Program and abstracts

Please click on a talk title for the abstract.

Day 1 - Thursday 27th September 2018

Time	Title	Presenter	Slides
08:30 – 09:00	Registration and Networking
09:00 – 09:30	Reflecting on thoughts and writing by G.A. McIntyre	Olena Kravchuk	Download
Reading slowly McIntyre's original paper [Australian Journal of Agricultural Research, 3:385-390, 1952] highlights the reasons for slow adoption of the ranked set sampling in agriculture and environment field practice and suggests ways of engaging biology researchers in a dialogue leading to a mutually productive extension of the RSS practice.
09:30 – 10:30	Ranked Set Sampling: rationale and historical development	Doug Wolfe	Download
When McIntyre first proposed the innovative idea of ranked set sampling in 1952, I doubt that he had any idea of the impact it would eventually have on the field of statistical inference. There was a flurry of statistical activity immediately following his seminal paper, including derivation of some basic properties of mean estimation based on ranked set samples, and there were scattered applications of the methodology. However, it would take another thirty years of virtual dormancy in the field before the spark from Stokes and Sager (1988) would reignite interest in this novel approach to sampling. Additional methodological papers followed fast and furious and active research in the field continues today. In this talk we briefly discuss these early efforts, but our emphasis will be on the general structure of ranked set sampling and the unique properties of statistical procedures utilizing RSS data. Seldom has a statistical concept/idea presented itself with so many interesting and unique opportunities for methodological development. In the second part of the presentation, we highlight the special features and attributes of RSS that lead to both its flexibility and its effectiveness, as well as address a few issues of concern when using this approach to sampling. We then briefly discuss a number of statistical settings where RSS procedures have been studied. In the final part of the talk, we describe some RSS-inspired offshoots that have also led to substantial research development. The full extent of the impact on statistical methodology that McIntyre inspired in 1952 is yet to be realized.
10:30 – 11:00	Morning tea break
11:00 – 12:00	Design Based Inference in Ranked Set and Judgment Post Stratified Sampling Designs	Omer Ozturk	Download
In this talk, we will consider design based inference in finite populations and experimental designs. In finite population setting, we construct probability proportional-to-size (pps) ranked-set and judgment-post-stratified sampling designs. These sampling designs select the units with selection probabilities proportional to their size and then induces further stratification based on their relative positions in a small comparison set. Hence, the information content of the samples are increased by inducing additional structure in the sample based on their position information in the comparison sets and the use of unequal selection probabilities. We extend the pps-ranked-set (ppsrs) and pps-judgment-post-stratifed (ppsj) sampling designs to stratified populations, and construct unbiased estimators for the population mean, total, and their variances. The new sampling designs are applied to apple production data to estimate the total apple production in Turkey. The second part of the talk, we introduce a special design, order restricted randomized design, which uses available subjective information in a small set of experimental units to create a judgment ranked blocking factor. The design then uses a randomization scheme that relies on these subjective judgment blocks to assign the treatment levels to experimental units. Since the randomization is tied to subjective ranking of the units prior to performing the experiment, the proposed scheme fits well within the general framework of ranked set and jps sampling methodologies. On the other hand, unlike rss and jps, it uses all the units within each set with some restricted randomization to treatment combination. Such an assignment induces positive dependence among within set units, but the restrictions on randomization translate this positive dependence into a variance reduction technique. We will provide some examples with this design and discuss some logistical and practical issues in the implementation of the design.
12:00 – 12:30	Embedding Ranked Set Sampling in design of experiment	Richard Jarrett	Download
In this talk, we consider how ranked set sampling might be used on the plots within a designed experiment. Sampling from a plot might be thought of as sampling from either an infinite or a finite population. Either way, with the correct sampling method, an analysis of variance can be used to identify within and between unit variances, and to determine the efficiency of the method. Simulations are used to show how the efficiency of the method drops away as the correlation between measurement variable and classifying variable reduces. Finally, we show that it is possible to do ranked set sampling where only one sample is taken from each plot.
