Project

Maricopa Phenotyping

Goal: Understand the GxE factors that drive crop phenotypic expression
Identify sensing technology to non-destructively quantify plant traits in the field
Deploy positioning systems to spatiotemporally suspend sensing arrays across field plots
Develop computational pipelines to derive biological information from electronic sensing

Methods: Agricultural Plant Science, Data Analysis, Data Science, Phenotyping, Agriculture Research, Crop Breeding, Scientific Visualization, field phenotyping, proximal sensing, plant sensing, imaging, agriscience, high throughput phenotyping, in-field phenotyping, crop science, electronic sensing, crop modeling

Date: 31 March 2015

Updates

0 new
76
Recommendations

0 new
11
Followers

0 new
37
Reads

0 new
1675

Project log

Matthew M. Conley
added a research item
Using a lightweight and narrow wheeled electric-drive field proximal sensing platform, local plant and environmental measurements of cotton and other crops were collected on the Maricopa Agricultural Center farm as part of the Arid Land Agricultural Research Center’s plant phenotyping program. Beginning in 2012, the single pass continuous scan high frequency approach to measure agricultural crops has developed into sensing the analog tri-metric of plant health status, which consists of canopy temperature, canopy height and canopy spectral reflectance. The method also captures environmental micro-energetics such as vertical air temperature and humidity profiles, incoming solar radiation, plus wind speed and direction. Positioning additional sensors in a low mounted and side angled view has also complimented the typical above canopy nadir sensor views. Fusing of the metrics has resulted in single term descriptive calculations and vegetation signature indexes. Color imagery has provided inputs for plant emergence timing, cover estimations, and relations of vigor, as well as providing input for structure from motion surface and volumetric modeling. Taken together, the data rich amalgamation of proximal measurements offers a viable method for biological discovery, physically positioned between hand-collected and unoccupied aerial derived sampling.
Matthew M. Conley
added an update
High-throughput plant phenotyping in field conditions using proximal sampling is useful to support agricultural research by the encoding of plant genetic response to environment and management influence. Proximal Sensing Carts employed by the USDA in Maricopa Plant Group Phenotyping Team have successfully measured multiple plant traits and environmental variables in field conditions by typically focusing on a two-meter bounded volumetric scanning measurement space.
The presentation, “F113 2021 Wolverine Run4 Data Image Review”, shows a phenotyping method process applied to a short turf grass, where we collected initial test data to aid research experiment development and method tuning. Included are examples of sensor data returns, where the sensor performance and associated context are visible.
Importantly, examples of GoPro Hero7 Black action camera imagery are presented. The limit of effective image geometric and color resolution was examined to verify if the inexpensive camera technology would be sufficient to compute plant cover area and height. Basically, results showed that plants and soil were geometrically characterized with about 1 mm ground sampling resolution in a single 2-meter distant image, but that optical aberration inherent between images and cameras with different views, resulted in spatial errors of about 2.5 cm across a 1 m plot space. Also, the pixel colors returned did show increased base spatial error beyond 1 mm, apparent at the edge of green leaf and brown soil transitions where color bleed-over and a yellow color artifact was detected. Of note is that image pixel saturation increase or color shift did not harm Metashape structure from motion model formulation. Instead, increased pixel color tended to increase the number of pairwise matched model tie-points.
 
Matthew M. Conley
added an update
Single-pass continuous high frequency two-meter proximal photographic sampling of agricultural plants grown in field conditions can resolve color and 3D geometric phenotypic traits, for example, if an adequate view angle, image overlap, and resolution are provided as input to modern structure from motion software. The USDA in Maricopa Plant Group Phenotyping Team presentation, “Summer Proximal Imagery Report”, notes an approach where multiple GoPro Hero 6 and 7 Black GPS action cameras were mounted to a moving Proximal Sensing Cart, and then set to continuously record plant canopy color and structure across more than one acre of two crop rows for cotton and soybean plants growing in experimental plots of the University of Arizona Maricopa Agricultural Center farm, USA.
 
