The potential of unmanned aerial vehicles and field robotics for capturing key agronomic characteristics in the field
Author: Brooke Schofield, Larn McMurray, Christine Walela (SARDI Clare), Rohan Kimber, David Brunton and Kathy Ophel Keller (SARDI Waite) | Date: 09 Feb 2016
Background
In grains research continual monitoring and observation is required to maintain and quantify the performance of crop trials. These trials often consist of thousands of plots, which makes ground measurements and monitoring by visual estimates and handheld devices time consuming. Additionally, there is a potential for data bias to be introduced when plots are assessed by an inexperienced operator or by different operators. One method to obtain rapid measurements from a large number of plots and avoid such bias is to capture data autonomously from highly accurate sensors fixed to robotic platforms.
This research outlines the potential suitability and accuracy of incorporating both ground-based and aerial platforms into crop research trials. This project was designed to undertake research activities that evaluate the application of an autonomous field robot and assess a UAV for automated capture and processing of data aligned to phenotypic traits of interest within selected pulse and cereal field trials.
Key properties analysed include Normalised Difference Vegetation Index (NDVI), crop heigh,t and canopy growth rate, all of which are conventionally measured on-ground for each plot. Image capture using a field robot or UAV could be processed to distinguish and measure agronomic data as accurately as traditional on-ground measurements. This may reduce random error and saving human time, labour, and cost.
The objective of the study is two-fold:
- To evaluate the utility, efficiency, and practicality of using an autonomous ground-based system for acquiring and processing data where the data is already known to be useful, but is labour intensive and time consuming to acquire.
- To investigate whether the quality of information generated is an improvement or is value-added when compared with current data measurement practices in terms of coverage, resolution, precision, accuracy, and repeatability.
Methodology
Study site
Eight individual trials, including four cereal trials and four pulse trials, were sown at the Turretfield Research Centre in the lower Mid North of South Australia (SA). These trials were specifically designed with varied agronomic traits and objectives in order to evaluate the capability of the UAV and robot imagery. Plot size in all experiments was 1.35m (six rows of 12 inch spacing) by 10m with three replicates (Table 1).
Table 1: Trial site details.
Trial name | Crop and size | Objectives to | Field |
---|---|---|---|
Plant Density | Two varieties of peas, lentils, chickpeas. Five rates. | Resolution of plant densities and height. | NDVI, height, yield, biomass, flowering score, lodging. |
Nodulation Failure | One variety of chickpea, three treatments. | Differences in plant height, crop growth parameters | NDVI, height, yield, biomass. |
Pulse Disease | Two varieties of beans, chickpeas. Three treatments. | Severity and extent of disease. | Disease scores, NDVI, yield, height. |
Pulse Herbicide | Two varieties of peas and lentils, five treatments. | Severity and extent of herbicide treatments and damage. | NDVI, damage scores, height, biomass, yield. |
Growth Regulation | Two varieties of wheat and barley, two treatments. | Differences in plant height, crop growth parameters. | NDVI, height, yield. |
Cereal Herbicide | Two varieties of wheat and barley, three treatments. Severity and extent of herbicide treatments and damage. | Severity and extent of herbicide treatments and damage. | NDVI, height, yield, damage scores. |
Cereal Disease | Two varieties of wheat and barley, three treatments. | Severity and extent of disease. | NDVI, height, yield, disease scores. |
Seeding Rate | Two varieties of wheat and barley, four rates. | Resolution of plant densities and height. | Biomass, NDVI, height, yield. |
The UAV was also flown over additional agronomic research trials run by the SARDI New Variety Agronomy group located at Tarlee and Mallala and the robot at Mallala and Pinery.
Ground measurements - manual capture
Manual in-field measurements including plant height, biomass, NDVI, and disease scores were carried out as near as possible to the flights in order to validate the UAV method. A spectroradiometer equipped with an active sensor (GreenSeeker® handheld crop sensor) was used to measure the average NDVI for each plot.
A one metre biomass sample was taken in for each plot in selected trials. For the fresh biomass the samples were weighed in the field. A subsample was then taken and dried in an oven at 60◦C for 60 hours and weighed again to provide a dry biomass weight.
Plant heights were taken from the base of the plant to the tip of the highest leaf for four plants in each plot. These values were then combined to give an average height as well as some indication of variability within each plot.
In addition to quantitative measurements, plots were also assessed and scored visually for parameters such as disease, herbicide damage, weed presence, and health.
Data capture - unmanned aerial vehicle (UAV)
The fixed wing UX5® Unmanned Aircraft System developed by Trimble (Trimble Navigation, Ltd, California) was used to carry out the flights. Five flight campaigns were carried out during the 2015 growing season at the Turretfield, Tarlee and Mallala sites.