12:30 – 13:00	Commonalities between Latin Squares Sampling and Ranked Set Sampling	Ray Correll	Download
Latin square sampling is a technique described by Cochran. In common with Ranked Set Sampling (RSS) it requires there to be n2 sampling units. The units are typically defined in the field by an n-by-n grid (but not necessarily a square grid). A Latin square design is then placed on the grid, and the units with the same letter (e.g. all n B’s) are then sampled. In a situation where there is a gradient across the columns the rows are then ranked sets. In Latin square sampling the choice of units within the set is predetermined so that is a case of RSS. The mean of the Bs is not dependent on their order, so the condition of a steady gradient can be relaxed. The sets could also be defined with the gradient across the rows. Latin square sampling is therefore a 2-dimensional ranked set sampling. This small study highlights similarities between experimental design and sampling. The effectiveness of the Latin square sampling is illustrated from sampling for disease in sugar cane
13:00 – 14:00	Lunch
14:00 – 17:30	Round-table discussion at Aroma Café, Waite Campus, Uni of Adelaide	McLeod House Rms G1B and G1C
14:00 – 14:30	Working together on creating Ranked Set Sampling designs: researcher-statistician knowledge transfer	Olena Kravchuk
Designing a robust and efficient plan for collecting research data in field, glasshouse and laboratory research is a task that requires close collaboration between biologists, statisticians and software and hardware engineers. Discipline-specific principles of data-mining and reasoning with data have to be clearly communicated between the parties involved. In a series of short exercises here, we work through statistical principles of design-based sampling and measurement error typical in field sampling.
14:30 – 17:30	Round-table discussion with Agricultural and Biological Scientists and grain growers	Facilitator: Dr Belinda Cay, AgCommunicators
16:50 – 17:30	Refreshment and networking
18:30 – 21:30	Symposium dinner at Edinburgh Hotel, Mitcham	Google Maps Location

Day 2 - Friday 28th September 2018

Time	Title	Presenter	Slides
09:30 – 10:30	Ranking methodologies in Ranked Set Sampling	Amer Al-Omari	Download 1 Download 2
The ranked set sampling (RSS) was first suggested by McIntyre [Australian Journal of Agricultural Research, 3:385-390, 1952] for mean estimation as efficient method compared to the well-known simple random sampling (SRS) method. Various modifications of the usual RSS since 1952 have been suggested to estimate the population parameters with some applications. This talk will be about the RSS and multistage extreme ranked set sampling (MERSS) methods (Jemain et al. 2007, DOI:10.3844/jmssp.2007.58.64) in estimating the population mean.
10:30 – 11:00	Morning tea
11:00 – 11:30	Improvements to Ranked Set Sampling	Jennifer Brown, Abdul Haq	Download 1 Download 2
Ranked set sampling (RSS) has been extended in various ways since it was first suggested by McIntyre in 1952. The original design remains an efficient survey method when compared with simple random sampling. It has potential application to a wide range of studies in environmental management, ecology, and agriculture, where exact measurement of a selected unit is either difficult or costly and time-consuming. Various modifications to RSS have been suggested. Here we review some of these modifications, including paired RSS, double RSS, multistage RSS, partial RSS, mixed RSS and paired double RSS. We conclude with some comments on the comparison of these methods and opportunities for their use in biology field studies.
11:30 – 11:50	Case-studies of typical agronomy sampling	Peter Kasprzak, Ray Correll	Download 1 Download 2
Two current industry standard sampling protocols were observed in mid-2018 in order to help identify possible improvements and to determine if Ranked Set Sampling (RSS) was appropriate. The first in Ouyen Victoria looked at soil grid sampling to investigate soil nutrient and moisture levels resulting from treatments in a randomised complete block design. 62 plots were randomly allocated over 4 blocks in sandy soil. The second involved investigating seed emergence counts in Roseworthy Adelaide from different seeding methods. 66 plots were randomly allocated 11 treatments in a randomised complete block design. In each case a recommended industry protocol was followed which was critically analysed to determine if it was indeed answering the required questions driving the experiment, and whether there was room for improvement in terms of accuracy and efficiency. In each case it was determined that there was room for improvement with respect to the current protocol and available tools. RSS was determined to not be suitable with current tools for this case of soil sampling, but exciting opportunities exist for RSS in seed emergence sampling, involving current precision agricultural practices.