Matthew M. Conley
added a research item
High-throughput plant phenotyping in field conditions using proximal sampling is useful to support agricultural research by the encoding of plant genetic response to environment and management influence. Proximal Sensing Carts employed by the USDA in Maricopa Plant Group Phenotyping Team have successfully measured multiple plant traits and environmental variables in field conditions by typically focusing on a two-meter bounded volumetric scanning measurement space. The presentation, “F113 2021 Wolverine Run4 Data Image Review”, shows a phenotyping method process applied to a short turf grass, where we collected initial test data to aid research experiment development and method tuning. Included are examples of sensor data returns, where the sensor performance and associated context are visible. Importantly, examples of GoPro Hero7 Black action camera imagery are presented. The limit of effective image geometric and color resolution was examined to verify if the inexpensive camera technology would be sufficient to compute plant cover area and height. Basically, results showed that plants and soil were geometrically characterized with about 1 mm ground sampling resolution in a single 2-meter distant image, but that optical aberration inherent between images and cameras with different views, resulted in spatial errors of about 2.5 cm across a 1 m plot space. Also, the pixel colors returned did show increased base spatial error beyond 1 mm, apparent at the edge of green leaf and brown soil transitions where color bleed-over and a yellow color artifact was detected. Of note is that image pixel saturation increase or color shift did not harm Metashape structure from motion model formulation. Instead, increased pixel color tended to increase the number of pairwise matched model tie-points.
Matthew M. Conley
added a research item
Single-pass continuous high frequency two-meter proximal photographic sampling of agricultural plants grown in field conditions can resolve color and 3D geometric phenotypic traits, for example, if an adequate view angle, image overlap, and resolution are provided as input to modern structure from motion software. The USDA in Maricopa Plant Group Phenotyping Team presentation, “Summer Proximal Imagery Report”, notes an approach where multiple GoPro Hero 6 and 7 Black GPS action cameras were mounted to a moving Proximal Sensing Cart, and then set to continuously record plant canopy color and structure across more than one acre of two crop rows for cotton and soybean plants growing in experimental plots of the University of Arizona Maricopa Agricultural Center farm, USA.
Matthew M. Conley
added a research item
Proximally sampled Structure from Motion multi-angular photogrammetric optical modeling has proven useful in architectural and physical archeological research. Yet in agricultural plant phenotyping the method is less mature, where the optimum camera view angle and distance from target to capture a complicated and often obfuscated biologic of living plant canopies in the field is less understood. The USDA in Maricopa Plant Group Phenotyping Team investigated several photographic capture approaches and found success in point-cloud modeling of cotton experimental plants growing forty inches apart outdoors, by using a single GoPro GPS action camera positioned above and beside rows of the crop plants. The presentation offered, “GoPro image captures for Metashape SfM point cloud modeling”, shows initial results, where the camera was manually moved around plants to test which view angle and how many images would be required to create color pixel tie-point models representing the cotton plants, as computed in Agisoft Metashape Pro software. Of interest was the side view reconstruction, which can be difficult to obtain from a typical more remote un-occupied platform aerial vantage. Additionally, camera views at ninety degrees to the plant rows resolved the understory fruiting bodies best in a single image, but they did not stitch together well and rather introduced unwanted background sky pixels.
Matthew M. Conley
added a research item
It is common in contemporary environmental data acquisition, for a UAS platform to collect RGB type imagery of agricultural experimentation and provide data input to compute orthomosaic composites of field plots, or basic volumetric models via multi-angular structure from motion derived 3D color point clouds. Acres of ground may be sampled in minutes and results are generally good to support experimental biological targets. However, the resultant optical data constructions can be somewhat low resolution unless a robust and therefore often heavy and expensive camera is used which increases operational cost. Proximal terrestrial RGB image collection effectively offers high optical resolution by suspending an inexpensive camera close to the ground targets, at perhaps two meters distant. Closer proximity allows increased pixels on target, but it does require more images to generate the orthomosaic and cloud models of the experimental plot areas, and a subsequent increased time to cover the acreage. The USDA in Maricopa Plant Group Phenotyping Team presentation, “Multiple GoPros in nadir view SfM Metashape initial model compute Soybean”, is offered as example of our first field image collections from Avenger Rig in 2019 that showed high frequency proximal sampling viability in using inexpensive GoPro HERO 6 Black action GPS cameras and the Agisoft Metashape software to optically model a complicated soybean canopy across two acres of experimental plots. This image test method was conducted in addition to our regular roaming geolocated tri-metric of plant health status and environmental micro-meteorological phenotyping data acquisitions, and therefore compliments those analog point data streams with photographic color grids.
Matthew M. Conley
added a research item
It is now somewhat common for simulation modeling in silico of agricultural plants to represent genetic, environmental, and managerial influence and compute a predicted vegetative biological growth construct. Yet to effectively create or validate the programmatic parameters employed, there is need for an accurate measurement of the environmental status within which the plants of known genetics are grown. Often a standard weather station is used to measure the micro-meteorological phenomena near the genotypic accessions, where a course time granulation of basic environmental information is attributed as input to the numeric construct. However, additional parameters at higher temporal frequency, that include soil and surface energetic status, and arrayed samples distributed in a vertical spatial profile may add to an energy balance and thermodynamic spatial understanding, and perhaps further support compute simulation. The USDA in Maricopa Plant Group phenotyping team presentation note, “EOS IRT stations F119 2021”, the third in a series of three, shows images and end of season example data from two modified weather stations recording soybean plants. The two stations each measured continuous 5Hz non-contact surface temperatures of a single row of plants from both a nadir and angled view, along with air temperature and humidity, solar radiation, soil moisture and temperature in the upper soil zone, and contact temperatures from several type-T thermocouple junctions placed in the soil at different depths and placed both on the soil bed and in the soil furrow. Photographic images and timeseries charts are presented to show some initial result raw data, and the method of measurement approach tested. This preliminary investigation involved a test setup of high frequency environmental and plant sampling that recorded most of the 2021 summer season via two station locations, each measuring the same soybean variety. The approach allowed generation of a dataset that included repeated measures of dynamic and changing plant growth conditions. Therein, multiple ambient sub-surface wetting and weather events resulted in substantial thermal and latent heat fluctuations. These patterns appeared to resolve micro-environmental and phenotypic differences likely to support crop growth simulation modeling. Measurements were collected in the spatial and temporal context of related roaming proximal phenotypic platforms, UAS derived imagery, and robust physical sampling, among other experimental parameters. We considered adding additional metrics to the IRT stations, such as the optical reflectance NDVI and PRI vegetation indices, and a vertical profile sampling of air temperature and humidity, as well as wind speed and soil matric potential. We also considered understory solar interception sampling and/or up looking imagery. Many hand-held images were taken to investigate a view of the changing vegetative, mineral, and environmental measurement areas, and to image the biological target from different angles and positions. Therefore, what measurement approach would best compliment generic and collaborative phenotypic modeling? What total parameters are needed, and at what physical locations should measurements be taken? What temporal and spatial granularity and accuracy is desired?
Matthew M. Conley
added a research item
Simulation modeling in silico of agricultural plants, to represent genetic, environmental, and managerial influence, and compute a predicted growth of the vegetative biological construct is now somewhat common. Yet to effectively create or validate the programmatic parameters employed, there is need for an accurate measurement of the environmental status within which the plants of known genetics are grown. Often a standard weather station is used to measure the micro-meteorological phenomena near genotypic accessions, where a course time granulation of basic environmental information is attributed as input to the simulation model. However, additional parameters at higher temporal frequency, that include soil and surface energetic status, and arrayed samples distributed in a vertical spatial profile, may add to an energy balance and thermodynamic spatial understanding, and perhaps further support compute simulation. The second in a series of three USDA in Maricopa Plant Group phenotyping team presentation notes, labeled “F119 2021 IRT stations mid-season report”, shows example data from a pair of modified weather stations recording soybean row-crop plants. The two stations each recorded 5Hz non-contact canopy surface area temperature emissions from a single row of plants, in both a nadir and angled view, along with local air temperature and humidity, incoming solar radiation, soil moisture and temperature at 20cm, and temperatures from several type-T thermocouple junctions placed in the soil at different depths. With a focus on thermal metrics, questions presented were: At what spatial locations does temperature sampling best support crop simulation modeling? For which soil depths, and how many total measurements are most valuable? Is a spatial profile of air temperatures or leaves within the canopy desired? How does the non-contact surface temperature recordings of infrared radiometers compliment the contact temperatures of thermocouple wire junctions? Can an angled radiometric thermometer better sample heat emission from a row of plants than a nadir view?
Matthew M. Conley
added a research item
The simulation modeling in silico of agricultural plants represents genetic, environmental, and managerial influence, to compute the predicted growth of a vegetative biological construct. Yet to effectively create or validate the programmatic parameters employed, there is need for accurate measurement of environmental status within which plants of known genetics are grown. Often a standard weather station is used to measure micro-meteorological phenomena near the genotypic accessions, where a course time granulation of basic environmental information is attributed as input to the simulation modeling. However, additional parameters at higher frequency, that include soil and surface energetic status, and repeated samples distributed in a vertical spatial profile, may add to energy balance and thermodynamic understanding and perhaps further support compute simulation. The USDA in Maricopa Plant Group phenotyping team presentation note, “IRT stations initial setup”, the first in a series of three, shows initial setup notation of a modified weather station placed in the field of a 2021 soybean experiment conducted at the Maricopa Agricultural Center, USA. The station recorded 5Hz non-contact canopy area surface temperature of a single row of plants from both a nadir and angled view, along with air temperature and humidity, solar radiation, soil moisture, and temperatures from several type-T thermocouple junctions placed at the soil surface and on the underside of leaves. The question presented was, what metrics would best support crop simulation modeling and how best to sample them?
Matthew M. Conley
added a research item
The sustainable production of upland cotton, an economically important fiber crop, is threatened by changing environmental factors including high temperatures and low-soil water content. Both high heat and low-soil water can reduce net photosynthesis resulting in low fiber yields or poor fiber quality. Leaf chlorophyll content has a direct relationship with photosynthetic rate. Understanding how high heat and low-soil water affect chlorophyll content can identify opportunities for breeding improvement that will lead to sustainable fiber yields. A two-year field trial located in Maricopa Arizona measured leaf chlorophyll content, available soil water, ambient air temperatures, and cotton growth measurements collected by a high-clearance tractor equipped with proximal sensors. The results showed that low-soil water significantly increased leaf chlorophyll content, while high temperatures significantly reduced content. Structured equation modeling revealed that cotton may divert available resources to leaf area and chlorophyll content for the production of photosynthates during periods of high temperatures.
Matthew M. Conley
added a research item
Dr. Kevin Bronson provides a dataset representing the third of three consecutive years of cotton and nitrogen management experimentation in Field 113 of the Maricopa Agricultural Center, Arizona USA. Included is an intermediate analysis mega-table of correlated and calculated parameters, laboratory analysis results generated during the experimentation, plus high-resolution plot level intermediate data analysis tables of SAS process output, as well as the complete raw data sensor recorded logger outputs. This third year of F113 cotton experimentation includes a large utilization of nitrogen-15 isotope tracing to support evaluation of nitrogen use and uptake. The UC Davis Stable Isotope Facility (SIF - https://stableisotopefacility.ucdavis.edu/) provided laboratory analysis of samples to determine isotope percent recovery. https://data.nal.usda.gov/dataset/bronson-files-dataset-10-field-113-2018-cotton Visible in the cover image is Dr. Bronson operating his Hamby proximal sensing rig in the field.
Matthew M. Conley
added a research item
Dr. Kevin Bronson provides a dataset representing the second of three consecutive years of cotton and nitrogen management experimentation in Field 113. Included is an intermediate analysis mega-table of correlated and calculated parameters, laboratory analysis results generated during the experimentation, plus high-resolution plot level intermediate data analysis tables of SAS process output, as well as the complete raw data sensor recorded logger outputs. This dataset includes a large utilization of nitrogen-15 isotope tracing to support evaluation of nitrogen use and uptake. The UC Davis Stable Isotope Facility (SIF - https://stableisotopefacility.ucdavis.edu/) provided laboratory analysis of samples to determine isotope percent recovery. https://data.nal.usda.gov/dataset/bronson-files-dataset-9-field-113-2017-cotton
Matthew M. Conley
added a research item
Dr. Kevin Bronson provides a dataset representing the first of three consecutive years of cotton and nitrogen management experimentation in MAC farm Field 113, Arizona USA. Included, is an intermediate analysis mega-table of correlated and calculated parameters, laboratory analysis results generated during the experimentation, plus high-resolution plot level intermediate data analysis tables of SAS process output, as well as the complete raw data sensor recorded logger outputs.
Matthew M. Conley
added a research item
Dr. Kevin Bronson provides a second experiment year of Field 13 nitrogen and water management in cotton agricultural research data for compute, including notation of field events and operations, an intermediate analysis mega-table of correlated and calculated parameters, and laboratory analysis results generated during the experimentation, plus high-resolution plot level intermediate data analysis tables of SAS process output, as well as the complete raw data sensor recorded logger outputs. https://data.nal.usda.gov/dataset/bronson-files-dataset-7-field-13-2015
Matthew M. Conley
added a research item
Dr. Kevin Bronson provides another nitrogen and water management in cotton agricultural research raw dataset for compute. Dr. Bronson modified an old high-clearance Hamby 6000 rig, adding a tank and pump with a rear boom, to perform precision liquid N applications. A Raven control unit with GPS supplied variable rate delivery options. The 12 volt Holland Scientific GeoScoutX data recorder and associated CropCircle ACS-470 sensors with GPS signal, was easy to mount and run on the vehicle as an attached rugged data acquisition module, and allowed the measuring of plants using custom proximal active optical reflectance sensing. The HS data logger was positioned near the operator, and sensors were positioned in front of the rig, on forward protruding armature attached to a hydraulic front boom assembly, facing downward in nadir view 1 m above the average canopy height. A 34-size class AGM battery sat under the operator and provided the data system electrical power supply.
Matthew M. Conley
added a research item
Dr. Kevin Bronson provides a second year of nitrogen and water management in wheat agricultural research dataset for compute. Ten irrigation treatments from a linear sprinkler were combined with nitrogen treatments. This dataset includes notation of field events and operations, an intermediate analysis mega-table of correlated and calculated parameters, including laboratory analysis results generated during the experimentation, plus high resolution plot level intermediate data tables of SAS process output, as well as the complete raw data sensor records and logger outputs.
Matthew M. Conley
added a research item
Dr. Kevin Bronson provides this unique nitrogen and water management in wheat agricultural research dataset for compute. Ten irrigation treatments from a linear sprinkler were combined with nitrogen treatments. This dataset includes the complete raw sensors records and logger outputs. In the picture, Mr. Matt Hagler is shown starting the GeoScout device so to record AOS reflectance data from the PSCM1 platform 01/04/2013.
Matthew M. Conley
added a research item
Data is on the National Agricultural Library’s, USDA Agricultural Data Commons, made available by Dr. Kevin Bronson, with intention to support machine read and compute awareness. In the dataset cover image, the Proximal Sensing Cart Mark1 is seen suspending a High-Throughput Plant Phenotyping sensing cluster in proximal environmental data acquisition operation. Views of first-year Guayule bush in flower are shown looking down crop rows, from the side, and in nadir. In the distance, older larger plants are visible. Parthenium argentatum is valued as a commodity plant today - https://www.bridgestone.com/technology_innovation/natural_rubber/guayule/ This “new crop” holds rich history and continued promise -https://doi.org/10.1016/j.indcrop.2015.03.042 Dr. Diaa Eldin M Elshikha is credited with recent and comprehensive study of this important rubber bush, integrating a wide depth and breadth of research disciplines and experts understanding this genetic asset. https://www.researchgate.net/publication/351244705_Nitrous_oxide_emissions_N_uptake_biomass_and_rubber_yield_in_N-fertilized_surface-irrigated_guayule https://www.researchgate.net/publication/332870724_High_guayule_rubber_production_with_subsurface_drip_irrigation_in_the_US_desert_Southwest https://www.researchgate.net/publication/353225626_Direct_seeded_and_transplanted_guayule_crop_coefficients_and_estimation_using_remotely_sensed_vegetation_indices https://www.researchgate.net/publication/343059008_Using_RGB-based_vegetation_indices_for_monitoring_guayule_biomass_moisture_content_and_rubber https://www.researchgate.net/publication/344754973_GENE_EXPRESSION_OF_GUAYULE_FIELD_PLANTS_UNDER_DROUGHT_STRESS_A_COMPARATIVE_RNA-SEQ_STUDY https://www.researchgate.net/publication/344754809_GUAYULE_GROWTH_AND_YIELD_OVER_TIME_AT_TWO_LOCATIONS_AT_HIGH_AND_LOW_IRRIGATION_TREATMENTS https://www.researchgate.net/publication/337367872_DORMANCY_AND_THE_GUAYULE_Parthenium_argentatum_G_SOIL_MICROBIOME https://www.researchgate.net/publication/337367637_ASSESSMENT_OF_IRRIGATION_REQUIREMENT_FOR_GUAYULE_USING_WINDS_MODEL https://www.researchgate.net/publication/316606450_A_comparative_life_cycle_assessment_of_flood_and_drip_irrigation_for_guayule_rubber_production_using_experimental_field_data https://www.researchgate.net/publication/353247094_Yield_responses_of_furrow-irrigated_guayule_using_deficit_irrigation
Matthew M. Conley
added a research item
Data on the National Agricultural Library’s, USDA Agricultural Data Commons, made available by Dr. Kevin Bronson, with intention to support machine read and compute awareness. The dataset cover image, shows the University of Arizona’s Maricopa Agriculture Center field number 13, in USA, with cotton plants growing during Dr. Bronson’s 2013 summer experimentation season. In the lower-left corner of the image, an early PSCM1 platform is shown suspending our first sensor cluster configuration. In the lower-right, an ArcMap field interpolation of plant height examples processing done at the time. Above in the upper-right, two vantages looking down the crop row signify field geometry and condition as photo-documentations. Center in the image, behind crop rows of the young cotton plants, the USDA, ARS, Arid-Land Agricultural Research Center is visible in the distance. https://data.nal.usda.gov/dataset/bronson-files-dataset-2-field-17-2013
Matthew M. Conley
added a research item
Data on the National Agricultural Library’s, USDA Agricultural Data Commons, recently made available by Dr. Kevin Bronson, with intention to support machine read and compute awareness. Nitrogen and water crop plant management experimental detail is reported, exampling also the methodology of the Proximal Sensing Cart platform and tri-metric sensing approach for high-throughput plant phenotyping in field conditions. https://data.nal.usda.gov/dataset/bronson-files-dataset-1-field-17-2012
Matthew M. Conley
added a research item
Remote-sensing using normalized difference vegetation index (NDVI) has the potential of rapidly detecting the effect of water stress on field crops. However, this detection has typically been accomplished only after the stress effect led to significant changes in crop green biomass, leaf area index, angle and position, and few studies have attempted to estimate the uncertainties of the regression models. These have limited the informed interpretation of NDVI data in agricultural applications. We built a ground-based sensing cart and used it to calibrate the relationships between NDVI and leaf water potential (LWP) for wheat, corn, and cotton growing under field conditions. Both the methods of ordinary least-squares (OLS) and weighted least-squares (WLS) were employed in data analysis, and measurement errors in both LWP and NDVI were considered. We also used statistical resampling to test the effect of measurement errors of LWP on the uncertainties of model coefficients. Our data showed that obtaining a high value of the coefficient of determination did not guarantee a high prediction precision in the obtained regression models. Large prediction uncertainties were estimated for all three crops, and the regressions obtained were not always significant. The best models were obtained for cotton with a prediction uncertainty of 27%. We found that considering measurement errors for both LWP and NDVI led to reduced uncertainties in model coefficients. Also, reducing the sample size of LWP measurement led to significantly increased uncertainties in the coefficients of the linear models describing the LWP-NDVI relationship. Finally, potential strategies for reducing the uncertainty relative to the range of NDVI measurement are discussed.
Matthew M. Conley
added an update
Wolverine PSC carries our USDA Plant Group, Maricopa Phenotyping team, forefront proximal phenotyping methods applications, where acquisition of our phenotyping tri-metric of plant health status is combined with ambient environmental energetics sampling and imagery. Sensors are affixed to the rover platform either in a static location, or on a front rack with sliding adjustment on vertical rails.
We are currently sampling two crop rows at one time, including a redundancy in both passive and active multi-spectral reflectance. And our newest methods include sampling air in a vertical array and low side views.
Acquired together at 5Hz, all data is logged across four electronic devices linked by common GPS UTC.
GPS stamped RGB video and images provide more sampling context and analysis input.
Shown here in Field 119 of the Maricopa Agriculture Center Farm USA, Wolverine PSC prepares for its third run of Soybean season, sampling the experimental plots of Dr. Thompson and Dr. Herritt.
 
