Dr Katy Martin Rainey was recently appointed as

  • Slides: 40
Download presentation
Dr. Katy Martin Rainey was recently appointed as the new director of the Purdue

Dr. Katy Martin Rainey was recently appointed as the new director of the Purdue University Soybean Center by Dr. Karen Plaut, the Glenn W. Sample Dean of the College of Agriculture. Rainey, an associate professor of agronomy, specializes in soybean genetics and plant breeding. Dr. Katy Martin Rainey is an Associate Professor of Plant Breeding and Genetics in the Purdue University Department of Agronomy. She studies genetic improvement of soybeans for increased yield and better quality using multidisciplinary approaches. Dr. Rainey focuses on integrating diverse sources of information to demonstrate new approaches to soybean breeding. She collaborates with geneticists, agronomists, economists, engineers, and other soybean breeders in the public and private sectors to do this. Dr. Rainey has been breeding soybeans for over 10 years and has released specialty cultivars. For yield improvement, she focuses on dissection of yield into components traits in productive environments. This includes goals to describe new traits associated with yield that can measured using new technologies such as precision and high-throughput phenotyping. She also explores how to predict yield and maximize gain from selection in soybeans using the components of phenotypic variance, high density markers, and mixed models for analyses. Katy Rainey, Associate Professor of Agronomy Purdue University

Managing UAS Imagery for Developmentally-Driven Decision Making Dr. Keith Cherkauer UAS: Unmanned Aerial Systems,

Managing UAS Imagery for Developmentally-Driven Decision Making Dr. Keith Cherkauer UAS: Unmanned Aerial Systems, aka drone.

The number of plots for multi-environment yield trials in the Rainey lab in 2019,

The number of plots for multi-environment yield trials in the Rainey lab in 2019, a typical year. A total of 14, 000 plots evaluated annually at three location in IN 6, 000 plant rows 3, 000 preliminary yield trial plots 3, 000 plots other experiments 1000 AYT plots Specialty traits, genetic architecture, phenomic inference and characterization of diverse germplasm. Advance yield trials PLANT ROWS: the first stage in soybean breeding that selected inbred plants are grown in rows/plots.

The benefits of Predicting Genetic Yield Potential More accurate selections. Screen more material. -

The benefits of Predicting Genetic Yield Potential More accurate selections. Screen more material. - Increase the probability of identifying rare variants. More effective field operations generally. - Risk management. - Focused effort. Where i is the standardized selection differential r is the selection accuracy �� A is the square root of the additive genetic variance T is the length of time to complete one breeding cycle GENETIC YIELD POTENTIAL: the yield of optimal or ideal genotypes in a target environment, across major growing regions and environments

K. M. Rainey is co-founder of Progeny Drone, Inc. Progeny Drone Inc. Dr. Anthony

K. M. Rainey is co-founder of Progeny Drone, Inc. Progeny Drone Inc. Dr. Anthony Hearst (CEO) (Full Time) Ph. D - Drone Image Processing & Software Development ahearst@progenydrone. com https: //www. progenydrone. com/ Dr. Katy Rainey (CTO)

Define the Data You Need Define your objectives. - Genetic analyses needs many observations.

Define the Data You Need Define your objectives. - Genetic analyses needs many observations. - Ergo robust UAS platforms for row crops. - RGB: growth analyses, canopy features. - MS: color changes (i. e. senescence). - RGB + Thermal: enahanced yield prediction? Define the relevant phenology. Define your minimal spatial and temporal resolution. Always at least 80% overlap. RGB: red, green and blue; images from a typical camera. MS: multispectral; sensor data quantifying reflectance of spectral bands. PHENOLOGY: plant life cycle events influenced by seasonal and interannual variations in weather and climate. LONGITUDINAL TRAITS: traits recorded multiple times during a season that have continuously-variable phenotypes.