11:50 – 12:10	Multiphase design and linear mixed model analysis of NIR scanning data	Sharon Nielsen	Download
Near infrared (NIR) spectral analysis is a rapid, non-destructive assessment tool widely used to predict or describe chemical properties of a material under investigation. Reflectance values from the NIR region of the electromagnetic spectrum are measured on samples of the material. Different wavelengths in the NIR spectrum will interact with specific chemical bonds (usually O-H, N-H and C-H), resulting in high absorbance of the electromagnetic energy at that wavelength. This interaction results in energy absorbance at related wavelengths producing highly correlated absorbance information referred to as overtones and combinations. These relationships and the presence of serial correlation in the spectra mean that multicollinearity is a major problem for the statistical analysis of NIR spectra if traditional linear models are used. Therefore, partial least squares regression and principal components regression became popular for forming calibrations based on NIR spectral data even though several steps are required in this process. My research has led to the development of graphical tools for the visualization of NIR spectra. I have also explored the multiphase aspects of spectral studies, using two spectral studies that have been designed and analysed using rigorous scientific methods. The analysis of these studies has been achieved in the linear mixed model framework where the correlation of the spectra between wavelengths can be modelled appropriately. Additionally, the variances associated with the strata of a spectral study have been estimated. The importance of a statistical design for the collection of spectral data has been highlighted in the results from these studies.
12:10 – 12:30	Big data and HPC in grains research	Andy Timmins	Download
In recent years there has been an explosion in data which is related to agricultural research. This data deluge originates from a variety of sources such as sensors, satellites, drones and next generation sequencing to name but a few. Big data is a term that describes the large volume of data – both structured and unstructured – that is being generated on a day-to-day basis. So, what does the term “big data” actually mean? According to Wolfert et al. (2017), ‘a unifying definition of big data is difficult to give, but generally it is a term for data sets that are so large or complex that traditional data processing applications are inadequate’ [Agric Syst 153:69–80]. Laney (2001) introduced the concept of the three Vs which has become a popular framework for thinking about big data: Volume (quantities of data that are being generated), Velocity (the frequency at which data is generated, captured, and shared), Variety (wide variety of data which comes in all types of formats – from structured, numeric data in traditional databases to unstructured text documents) [META group research note 6.70 (2001): 1]. It really does not matter which definition of big data you choose to use. What is important is what you do with the data that is being collected. This is the problem facing researchers working on grains because processing big data is not a trivial task. It is essential to have adequate computational processing power to “wrangle” the data. Storing, managing and analyzing the data present many challenges in terms of making sense of these vast data sets in a timely and accurate fashion. This is where high-performance computing (HPC) comes to the rescue. The focus of this talk will be on providing examples of the types of data that are becoming available to agricultural researchers and to discuss the cloud-based HPC platform that has been adopted by our research group to “crunch” these massive data sets.