Matthew M. Conley
added a research item
Wolverine PSC carries our USDA Plant Group, Maricopa Phenotyping team, forefront proximal phenotyping methods applications, where acquisition of our phenotyping tri-metric of plant health status is combined with ambient environmental energetics sampling and imagery. Sensors are affixed to the rover platform either in a static location, or on a front rack with sliding adjustment on vertical rails. We are currently sampling two crop rows at one time, including a redundancy in both passive and active multi-spectral reflectance. And our newest methods include sampling air in a vertical array and low side views. Acquired together at 5Hz, all data is logged across four electronic devices linked by common GPS UTC. GPS stamped RGB video and images provide more sampling context and analysis input. Shown here in Field 119 of the Maricopa Agriculture Center Farm USA, Wolverine PSC prepares for its third run of Soybean season, sampling the experimental plots of Dr. Thompson and Dr. Herritt.
Matthew M. Conley
added an update
As a complement to the more robust gridded acquisitions of now digital imagery, one may compare the numerically lithe analog type signals derived from single-point environmental sensors. Because often a single field of view sensor can effectively achieve a higher accuracy in sensing points, than can a single camera pixel, due to the larger physical size of the electronic detection elements utilized in analog type industrial or environmental sensor products designed to deliver one detection value.
Additionally, data overhead is greatly reduced when dealing with ASCII text table files of floating-point environmental variables represented in two dimensions, rather than the gridded or cubic radiometric inputs from imagery that lead to multi-dimensional intrinsically covariant and complicated structures.
To find a separate measure of statistical process employed on complex modern agro-environmental imagery compute, compliment high-frequency analog-type proximal scanning allows experimental re-sampling of the image targets, typically within a larger temporal and more focused spatial context; where the analog sensor can be left to run continuously at 1-20Hz, and also sample dial ambient features framing the experimental plot plant measurements.
 
Matthew M. Conley
added an update
“This data was collected using a proximal sensing cart on an upland cotton population grown in Maricopa, Arizona in 2018. The data contains spectral reflectance from Crop Circle ACS-470 sensors, canopy height from Pepperl+Fuchs UC2000 ultrasonic transducers, and canopy temperature from Apogee SI-131 infrared thermometers. Cart position was recorded by an A101 Hemisphere GNSS receiver, and heading was recorded by a VectorNav VN-100 inertial measurement unit. All data can be matched by the timestamp provided by the Campbell Scientific CR-3000 data logger. Data was collected at a rate of 5Hz with analog signals recorded as 0-5V potentials.”
 
Matthew M. Conley
added a research item
Plant height in field crops like cotton (Gossypium hirsutum L.) have long been of great interest to agronomists and plant breeders. Canopy height sensors can replace laborious point hand sampling and enable site‐specific management. Study objectives were to compare manual plant height measurements with ultrasonic distance sensor measurements in subsurface drip‐irrigated (SDI) cotton, and to assess the effects of nitrogen (N) and water management. Two Honeywell 943 sensors were used to weekly estimate canopy height in an N and water management study in Maricopa, AZ, USA from 2016 to 2018. Hand measurements of plant height‐ultrasonic‐sensed height correlation increased with growth stage with R2 of 0.92 to 0.99 at mid bloom. Root mean square errors (RMSE) ranged from 0.9 to 4.0 cm. Canopy height was strongly affected by N and water treatments, starting at first square. This rapid‐response ultrasonic distance sensor is accurate and has great potential as a canopy height measuring instrument in cotton. This article is protected by copyright. All rights reserved
Matthew M. Conley
added an update
A light-weight high-clearance metal frame sits atop four bicycle wheels. The resultant narrow wheel form factor and low total rig weight provide for low plant disturbance when traveling through experimental plots of agricultural row crops. An easy fabrication derived from common products and materials cost hundreds of dollars; this allows maximum resource allocation to support the sensing, rather than the platform. A research-grade PSC can be pushed by hand, or motorized and controlled remotely.
One strength of the PSC modality is static positioning. Where the rig can suspend sensors with fixed proximal views for short to medium periods of time (minutes to days) and record specific plot data for the purposes of environmental model creation or validation.
Link to examples of Maricopa Phenotping PSCs
 
Matthew M. Conley
added a research item
Abstract Field-based high-throughput plant phenotyping (FB-HTPP) has been a primary focus for crop improvement to meet the demands of a growing population in a changing environment. Over the years, breeders, geneticists, physiologists, and agronomists have been able to improve the understanding between complex dynamic traits and plant response to changing environmental conditions using FB-HTPP. However, the volume, velocity, and variety of data captured by FB-HTPP can be problematic, requiring large data stores, databases, and computationally intensive data processing pipelines. To be fully effective, FB-HTTP data workflows including applications for database implementation, data processing, and data interpretation must be developed and optimized. At the US Arid Land Agricultural Center in Maricopa Arizona, USA a data workflow was developed for a terrestrial FB-HTPP platform that utilized a custom Python application and a PostgreSQL database. The workflow developed for the HTPP platform enables users to capture and organize data and verify data quality before statistical analysis. The data from this platform and workflow were used to identify plant lodging and heat tolerance, enhancing genetic gain by improving selection accuracy in an upland cotton breeding program. An advantage of this platform and workflow was the increased amount of data collected throughout the season, while a main limitation was the start-up cost.
Matthew M. Conley
added an update
Waking within the brave world of Maricopa Phenotyping, “best genetics” baby cotton plants emerge from drip irrigation tube embedded, fine loam raised soil beds.
Marking the start of our summer season with staggered plantings, associated new proximal phenotyping hardware is set to deploy in our traditional trial Field #119 as part of Dr. Thompson’s larger body of research, including her pre-breeding selection evaluations.
What do you see in the video slideshow presentation linked below?
Your comments are welcome.
Presentation Cotton Show
 
Matthew M. Conley
added a research item
Waking within the brave world of Maricopa Phenotyping, “best genetics” baby cotton plants emerge from drip irrigation tube embedded, fine loam raised soil beds. Marking the start of our summer season with staggered plantings, associated new proximal phenotyping hardware is set to deploy in our traditional trial Field #119 as part of Dr. Thompson’s larger body of research, including her pre-breeding selection evaluations. Here a visual-centric presentation is offered as a pictorial approach, in loosely documenting general experimental contextual elements about a month after planting, and perhaps allowing additional unique vantage based observations. I am looking to evaluate the merit of technically subjective based basic graphic presentation for purposes of phenotyping team conceptual development and quality control. So basically a quick image set and a report would be derived once a week, which could then be viewed by field-remote collaborators, and serve as a basic visual record. What do you see in the attached video slideshow? Your comments are welcome.
Matthew M. Conley
added an update
The micro-metrological weather sensing station has long been a standardized environmental reference collector in understanding localized energy dynamics.
Typically values are aggregated to a granularity of minutes. And yet we know that environmental fluctuation occurs quickly with significant effect as shown with the Eddy Covariance approach.
However, it is important to be aware that there is also a primary atmospheric spatial observance easily obtained using an up-looking color camera. Exampled in the linked time-lapse video “WST_M1.mp4”, water vapor in the sky is apparent above a test setup mobile weather station as it records an array of sensors at 1Hz.
In the past, I have successfully used imagery to support the analysis of concurrent analog data, and moreover suggest the two are indeed compliments. For instance, by spatially separating radiometers and using a camera view of the sky, individual sensor difference could be explainable, when values are referenced to physical known model and a clear sky radiation signal.
The mobile station example shown in this video was presented previously wherein a GPS enabled video camera is listed as one of its exampled components. Here, the station was set on a gravel island at our location parking lot. At this location I made final mechanical observations and took a single position static comparison dataset. Part of that final dataset was analyzed by students of the ASU engineering undergraduate program as a class project (thank you Henry Shrock).
 