Implications for defining your minimal spatial and temporal resolution. There is a LOT of

Implications for defining your minimal spatial and temporal resolution. There is a LOT of genetics and selection to be done on growth analyses of phenotypes from RGB images. -Recognize data snobbery. -Do Genetics Now! Collect images and other data when developmentally relevant. -This is dynamic. From Fabiana Moreira Each site and each trait has different requirements resolution. -Not helpful to standardize platforms or data collection protocols. Stand Counts: 1 cm/pixel Canopy Cover: 4 cm/pixel Biomass Accumulation: dynamic

Distribution of average canopy coverage of the checks by days after planting for progeny

Distribution of average canopy coverage of the checks by days after planting for progeny rows 2015 and 2016. How to describe/ prescribe a phenotyping protocol? F. F Moreira, Hearst, A. A. , Cherkauer, K. A. & K. M. Rainey. Improving the efficiency of soybean breeding with high-throughput canopy phenotyping. Plant Methods 15, 139 (2019). https: //doi. org/10. 1186/s 13007 -019 -0519 -4 CC/ACC: canopy coverage measured with UAS calculated as percentage pixels from canopy-segmented RGB plot clips/average canopy coverage is a phenotype derived from fitting CC observations to a model (logistic) and averaging the interpolated values

Correlation between interpolated daily canopy coverage values and yield in Soy. NAM. Agronomy 2018,

Correlation between interpolated daily canopy coverage values and yield in Soy. NAM. Agronomy 2018, 8(4), 51

Ground Reference Data and Metadata Do you need to develop a remote-sensing prediction equation?

Ground Reference Data and Metadata Do you need to develop a remote-sensing prediction equation? - Predict ground reference phenotypes from remote data with cross-validation. - You may need to calibrate both the spectral data and the prediction equation. Example biomass estimation pipeline Image Acquisition Image analysis Vegetation indices & canopy coverage From Fabiana Moreira Principal component analysis of Log biomass Linear regression with 10 -fold cross validation by environment Imagederived biomass for each DAP GROUND-REFERENCE DATA: ground observations of the phenotype to be estimated from remote-sensing, and metadata needed for calibration of prediction equations. Data Type Description Format Supplementary groundreference data. Includes the location of all ground control points and spectral targets. Field spectrometer + thermal target data. Varies, but mostly ASCII text data files. Size: < 10 GB Short-term Storage Field computer.

Ground Reference Data and Metadata Do you need to calibrate your spectral data for

Ground Reference Data and Metadata Do you need to calibrate your spectral data for reflectance? From Bilal Jamal Abughali SPECTRAL REFLECTORS, OTHER TARGETS OR REFLECTORS: ground features and fixed points visible in images for calibration of reflectance, color, temperature, height. REFLECTANCE PANELS: reflect at a specific and consistent percentage of light across the Visible and near Infra-red spectrum.

Ground Reference Data and Metadata Do you need georeferencing? - Maybe not. Just use

Ground Reference Data and Metadata Do you need georeferencing? - Maybe not. Just use range, row coordinates (needed for GPS planters). - Experiments blocked in field by UAS objective to minimize flight number. - Merge all data from range, row coordinates. Include borders. Minimum features and metadata: - GPS coordinates of field corners and/or blocks (arranged by plot size). - Visible flagging or feature in first and last plot. https: //bmspro. io/1357 GROUND CONTROL POINTS (GCPS): ground features visible in images that provide fixed points for geo-referencing.

Free Flight Data Management with Drone Logbook Data Type UAS Flight Info Description Format

Free Flight Data Management with Drone Logbook Data Type UAS Flight Info Description Format Flight parameter inputs, and Varies depending on data actual flight positions, stream, but primarily ASCII including GPS locations of text files. all images. Size: < 1 GB Short-term Storage UAS data card to field computer.

UAS and Ground Data Collection & Phenology Visualization Planting Date 2018 -05 -17 2018

UAS and Ground Data Collection & Phenology Visualization Planting Date 2018 -05 -17 2018 -05 -22 2017 -05 -31 Traveling phenomics teams will be ineffective.