12:30 – 14:00	Lunch and tours at the Waite Campus
13:00 – 13:30	The Plant Accelerator tour: https://www.plantphenomics.org.au/about-us/	Dr Chris Brien
13:30 – 14:00	The Waite Arboretum tour: https://www.adelaide.edu.au/waite-historic/arboretum/	Dr Kate Delaporte
14:00 – 14:10	Coombe vineyard as a teaching tool for sampling	Olena Kravchuk	Download
Undergraduate teaching in Agriculture and Biological Sciences has a strong focus on project-based learning. This method of learning is also efficient in professional development programs for researchers. Incorporating field research facilities in undergraduate teaching, postgraduate research training and professional development programs provides a basis for such project-focussed teaching. In this talk, we share the story of transforming the Coombe vineyard at the Waite campus of the University of Adelaide into an education tool for teaching the principles and techniques of sampling. The vineyard was planted about 30 years ago. The block allocated for this project contains 352 individual vines of eight rootstocks of Shiraz. The vines are planted in adjacent pairs of panels of two plants of each rootstock, 16 panels per row, 11 rows, each row containing a complete and balanced set of rootstocks. The sampling population is scalable: 88 groups of four adjusted vines of same rootstock, 176 panels of two vines of same rootstock, 352 individual vines, about 352 x 30 shoots, about 352 x 50 bunches, and so on till one is interested in individual leaves or berries. In 2017 – 2018, teaching modules for sampling principles have been developed and incorporated in an undergraduate course for the 3rd year students in BAgScience and in a workshop for researchers at the Australasian Soil Disease Conference (Adelaide, 2018). In discussions with teachers and trainers, participants designed, conducted and analysed descriptive sampling for assessing the average circumference of plant trunk. GenStat (v.17) software was used in the undergraduate project, and the survey and sampling R packages in the research workshop. We are commenting here on the teaching potentials of this project-based approach.
14:10 – 14:20	About assessment performance in the visual estimation of areas of grape bunch defects	Olena Kravchuk, Eileen Scott	PMapp page
Developing tailored solutions for field sampling in industry and research would bring most benefit to ensuring the quality and traceability of the data collected. This talk presents a reflection on the conversations between statisticians and the wine industry steering group in a research project led by Prof Scott (Professor of Plant Pathology, University of Adelaide) and sponsored by Wine Australia (Australian Government), 2014 - 2017. A series of round-table dialogues facilitated by Prof Scott allowed the statisticians to understand requirements, expectations and constraints by the industry, and the industry representatives to engage with the statistical measures of the assessment quality, tailored to the practice of assessors training. An online tool for area estimation was developed as part of the overarching project and is now available online at www.pmassessment.com.au for grape assessment training and other use as appropriate.
14:20 – 14:30	PMapp functionality improved and new smart-phone app for assessing grapevine diseases, pests and disorders	Eileen Scott
Diseases, pests and disorders reduce yield and quality of grapes harvested. Damage may be assessed during the growing season to support decisions about treatment options and close to harvest to inform decisions about quality and price. Many Australian wineries set thresholds for contamination or severity of damage at 3-5%, above which consignments of grapes may be downgraded or rejected. Assessment is typically based on visual inspection. A smart-phone application, PMapp, was released in 2015-16 to facilitate training and in-field assessment of powdery mildew on bunches. Feedback indicated that accuracy of the GPS was insufficient and that the app was being used to assess various diseases and pests in addition to powdery mildew. An updated PMapp has been developed with improved GPS capability, typically better than 10 metres, depending on network coverage. A new app, Grape Assess, offers all the capability of the improved PMapp and allows the user to record assessments of multiple diseases and disorders, including bunch rot, downy mildew, insect damage and sunburn. Grape Assess allows the user to customise each assessment by selecting the plant part and severity categories that best suit their requirements. Both apps are linked to online resources for assessing powdery mildew www.pmassessment.com.au.