Matthew M. Conley
added a research item
The micro-metrological weather sensing station has long been a standardized environmental reference collector in understanding localized energy dynamics. https://www.researchgate.net/publication/339945786_AzMET_Weather_Station_tribute_presentation Typically values are aggregated to a granularity of minutes. And yet we know that environmental fluctuation occurs quickly with significant effect as shown with the Eddy Covariance approach. However, it is important to be aware that there is also a primary atmospheric spatial observance easily obtained using an up-looking color camera. Exampled in the linked time-lapse video “WST_M1.mp4”, water vapor in the sky is apparent above a test setup mobile weather station as it records an array of sensors at 1Hz. In the past, I have successfully used imagery to support analysis of concurrent analog data, and moreover suggest the two are indeed compliments. For instance, by spatially separating radiometers and using a camera view of the sky, individual sensor difference could be explainable, when values are referenced to physical known model and a clear sky radiation signal. The mobile station example show in this video was presented previously wherein a GPS enabled video camera is listed as one of its exampled components. Here, the station was set on a gravel island at our location parking lot. At this location I made final mechanical observations and took a single position static comparison dataset. Part of that final dataset was analyzed by students of the ASU engineering undergraduate program as a class project (thank you Henry Shrock). https://www.researchgate.net/publication/336568914_ALARC_PlantGroup_Mobile_WST_2019_example
Matthew M. Conley
added a research item
Humans resolve the physical environment largely through ocular construction; and a digital picture grid is a function of 2D table projections, in that an exposed color space value is eventually assigned to each spatially discrete pixel observed. Good and useful representations of target physical characters are subsequently created. Moreover, a simplified re-characterization and annotation of image pixel regions serve to segment or classify instances of similarity or quantification, supporting analysis. That is, understanding what is pictured. In the field phenotyping of plants, four basic image regions of character are present in nadir oriented medium field of view full solar illuminated proximal imagery: 1) Sunlit plant, 2) Shaded plant, 3) Sunlit soil, and 4) Shaded soil, here suggested that they are quadrilateral segments of phenotyping imagery. The attached presentation video examples taking a nadir view cotton phenotyping image taken from an iPad mini3 tablet camera, and where ImageJ software is used to demonstrate a basic image segmentation approach.
Matthew M. Conley
added a research item
A tribute presentation in honor of the AzMet weather station in Maricopa Arizona, USA.
Matthew M. Conley
added an update
Matthew M. Conley
added a research item
High-throughput electronic plant phenotyping in field conditions and within a proximal distance on terra, grows now beneath the more remote and common aerial collections supported by small cameras and advancing software and machine image technologies. In this context, current researchers may employ advantages of proximal measurement, such as in achieving higher resolution or increased physical interface tandem UAS. The attached 2020 Plant Group Phenotyping Team poster entitled, “Proximal Terrestrial Plant Phenotyping.pptx”, is presented as an overview of the 2019 technical program, core data acquisition status, where the Avenger rig, Professor and Wolverine Proximal Sensing Carts, and a mobile energy balance weather station are used along with imagery, to feed a developing proximal terrestrial phenotyping data method that supports multiple scientific objectives and involves multiple investigators.
Matthew M. Conley
added a research item
Cotton flower images for example
Matthew M. Conley
added a research item
In the phenotyping of field plants used as vegetative producer crops, UAS color or extended spectral imaging is effective, and recommended, for rapid canopy sensing and managerial decision support product input; and also to enable course level modeling. However more proximal imaging is recommended to detect the fine, weak, or transient signal biotic characters which are often lost in an amalgamation of information, such as when plants are viewed from a higher vantage. Moreover, consider the opportunity of in-field leaf transmittance imaging, herein presented as a program novel cotton leaf cell resolution level method.
Matthew M. Conley
added a research item
A picture is worth a thousand words, and my intention here is to offer some example viewpoint pictures as food for thought around next phenotyping data captures.
Matthew M. Conley
added a research item
Peltier thermoelectric cooling (TEC) technology was utilized to cool Nikon N1 cameras on Professor Proximal Sensing Cart (PSC), in F119 summer 2019 cotton, successfully extending the operation time of the cameras in the extreme heat encountered during mid-day plant phenotyping collections.
Matthew M. Conley
added an update
Great opportunities in this our new year 2020, regarding high-frequency throughput electronic phenotyping of plants and their associated ecosystem level environmental parameters of growth.
We talented humans benefit from a continued progress in compute model metrics fusion across the dynamic radiometric and volumetric datasets we gather, typically from aerial and other roaming physical extension platforms, and including the expanding imagery acquisitions derived using now robust and accessible RGB discriminated reflectance-based optical camera technology and among other sensing apparatus. This today style micro-detection tooling allows discovery rich high-resolution multi-angular structural composition of unique biological models in-silica. Taken together, emerging platforms of machine learning and anthropogenic omics in understanding our ambient system patterns and functional processes opens a brave new being in a neoteric decade.
Big honor and universal goodwill to researchers supporting shared ethical, rational, and scientific understanding.
Shown in the attached picture “NewYear2020.JPG”, is a winter sunrise viewed from the Maricopa research farm.
 
Matthew M. Conley
added a research item
Threshing method note to support computer simulation modeling and validation.
Matthew M. Conley
added a research item
The typical technical steps employed in operating the Avenger Rig field phenotyping platform in 2017 are listed.
Matthew M. Conley
added 2 research items
List of 27 data processing jobs supporting the Avenger Rig field captures
A list of 69 topics for discussion, in order to fully understanding the scope of Avenger Rig field phenotyping service.
Matthew M. Conley
added an update
Our first Avenger Rig, we pet named it “Captain America”, during the ALARC Plant Group technical Phenotyping Team program development of the terrestrial high-throughput analog “tri-metric” plant health measurement in the context of agricultural research which has operationally spanned 2015-2019.
In 2017, I outlined the typical technical steps in running phenotyping field collections. Please see the document, “USDA ARS 2017 FB2.pdf” attached. I suggest they are overly cumbersome when compared to now day imagery-based UAS operations, albeit I find this technical approach effective in the research setting, even today.
To date the Avenger Rig technical program has evolved substantially under the leadership of Dr. Dyer and Dr. Thompson, to include new compute, imagery and database. We have now developed a working team where Dr. Thompson and Mr. Roybal have mentored students to support the scientific research.
 
Matthew M. Conley
added a research item
A general approach note to the measurement of soil moisture in an agricultural phenotyping field space.
Matthew M. Conley
added a research item
Nitrogen fertilizer use efficiency (NUE) is low in surface-irrigated cotton (Gossypium hirsutum L.), especially when adding N to irrigation water. A NO3 soil-test algorithm was compared with canopy reflectance-based N management with surface- overhead sprinkler-irrigation in central Arizona. The surface irrigation studies also compared fertigation of N fertilizer with knifing-in of N and the addition of a urease and nitrification inhibitor (Agrotain Plus, Koch Agronomic Services, Wichita, KS) to urea ammonium nitrate (UAN). Cotton lint and seed yields responded positively to N fertilizer in all four site-years. Recovery efficiency (RE) of N at low N fertilizer rates (60 to 76 kg N ha–1) ranged from 21 to 61% with surface irrigation and from 81 to 97% with overhead sprinkler irrigation. Deep percolation below 1.8 m was 4 to 11% of applied surface irrigations and rain, but was undetectable in the overhead sprinkler. Leaching of NO3 was apparently the largest N loss pathway in the surface-irrigated system. Fertigating UAN into surface irrigation resulted in similar lint yields and RE as knifing UAN. Use of Agrotain Plus with UAN gave similar yields and RE as using UAN alone. Reflectance-based N management using normalized difference vegetation index-amber (NDVIA) saved 50% of N fertilizer of the full soil-test based dose without a yield reduction in three of four site- years. Nitrogen fertilizer was over-prescribed with the soil-test-based treatment. This may have been due to not accounting for N mineralization, which the reflectance method indirectly measures.
Matthew M. Conley
added a research item
My target is an accurate volumetric soil moisture profile determination, parameterizing the crop available water, within and across a soil suspension medium. Sensing the soil moisture, spanning from a z=0 surface, to vertically below the root zone. The apparent water infiltration rate and the root penetration appearance are additional phenomena that help resolve a plant soil atmosphere continuum understanding.
Matthew M. Conley
added an update
Due to intrinsic variability in environmental systems, multiple and redundant measurements are recommended in order to capture accurate phenotyping growth system energetics information.
A micro-meteorology type portable weather station suspending an array of sensors is useful in characterizing ambient environmental conditions for one or multiple places in time series.
Attached herein, a modified Campbell Scientific type mobile weather station is shown temporarily positioned in the bed of a pickup truck as a test example setup. The data acquisition system is based on my model TRACE WST optimized DAQ, including system performance and power metrics linked to 1 Hz sensed primary data values.
Please see the project update named, “A CR3000 micro-logger used in field phenotyping”, for the previous method note document covering the Campbell Scientific portion.
My intention here is to support field phenotyping plant measurements by delivering increased accuracy in environmental growth condition recording as part of a standardized methodological approach.
There are many nice weather station products and approaches. I am open to discussion around learning and development.
Additional notes follow
My typical technology is the Campbell Scientific selections: https://www.campbellsci.com/grws100
Please note that there are some very interesting new products available like the PheNode https://www.agrelaeco.com/
However, a mobile weather station with wheels allows a spatial dimensionality capture.
A favorite example of mine is Larissa Larsen’s bike adaptation at University of Michigan https://www.citylab.com/solutions/2016/02/a-weather-station-with-wheels/433986/
Arizona State University’s Ariane Middel did a really nice dual planar net radiation setup on her mobile rig https://actreesnews.org/alliance-for-community-trees-news/mobile-weather-station-maps-comfortable-route/
Or examples of DIY Stevenson screens:
Robust industrial examples include:
An alternative model:
This research-oriented information is provided for academic educational purposes and is not an endorsement of any company or product.
 