Raw Drone Imagery is not “the data”. Data Type Description Original imagery Raw collected

Raw Drone Imagery is not “the data”. Data Type Description Original imagery Raw collected by the UAS cameras, Imagery includes RGB, multispectral and TIR. Format RAW or other uncompressed image file format. Actual format will be camera dependent. Size: 100 s of GB per day, size varies cameras used. Cumulative 5 -7 TB per season. Short-term Storage Camera memory card to field computer.

Don’t quantify phenotypes from orthomosaic “pretty pictures”. Multi-layer Mosaics Extract multiple replicate images of

Don’t quantify phenotypes from orthomosaic “pretty pictures”. Multi-layer Mosaics Extract multiple replicate images of individual plots from the raw, overlapping frame photos. PLOT CLIPS: Extracted smaller images of plots from raw photos. Orthorectified and labeled images of individual plots of uniform pixel dimensions. Could be calibrated, segmented, binary, etc. ORTHORECTIFIED: effects of image perspective (tilt) and relief (terrain) are removed from the image data.

o Dateof. Flight_Platform. Camera. Location. Experiment o 190709_e. BSODAACRESoy Data Type Description Format Processed

o Dateof. Flight_Platform. Camera. Location. Experiment o 190709_e. BSODAACRESoy Data Type Description Format Processed UAS Imagery Intermediate labeled image products, potentially atmospherically corrected or segmented. Various image file formats. Size: Variable from 1 to 100 s of GB. Shortterm Storage Local computer hard drive.

Advantages of replicate labeled plot clips of uniform pixel dimensions. Preserve raw image colors

Advantages of replicate labeled plot clips of uniform pixel dimensions. Preserve raw image colors & sharpness. - Avoids mosaicking errors & color distortion. Trace to original raw photo, which provides position of UAS when photo was taken (cardinal orientation, altitude, pitch, roll, yaw). Can be filtered, discarded and subset for quality control. - QC: blurriness and location of a plot within the photo. - Photo perspective (cardinal) relative to plot, i. e. can correct for glare. Can be batch processed. Provide standard deviation of phenotype. - Enable statistical analyses & quality control. - Quantify precision & improve accuracy. Avoids HPC and cloud computing. Replicate plot image 1 Can can be filtered, discarded 2 3 4

Software Outputs ✕ No RTK-GPS ✕ No ground control points ✕ No digital terrain

Software Outputs ✕ No RTK-GPS ✕ No ground control points ✕ No digital terrain models ✕ No polygons or shape files ✕ No orthomosaics ✕ No cloud-computing ✕ No high-performance computing üAutomated ü 10 -20 min üAt fields edge üImagery QC at site üPhenotype QC at site üDecision-making at site K. M. Rainey is co-founder of Progeny Drone, Inc.

Software Outputs K. M. Rainey is co-founder of Progeny Drone, Inc. The replicate phenotypes

Software Outputs K. M. Rainey is co-founder of Progeny Drone, Inc. The replicate phenotypes quantified from plot clips. REFERENCE IMAGE/ REFERENCE VALUE: the plot clip taken from the photo that was most nadir/ the phenotype quantified from the reference image. NADIR VIEW: the camera is directly above the plot.

Software Outputs K. M. Rainey is co-founder of Progeny Drone, Inc. Heat map of

Software Outputs K. M. Rainey is co-founder of Progeny Drone, Inc. Heat map of plots with real-time UAS phenotypes Range Median canopy coverage Generating this map at the field’s edge allows you to scout, QC or annotate outliers, or quickly develop a subsampling strategy, or make selections immediately. Row

Soybean Progeny Row Selection Experiment Progeny Rows 2015 Progeny Rows 2016 Fabiana Moreira, A.

