Exploring Maize Resilience Through Genetics, Phenomics, and Canopy Architecture

Addie Thompson; Assistant Professor, Crop and Soil Science, Michigan State Author

12/18/2023 Added

4 Plays

Description

Cohost with CROPS, a graduate student and postdoc group funded and supported through the Center for Plant Science Innovation. Social following the seminar.

Searchable Transcript

Search:

[00:00:00.780]The following presentation
[00:00:02.220]is part of the Agronomy and Horticulture Seminar Series
[00:00:05.790]at the University of Nebraska-Lincoln.
[00:00:08.670]So folks, let's get settled in, get started.
[00:00:11.760]It's my real pleasure this afternoon
[00:00:13.260]to be introducing our seminar speaker
[00:00:15.150]for the Ag and Hort Departmental Seminar Series,
[00:00:17.370]Professor Addie Thompson, Michigan State University.
[00:00:20.490]Now, Addie grew up, what, about an hour away from here
[00:00:22.590]in Hamburg, Iowa,
[00:00:24.600]got her undergraduate degree from Iowa State University,
[00:00:27.600]PhD, working with Gary Muehlbauer at Minnesota,
[00:00:33.810]studying the developmental genetics of maize.
[00:00:37.080]Moved over to Purdue University.
[00:00:38.820]Did a postdoc with Mitch Tuinstra,
[00:00:40.740]one of the original RPE TERRA phenotyping projects,
[00:00:43.920]a lot of drone work, also international work.
[00:00:46.230]I think heat-tolerant...
[00:00:47.250]Was it heat-tolerant or drought-tolerant?
[00:00:49.040]Drought.
[00:00:49.873]Drought-tolerant maize for Africa.
[00:00:51.600]Was then hired at Michigan State University
[00:00:54.300]as a member of the Plant Resilience Institute?
[00:00:57.540]I actually joined it later, but yeah.
[00:00:58.530]Okay. Hired at Michigan State,
[00:01:00.690]joined the Plant Resilience Institute,
[00:01:03.032]now manages a program funded with NSF, USDA, and DOE grants.
[00:01:09.000]Collaborates with professors
[00:01:10.770]across the Midwestern Corn Belt, as well as private sector,
[00:01:16.560]IBM, number of seed companies,
[00:01:19.110]different private sector collaborations,
[00:01:21.660]and is gonna tell us about a wide range of things.
[00:01:24.840]So with that, please help me
[00:01:26.610]in welcoming Professor Addie Thompson.
[00:01:34.080]Thank you, James.
[00:01:34.913]That was a very thorough and entertaining introduction.
[00:01:37.950]I appreciate that.
[00:01:39.720]Thank you also for the invitation,
[00:01:41.520]both from PSI and the department and CROPS,
[00:01:45.420]so the Agronomy and Horticulture Department,
[00:01:48.630]Plant Science Institute, and the CROPS group.
[00:01:50.730]I actually got two invitations to come, which was great,
[00:01:53.580]and I made sure I didn't come twice,
[00:01:55.200]although that would've been fun.
[00:01:57.451]And also to Mike Tross, shout out,
[00:01:58.800]because he's been driving me all over the last couple days,
[00:02:00.990]and that's been really, really helpful.
[00:02:03.270]It's great to be here.
[00:02:04.620]It's great to be in a place
[00:02:05.970]that loves corn almost as much as I do.
[00:02:08.250]This is my hotel. That is my hotel room.
[00:02:11.760]The curtains, the bathroom, the wallpaper. It's amazing.
[00:02:16.020]And above my bed, there's this map of Nebraska.
[00:02:19.710]I've zoomed in on the southeast corner,
[00:02:21.600]because as James said, you're in Lincoln here,
[00:02:23.700]obviously that's where we are now.
[00:02:25.290]This is where I grew up. So right there, pretty close by.
[00:02:28.500]And because of that,
[00:02:29.970]I actually had kind of a fun random opportunity.
[00:02:33.660]My parents, hi, are gonna see one of my talks
[00:02:37.080]for the first time in my life.
[00:02:39.060]So, (chuckles) they just happened to be so close by.
[00:02:43.950]Yes, that's great.
[00:02:45.030]I said if they fall asleep, they just can't snore.
[00:02:47.310]So same rule applies to everybody else.
[00:02:51.900]Okay. Currently I am at Michigan State University.
[00:02:54.555]One of the goals of our program,
[00:02:55.980]the main overarching theme is improving plant breeding.
[00:03:00.240]I am quick to point out
[00:03:01.890]that I'm not a bonafide plant breeder
[00:03:04.230]because we don't really release varieties.
[00:03:06.180]Michigan State's not really set up to do that.
[00:03:08.220]I was the first maize geneticist there, really,
[00:03:10.230]for the last 30, 40 years before I came there.
[00:03:13.830]So I was starting a new program from scratch
[00:03:15.810]without any germplasm or setup.
[00:03:20.610]So most of my focus is on maize.
[00:03:22.110]We do also do some work with sorghum.
[00:03:23.970]and our main interests are in genetic architecture
[00:03:27.150]of quantitative traits,
[00:03:28.260]and that's various quantitative traits.
[00:03:29.767]Improving approaches for predictions,
[00:03:32.850]so we utilize genetics and genomics,
[00:03:34.710]high throughput phenotyping, as James mentioned,
[00:03:37.278]learning about kind of G by E by M interactions,
[00:03:39.930]and then breeding program simulation and optimization.
[00:03:43.080]So we do a lot of germplasm characterization
[00:03:45.193]as well as population development.
[00:03:48.630]So you could consider that like pre-breeding.
[00:03:51.060]What this looks like in terms of the program,
[00:03:52.740]we do a lot of computational work,
[00:03:54.420]but we also do a lot of field work.
[00:03:55.860]So we have nurseries and observation plots.
[00:03:58.410]Someone asked me the other day how many acres this was,
[00:04:00.537]and I actually hadn't calculated it,
[00:04:02.610]but it's around 18 acres a year that we manage.
[00:04:04.440]So it really is like a breeding program, but to do research.
[00:04:08.340]So I'm gonna start out with a little bit of the background
[00:04:12.150]just so that we're all on the same page,
[00:04:14.233]though I'll warn you,
[00:04:15.180]I'm gonna move through that pretty fast
[00:04:16.920]because I have a lot of cool science
[00:04:18.540]that I wanna talk about.
[00:04:19.620]So I have a lot of assumptions
[00:04:20.790]about the knowledge level in this room.
[00:04:23.220]We'll see.
[00:04:26.580]And then I'll focus on a few
[00:04:28.860]of the different research projects that we have going on,
[00:04:31.860]and hopefully we wrap up within an hour.