Matthew M. Conley
added a research item
Due to intrinsic variability in environmental systems, multiple and redundant measurements are recommended in order to capture accurate phenotyping growth system energetics information. A micro-meteorology type portable weather station suspending an array of sensors is useful in characterizing ambient environmental conditions for one or multiple places in time series. Attached herein, a modified Campbell Scientific type mobile weather station is shown temporarily positioned in the bed of a pickup truck as a test example setup. The data acquisition system is based on my model TRACE WST optimized DAQ, including system performance and power metrics linked to 1 Hz sensed primary data values.
Matthew M. Conley
added an update
Avenger platform is our current flagship terrestrial proximal phenotyper, and I have a plan to upgrade acquisitions. Regardless, since 2015 this platform has served very well in generating a large body of standardized phenotyping data, which I have been able to leverage in developing a data method and hope to present.
Federal Labs .org article: “…mission is to develop sustainable agricultural systems, protect natural resources, and support rural communities in arid and semi-arid regions through interdisciplinary research.”
Kansas State University, NSF phenotyping workshops partner website: http://www.fieldphenomics.org/research/vehicles
 
Matthew M. Conley
added an update
Wolverine PSC examples a value terrestrial plant phenotyping platform application, where a welded steel frame atop four bike wheels is RC driven, and propels four adjustable aluminum T-Slot sensor arms. In 2019 cotton, we suspend our model analog plant health data acquisition array and cameras in experimental field plots.
Thank you platform operator Mr. Jacob Goldman.
 
Matthew M. Conley
added an update
“An easy-to-customize, low-cost, low disturbance, motorized, and adjustable proximal sensing cart for field-based high-throughput phenotyping is described. General dimensions, motor specifications, and a remote operation application are given. The cart, named "Professor", supports mounting multiple proximal sensors and cameras for characterizing plant traits grown under field conditions. Professor easily adapts to multiple sensor configurations supporting detection of multiple target traits and has two axes of adjustable clearance by design. Professor is useful as a field-based phenotyping platform and offers a framework for customized development and application.” – Thompson et. al.
 
Matthew M. Conley
added an update
We successfully adapted thermoelectric cooling to maintain an 80F ambient environment around a field spectrometer. The spectral radiometer instrument was secured inside a customized polystyrene beaded shipping box from the lab. We mounted the box low in order to reach small plants. Using our new push cart and temporary wooden rail set and slide table, rolling proximal access to in-field plots was achieved for the instrument, computer, and batteries. Our second field collection is pictured which lasted about two hours.
 
Matthew M. Conley
added a research item
Thermoelectric cooling is used in field phenotyping via application of the inefficient Peltier effect, where system exhaust heat moves to the environment and a deployed radiometric sensor device target is thermostatically controlled.
Matthew M. Conley
added a research item
The Nikon N1 aw1 GPS action camera is set up for field phenotyping by locking exposure settings and view angle. In this video, Image exposure settings recommendations are offered for F-Stop 11 and incorporating a new physical solar interceptor intended to decrease image contrast.
Matthew M. Conley
added an update
Triangle, as in trigonometry, or a triangulated irregular network, so useful and fun.
Right Triangle, described by Pythagoras, looks to be a great design model for our next step PSC platform. We included a top cage to connect two right triangle side frames with wheels as a mockup and it was a success that built upon an innovative previous Isosceles triangle design.
Therefore the right triangle is seen as a design mode to extend a Proximal Sensing Cart crop plant vertical clearance height accommodation beyond 2 meters, while retaining a 2 meter rig width.
 
Matthew M. Conley
added an update
There is a promising 2019 summer field season unfolding in Maricopa Phenotyping. One example, in MAC Field #13 young cotton varieties are growing well, elongating after a grass cover rotation and direct seeding into a no-till soil.
Over several years Dr. Thorp and Mr. Hagler have developed effective precise sprinkler control of the water and nitrogen delivered to Field #13 plants. Pictured are cotton plants located in experimental plots inside the Field #13, and under a robust nitrogen investigation conducted by Research Leader Dr. Bronson.
Notably Dr. Thorp models leadership by integrating several additional collaborative experiments, including powerful new investigation by Dr. Thompson and Dr. Pauli, into his already successful agronomic management, and imagery / reflection based evapotranspiration modeling research.
Big data high throughput phenotyping takes a step forward this season, as our largest yet constellation of UAS borne cameras scan MAC fields, supporting orthomosaic map rectification and SfM point cloud processing pipelines.
 
Matthew M. Conley
added a research item
Plant height is a morphological characteristic of plant growth that is a useful indicator of plant stress resulting from water and nutrient deficit. While height is a relatively simple trait, it can be difficult to measure accurately, especially in crops with complex canopy architectures like cotton. This paper describes the deployment of four nadir view ultrasonic transducers (UTs), two light detection and ranging (LiDAR) systems, and an unmanned aerial system (UAS) with a digital color camera to characterize plant height in an upland cotton breeding trial. The comparison of the UTs with manual measurements demonstrated that the Honeywell and Pepperl+Fuchs sensors provided more precise estimates of plant height than the MaxSonar and db3 Pulsar sensors. Performance of the multi-angle view LiDAR and UAS technologies demonstrated that the UAS derived 3-D point clouds had stronger correlations (0.980) with the UTs than the proximal LiDAR sensors. As manual measurements require increased time and labor in large breeding trials and are prone to human error reducing repeatability, UT and UAS technologies are an efficient and effective means of characterizing cotton plant height.
Matthew M. Conley
added an update
Today, the Proximal Sensing Cart platform (PSC) is an attractive terrestrial field phenotyping apparatus supporting low cost, low disturbance, Field-Based High Throughput electronic Plant Phenotyping (FB-HTPP) and micro-meteorological measurement. Although there are several important considerations in design and operation of a successful phenotyping PSC platform, perhaps the most important element is the wheel. Where the rubber meets the road as it were.
Humans walked upright, then used a wheel. The Mesopotamian potter’s wheel around 3500 BC, and spoked wheels about 2000 BC. Wire tension suspension wheels were first proposed by George Cayley in 1808, with modern tangential spoking allowing transfer of torque between a typical bicycle wheel’s extruded abutted circular rim and alloy hub.
In Maricopa Phenotyping, we found that the bicycle wheel is well suited to successful, on the ground FB-HTPP, such as when used in quartet on a cart, where it can support a 3 meter wide multi-row sensor array and up to 400 pound payload, all on narrow wheels which slide through plant rows of crops. The bike wheel provided us traction in electric drive, steering when swiveled or skid, did not bog down in light mud, and could dig through soft soil berms when powered; a most effective form factor indeed, and so highly developed.
Yet all wheels are not equal. Now days a bike wheel is an assembly of semi-regular components. Maximum value in this phenotyping program element was desired, so we turned to local bike expert, Mr. Vince San Nicolas from Global Bikes of Ahwatukee, and he specified and assembled a custom wheel configuration with field phenotyping in mind. He took the time to review product catalogs and make his best estimations of value in an iterative and communicative process. This public / private spontaneous phenotyping collaboration may be an early first instance of a commercial bicycle technical expert supporting American and global climate resiliency in this way.
PSC wheels were selected for strength, while still retaining the standard narrow width factor of a common off-road bicycle tire (this is the minimum effective track width in our loam soil). We use a typical disk brake type hub that accepts bolts, so we can affix a drive sprocket. Vince installed an old school axle with long threaded studs so we can easily mate to our fork assemblies using nuts. He also selected specific tubes and tires, and moreover provided what I was looking for to develop the PSC program element, which was an examination of application in product component selection for phenotyping, however generated from a technical commercial viewpoint. Not only can Mr. San Nicolas offer us ongoing technical and mechanical service and component sourcing from his shop at Global Bikes as is typical, such as to tune spoke tension on a wheel, but he has vast product application and maintenance experience running service for the local bike community daily. There is a vibrant industry around new highly designed bicycle products.
Please find the following list of bicycle wheel parts for phenotyping Proximal Sensing Carts as identified for 2019 field testing and program deployment.
USDA does not endorse companies or products, trade names are given for education and research support
Wheel - WHL FT 26x1.5” & WEI XM280 disk BK 36 WM MT2000 6B BK 100mm 14gBK (640168 JBI)
Axle - #14 solid axle 9x1x155mm (HU8257 QBP)
Tube - Thorn Resistant S/V TR T/R 26x1.75-2.4
Tire – 26x1.95 BK/BK Alphabite K83 C-1040N Sunlite
Thank you to Dr. Dyer, Dr. Thompson, Ms. Parks and special thanks to Mr. Vince San Nicolas.
 
Matthew M. Conley
added an update
Seeing is believing, as in an optical perception. I see what you mean. Because visual understanding drives biologic external material awareness, with human devotion to visual cortex as almost a third.
I am pondering visual structures, as Avenger Rig resumes with the 2019 collection of proximal tri-metric phenotyping data supporting the third year of a Camelina field trial led by Dr. Abdel-Haleem.
 