The first flight, on 25 May 25, acted as a bare earth flight and was used for the creation of crop surface models (CSMs) as a method for measuring crop height. Further acquisitions were taken on July 27, August 24, October 16, and November 11.
NDVI is an index which measures the difference between the visible light absorbed by green vegetation and the amount of solar energy they reflect in the near infrared region of the spectrum. NDVI provides a measure of overall plant greenness, condition, and density. Healthy, green vegetation have NDVI values of up to one, but NDVI decreases as plants exhibit water stress, disease, or die. Unhealthy or necrotic vegetation and bare soil have NDVI values around zero, while water often shows negative values.
Both a RGBA (red, green, blue, alpha) true colour image and a colour infrared image (required for calculating NDVI) were obtained for each date. The NDVI images were used to visually infer the overall plant health for each plot. It was also analysed using ArcGIS software in order to extract a mean NDVI value for each plot.
In addition, evaluation of a multi-spectral sensor fitted to a quadcopter was assessed by the University of Adelaide’s Unmanned Research Aircraft Facility (URAF). They conducted additional imaging activity in September and will provide data for additional comparisons between platforms and field robotic imaging techniques.
Data capture - autonomous field robot
The Australian Centre for Field Robotics (ACFR) has built a ground-based robot called the Ladybird which offers both established and novel forms of data with high resolution and broad coverage. The Ladybird is equipped with several sensors including stereo visual, thermal, panospheric and hyperspectral cameras, light detection and ranging (Lidar), as well as a highly accurate inertial global positioning system.
Data capture was carried out with the Ladybird at the Turretfield, Pinery, and Mallala sites. Two field campaigns were conducted by the ACFR on August 17-21 and September 23-25 at the three field research trials at the three sites. The capture by the Ladybird and UAV was coordinated with routine field-based data collected by SARDI staff in the Pulse and Cereal Pathology group (Waite campus) and the Crop Evaluation Group (Clare campus). This collaborative study also included John Weiss (Victoria Department of Economic Development, Jobs, Transport and Resources) as part of a Plant Biosecurity CRC project (PBCRC project 2135) to optimise the use of UAV platforms for plant biosecurity surveillance, when detailed disease assessments were collected in August and September during field scanning by the Ladybird and UAVB platforms.
The robot also collected NDVI as well as height and true-colour imagery. Data collected from the robot is being analysed and processed by the researchers at ACFR.
Preliminary results and discussion
Methods of measuring NDVI
In general, the UAV NDVI results were lower than those measured with the Greenseeker®. However, the UAV consistently gave lower amplitudes across trials and acquisition dates (Table 2). The average ground-measured Greenseeker® NDVI for each plot was compared with that obtained from the UAV imagery in order to obtain a relationship between the two (Figure 1). If a solid relationship can be numerically defined between the two methods of measuring NDVI, it may be possible to infer the Greenseeker® NDVI when only the UAV method is used (and vice versa).
The change of NDVI as the plant grows from a developing seedling through to a green plant and then to maturity can be seen from the graphs. All crops start off with a relatively low NDVI. This is due to the bare soil having an influence within the low density seedlings. As the growing season progresses, the crops have a higher leaf area and are green and healthy, hence displaying a higher NDVI. Finally, as the crops begin to mature and turn yellow, a low NDVI is again achieved.
It is generally accepted that a cereal plant will exhibit its highest ‘greenest’ at Zadoks growth stage 59, corresponding to the stage immediately prior to the onset of anthesis (flowering). The data was split up into the dates prior to and post the onset of anthesis which allowed a trendline to be fitted to the data representing the NDVI trend during plant growth and during plant maturity.
NDVI as measured by the Greenseeker® and UAV for wheat (Figure 1a) and barley (Figure 1b) shows a strong linear relationship as the plant grows greener. Once these crops begin to mature and lose their greenness, the NDVI values begin to decrease, but follow a polynomial relationship. The lentil density (Figure 1c) and pea density (Figure 1d) NDVI data also shows this trend (a linearly increasing NDVI before hitting the threshold at maximum greenness, at the onset of anthesis, and then decreasing according to a polynomial relationship).
The NDVI data from the beans (Figure 1e) shows a weak linear relationship. However, it is greatly lacking data so the results do not have a high confidence. The chickpea NDVI data (Figure 1f-1h) has a very different relationship to the wheat and barley data. The chickpea NDVI values follow a third-order polynomial relationship and are increasing throughout the season without hitting a threshold and decreasing. This is because the chickpeas matured later than the cereals and their maturity phase was not captured by the UAV or Greenseeker®. However, it is unlikely that the change in NDVI of chickpeas throughout the season actually follows a third-order polynomial relationship. It is more likely to exhibit a linear or second-order polynomial relationship similar to the wheat, barley, pea density, and lentil density data. Further data is required to prove or disprove this hypothesis.