[00:04:34.974]Okay, so I mentioned we're really interested
[00:04:37.530]in plant breeding research,
[00:04:39.120]and when you're doing plant breeding,
[00:04:40.920]you're trying to select individuals from populations
[00:04:43.830]to then form the next generation of your program
[00:04:46.440]to progress whatever your trait of interest is, right?
[00:04:49.260]Higher yield, better resistance
[00:04:50.850]to abiotic or biotic stresses, et cetera,
[00:04:53.880]or quality and nutrition.
[00:04:55.440]But how do we choose the best plants?
[00:04:57.690]Well, variety trials are a good way to do this, right?
[00:05:02.130]And I teach the first semester of plant breeding
[00:05:05.460]in a state where corn is not necessarily the main focus.
[00:05:11.310]Michigan is the second most agriculturally diverse landscape
[00:05:14.070]after California.
[00:05:15.180]So the students there
[00:05:16.110]study all kinds of different interesting plants.
[00:05:19.140]So I try to incorporate examples
[00:05:20.670]from things that are not corn.
[00:05:22.080]So this is sugar beets, happens to be.
[00:05:24.810]So the best approximation for a variety's performance
[00:05:27.780]is to grow lots of replications of that variety
[00:05:30.840]across a bunch of locations throughout what we would call
[00:05:32.910]your target population of environments, TPE,
[00:05:35.760]over the course of many years,
[00:05:36.870]and then directly measure the traits of interest,
[00:05:38.850]and make a lot of crosses, right?
[00:05:41.010]But that costs a lot of time and money.
[00:05:43.080]So taking shortcuts requires us
[00:05:45.420]to come up with some new tools.
[00:05:47.042]So maybe a reasonable approximation could be
[00:05:51.030]to replicate genes as opposed to whole varieties,
[00:05:54.394]borrow information from relatives,
[00:05:56.273]maybe use crop growth models to make different predictions
[00:06:00.540]across environments and climates,
[00:06:02.640]and then statistically model correlated or related traits,
[00:06:05.730]preferably, hopefully ones that are easier or cheaper,
[00:06:08.700]more accurate and predictable and more heritable.
[00:06:11.130]And then instead of lots of crosses,
[00:06:12.810]potentially you could be making targeted genetic changes.
[00:06:15.510]So this is going to require
[00:06:17.370]and does require some different technologies.
[00:06:20.370]Genotyping, parameterization.
[00:06:23.220]Trending data and fancy math.
[00:06:24.480]You can tell that that's not gonna be the focus
[00:06:26.100]of my talk today.
[00:06:27.330]Other phenotypes and gene editing technologies.
[00:06:30.060]So I am gonna focus on a few of these different aspects,
[00:06:35.200]but from a broad sort of view,
[00:06:40.500]sort of the not quite 40,000 feet, but maybe maybe 400 feet,
[00:06:44.400]which is about where would a drone be flying.
[00:06:48.960]All right, so I'm gonna give you a crash course
[00:06:51.030]in quantitative genetics,
[00:06:52.170]followed by a crash course in phenotyping,
[00:06:54.960]but they're each gonna take about two minutes.
[00:06:57.720]So, Genome-Wide Association Study.
[00:06:59.970]This is one of the tools that we use
[00:07:01.239]to combine genotypic and phenotypic data to identify loci
[00:07:05.370]that are significantly impacting our traits of interest,
[00:07:07.860]whatever those may be.
[00:07:09.000]So imagine a chromosome,
[00:07:11.910]and there are some genes on this chromosome.
[00:07:14.430]And then we have genotyping information,
[00:07:17.910]usually from SNPs, so single nucleotide polymorphisms,
[00:07:21.900]where we've scored across our population these SNPs,
[00:07:25.096]and then for each of those throughout the population,
[00:07:28.020]we say, okay, individuals with an A
[00:07:30.180]versus individuals with a G, what is their phenotype?
[00:07:32.520]So maybe this group, it's here, it's here.
[00:07:35.130]So we would say, okay, this SNP is significantly associated
[00:07:37.980]with that trait because we see this difference,
[00:07:40.035]whereas over here, this one is really not.
[00:07:42.292]That SNP doesn't have an impact on the trait.
[00:07:45.210]You can extend this out to genomic prediction,
[00:07:46.980]which is where you use all of your markers
[00:07:50.250]to estimate the effects on the trait of interest.
[00:07:52.606]So this uses linear modeling,
[00:07:54.510]and you're estimating each marker's numeric effect.
[00:07:56.640]So given our same genome here, let's say we estimate
[00:07:59.760]that this has an impact of 1.7,
[00:08:01.740]this would be something very close to 0.
[00:08:04.260]But you do this for every marker,
[00:08:05.371]and then you score individuals based on their genotypes
[00:08:08.475]in each case, sum them all up,
[00:08:10.830]and that gives you
[00:08:11.663]what we would call the genomic estimated breeding value.
[00:08:14.160]You may have heard that phrase before, that acronym.
[00:08:18.000]So then you would test the accuracy
[00:08:19.800]on individuals with known phenotypes, choose the best model,
[00:08:22.620]apply that to somewhere where you have unknown phenotypes,
[00:08:25.500]but you've done the genotyping.
[00:08:27.090]So when you do that in a breeding program
[00:08:29.940]to make breeding decisions,
[00:08:30.990]that's what we call genomic selection.
[00:08:32.794]Okay. So now you're experts in quantitative genetics.
[00:08:36.028]There are really good applications for this.
[00:08:38.850]There's advantages, but there's also limitations.
[00:08:40.980]So this is good for applied questions
[00:08:43.350]if you wanna do marker and loci discovery
[00:08:45.090]or trait prediction.
[00:08:46.590]For basic biology, like causal gene discovery,
[00:08:49.320]it can be less straightforward,
[00:08:51.270]although given really large populations
[00:08:53.190]with really high marker coverage, it can still be possible.
[00:08:56.370]But you're limited by accuracy, precision,
[00:08:58.440]and heritability of the trait,
[00:08:59.430]as well as the population; type, diversity, recombination,
[00:09:02.460]structure, kinship, et cetera;
[00:09:04.112]what sort of markers you have;
[00:09:05.430]and then the architecture of the trait itself.
[00:09:08.160]It also doesn't consider effects of the environment.
[00:09:10.800]So if you're trying to run a genomic prediction study
[00:09:12.750]to predict into other environments,
[00:09:14.970]that becomes incredibly challenging, if not impossible.
[00:09:17.757]So, another approach. Moving on to plant phenomics.
[00:09:21.900]Here's where we're measuring features,
[00:09:23.667]and I use the word features instead of traits
[00:09:27.540]because sometimes we don't know what the trait is
[00:09:30.150]that we're measuring per se.