Soybean Progeny Row Selection Experiment Progeny Rows 2015 Progeny Rows 2016 Fabiana Moreira, A. Hearst, K. Cherkauer, and K. M. Rainey (2019) Improving the efficiency of soybean breeding with phenomic-enabled canopy selection (Under Revision)

F. Moreira et al. Soybean Progeny Row Selection Experiment Year 3 Advanced Yield Trials

F. Moreira et al. Soybean Progeny Row Selection Experiment Year 3 Advanced Yield Trials AYT early 17 AYT late 17 Rank AYT early 18 AYT late 18 Yield, Yield|ACC 1 Yield, Yield|ACC 2 Yield, Yield|ACC ACC, Yield|ACC 3 Yield, Yield|ACC 4 Yield, Yield|ACC ACC, Yield|ACC 5 Yield|ACC ACC, Yield|ACC 6 ACC, Yield, Yield|ACC ACC, Yield|ACC 7 ACC Yield, Yield|ACC 8 Yield, Yield|ACC Yield ACC, Yield|ACC 9 Yield, Yield|ACC 9 8 Yield, Yield|ACC 3 3 8 Yield 5 Yield 26 Yield|ACC 10 29 ACC 10 8 5 Yield|ACC 2 6 8 Yield|ACC 0

Model using the Green Ratio Validation of the model using an Vegetation Index (GRVI)

Model using the Green Ratio Validation of the model using an Vegetation Index (GRVI) to predict independent sample. biomass. From Fabiana Moreira Logistic growth curve of biomass for each genotype from emergence until 125 days after planting (DAP).

Correlation between interpolated daily canopy coverage values and yield in Soy. NAM. How to

Correlation between interpolated daily canopy coverage values and yield in Soy. NAM. How to pick “a day”? Agronomy 2018, 8(4), 51

“Results showed improvements in the predictive ability for yield with respect to those [genomic

“Results showed improvements in the predictive ability for yield with respect to those [genomic selection] models that solely included genomic data. These relative improvements ranged [27 – 165%]. ” “Similar improvements were observed…when the reduced canopy information for days 14– 33 was used to build the trainingtesting relationships, showing a clear advantage of using phenomics in very early stages of the growing season. “ Agronomy 2018, 8(4), 51

Canopy Coverage Changes from Herbicide Injury: 0 -99 Days after Treatment with 0%, 30%

Canopy Coverage Changes from Herbicide Injury: 0 -99 Days after Treatment with 0%, 30% and 60% Glyphosate Rates Days After Treatment 1. Automate and quantify herbicide injury ratings. 2. Genetics of herbicide resilience in soybean. 3. Calibrate UAS sensitivity for detecting stress for targeted management.

Canopy Greenness Changes from Herbicide Injury: 0 -99 Days after Treatment 30% Glyphosate Rate

Canopy Greenness Changes from Herbicide Injury: 0 -99 Days after Treatment 30% Glyphosate Rate Canopy Greenness Nontreated 60% Glyphosate Rate

CIE + VID MLM for MS In development in the Cherkauer lab The work

CIE + VID MLM for MS In development in the Cherkauer lab The work flow of the first module, Crop Image Extraction (CIE) The work flow of the second module, Vegetation Indices Derivation. A fast image indexing and customized Vegetation Indices (VI) derivation. MLM: Multi Layer Mosaic, analysis technique using all raw UAS images

Mesh from Point Cloud by Delaunay triangulation Dr. Monica Herrero-Huerta Rainey Lab Herrero-Huerta and

Mesh from Point Cloud by Delaunay triangulation Dr. Monica Herrero-Huerta Rainey Lab Herrero-Huerta and Rainey (2019) High Throughput Phenotyping of Physiological Growth Dynamics from UAS-based 3 d Modeling In Soybean

Biomass Estimations POINT CLOUD MESH BIOMASS ESTIMATION DAP 44 DAP 56 Monica Herrero-Huerta, and