Matthew M. Conley
added an update
How do we delineate an appropriate theoretical standardization for imagery based environmental awareness?
Please contribute to an RG discussion
Some steps in optical perception acquisition:
1. Camera detector type
2. Radiometric bands detected, their number and band-widths
3. Lensing structure and optical ray clarity
4. Camera distance and sampling resolution relative to a signal of interest
5. Photographic exposure settings, such as aperture, shutter speed, and ISO
6. Image capture rate, and file format as RAW or loss of information
7. Data calibrations, is there a control panel in image
8. Illumination, contrast, shade, diffuse solar is best
9. Camera position and view angle, GPS location, IMU accelerometry
10. EXIF meta information attached to each image file
11. Camera data throughput and storage
12. Image transfer for backup and compute
13. Geometric corrections
14. Radiometric corrections
15. NIR or hyperspectral options
16. Principal components analysis
17. HSB color space analysis
18. RGB color space analysis
19. Segmentations and clustering models
20. Change detections
 
Matthew M. Conley
added an update
Field phenotyping methods of environmental science have become largely dependent on electrical and compute technology for common high throughput electronic now day sampling. As an extension of human vision, the optical camera device has shown arguably the most promise in characterizing biological phenomena via a signature of gridded radiance. Concurrently, materials and robotics engineering development has allowed camera device travel from vantages of geo-orbit to more proximal suspension. However interpretation of the quality data derived therein remains problematic.
How do we best interpret a biological picture?
Shown is an example gif image with bulk green pixel selection, taken from a GoPro6 camera about 3 meters above and facing the ground, mounted on a diagonally oriented wood stick, rigged off the Wolverine PSC during Dr. Thompson’s F119 cotton third 2018 data collection.
 
Matthew M. Conley
added an update
I’m holding gratitude, for participating in the process of agricultural phenotyping, as our world moves deeper into ecological change; watching an accelerating technology based data generation, and seeking to understand more results; wanting to provide answers and solutions, knowing this is a riddle larger than myself; looking upward to researchers and people of vision, compassion, humility and power.
The 2018 year of Maricopa Phenotyping brought large program changes, with the rise of major PSC operational platforms connected to a program wide database and quality control “level zero” data processing pipeline, additional field data collection and methodology development progress, a new big weather station deployment, and much more.
Thank you phenotype researchers in Maricopa, across the country and around the globe.
 
Matthew M. Conley
added an update
Maricopa Phenotyping saw great success in the 2018 summer season, with the roll-out of Wolverine Proximal Sensing Cart as a primary phenotyping data collection platform informing cotton pre-breeding operations. Wolverine PSC supplanted Avenger rig as part of a test case in understanding the feasibility and quality of phenotyping operations using remote controlled, lightweight, narrow wheeled, metal framed electric carts to derive tri-metric type field data collections. Initially there was some discussion skeptical of non-Avenger terrestrial phenotyping, however Dr. Thompson has shown in her 2018 collections that our forward looking platform vision continues to appear sound and attractive.
Demonstrable success has encouraged other researchers to implement similar technologies within our research group. Moreover, we support external phenotyping research via PSC platforms and field methodologies through technology transfer, project collaboration, or consultation of ideas with interested parties.
I feel Wolverine PSC and this work examples a now alternate high-throughput field-based phenotyping mythology; which is a relatively slow travel, high resolution, proximal terrestrial environmental sensing, best suited for static and precision investigation. The major platform ambient plant growth condition tolerance strength is high wind, and its major weakness is high soil moisture. The terrestrial application strength is precision location and temporal duration, while weakness is major coverage limitation.
The slow ground-rigs are not expected to traverse the large acreage needed for breeding trial field plots. Aerial imagery seems better suited for such large scale population mapping in a landscape. However the expectation is that aerial systems would direct the higher resolution proximal samplings and that mobile platforms on the ground will carry supporting equipment as needed to project physical operations.
 
Matthew M. Conley
added an update
In a round two of testing imagery collections, Professor PSC was outfit with Nikon GPS cameras to collect multiple view angles. Cameras were triggered by a Campbell Scientific logger in order to achieve precision control and temporal integration with phenotyping tri-metric data. Camera overheating was addressed with solar shading and mechanical cooling. Camera exposure settings and field of view were considered with PhotoScan input in mind.
Pictured is ASU student Henry Shrock operating Professor in Dr. Thompson’s cotton experimental plots. In the foreground are Vernonia bushes in flower, an ice-cream attractor plant being used to assess factors of entomological agronomics.
 
Matthew M. Conley
added an update
How do I know my sensor is accurate, or which of the same two sensors I would trust more?
In Field-based plant phenotyping, an apparent sensor bias may be effectively measured using a powered control chamber where the ambient condition is manipulated across a range of expected field environmental conditions and while sensor measurements are recorded. For example, four new air temperature and relative humidity sensors were fixed loosely clustered together within an airspace of about nine inches inside our large Maricopa Phenotyping Plant Group’s calibration room, and the room temperature was adjusted. Results show a slight apparent bias in temperature that persisted even when the sensor order was shuffled twice, and the sensor cluster mounting location changed once. This simple type of sensor comparison allows individual sensor operational or positional selection or rejection based on demonstrated apparent performance; and a characterization of apparent sensor behavioral personality traits, allows a non-standard compensation action option, and an increased description potential. Highest performing sensors can be selected for primary measurement operations, while the lowest individual performers may require manufacturer refurbishment or operational removal least they may be relegated to lowest value measurements. Furthermore, sensors may respond differently regarding thermal influence, geometric relation, or communication error rate. A consideration or compensation may be provided to normalize a dependable and acceptable sensor induced artifact, once that specific sensor behavior can be adequately characterized and documented.
In Maricopa Phenotyping, we typically resolve sensing product acting within tolerance, however we have detected grounding noise, and a calibration offset, in sensing hardware before it was otherwise known. In other projects, I’ve seen a bad ARCTAN2 calculation in an OS and we have developed vendor parallel sensor calibration protocol in support of product quality. Commonly, sensors are verified or “spot-checked” in a way that is specific to the experimentalist and situation so as to gain confidence that a critical unknown error is not in play. Often so much pre-experiment data is generated that further analysis may take place if subsequent evidence warrants, so if I have a suspicion let’s say, then a historic numeric reference can be queried to support understanding.
I suggest the phenotyping data acquisition system sensing package be constructed with sensors connected, and run in a calibration control space, for an “as-built” recording, ideally across a range of temperature or other significant factors of interest. The range of the factor of interest should span beyond that factor’s range in the expected experimental condition. Next the data package is taken to the field for the actual experimentation recording. However, come the end of the experiment or a major recording cycle, the data package could be brought back to the same calibration room and recordings would verify sensor stability. If a protected standard sensor is used that remains in the calibration room, then even more confidence is possible in continuity of apparent sensor accuracy relative to the experimentation.
For another method, intermittently line up all your sensors on an outside, small artifact, ridged, and regular rack, with adjustable arms, so each sensor can be leveled and placed at the same height. Collect sensor array data for several days outdoors, hopefully with large swings in environmental condition. A sensor comparison individual sensor unit to unit and then across time can detect anomalous behavior, and otherwise value add confidence. Past method included yearly sensor comparisons with inclusion of standard sensors which were only used during the comparison operations.
Lastly, in order to get a quick sensor check, I have just placed a sensor package close to a weather station or other acquisition system and recorded data as a secondary field comparison approach.
 
Matthew M. Conley
added an update
The acronym UTC stands for Coordinated Universal Time, which is a universal time coordinate value originating locally on the Prime Meridian in Greenwich England. UTC is based on the momentary terrestrial derived International Atomic Time (TAI), and uses leap seconds to hold temporal accuracy at a unit of one second relative to the true Universal Time (UT), whereby Earth rotation is measured relative to quasars the moon and other targets, to then compute the relative mean solar time at zero degree longitude. UTC is broadcast in satellite L-band carrier waves and can be recorded, for example, as GNSS NMEA RMC or GGA type serial data strings as derived on a GPS receiver. Maricopa Arizona is seven hours behind UTC time, so local time is achieved by subtracting that difference.
It is an important consideration to always include UTC data recording into weather station, mobile platform and imagery based phenotyping data captures, as a true time value. The reception of UTC during acquisition allows subsequent disparate dataset comparison arrayed at one second of time in a series. Often there can be multiple data acquisition platforms collecting at the same location, or rather recording on roaming mobile rigs, or sensor networks, which allow simultaneous data streams at the same time but from different locations.
The typical computer generated central processing system oscillation derived onboard time value can drift by a second or more over days or months due to minor imprecision of the piezoelectric crystal used. However the fast CPU cycling and other dedicated precision pulse signals and onboard processing can extend temporal designation in data collections below the one second base UTC signal resolution. Data that over-runs the base temporal UTC timestamp with burst or high-throughput measurement can be distributed between true time pulse anchors based on the series granulation and a linear time interpolation. Regarding field-based terrestrial phenotyping mobile platforms moving 1-5 mph, I suggest that 1 Hz UTC anchors allow 100 Hz data capture spatial analysis of the biological signal of interest.
I would like collaboration to analyze some 100 and 20 Hz test data I have collected for trait significance and then re-sample in various ways at 1 and 5Hz to test statistical approaches.
 
Matthew M. Conley
added an update
A phenotyping sampling, such as to measure a field-based biological process once each second or faster continuously, including the ambient environment, and distributed in space as well as time, as with sensor arrays, networks, and digital imaging grids. Sampling opportunities begin before, and continue until after a plant life cycle, and they increase proportionally with sampling resolution. Herein, technical notes are presented as derived in Maricopa Phenotyping method development for a value approach to non-imagery data acquisition.
The Campbell Scientific CR3000 (or CR1000) micro data logger platform example:
· 0-5 volt electrical potential measurements are primary, with up to one third microvolt max resolution, current signals are resistor converted, pulse and serial communications are included
· The micro-voltage potential measurement ability enables thermocouples and other weak signal sensors such as the SI-131 IRT or SKYE radiometers without additional signal amplification
· RS-232 and SDI-12 protocol smart sensor serial data is a significant element of the typical phenotyping recording, however note that serial data streams are often slow, may have communication error, and that serial digital data can be recorded on inexpensive electronics
· Keep the program execution processing and measurement times below the loop scan interval to avoid data over-run skips, a fully loaded logger will easily run at 1Hz, while 5Hz speed on big measurement arrays may start to require operations optimization, consider 20Hz measurements as a good frequency to achieve a now days rich resolution time series phenotyping data stream
· Use the three lead wire differential measurement approach if possible (high, low, and common)
· If time allows, use the reverse difference scan option (for a second measurement comparison)
· Typically use the “_60Hz” long 16.67 ms integration time, but for speed use a 750 or 500 us settling time (250 minimum), note that the long time is for rejection of North American electrical noise, and that the short time may even benefit weak signal amplitude
· If the lesser option, two wire, single-ended measurement of signal and common is used, allow the onboard calibration background potential common reference with each measurement
· If target signal is stable within a differential measurement range relative to two program loops, the AutoRange option allows a possibility of better signal resolution by providing an intermittent voltage measurement range re-framing, C option should allow open channel detection
· System status and performance information is available from the status table for recording, I look for any skipped scans and then chart loop processing times to visualize system speed during events such as when binary tables are being allocated, intermittent system actions occur, or streaming serial information fluctuates, tracking all loop counts with variables can be helpful
· Program signatures, measurement offsets, calibration gains, and coefficients are poll-able from the system status table and can be recorded as additional sampling meta information
· If a PS200 power regulator is used, serial communication allows easy power data recording, the collective system load and charging currents are important in the system power management
· Connect the power reference leads and COM channel ground references to the protected power terminals “G” ports, and connect sensor shields and references to the ground symbolled reference ports (three bars in an upside down pyramid) with reduced electronic filtering artifact
· Don’t select the high dialectic constant PVC material in insulating sensor leads
Trade names are presented as examples for educational purposes and not endorsement
 