[00:09:31.851]We can model traits, we could predict traits,
[00:09:35.550]but sometimes we're measuring things
[00:09:37.110]that aren't really the trait itself.
[00:09:38.970]But the idea in my definition here
[00:09:41.070]is that we want to measure these
[00:09:42.210]with some sort of increased throughput
[00:09:44.430]or efficiency or detail.
[00:09:46.890]So what system you're going to use
[00:09:48.364]and how successful that's going to be
[00:09:50.067]will depend on your questions of interest.
[00:09:52.410]For today's purposes, I'm gonna focus on data
[00:09:55.800]collected from drone-mounted sensors,
[00:09:58.110]otherwise known as an unoccupied aerial system,
[00:10:03.997]and that will include a number of different sensors.
[00:10:07.078]I'm focusing on drone-based imaging,
[00:10:09.450]partly because for me that's the most relevant
[00:10:12.480]to a field scale application,
[00:10:15.001]but also because you guys
[00:10:16.500]have this really cool, awesome greenhouse.
[00:10:18.750]We don't, so I don't have that system. (chuckles)
[00:10:21.420]This is what I have.
[00:10:22.980]So RGB, I am explaining my acronyms here.
[00:10:26.820]So this is, it stands for red, green, blue.
[00:10:28.410]It's basically natural light.
[00:10:29.700]So taking pictures, basically what your eyes would see
[00:10:33.763]Multispectral, that's usually a smaller number of bands.
[00:10:38.070]In our case, we have five bands of multispectral imagery,
[00:10:40.635]and they're usually wider bands of light.
[00:10:44.100]So that could be within or outside of the visible spectrum.
[00:10:46.410]I think the five bands that we have are red, green, blue,
[00:10:48.570]near-infrared, red edge.
[00:10:50.040]So you're looking at different aspects
[00:10:52.740]of plant health, usually.
[00:10:54.870]Hyperspectral imager, this would be
[00:10:56.760]where you have a larger number of more narrow bands,
[00:10:59.427]and that spans more regions usually of the spectrum.
[00:11:02.630]Thermal would be for temperature,
[00:11:04.710]so that is actually still also part
[00:11:06.090]of the electromagnetic spectrum, just even more distant.
[00:11:08.404]And then, finally, LiDAR.
[00:11:10.470]So that stands for light detection and ranging.
[00:11:12.930]You've probably heard about this most
[00:11:14.490]in terms of self-driving vehicles.
[00:11:16.170]So the lasers, the spinning set of lasers,
[00:11:18.840]they go out, they bounce off an object,
[00:11:20.370]and it measures the time to return,
[00:11:21.872]which then gives you an estimate
[00:11:23.670]of where that point hit in space.
[00:11:25.680]So it maps geometry, essentially.
[00:11:28.320]So you can use that to drive a car,
[00:11:30.390]but you can also use it to look at canopies
[00:11:32.445]and structure of plants.
[00:11:35.550]So, applications, advantages and limitation here
[00:11:38.250]is that you could measure more individuals,
[00:11:40.710]more varieties, plots, reps, locations,
[00:11:42.840]phenotypes, timepoints, and features.
[00:11:45.350]You can oftentimes enable new information capture,
[00:11:48.060]so things that you can't see or quantify
[00:11:50.700]by the human eye or by hand.
[00:11:52.800]It can capture things that are still undefined information
[00:11:56.430]within that feature space.
[00:11:57.780]So it's up to you to figure out
[00:11:59.100]what it is that that's explaining.
[00:12:01.230]And it can be limited by the necessity
[00:12:04.770]of having diverse skill sets,
[00:12:06.570]because you need to be able to design the experiment,
[00:12:08.610]manage the trial, acquire data, process it,
[00:12:11.460]and do the analytics,
[00:12:12.293]and that's a lot of different skill sets
[00:12:13.809]that are equally important.
[00:12:16.380]Cost can be a limiting factor,
[00:12:18.856]and sometimes data compute, storage, and transfer
[00:12:21.540]can be limiting.
[00:12:22.440]We're lucky at institutions
[00:12:23.580]that we oftentimes have in-house,
[00:12:25.620]kind of high throughput computing,
[00:12:27.690]high performance computing.
[00:12:30.270]It can also be limited by varying interpretability.
[00:12:34.170]I mentioned that sometimes you're getting features
[00:12:35.910]you don't really understand.
[00:12:36.840]So sometimes you are able to figure out
[00:12:38.580]what it is that it's picking up on, sometimes you're not.
[00:12:41.970]But all of these different approaches
[00:12:46.200]are really getting at this idea of addressing bottlenecks.
[00:12:49.500]And I wanted to include this slide.
[00:12:51.540]It's actually borrowed from a talk that I gave
[00:12:53.456]to a group of really diverse graduate students
[00:12:57.727]across plants and animals at a symposium about bottlenecks.
[00:13:01.440]But I like this slide because it helps
[00:13:03.330]to think about what question you're asking
[00:13:05.970]and how to go about answering it.
[00:13:07.800]So it's kind of a conceptual flowchart.
[00:13:09.930]What is your problem?
[00:13:10.853]Is it people and skills
[00:13:12.720]or is it more about equipment and tools?
[00:13:14.790]If it's the people and skills,
[00:13:16.110]does somebody have that expertise?
[00:13:18.390]If they do, you could find a collaborator, right?
[00:13:21.390]If they don't, then you're gonna need to train and study
[00:13:23.520]and get the skills.
[00:13:25.290]If you're not sure, then ask.
[00:13:27.510]You gotta ask around and figure it out.
[00:13:30.210]For equipment and tools, if the tool exists, get it.
[00:13:33.450]If not, maybe you'll have to build it.
[00:13:35.430]If you're not sure, test it. Figure it out.
[00:13:37.980]This seems really straightforward,
[00:13:40.560]but it, to me, helps kind of conceptualize
[00:13:44.008]when you reach a stumbling block,
[00:13:47.970]what is the step that you should take?
[00:13:49.950]And oftentimes these things interrelate,
[00:13:52.560]because probably you don't necessarily have the skills
[00:13:54.720]to build a tool,
[00:13:55.770]but maybe there's an engineering person that does
[00:13:57.630]you could collaborate with.
[00:13:59.250]The other thing I like to emphasize here
[00:14:00.690]is that a lot of graduate programs tend to think about this,
[00:14:05.970]training and studying, acquiring the knowledge yourself,
[00:14:08.370]as the main outcome of the program.
[00:14:10.800]But I would argue
[00:14:11.633]that a lot of the interpersonal skills are really critical,
[00:14:13.980]because here you're thinking about critical thinking
[00:14:15.914]and knowledge and skills,
[00:14:17.310]but also you're learning about communication
[00:14:19.876]and the confidence to go to an expert in the field
[00:14:22.530]and ask them for advice.