The NDVI values measured in November did not follow the trend of the other dates. This data was collected once the crops had almost reached maturity and had low NDVI values. The above trends have only taken into consideration the crops prior to the start of maturity. This is a major limitation in the study method. Currently, the derived relationships and trendlines are only applicable to crops which are still dominated by green growth. This can be improved by collecting more frequent NDVI data during the different stages of maturity where the NDVI is changing rapidly.
Additional data is also required later in the season in order to capture the chickpea maturity stages. Since the chickpeas matured late in the season, no data was collected once their NDVI had started to decrease. Therefore, the overall trend of the chickpea NDVI evolution was not captured.
Further data is also required at early stages of the growing period in order to more accurately identify the change in NDVI. A large gap exists in many of the graphs between the July and August acquisition times. The trends would be improved if fortnightly measurements were taken rather than monthly.
NDVI, destructive, and ground-based measurements
A comparison was carried out between Greenseeker® NDVI measurements, destructive biomass cuts and yield (Figure 2). The NDVI measurements for September 24 were chosen for the comparison because they align with the timing of the biomass cuts. Overall, the data shows a linear relationship between the NDVI measured by the Greenseeker® on September 24 and the yield weight at harvest (Figure 2a) suggesting that a higher yield will be obtained from a plot with a higher NDVI value. The data should be further analysed to see whether the linear relationship also holds true for NDVI measurements taken on different dates.
The dry biomass weight was also compared to the Greenseeker® NDVI measurements in the hope of finding a relationship between the two, and hence be able to predict yield from earlier NDVI measurements. An exponential relationship exists between biomass and Greenseeker® NDVI (Figure 2b). However, it appears to vary amongst different crop types. As with the previous situation, further analysis across multiple dates is required as well as analysis of individual crop types in order to validate the results.
Finally, the dry biomass weights were compared to the final yields to determine whether a higher biomass necessarily leads to a higher yield. Overall, the data appears to be related by a polynomial function. However, the individual crop types should be separated and treated individually before solid relationships are implied.
Date | Trial | Cereal Seeding Rate | Cereal Disease | Cereal Herbicide | Cereal Growth Regulant | Herbicide Lentils | Herbicide Peas | Disease Beans | Disease Chickpea | Nodulation Chickpeas | Density Peas | Density Lentils | Density Chickpea |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
27-07 | GS NDVI | 0.26-0.60 | 0.33-0.55 | 0.25-0.53 | 0.23-0.46 | 0.21-0.27 | 0.27-0.34 | 0.33-0.43 | 0.18-0.24 | 0.19-0.23 | 0.18-0.50 | 0.17-0.33 | 0.14-0.29 |
GS Amplitude | 0.34 | 0.22 | 0.28 | 0.23 | 0.06 | 0.06 | 0.10 | 0.06 | 0.04 | 0.32 | 0.16 | 0.15 | |
UAV NDVI | 0.16-0.25 | 0.18-0.22 | 0.15-0.21 | 0.15-0.19 | 0.13-0.15 | 0.14-0.17 | 0.18-0.19 | 0.11-0.13 | 0.11-0.12 | 0.10-0.18 | 0.09-0.13 | 0.07-0.11 | |
UAV Amplitude | 0.09 | 0.04 | 0.06 | 0.04 | 0.02 | 0.03 | 0.01 | 0.02 | 0.01 | 0.08 | 0.04 | 0.04 | |
20-08 | GS NDVI | 0.59-0.89 | 0.74-0.86 | 0.47-0.84 | 0.55-0.82 | 0.20-0.60 | 0.32-0.72 | 0.63-0.72 | 0.31-0.48 | 0.37-0.47 | 0.34-0.74 | 0.35-0.72 | 0.17-0.53 |
GS Amplitude | 0.3 | 0.12 | 0.37 | 0.27 | 0.40 | 0.40 | 0.12 | 0.17 | 0.10 | 0.40 | 0.37 | 0.36 | |
UAV NDVI | 0.45-0.48 | 0.44-0.47 | 0.43-0.47 | 0.41-0.46 | 0.27-0.36 | 0.33-0.37 | 0.37-0.39 | 0.31-0.33 | 0.32-0.33 | 0.31-0.37 | 0.31-0.35 | 0.27-0.34 | |
UAV Amplitude | 0.03 | 0.03 | 0.03 | 0.05 | 0.09 | 0.04 | 0.02 | 0.02 | 0.01 | 0.06 | 0.04 | 0.07 | |
19-10 | GS NDVI | 0.14-0.39 | 0.24-0.44 | 0.23-0.46 | 0.18-0.41 | 0.11-0.63 | 0.29-0.42 | 0.43-0.60 | 0.54-0.66 | 0.59-0.73 | 0.18-0.32 | 0.34-0.50 | 0.41-0.67 |
GS Amplitude | 0.15 | 0.20 | 0.23 | 0.23 | 0.52 | 0.13 | 0.17 | 0.12 | 0.14 | 0.14 | 0.16 | 0.26 | |