Matthew M. Conley
added an update
The analog phenotyping tri-metric of plant height, color, and temperature, is not able to distinguish between a crop plant and a weed plant, granted it is an arbitrary distinction, so I am referring to the speciation. Shown, is a stunted and stressed soybean crop plant occluded by a larger and more vigorous Lamb’s quarter weed plant. The phenotyping electronic signature from this situation would be similar to that of a healthy crop absent the weed, in that the target tri-metric signature would be variably taller, greener, and cooler, than its environmental background.
I have encountered experimental crop fields with almost no weeds, and other experiments that were almost completely covered in weed canopy. Investigators who prepare their soil beds and practice generally good crop production management practice have reduced weed issues. To achieve best practice, investigators would manage weeds by removing seeds, sprouts and targeting established individuals as soon as possible, and as part of an on ongoing campaign, a seek and destroy effort, shared by all members of the experimental team. To that end, physically stretch your body and work up an ability to bend at the waist with your back and knees straight when stooping. I use a hoe with cut-down blade and long handle. To support process integrity, collectively review and agree to the rules of engagement with anybody who is going to cull a plant in your field.
When a weed is extracted, place the vegetation material outside the primary sampling transect. Likewise, if a soil dimple is created from the extraction of a weed root ball or other mechanical soil action, I suggest restoration of the soil surface integrity through smoothing and translocation of nearby soil. Keeping soil height and effect patterning consistent across a measurement season is crucial to better phenotyping spatial analysis of the target vegetation.
Do not let weeds develop in your experiment, or worst yet set seed. Weekly cycles of daily weeding are needed throughout an active growth season. Although contract manual labor weeding is attractive from a management perspective, be cautious of any non-technical contribution to experimentation. Higher cost investigators or trained technicians would also characterize plot and environmental conditions to value add the weeding operation.
Significant weeds can be removed manually, at measurement time, by simply walking in front of the measurement rig and clearing out the conspicuous weeds. I figure it is always better to manage a weed, rather than measure it with the crop, and that any biological void will collapse quickly.
Finally, I would like to be able to discern weeds and crops using basic phenotyping parameters such as temperature, light and displacement. Since weeds often exhibit vigorous and unique growth, I would like to detect those traits more clearly.
 
Matthew M. Conley
added an update
Showing development in our field phenotyping method after three years of Avenger operations, Wolverine electric drive Proximal Sensing Vehicle 2018 platform is the first Avenger analogue product.
Wolverine exhibits an improved thermal crop signature detection method within a micro-climatic envirotype capture that is germane to excellent phonemics applications. And our ACS-470 unique validation methodology, using thermal recording and control with standardized reflectance panels, is showing a superior quality control and results.
Three improved thermal based high value method applications on Wolverine PSV 2018:
1. Crop Circle sensor double insulation and type-t thermocouple (TC) based individual sensor body temperature tracking, with a standardized panel warm-up radiance detector characterization, and a verification pre / post field collection detector status bookending
2. Four nadir view low mount high accuracy narrow window insulated and factory shielded infrared radiometers, with secondary TC based solar shaded ambient air temperature recordings at each sensor location, the four TCs represent our first passive horizontal distributed mobile air array deployment
3. Active aspirated double insulated air temperature and Rh sensor, with a secondary integrated insulated and aspirated air temperature TC, plus the Garmin Bluetooth ANT temperature sensor
I honor Dr. Thompson’s leadership in producing Wolverine PSV which models our Plant Group field-based phenotyping team success. I think Wolverine PSV examples future terrestrial phenotyping apparatus; and I look forward to initial results and how this new program momentum will translate to additional discovery.
 
Matthew M. Conley
added a research item
Aspirated micro-environmental sensible heat sampling on a proximal phenotyping platform
Matthew M. Conley
added an update
USDA in Maricopa, Plant Group team, Avenger phenotyping program overview
LeeAgra AvengerPro modified rig has run as a terrestrial phenotyping platform, in a 2.0 type version enhanced mode, from 2015 to 2019, whereby the basic phenotyping analog 5Hz tri-metric, plant displacement, surface temperature, and multi-bandpass spectral sensing produced by initial program researchers in 2011 to 2014 (engineering by Mr. Strand), was upgraded relative to improved DAC, and power supply, and augmented with FLIR thermographic recording, LiDAR technology, navigational 9-axis recording, ambient micro-meteorological environmental sensing, and including a hodgepodge of consumer grade video and still imaging. To generate a total of about 200GB of data per hour at a 1.5mph typical travel speed, we included up to a dozen cameras; the single FLIR thermal imager resolving between 1-10Hz thermography has been the biggest data generator. Hyperspectral recordings, wind, and increased quality image capture is wanted. Where a third phase program upgrade is planned with new hardware and best practices to be implemented for an increased production target by spring of 2019. Of special interest to me is coding in LabVIEW for increased sampling, and integrating primary time coordinate structures and other data controls to output specific standard format and support modularized pre-processing function.
Base fusion elements in terrestrial phenotyping analog signal “tri-metric” sensing:
1. Plant height, as a function of biological form volumetric displacement
2. Plant temperature, as a function of target area surface thermal signature
3. Plant reflectance, as a function of apparent visible and NIR surface reflectivity
Plant height – displacement sensors (120-180KHz ultrasonic sonar, with NIR laser ToF)
· UC2000, ToughSonic14, and Honeywell ultrasonic displacement (with LidarLiteV3)
Plant temperature – non-contact thermometers
· Apogee SI-131 infrared radiometers
Plant reflectance – spectral reflectance sensors
· CropCircle ACS-470, [passive options from Skye or Meter Group]
Major data generating and support components around the tri-metric
  • Trimble R6 RTK GPS – precision geopositioning for data geospatial reference
  • VectorNav VN-100 AHRS – precision navigational MEMS device
  • SICK LMS511 LiDAR – ruggedized laser point cloud generator
  • FLIR A600 series thermal imagery – 600x480 grids
  • Campbell Scientific environmental logging – voltage and communications recording
  • Dynasys APU and power rectification – clean power generation and supply
Including good value phenotyping cameras: GoPro Hero6, Garmin VIRB XE, Transcend DrivePro220
The USDA does not endorse products or companies, all information is given for educational research support.
 
Matthew M. Conley
added an update
I look forward, to future coordinated platform actions, and how micro robotics will empower environmental measurement.
Sensors moving under and then inside canopies offer new territory to explore micro-condition and responses.
Present unmanned aerial system platform with camera has become extremely popular in field phenotyping. Innovative companies are offering phenotyping products and services both in operations and data. Leading products already interoperate across the agronomic system, allowing farmers or researchers UAS based imagery and GIS online compute knowledge products within hours of request (http://www.slantrange.com/).
Leading agricultural University programs now offer UAS pilot certification in tandem with agronomic study (https://polytechnic.purdue.edu/degrees/unmanned-aerial-systems).
Looking forward →
It is clear that robotic automation and visual perception based machine sensing will drive technology functions in the coming years. Both aerial and terrestrial platforms of dynamic and varied form will be common in this new paradigm. Moreover as new sensing technology emerge, new biological insight become available as compute tools and statistical theoretic evolve in parallel.
Therefore it is key to incorporate and coordinate new ideas, product and learning, while retaining process continuity, experimental trajectory, and research integrity, in order to ensure consistent discovery potential, and best practice field phenotyping implementation.
 
Matthew M. Conley
added a research item
Terrestrial high-throughput plant phenotyping in the field and subsequent first order data pre-processing for phenonomics
Matthew M. Conley
added an update
The USDA will advertise a Phenotyping Engineering and Research position 7/18 to 7/31/2018 (file attached). This is a full time PhD level hard-money permanent phenotyping research science position that will include a technical support employee so it is a really great opportunity.
My hope is that I could work supporting a new phenotyping program engineer and move our Maricopa Phenotyping operations to higher levels of success for which I know we are capable as a location, and with the institutional knowledge and resources we currently have at hand.
Plus Dr. Dyer is a Research Leader who empowers employees to greatness, allowing personal merit and inspiration to flower in an honest and just way, without managerial over-constraint.
Please take a moment to evaluate if you or a colleague who is a U.S. Citizen, would be interested in leading with excellence, the important field-based high throughput phenotyping of plants evaluating heat and drought stress for endpoints of crop management, breeding selection, and compute modeling.
 
Matthew M. Conley
added an update
Field based phenotyping of plants, using rapid non-contact electronic proximal sensing, has increasingly supported agronomic management, plant breeding, and biological modeling. Therein, common visible light camera imaging technology has assumed a place of primacy, in providing high value phenotypic characterization of plants in the field, by detecting color, shape, and an inherent spatial information across the detector grid.
Nowadays, consumer grade cameras can provide a phenomics useful high-throughput continuous and consistent image capture, including GPS information, however, it is important to select a camera product that fits one’s field phenotyping paradigm.
For a researcher with limited monetary resource, the inexpensive GoPro Hero6 action camera product is offering the following field phenotyping desired major performance elements:
· Environmentally sealed and miniature form factor
· High heat tolerance
· Long battery life
· 2Hz continuous image capture
· GPS UTC and other information populated to image EXIF
· Locked photographic exposure and white balance settings
· “Linear” low optical distortion field of view mode
· [RAW imagery capture possible]
Although the current Hero6 platform does not offer the IMU metrics and software editing functionality that Garmin VIRB product lines have shown, Hero6 succeeds in capturing proximal phenotyping value optimized field still imagery.
I have used the GoPro Hero6 to collect proximal nadir view field imagery of cotton, as mounted off a proximal sensing cart and resulting in about 14 thousand images over 2 hours of operation.
This collection occurred in mostly clear and still atmospheric condition, without the unit shaded from the sun, and in an ambient temperature above 100 F. The attached image is an example from that collection.
I am very happy with these initial performance results where basic field phenotyping image capture targets have been met.
The trade names and products mentioned are intended only for educational purpose, they are not endorsed by the USDA.
 