[00:14:24.090]You have a lot of creativity
[00:14:24.923]in thinking about how would I go about
[00:14:27.150]making or testing a tool for this problem, right?
[00:14:30.540]And of course, obviously you're always
[00:14:31.920]gonna have to weigh time and cost.
[00:14:33.840]So this is kind of the framework I like to think about
[00:14:35.970]in terms of improving plant breeding research,
[00:14:38.310]but really any research.
[00:14:40.320]So that ends your flash introduction.
[00:14:42.893]And I'm gonna move into, now,
[00:14:46.020]some of our research project spotlights.
[00:14:48.660]Like I said, this will be a subset
[00:14:49.811]of what we work on in the lab.
[00:14:51.930]So just for fun...
[00:14:56.310]So let's see.
[00:14:57.708]Current projects, we are working on maize nitrogen response
[00:15:02.060]in terms of genetics and phenomics.
[00:15:04.020]I will talk about that.
[00:15:05.120]The sorghum functional genomics and physiology
[00:15:07.772]is unfortunately not gonna be talked about,
[00:15:10.320]but I'd be happy to talk to anybody individually
[00:15:12.870]over pizza later.
[00:15:15.360]We'll touch on tar spot disease resistance a bit.
[00:15:17.688]Again, the phenolic compound accumulation,
[00:15:19.950]I won't be able to talk about much,
[00:15:21.300]but that's an interesting project
[00:15:22.710]where we're looking at phenolic compounds,
[00:15:24.420]both in some leaf, but mostly grain tissue
[00:15:26.783]in the two species.
[00:15:29.121]Phenomics of canopy architecture
[00:15:31.020]within the Genomes to Fields program is in the talk.
[00:15:35.880]This breeding program simulation and optimization is not.
[00:15:38.774]There is a preprint out. The program is called Piber Ops.
[00:15:43.170]It's all in Python,
[00:15:44.190]and it's a multi-objective genomic mating strategy.
[00:15:48.780]So you can check that out if you're interested.
[00:15:53.006]And yeah, so I'll talk first
[00:15:57.240]about the tar spot disease project.
[00:15:58.620]So this is one of my master's students primarily.
[00:16:02.790]Then for the nitrogen work, this will be Brandon Webster.
[00:16:06.540]Kinda giving my credits upfront so I don't forget.
[00:16:09.210]And then canopy architecture is mostly Zhongjie,
[00:16:12.690]who's a almost graduated PhD student
[00:16:15.360]who is looking for postdocs.
[00:16:19.920]Okay, if you have not yet heard of tar spot disease,
[00:16:23.160]it's a fairly new disease here in the Upper Midwest.
[00:16:26.790]It's caused by the pathogen Phyllacora maydis,
[00:16:29.795]and it is an ascomycete fungi.
[00:16:31.350]It's host is maize.
[00:16:32.700]Unfortunately, it's an obligate,
[00:16:33.900]and we can't actually inoculate for it, really.
[00:16:35.850]So that creates challenges.
[00:16:37.920]But it's spread by wind and water.
[00:16:39.900]It causes these raised shiny black spots on the leaves.
[00:16:42.960]They look like flecks of tar.
[00:16:44.430]If you go to scratch it, it won't scratch off.
[00:16:46.680]If it does scratch off, it's probably insect frass or dirt.
[00:16:50.850]It thrives in cool, humid conditions,
[00:16:52.301]and it can spread quickly and cause lots of yield loss.
[00:16:55.440]So this disease has actually been around
[00:16:56.940]for the last 100 years in Central and South America,
[00:16:59.070]but it's new here as of about 2015.
[00:17:02.670]So it can cause varying spot sizes
[00:17:04.530]with or without these necrotic lesions.
[00:17:06.330]They call 'em (indistinct)
[00:17:10.348]Okay, now are we back? Excellent.
[00:17:12.900]All right. Welcome back, everyone. (chuckles)
[00:17:17.130]So tar spot, it can quickly break down the canopy,
[00:17:19.470]and this becomes an issue both for forage quality
[00:17:23.100]but also for lodging.
[00:17:26.057]So this image here is from,
[00:17:28.740]they're taken about two weeks apart,
[00:17:30.930]and this area where we see all this brown
[00:17:33.600]is heavily infested.
[00:17:35.100]It spreads throughout the field.
[00:17:36.420]The whole field is infested, and this part starts lodging.
[00:17:38.790]So that's one of the big problems.
[00:17:42.390]This is not a, oh there we go. Okay.
[00:17:44.190]So it pulls the nutrients from the stalk
[00:17:46.260]and increases lodging.
[00:17:48.420]Now, the reason why I think this is at least
[00:17:50.220]of some interest to this audience
[00:17:52.110]is that this disease has spread significantly,
[00:17:55.290]and it is now in Nebraska.
[00:17:57.810]Yay! Sorry.
[00:18:02.880]I will say, this is incidents.
[00:18:05.040]It doesn't have any information on this graph
[00:18:07.260]about severity.
[00:18:08.280]And I can tell you that it is very, very low here.
[00:18:10.830]So what we see around the Great Lakes
[00:18:12.763]is all this extra moisture and humidity,
[00:18:16.650]and the leaves stay wet and we get lots of disease pressure
[00:18:19.470]and it causes problems.
[00:18:20.880]Out here, this is probably just a few spots.
[00:18:22.993]So someone found it to say it is here.
[00:18:26.359]That doesn't mean it's having any impact on yield yet.
[00:18:29.280]Now, you get a wet season in the future,
[00:18:31.410]you might be in trouble.
[00:18:33.960]So. Or irrigation.
[00:18:36.337]Or irrigation. Oh yeah.
[00:18:37.260]Yeah, no, management plays a huge role,
[00:18:38.670]and if you are somebody that likes to irrigate a little bit,
[00:18:41.436]you know, that's a little bit very frequently,
[00:18:44.640]that's gonna be a problem for sure.
[00:18:47.340]Yeah.
[00:18:48.660]So the work that I'm talking about
[00:18:50.250]was funded from the USDA through this
[00:18:52.410]what we call the Great Lakes Tar Spot Initiative.
[00:18:54.690]I figure, you know, go big, right? Make it sound cool.
[00:18:58.410]So I formed this collaboration
[00:18:59.910]among some groups around the Great Lakes.
[00:19:01.890]Like I said, that's where it's, the severity is the worst.
[00:19:04.072]And we decided that what we really wanted to do
[00:19:06.197]was to screen varieties for resistance.
[00:19:09.390]And so we chose about 800 diverse varieties.
[00:19:11.550]These were taken from the Wisconsin Diversity Panel
[00:19:13.530]in maize as well as the gem lines,
[00:19:15.480]so germplasm enhancement of maize,
[00:19:17.142]thinking that we would get some temperate
[00:19:19.530]and some tropical materials in combination.
