSE of plant height estimation over the growing season across all fields for the three methods: (A) Method-A: regular-GNSS with GCPs, (B) Method-B: RTK-GNSS with GCPs, and (C) Method-C: RTK-GNSS without GCPs.