Matthew M. Conley
added an update
Arizona cotton, King in 1920, one of Arizona’s “Five C’s” driving historic economic activity, those being copper, cattle, cotton, citrus and climate. Stemming from a Native and Egyptian genetic cross, Pima long-staple, high quality fiber cotton was bred into extra-long staple Supima, which continues to represent highest quality fiber product, as grown with minimal disturbance in an irrigated low desert.
In Maricopa Phenotyping, we utilize those cotton and climate factors to affect phenotype drought and heat stress on the plants in experimental plot design at the University of Arizona, Maricopa Agricultural Center farm.
Furthermore, Cotton has a complex canopy and fruiting architecture with substantial exterior organ optical occlusion of inner canopy components. We are working on angled vantages and creative approaches to resolve flowers and bowls. Yet how to characterize fiber quality in the field with a high throughput and negligible damage?
Pictured is new season cotton genetic expression, rising from a sprinkled fine loam soil. Growing phenotype defacto.
 
Matthew M. Conley
added an update
The veteran Proximal Sensing Cart Mark 1 phenotyper emerges into a new life of service as Deadpool.
Whilst mature spatial based RTK-GPS precision management field applications were evolving into first high-throughput field phenotyping methods on MAC farm by researchers such as Dr. Andrade-Sanchez and Dr. Gore, Dr. White lead creation of PSCM1, the Proximal Sensing Cart Mark 1 platform. Starting with the bicycle and square frame concept, we generated a square tube steel open box atop two bicycles. Adding an adjustable dual slider sensor arm, push handle, and our data acquisition equipment, allowed us many successful, rich and novel data collections supporting multiple projects, including workshop demonstrations.
The 2015 textbook, “Phenomics: How Next-Generation Phenotyping is Revolutionizing Plant Breeding”, features an image of yours truly with PSCM1 as an example primary platform, and I believe the initiation of the epic TERRA-REF is connected to PSCM1 early impact. The cart was published as, “A Flexible, Low-Cost Cart for Proximal Sensing” in Crop Science 2013. If the search term “proximal sensing cart” is supplied to Google image search engine, PSCM1 is often returned top of the list. Note there is a similar looking unnamed PSC on the Sequoia farm in Wilcox Arizona, which I was given opportunity to make for Dr. Lynch and his team operations.
[Please comment or message me if you feel the 2011-2016 PSCM1 work was useful to your research, so I may be more inclusive and accurate in my descriptive narrative.]
PSCM1 was cut in half 2018, the bike frames were eliminated, and the remaining top frame was mated to Mr. Conrad’s improved custom designed bicycle wheel assemblage with swivel rear wheels. New dual sliding sensor arms were installed which included our improved clamping tab design. Dr. Bronson has inspired a new operational life, through a nitrogen management collaboration with Dr. Lewis of Texas A&M University, for the cart emerges as Deadpool, moving into new research potential, and with new documentation presented by Dr. Thompson.
Having several years field phenotyping by hand, with push carts, and with tractor style light implement rigs, I believe lightweight, steel framed carts on bike wheels like Deadpool will continue to be effective in proximal environmental sensing and in supporting UAS.
Documentation on Deadpool is hosted by the USDA National Agricultural Library:
“A how-to-build guide for Deadpool, a proximal sensing cart”
 
Matthew M. Conley
added an update
The Professor, a proximal sensing cart developed for field phenotyping.
Professor is square in shape and framed in metal T-slot, with linear clamping adjustments and electric drive. Professor was born when Maricopa High School engineering path students were tasked with an adjustable 2m dual axis proximal sensing platform displacement target, and were provided hardware by USDA and guidance by Mr. Miksch. Professor was given back to USDA as a built product, collapsed and rolled into a pickup truck for transport to the MAC farm.
Once at the Southwest Phenomics Center, Mr. Conrad made adjustments for field deployment, expanded rig capability, and developed electric drive and controls. We added three remote triggered mirrorless DSLR cameras with onboard GPS receivers, and our typical phenotyping tri-metric including micro-climate sensing, as Dr. Thompson leads ideas of imagery capture and illumination control, and production of the HardwareX manuscript and Ag Library media.
 
Matthew M. Conley
added a research item
An easy-to-customize, low-cost, low disturbance, motorized, and adjustable proximal sensing cart for field-based high-throughput phenotyping is described. General dimensions, motor specifications, and a remote operation application are given. The cart, named Professor, supports mounting multiple proximal sensors and cameras for characterizing plant traits grown under field conditions. Professor easily adapts to multiple sensor configurations supporting detection of multiple target traits and has two axes of adjustable clearance by design. Professor is useful as a field-based phenotyping platform and offers a framework for customized development and application.
Matthew M. Conley
added a research item
This is supplemental material for, "A Flexible, Low-Cost Cart for Proximal Sensing" 2013 paper which describes the first iteration of our Proximal Sensing Cart Mark 1 build.
Matthew M. Conley
added a research item
Field-based high-throughput phenotyping is an emerging approach to quantify difficult, time-sensitive plant traits in relevant growing conditions. Proximal sensing carts represent an alternative platform to more costly high-clearance tractors for phenotyping dynamic traits in the field. A proximal sensing cart and specifically a deployment protocol, were developed to phenotype traits related to drought tolerance in the field. The cart-sensor package included an infrared thermometer, ultrasonic transducer, multi-spectral reflectance sensor, weather station, and RGB cameras. The cart deployment protocol was evaluated on 35 upland cotton (Gossypium hirsutum L.) entries grown in 2017 at Maricopa, AZ, United States. Experimental plots were grown under well-watered and water-limited conditions using a (0,1) alpha lattice design and evaluated in June and July. Total collection time of the 0.87 hectare field averaged 2 h and 27 min and produced 50.7 MB and 45.7 GB of data from the sensors and RGB cameras, respectively. Canopy temperature, crop water stress index (CWSI), canopy height, normalized difference vegetative index (NDVI), and leaf area index (LAI) differed among entries and showed an interaction with the water regime (p < 0.05). Broad-sense heritability (H2) estimates ranged from 0.097 to 0.574 across all phenotypes and collections. Canopy cover estimated from RGB images increased with counts of established plants (r = 0.747, p = 0.033). Based on the cart-derived phenotypes, three entries were found to have improved drought-adaptive traits compared to a local adapted cultivar. These results indicate that the deployment protocol developed for the cart and sensor package can measure multiple traits rapidly and accurately to characterize complex plant traits under drought conditions.
Matthew M. Conley
added an update
Often an investigator carries robust research idea and practical discernment, yet modest resource base for which to effectuate vision. In high throughput phenotyping of field plants, we meet this challenge by developing the Proximal Sensing Cart technology.
Storm, that’s our pet name for a small proximal sensing push cart used for field phenotyping research. The cart is made from 1.25” 16 gauge A36 welded tube steel and emerges as a modification of Dr. White’s cart specification for a workshop field data collection exercise rig. Storm occurred when Mr. Swartz raised the rig clearance factor during fabrication, from the previous Little Red Cart built as specified with 37” clearance, to Storm’s increased 51” vertical platform displacement. Storm is all square tube except for a round handle, bicycle wheels, and bicycle forks. We have retained the specific bicycle product technology as good application fit, being common, strong, narrow, and sleek in form. Due to an unrelated shop welding lock out, a reasonably sized, wide U shape 1” square tube, was pulled from our Location’s recycle bin, drilled, painted, and re-purposed as a sensor arm. This sensor arm suspended a modified phenotyping tri-metric detector array, configured as a single sensor row pseudo cluster setup modular data pack, and was used for NSF workshop demonstrations exercises in 2016 and 2017. Storm was also used to collect example data in Camelina, before Dr. Thompson collected the first real experimental data in core program Cotton investigations season summer 2017. Therein complementing our basic tri-metric, an 8’ 2x2” wooden stick was clamped to the metal frame and used to forward suspend cameras.
I believe the Avenger data set from this same experiment is of highest rank for our program and serves well to compare. Furthermore, I suggest that when produced correctly, sensor results will be nominally platform specific. Therefore the steel push cart with bike wheels is inexpensive, reliable, and allows maximum research resources to be utilized in data acquisition or compute.
Please see our publication highlighting Storm and method, “Deploying a proximal sensing cart to identify drought adaptive traits in upland cotton for high-throughput phenotyping”, Frontiers 2018. I am very grateful to be published with these authors in what is to me a future popular platform and base phenotype detection application.
Dr. Thompson is providing in the National Agricultural Library, Ag Data Commons, the CAD files, parts lists and videos describing our rigs. These are media for which experimentalists can benefit in developing their own lightweight electric phenotyping platforms. Contact me for collaboration with my Research Leader Dr. Dyer and specific program support in developing value optimized field based high throughput phenotyping proximal sensing platforms.
The Storm proximal sensing push cart provides a frugal platform which supports a rich and refined high throughput phenotyping method application and subsequent valuable information empowering knowledge results.
 
Matthew M. Conley
added an update
Cotton seeds are going in the ground, signally the approach of another HTP summer season. And I am not quite ready to measure, so good luck to me in meeting my targets from now to year end. This time I have lots of additional projects to deploy which is good, and it is challenging to handle.
One suggestion in terrestrial field phenotyping, start early. In Maricopa, we like to collect a surface LiDAR based soil terrain early, perhaps right after planting. Soil settling and plant emergence can be captured soon after first irrigation with a subsequent scan. I keep watch on my normal 2-10 cm of wheel track soil compression (depending on soil type and water management) by repeating the same tracks so as to be incorporated into experimental design. I am thinking that by temporally front-loading my wheel track soil compaction artifact in season, I reduce the possibility of plant disturbances. Finally by tracking soil z-plane = 0 cm, and plant bed to sensor displacement accurately, a more accurate crop height observation is obtained.
Although electronic sampling is more regular and faster than hand sampling by an expert person, the error in displacement sensing for ultrasonic derived crop height graduated by cm can vary between the measurements of a 2 m tall cotton plant, versus a 10 cm short one, given sampling from a distance of about 1 m. This effect is not as pronounced when experts use eye-balls and meter sticks appropriately. Conversely, if an expert looked at a ruler measuring a plant from a high or low inclination, that person might introduce a similar effect optical illusion induced error, by recording some offset tick mark in view instead of the true z parallel one.
 
Matthew M. Conley
added an update