[00:19:22.668]We took weekly ratings and then recorded other phenotypes.
[00:19:25.325]And what I can tell you from that
[00:19:27.601]is that we did find resistance, which is great.
[00:19:33.120]So we screened this germplasm.
[00:19:34.320]We did identify some resistant varieties.
[00:19:37.740]We saw a little bit slower accumulation of disease
[00:19:41.300]in the tropical varieties,
[00:19:42.930]with the caveat that they were pretty devastated by rust.
[00:19:46.410]So it's a trade-off, you know,
[00:19:48.120]if you get something from another region,
[00:19:49.644]maybe it has resistance to diseases that are native there,
[00:19:52.560]but not so much to ones that are here.
[00:19:53.880]So we did find some resistance in temperate varieties
[00:19:56.580]that were still also rust-resistant, which was nice.
[00:19:59.310]We mapped genetic loci contributing to that variation,
[00:20:01.920]and then started back crossing the resistant lines
[00:20:04.133]into some elite temperate field and sweet corn lines,
[00:20:07.020]some collaborations in sweet corn as well.
[00:20:10.860]Then we also validated that
[00:20:12.090]by mapping it in an external MAGIC population,
[00:20:14.660]so we mapped QTL,
[00:20:17.720]and then started developing some,
[00:20:21.570]so we developed some new populations for study.
[00:20:24.000]We created some double haploid varieties
[00:20:26.250]using our resistant lines
[00:20:27.990]crossed to a couple of different other parents,
[00:20:31.410]and then we're back crossing subsets of these
[00:20:33.474]to the parents to study heterotic effects.
[00:20:34.350]I would love to tell you really cool results
[00:20:36.180]from that part of the study,
[00:20:37.417]but unfortunately, since we can't inoculate,
[00:20:41.283]we're reliant on natural infection,
[00:20:43.260]and we just haven't had really bad tar spot years
[00:20:45.630]the last couple of years.
[00:20:46.980]So, I joke that I created such resistant varieties,
[00:20:50.190]I just can't get them to be infected,
[00:20:51.840]but it's not actually the case.
[00:20:54.990]So then we're also using these populations as a test bed
[00:20:57.150]for our phenomics technologies,
[00:20:58.950]with the idea that hopefully
[00:21:03.960]detection and prediction for management decisions.
[00:21:09.480]Okay, thank you all, both in the room and online,
[00:21:14.430]for your patience
[00:21:16.020]with all of our technical exciting challenges.
[00:21:19.260]And thank you to the people in the front row.
[00:21:23.400]Thank you for my helpers. Yeah.
[00:21:27.150]Okay, so the idea, the hope eventually
[00:21:29.940]is that we would be able to improve detection
[00:21:31.890]of onset and severity through the use
[00:21:33.960]of some of these different types of imaging.
[00:21:36.330]So this is really an example to introduce you to the idea
[00:21:41.190]of vegetative indices, honestly.
[00:21:43.140]So the top image that you're looking at is what we call RGB.
[00:21:46.170]So that's just our, what your eyes would see from the sky,
[00:21:48.980]and the bottom is what you would see
[00:21:50.670]if you were looking at it
[00:21:51.810]through the lens of a vegetative index.
[00:21:56.040]So the idea is that you could collect data
[00:21:57.660]throughout the season,
[00:21:58.493]and then relate that back to onset and severity of disease
[00:22:00.810]to try and identify it before it's visible in the field.
[00:22:03.896]This has led into,
[00:22:05.779]along with our sort of phenolic compound data
[00:22:09.030]that I'm not talking about right now,
[00:22:10.712]another iteration of this tar spot project,
[00:22:14.520]where we're relating phenolic compound accumulation
[00:22:16.405]to tar spot resistance.
[00:22:18.420]And part of that
[00:22:19.253]comes with this like tiered hyperspectral phenolic approach,
[00:22:22.740]where we're using leaf tissue and leaf-level hyperspectral
[00:22:27.900]and then drone-level hyperspectral imagery,
[00:22:29.820]and hopefully then validating
[00:22:32.490]and testing some of our candidate resistance genes.
[00:22:35.198]And this is a collaboration here.
[00:22:38.400]So this is what the leaf hyperspectral data looks like,
[00:22:41.468]and this is a teaser for the LiDAR data,
[00:22:44.550]which I'll talk about a a little bit later.
[00:22:46.980]But for now, although tar spot is awesome,
[00:22:52.200]I'm gonna move on to talk a little bit
[00:22:53.610]about what we're working with
[00:22:55.020]on the genetics of nitrogen response in maize.
[00:22:58.260]And this is an image I'm showing you as an example
[00:23:00.736]from two months ago,
[00:23:04.920]showing some of our hybrid and inbred varieties.
[00:23:08.430]So this would be the parents of the hybrids
[00:23:09.960]that are growing here
[00:23:11.010]under plus and minus nitrogen conditions.
[00:23:12.840]And you can already see with your naked eye
[00:23:14.370]at this stage of the growing season, those differences.
[00:23:19.020]So, as I'm sure you are very well aware,
[00:23:21.420]nitrogen fertilizer is an expensive input
[00:23:23.550]with a lot of price volatility.
[00:23:25.181]This is information pulled September 29th.
[00:23:28.763]I actually don't know where it's at right this minute,
[00:23:31.680]but I can say that when this project started
[00:23:34.487]we were like right at one of these peaks.
[00:23:37.260]So it was a very pertinent time to be looking at nitrogen.
[00:23:43.463]And one of the things that was of interest to us
[00:23:45.510]is that it's actually kind of genetically dependent
[00:23:48.750]on how much nitrogen you're going to apply
[00:23:52.320]and when it will be utilized,
[00:23:53.160]because both of those differ by hybrid.
[00:23:54.900]So you have some varieties that will respond very highly,
[00:23:58.500]others that maybe might grow better under low nitrogen,
[00:24:03.180]but then they don't respond as much
[00:24:04.500]to the addition of supplemental nitrogen.
[00:24:06.660]And part of this is that it varies
[00:24:08.744]when the plant is actually taking up nitrogen from the soil.
[00:24:15.000]So the question really was how do farmers know
[00:24:17.370]how much to use for a new hybrid?
[00:24:20.010]What do you guess, right,
[00:24:21.150]for your amount that you're gonna use?
[00:24:25.650]Excuse me.
[00:24:26.863]So the questions for the project
[00:24:27.699]were to understand some of the response to nitrogen
[00:24:30.889]on a physiological level and a biochemical level as well.
[00:24:35.070]So what is it that's changing under nitrogen stress
[00:24:37.050]and throughout development?