UAV NDVI | 0.28-0.39 | 0.35-0.40 | 0.35-0.40 | 0.32-0.39 | 0.20-0.37 | 0.26-0.29 | 0.34-0.38 | 0.38-0.40 | 0.37-0.39 | 0.21-0.27 | 0.29-0.33 | 0.32-0.36 | |
UAV Amplitude | 0.11 | 0.05 | 0.05 | 0.07 | 0.17 | 0.03 | 0.04 | 0.02 | 0.02 | 0.06 | 0.04 | 0.04 | |
11-11 | GS NDVI | 0.14-0.16* | n/a | n/a | n/a | 0.12-0.18* | 0.12-0.15* | 0.17-0.19* | 0.23-0.37 | 0.25-0.35 | 0.13-0.17 | 0.16-0.23 | 0.18-0.31 |
GS Amplitude | 0.02* | n/a | n/a | n/a | 0.06* | 0.03* | 0.02* | 0.14 | 0.10 | 0.14 | 0.07 | 0.13 | |
UAV NDVI | 0.12-0.41 | 0.11-0.39 | 0.11-0.38 | 0.13-0.39 | 0.19-0.28 | 0.18-0.23 | 0.34-0.43 | 0.23-0.29 | 0.25-0.30 | 0.23-0.29 | 0.29-0.35 | 0.22-0.33 | |
UAV Amplitude | 0.29 | 0.28 | 0.27 | 0.26 | 0.09 | 0.05 | 0.09 | 0.06 | 0.05 | 0.06 | 0.06 | 0.11 |
*= incomplete data set
GS= Greenseeker®
Figure 1: Relationship between the average ground-measured Greenseeker® NDVI and that obtained from the UAV imagery.
Figure 2: Comparison between Greenseeker® NDVI measurements, destructive biomass cuts and yield.
Further results are yet to be deduced, but the following outcomes are expected to be achieved from this research:
- A greater understanding of the capacity for autonomous ground based platforms and UAVs as potential labour-saving devices for agricultural studies and field based trials and the possible trade-offs in terms of resolution, coverage and accuracy.
- A greater understanding of the quality of information that can be provided by the robot and UAV, particularly in comparison with existing manual (labour intensive) approaches.
- Ideas for next steps towards the adoption of new technology in this field.
Conclusion
This work reports a proof of concept exercise on a process to automate broad-scale crop evaluation in order to reduce time spent undertaking ground-based fieldwork and to remove any observer bias that may be present in results. It provided the first opportunity to apply the Ladybird technology developed by the ACFR to field crop research trials. Comprehensive reporting of outcomes, including the Ladybird data, will be included in reports to be finalised in March 2016.
Acknowledgments
GRDC and SAGIT are gratefully acknowledged for their support via the internship projects (SAGIT project S1513 and GRDC project DAS00142) as are the SARDI New Variety Agronomy groups at Waite and Clare for the use of existing agronomic research trials from GRDC-funded projects DAS00113, DAV00113 and DAS00100.
Martin Peters (FarmingIT) undertook all the UAV flights, performed the image post-processing, and created the CSMs. James Underwood and Alex Wendel (ACFR) were responsible for completing the field campaigns with the Ladybird robot. This was a large component of the project and the effort and work of Martin, James, and Alex is strongly acknowledged and appreciated. Acknowledgement and thanks is also given to Kenneth Clarke and Ramesh Raja Segaran (Adelaide University) for performing the quadcopter flight at Turretfield. A final thanks and acknowledgment is given to John Weiss (DEDJTR) for assisting with detailed disease scoring at Turretfield.
Useful Resources
Glossary of abbreviations and key terms
• UAV: Unmanned aerial vehicle.
• NDVI: Normalised difference vegetation index.
• CIR: Colour-infrared image. Image taken across the visible and near-infrared region of the spectrum.
• RGB: A colour image providing a true colour representation.
• CSM: crop surface model. A 3D cloud data set representing the surveyed area. Measured by the UAV during flight.
• URAF: Unmanned Research Aircraft Facility. The University of Adelaide.
• ACFR: Australian Centre for Field Robotics. The University of Sydney.
• DEDJTR: The Department of Economic Development, Jobs, Transport and Resources, Victoria Government.
• Hyperspectral: information collected over hundreds of narrow bands throughout the electromagnetic spectrum.
• Multispectral: information collected over a few large bands within specific regions of the electromagnetic spectrum.
• Lidar: technique measuring distance by emitting a laser and analysing the returning light.
Contact Details
Brooke Schofield
GRDC Project Code: DAS00142,
Was this page helpful?
YOUR FEEDBACK
To protect your privacy, please do not include contact information in your feedback. If you would like
a response, please contact us.