Biomass Estimations POINT CLOUD MESH BIOMASS ESTIMATION DAP 44 DAP 56 Monica Herrero-Huerta, and K. M. Rainey. ‘High Throughput Phenotyping of Physiological Growth Dynamics from UAS-Based 3 D Modeling in Soybean’. (2019) ISPRS-International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences 4213

3 D Processing: Height Point Clouds Monica Herrero-Huerta, and K. M. Rainey. ‘High Throughput

3 D Processing: Height Point Clouds Monica Herrero-Huerta, and K. M. Rainey. ‘High Throughput Phenotyping of Physiological Growth Dynamics from UAS-Based 3 D Modeling in Soybean’. (2019) ISPRS-International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences 4213

Do Genetics Now - Our approach is to use robust and inexpensive RGB data

Do Genetics Now - Our approach is to use robust and inexpensive RGB data collected over multiple environments and sampling dates. We observe critical time points that quickly change. - Crop scientists and agronomists themselves should progress with applications of these new and valuable data. - Emerging technologies will follow in the wake of that progress. Thank you!

Software Inputs K. M. Rainey is co-founder of Progeny Drone, Inc.

Software Inputs K. M. Rainey is co-founder of Progeny Drone, Inc.

From Fabiana Moreira Temporal Genetic Variation in Soybean Biomass Dr. Luiz Brito Asst. Professor

From Fabiana Moreira Temporal Genetic Variation in Soybean Biomass Dr. Luiz Brito Asst. Professor Purdue Animal Science Ph. D: Guelph March 16

Temporal Genetic Variation in Soybean Biomass Top 20 SNP effects over time for the

Temporal Genetic Variation in Soybean Biomass Top 20 SNP effects over time for the Quadratic Legendre polynomial coefficient. C From Fabiana Moreira

Genetic Architecture of Soybean Canopy Coverage In Soy. NAM Source Material: Soy. NAM Genetic

Genetic Architecture of Soybean Canopy Coverage In Soy. NAM Source Material: Soy. NAM Genetic Architecture of Soybean Yield and Agronomic Traits (2018) Brian W. Diers, et al. G 3: GENES, GENOMES, GENETICS October 1, 2018 vol. 8 no. 10 3367 -3375 Dr. Alencar Xavier Corteva Agri. Sciences Purdue Agronomy

Alencar Xavier, B. Hall, A. A. Hearst, K. A. Cherkauer, and K. M. Rainey

Alencar Xavier, B. Hall, A. A. Hearst, K. A. Cherkauer, and K. M. Rainey (2017) Genetic architecture of phenomic-enabled canopy coverage in Glycine max. Genetics 206(2): 1081– 1089. Days After Planting GWAS for canopy coverage. Genomic regions significantly associated with early-season soybean canopy coverage observed or estimated for each day from 14 to 56 days after planting. Chromosome -log(p-value)

F. Moreira et al. Soybean Progeny Row Selection Experiment Mean comparison among selections: Yield

F. Moreira et al. Soybean Progeny Row Selection Experiment Mean comparison among selections: Yield with R 8 as covariate from year 2 preliminary yield trials (PYT) Progeny Row Selection Categories

How to reduce cycle time, T? 1. Multiple generations per year 2. Early generation

How to reduce cycle time, T? 1. Multiple generations per year 2. Early generation selection of parents • F 2 -F 3 family testing • Genomic Selection/Prediction • Genomic Selection from a Recurrent Pool 3. Phenomic yield prediction: decide best performers early in the season and use as parents Three barriers to GS in Crops The extension of GS to the case of across cycles has been a challenge, mainly due to the low predictive accuracy resulting from two factors: 1. reduced genetic relationships between different families 2. augmented environmental variances between cycles 3. plant breeding phenotyping accuracy and efficiency needs improvement GS + HTPP: 2 basic approaches 1. reduce training data collection for GS on HTP trait directly 2. secondary trait data used in combination with GS for primary trait