[00:24:38.550]What is it that's unique or different about those hybrids
[00:24:40.710]that have high versus low response?
[00:24:43.590]Can we define an N response type for different hybrids
[00:24:46.020]and how stable is that across different environments?
[00:24:48.990]And then ultimately the goal
[00:24:50.340]is can we identify some genetic markers
[00:24:52.140]for nitrogen response?
[00:24:54.150]But what we found, we grew three seasons.
[00:25:00.030]28, 21, 22. Yes, three seasons.
[00:25:02.850]And we're marking here across 16 hybrids,
[00:25:05.280]so this is a set of four females, four males,
[00:25:10.170]the yield response to supplemental nitrogen.
[00:25:12.230]So there is high variability in hybrid response
[00:25:15.480]which comes from a combination of genotype, environment,
[00:25:18.780]and the interaction.
[00:25:21.503]At the same time
[00:25:22.336]we also looked at leaf nitrogen content over time,
[00:25:24.810]and what we saw is that overall
[00:25:29.189]it kind of looks like the slope is similar,
[00:25:31.470]but if you break this down and look at different hybrids,
[00:25:33.563]maybe that's not quite the case.
[00:25:35.957]We don't know yet though
[00:25:37.410]if that's because they're still taking up nitrogen
[00:25:40.101]later in the season,
[00:25:41.310]or if they're remobilizing it at a different rate.
[00:25:43.980]That's questions we've not gotten to yet.
[00:25:48.450]But since one of our interests is in phenomics,
[00:25:50.790]one of my students has been working on
[00:25:52.030]using these LeafSpec hyperspectral scans
[00:25:56.400]to model nitrogen content
[00:25:58.770]so we can predict pretty, pretty accurately.
[00:26:01.850]He has done this for a number of other elements as well,
[00:26:09.720]and it works to varying degrees.
[00:26:12.300]Water content is one
[00:26:13.260]that's a really, really good correlation.
[00:26:17.160]But ideally we could expand this out into our drone work.
[00:26:22.260]So far he's been working with the RGB data,
[00:26:24.240]so he has been showing
[00:26:25.830]that we do see differences from the sky.
[00:26:28.745]This is an example of plant height
[00:26:31.920]across growing degree days, so thermal time.
[00:26:34.787]This would be the timepoint where we applied the side-dress.
[00:26:37.740]That's our nitrogen fertilizer treatment
[00:26:39.330]is at the side-dress.
[00:26:40.230]And so we see that the height starts to differentiate
[00:26:43.020]about 27 days after the application,
[00:26:46.260]and then remains significant.
[00:26:50.730]He's looked at a number of other traits as well,
[00:26:52.800]but something I wanna highlight out of this work
[00:26:55.326]is that the trait correlations we see
[00:26:58.800]between our remote sensing and our field-based measurements
[00:27:02.490]can change under different treatments.
[00:27:05.400]So correlating remote sensing with field traits
[00:27:08.310]across the treatments would be one question,
[00:27:10.590]but correlating it within the treatments,
[00:27:11.880]you're gonna get different answers.
[00:27:14.730]So I'll highlight here the correlation
[00:27:17.940]of the grain nitrogen with yield
[00:27:21.240]differs in the two different treatments.
[00:27:23.970]And another interesting thing that he was finding
[00:27:26.820]is that he sees that the standard deviation
[00:27:31.710]of where the height is reaching,
[00:27:33.600]so how far down the point cloud is gonna go into the canopy
[00:27:36.992]also varies,
[00:27:38.171]in that you see a higher negative correlation with yield
[00:27:43.410]in the low nitrogen treatment only,
[00:27:46.140]which kind of makes sense.
[00:27:48.090]So I will give you a link if you're interested
[00:27:52.290]in looking at the biochemical and physiological responses.
[00:27:58.530]The phenomics portion that I just showed you
[00:28:00.270]is yet to come out, but this part you can look up.
[00:28:04.260]So what we've seen is that the supplemental nitrogen
[00:28:06.600]is maintaining photosynthetically-active leaf tissue,
[00:28:09.150]but to varying degrees in different hybrids.
[00:28:10.800]So we saw hybrid-specific changes
[00:28:13.140]in the agronomic traits that you would expect,
[00:28:17.460]but also in phenolic compound accumulation,
[00:28:19.693]chlorophyll levels, as well as carotenoids,
[00:28:22.380]plastoquinone and phylloquinone.
[00:28:23.907]And then also one of our collaborators was interested
[00:28:27.120]in cellular macrostructure and ultrastructure.
[00:28:29.940]So he was looking at plastoglobule size and abundance,
[00:28:33.000]cell wall thickness, and even plastoglobule shape.
[00:28:35.760]So he found some weird like crescent-shaped plastoglobules.
[00:28:38.955]As well as gene expression.
[00:28:40.920]So if you're interested in a gene expression dataset
[00:28:43.374]on different responses to nitrogen,
[00:28:45.840]that would be a good place to go.
[00:28:48.360]So I showed you this graph of plant height
[00:28:50.700]in these different treatments, and that's cool.
[00:28:54.510]I love it. It's great.
[00:28:55.770]But it's really just one piece of the picture.
[00:29:01.410]And what I really would love to be able to do
[00:29:03.990]is to kind of quantify the structure
[00:29:06.810]of all the leaves throughout the canopy
[00:29:08.310]beyond just what's the overall height of the canopy.
[00:29:11.040]'Cause really it's the leaves, you know,
[00:29:13.140]the leaves, the placement, the area, the number,
[00:29:15.440]that are doing the work,
[00:29:17.490]taking in the light and making the food,
[00:29:20.670]not just the height of the plant.
[00:29:22.530]So, again, back to this big question,
[00:29:25.767]we want to predict how a variety's gonna perform
[00:29:29.220]in order to assess its usefulness.
[00:29:31.080]As I said before, we can use genetics to predict phenotype,
[00:29:34.281]but it doesn't perform well in new environments.
[00:29:36.757]So something that we've been thinking about
[00:29:38.760]for a number of years now and are not, what we're working on
[00:29:43.590]is that maybe we could use this concept
[00:29:46.710]of physiological modeling, so crop growth models,
[00:29:48.660]to simulate varieties into different environments.
[00:29:51.780]The challenge is that that's gonna require
[00:29:53.680]a lot of tedious hand measured phenotypes
[00:29:55.730]to parametize the models.
[00:29:56.910]'Cause as geneticists,
[00:29:57.870]we wanna do it on a gene type specific basis,
[00:29:59.940]so this would be running hundreds of crop growth models
[00:30:02.340]to simulate how these will grow.
[00:30:05.820]So the big question we had
[00:30:06.720]is can we measure those phenotypes some other way?