14:30 – 14:50	Use of drones to non-destructive assessment in wheat field trials	Rhiannon Schilling	Download
Subsoil constraints prevent current elite wheat varieties from achieving their grain yield potential. This study uses drones to non-destructively assess the growth and spectral reflectance of wheat in field trials with dispersive subsoils. Trials at Mallala and Roseworthy were imaged using 3DR Iris+ quadcopters fitted with a Sony RX100 III red-green-blue (RGB) camera and a MicaSense® multispectral camera at stem elongation and anthesis. Both trials had 41 bread and durum wheat varieties and a destructive trial of 12 bread wheat varieties. Values for canopy biomass, height and spectral reflectance indices were derived from drone images for each plot. Ground measurements of plant height, leaf greenness and multispectral information using a FieldSpec® were recorded for each plot. Shoot biomass cuts were collected in the destructive trial at both time-points. Variation between wheat varieties for height, biomass and multispectral indices was detected at both sites. Plant height measured using a ruler was correlated to canopy height obtained from RGB images with an R2 = 0.92 at Roseworthy and R2 = 0.61 at Mallala. The fresh weight of biomass from destructive plots was correlated to the canopy biomass derived from RGB images with an R2 = 0.62 at Roseworthy and R2 = 0.75 at Mallala. Changes in plant height and biomass between stem elongation and anthesis suggest drones fitted with RGB cameras will be useful to non-destructively measure wheat growth through time in the field. Further research is underway to measure additional time-points of biomass throughout the growing season.
14:50 – 15:10	Shiny RSS app for a dialogue with field researchers	Peter Kasprzak	Download
It is well recognised that collaboration between statisticians and researchers is needed for designing a research study in which statistical reasoning can be proficiently exercised. Shiny is an R package that allows the creation of interactive web based applications without the need for other computing languages. Non-statisticians can input parameters into a pre-programed R framework and view the results in real time. Without the need for extended face-to-face consultations, the opportunity is enabled to guide a dialogue with field researchers through designing real-life sampling protocols, as well as to gather their expert knowledge for statistical simulations. Such a dialogue with field researchers can be opened in a space where commonly accepted generic protocols are seldom revisited. In this talk we are presenting an example of a Shiny app, which can also compare various sampling schemes, including balanced Ranked Set Sampling. We discuss how this approach can re-elevate the conversation about how data is gathered and alternative solutions. We suggest that an immediate extension for this collaborative dialogue tool is to enable the user to make use of already in place data gathering automated tools. The latter, we believe will provide direct benefits to implementing RSS in the practice of field research.
15:10 – 15:40	Afternoon tea
15:40 – 16:00	Latency of Fruit Yield Response to the Environmental Factors	Myung Hwan Na	Download 1 Download 2
This presentation points out the importance of incorporating the latency of response in the prediction model of fruit yield. The plant does not respond to changes in environmental factors immediately. To illustrate this, the response time of crop yield to a variety of environmental factors, including temperature, solar radiation and humidity, etc. is studied in particular. The field data used in this research is collected on farms so as to minimize the negative influence of uncontrolled factors on the accuracy of statistical analysis. Five farms participating in this research are located in South Korea.
16:00 – 17:00	Round-table discussion on sampling and survey methodology in the practice of agriculture research and education in South-East Asia	Ray Correll, Chris De Ieso, Oula Bouphakaly, Khonesavanh Phialathounheuane, Bounsanong Chouangthavy	Download
17:00 – 17:10	Symposium closing
18:00 – 19:00	Farewell drinks & science networking
19:30 – 21:30	Webinar, reflecting on the discussions in the Symposium To be held mid-late October	Facilitators: Jennifer Brown and Olena Kravchuk

Day 3 - Saturday 29th September 2018

Time	Title	Presenter
09:30	Depart Hotel	Majestic Roof Garden Hotel
09:45	Meet at St Saviours Church	St Saviours Church
11:00	Vineyard visit and talk about sampling for phylloxera (soil coring – placing of core – targeted sampling not representative but good for quarantine)	Ray Correll
12:00	Field site inspection – discuss philosophy of trial design. emergence sampling, dry weight sampling, harvest data, extrapolation from plots to field	Ray Correll
13:00 – 14:00	Lunch at Sedan
14:00 – 14:30	Examine hay bailing and sampling issues	Ray Correll
14:30 – 14:45	Sampling for rare weeds	Ray Correll
15:00	Palmer Field site, soil core sampling, sampling for patchiness, on-off row sowing with GPS tracking	Ray Correll
16:00	Afternoon tea	St Saviours Church
17:00	Return to Hotel