[00:30:10.200]When I started at Michigan State,
[00:30:11.820]I created sort of a flow chart
[00:30:15.245]of my concept of my ideal world here, right?
[00:30:22.260]None of this, don't take a picture of this
[00:30:23.970]and think that it's the truth, right?
[00:30:25.050]This is my imagination.
[00:30:27.810]This is what I would like to think would be possible.
[00:30:31.140]But we didn't really know
[00:30:33.215]whether any of this was true or not.
[00:30:37.262]But my dream would be that we could use some sensors
[00:30:41.550]to measure traits that we begin to combine
[00:30:44.901]to model other response traits,
[00:30:47.280]maybe combine that with environmental information,
[00:30:50.100]weather and soil, rainfall, so on,
[00:30:52.671]to then predict the composite traits.
[00:30:54.960]And the reason this is exciting to me as a geneticist
[00:30:57.540]is because a lot of the traits that are over here
[00:31:00.960]would be presumably higher heritability,
[00:31:03.777]more influenced by genetics
[00:31:05.340]and less influenced by the environment,
[00:31:07.260]so that we get better predictions here
[00:31:08.265]by including sort of the process-based modeling.
[00:31:11.580]If you go from genes to yield,
[00:31:14.130]you're gonna get a different answer in different years
[00:31:16.230]depending on what's limiting the yield, right?
[00:31:18.390]But if you use genes to predict things like this
[00:31:21.000]and then use the weather data to predict the outcomes,
[00:31:23.520]which is what's really more biologically meaningful,
[00:31:26.133]our thought was that could be better.
[00:31:28.530]But again, we didn't know. It was a guess.
[00:31:30.210]And the really, really big guess here,
[00:31:31.830]I had some experience with RGB, hyperspectral, and thermal
[00:31:35.220]to know that these were, I was more competent in these,
[00:31:38.460]but this LiDAR has a lot of red lines coming out of it,
[00:31:41.460]going to a lot of different things
[00:31:42.750]that I was not sure about.
[00:31:44.220]So that was the big question we had.
[00:31:46.440]And ended up forming a collaboration
[00:31:49.382]focused on plant phenomics across the plant sciences
[00:31:55.200]and computational sciences.
[00:31:56.790]And one of the many projects that we worked on
[00:32:01.020]was this project with the Wisconsin Diversity Panel,
[00:32:03.900]where we had 761 different inbred varieties,
[00:32:07.350]two replications.
[00:32:08.787]It was pretty small plots, an average of 40 plants per plot,
[00:32:11.340]but it was over 1,500 plots.
[00:32:13.161]I'm gonna point out that this is not an image stitch issue.
[00:32:17.340]This was a planter issue,
[00:32:18.660]that we did have a little bit of an offset.
[00:32:20.700]I know that it's not sexy to have crooked lines,
[00:32:24.000]but, you know, gotta do what you gotta do.
[00:32:26.790]So in each of those plots we measured stand count,
[00:32:30.300]and anthesis and silking dates,
[00:32:31.470]but then in each plot we also marked two plants
[00:32:34.639]that seemed representative, had good stand.
[00:32:37.413]We put these little pink stakes here,
[00:32:39.960]and we actually labeled the leaves as they came out,
[00:32:44.626]and we got leaf number traits and then leaf area traits
[00:32:49.545]as well as the height traits.
[00:32:52.800]And the reason we were so focused on this
[00:32:55.320]as our metrics for canopy architecture writ large
[00:32:59.520]was I had previously done work in sorghum,
[00:33:03.916]which is actually phenotypically quite similar to maize
[00:33:06.780]when you've just got the one stalk.
[00:33:08.580]I mean, it does tiller, but what we knew from that work
[00:33:12.183]was that if you were to plot out the leaf area
[00:33:15.600]of each leaf at its expanded size,
[00:33:20.820]it's really, really nicely modeled
[00:33:23.040]by a pretty simple power function.
[00:33:25.380]It's such that if you know the leaf number
[00:33:27.600]of the largest leaf, the total leaf number,
[00:33:30.990]and the area of the largest leaf,
[00:33:32.760]you can parameterize this power function.
[00:33:34.590]And the blue dots are what we've measured.
[00:33:36.488]This reddish orange line is the power functions model.
[00:33:40.200]And I don't think I'm ever gonna have anything
[00:33:41.940]that fits this well ever again.
[00:33:43.950]So that was our logic in measuring those particular traits.
[00:33:48.990]And those are also things we could use potentially
[00:33:50.790]to parametize this physiological model.
[00:33:54.090]So we took UAV data,
[00:33:55.170]and I'm gonna focus for this talk on the LiDAR data,
[00:33:58.020]'cause again, that was the thing that we weren't sure about.
[00:34:00.930]So we took seven dates throughout the growing season,
[00:34:03.240]and this is what it looks like as just the plots.
[00:34:09.120]And those were then adjusted to elevations
[00:34:11.551]like relative to ground level, of course.
[00:34:14.010]So this is what the seven scans would look like
[00:34:15.840]over the course of a single growing season for one plot.
[00:34:18.846]What we ended up doing,
[00:34:20.100]this was a collaboration with Dan Morris,
[00:34:21.570]who's over in Computer Vision at Michigan State,
[00:34:23.910]and he created these voxels for each of the plots.
[00:34:26.940]So each voxel represents the likelihood
[00:34:30.180]that there's a point that falls into that space,
[00:34:32.790]and that allowed him to be able to use this as inputs
[00:34:35.010]into a convolutional neural net,
[00:34:36.630]which then let us make these predictions.
[00:34:39.180]So he actually divided each of the plots in half
[00:34:41.490]and had four samples per genotype.
[00:34:43.410]But I'll be quick to note that when doing the holdout sets
[00:34:46.320]in our cross validations,
[00:34:47.310]we held them out the level of genotype
[00:34:49.410]so that we're not being unfair, and, you know, inserting.
[00:34:52.890]I knew that was gonna be a question if I didn't say it.
[00:34:54.780]All right, the fun thing was it worked pretty well,
[00:34:59.160]and this is a very complicated figure,
[00:35:01.740]so I'll apologize in advance, but I'll walk you through.
[00:35:04.950]So each of these colored bars for a given trait,
[00:35:08.370]we'll just look at anthesis right now,
[00:35:10.158]represents the prediction accuracy
[00:35:12.450]of using that individual date to predict that trait,
[00:35:15.900]and the solid bars are if you use only that date.
[00:35:18.540]These lighter bars are
[00:35:19.680]if you use everything up until that date.
[00:35:22.050]So clearly for things like flowering time,
[00:35:24.450]you benefit a lot in accuracy
[00:35:26.490]from being able to use all the data going into it
[00:35:29.250]so you see the change.
[00:35:30.600]So that makes sense, right?
[00:35:31.974]For things like height, it doesn't matter as much.
[00:35:34.794]So you can pick a date here.
[00:35:38.040]July 30th is probably sufficient
[00:35:39.870]to actually estimate end of season plant height.
[00:35:43.020]So the other thing to note on this figure
[00:35:46.410]is that we weren't sure how impressed to be
[00:35:50.550]with some of these results.
[00:35:53.730]So we took correlations between our different replications
[00:35:57.036]so we'd have a sense
[00:35:58.560]of what is the best we could possibly do,
[00:36:02.417]because we're modeling all of these against our,
[00:36:04.590]what we call ground truth, but it's not really truth,
[00:36:06.750]it's some human that's measured it
[00:36:08.250]and there's error involved there.
[00:36:09.570]So the best you could do
[00:36:10.980]is getting as close as what that would get.
[00:36:16.732]Dan Morris also tried an ablation over voxel size.
[00:36:24.030]So he had chosen a voxel size, but hadn't really...
[00:36:29.670]We wondered how much that impacted the results.
[00:36:32.400]And it turns out not very much.
[00:36:35.430]It doesn't actually matter so much what your voxel size is,
[00:36:39.540]you still get pretty good results.
[00:36:42.180]And this was surprising at first,
[00:36:43.894]because we were thinking about this
[00:36:46.054]kind of like conceptually like resolution,
[00:36:48.630]but it's not really, because your resolution,
[00:36:51.352]the information you're getting is still the same.
[00:36:54.005]Each voxel is is not a binary trait.
[00:36:58.440]It's a probability numeric trait.
[00:37:01.230]And so really it's kind of unfair
[00:37:03.510]to think about this as a resolution.
[00:37:05.639]So the voxel size was not as important as we thought
[00:37:08.340]for that reason.
[00:37:10.650]One of my students also decided to compare,
[00:37:12.780]just out of curiosity.
[00:37:14.640]If you were gonna be making predictions
[00:37:15.960]based on genomic prediction or the LiDAR traits,
[00:37:20.044]how close you would get.
[00:37:23.006]In most cases those are very similar.
[00:37:25.975]In the cases of height, ear height and plant height,
[00:37:29.460]LiDAR data wins hands down,
[00:37:31.800]which isn't actually too surprising.
[00:37:34.890]But we did have some traits
[00:37:37.140]that were not super well predicted either way,
[00:37:39.480]but LiDAR does not pick up on leaf width very well.
[00:37:44.040]I can live with that.
[00:37:45.060]So we were happy that we could collect data and it worked,
[00:37:49.140]and surprised that it did as well as it did.
[00:37:51.480]So I think this is encouraging
[00:37:52.469]in my grand master plan, my scheme of things,
[00:37:57.330]to be able to use high throughput phenotyping data
[00:38:00.608]to get at some of these parameter type traits
[00:38:03.380]that then we can also replace with genomic prediction
[00:38:08.232]to feed into the process-based models
[00:38:10.380]to get at our end results, which would be like yield.
[00:38:13.620]Okay, we're at the point now where I wanna talk a little bit
[00:38:16.110]about this like current-to-future work.
[00:38:19.110]That doesn't mean we haven't done it yet,
[00:38:20.250]it just means it's not a complete story,
[00:38:22.260]but it's getting there.
[00:38:23.725]And then mention some educational partnership opportunities.
[00:38:28.650]So I mentioned that we had collected this dataset
[00:38:30.930]in the Wisconsin Diversity Panel.
[00:38:33.060]We also collected similar datasets
[00:38:34.860]for our Genomes to Fields sites.
[00:38:37.350]And you guys are one of the participating sites.
[00:38:40.260]I think Jinliang manages it here, if that's correct. No?
[00:38:44.850]James. Oh James.
[00:38:47.070]I didn't know if it was one of you or both, or anyway.
[00:38:50.940]There are several participating universities.
[00:38:52.920]I don't know what the current number is.
[00:38:54.570]This is the 2019 that I'm showing here.
[00:38:57.138]But we painstakingly measured thousands of plants
[00:39:01.345]with this idea
[00:39:02.610]that if you could collect informed and targeted phenotypes
[00:39:06.390]and then train your genomic prediction models
[00:39:08.160]on those key traits
[00:39:09.030]that aren't as responsive to the environment,
[00:39:11.010]parameterize the crop growth model
[00:39:12.252]using the predicted parameters,
[00:39:14.460]which means you could also include genotypes
[00:39:17.910]you haven't created yet,
[00:39:20.010]then use your process-based model
[00:39:22.320]to integrate these other components,
[00:39:24.300]that would let you basically predict unobserved genotypes
[00:39:27.540]into unobserved environments,
[00:39:29.040]which is what we're really excited about.
[00:39:31.014]And this is something that is in some ways being done
[00:39:39.780]in industry currently,
[00:39:41.220]though not really the way that we're proposing it.
[00:39:44.580]'Cause I think that we're kind of doing it in a crazy way.
[00:39:49.620]I've never been one to be scared by a lot of work.
[00:39:52.440]So we measured a lot of traits. (chuckles)
[00:39:56.880]We took leaf appearance rates
[00:39:59.040]based on taking counts of the leaves
[00:40:01.350]at different timepoints across the season,
[00:40:03.510]and then the traits that I mentioned before,
[00:40:06.085]as well as UAV based imagery, leaf area,
[00:40:10.770]and then we derived this power function
[00:40:12.510]based on those measurements.
[00:40:13.740]We didn't bother measuring the area of all the leaves;
[00:40:15.540]it was just the one.
[00:40:18.233]And one of the things,
[00:40:19.350]one of the low hanging fruits that we asked,
[00:40:21.150]questions we asked right away
[00:40:23.670]was if we just take a standard vegetative index,
[00:40:26.203]what can we capture?
[00:40:28.680]And what we found is that early in the season,
[00:40:32.910]NDVI does a good job at capturing things
[00:40:34.770]like leaf area index and stand count,
[00:40:36.600]but later in the season, down here,
[00:40:39.540]it's pretty unique information.
[00:40:40.890]Across all these traits we measure,
[00:40:42.030]we weren't really getting at what was happening
[00:40:44.460]with the reflectance,
[00:40:45.293]which makes some sense if you think about it,
[00:40:47.910]because we were measuring mostly architecture
[00:40:49.371]and not biochemical properties of the leaves.
[00:40:54.000]So showing this another way,
[00:40:56.820]we took all of our measured and modeled traits
[00:41:00.240]to try and predict in NDVI.
[00:41:02.670]And as you would expect,
[00:41:03.503]we had higher accuracy earlier in the season,
[00:41:05.610]and then later it dropped off.
[00:41:08.220]We could, however, get a pretty good estimate
[00:41:11.460]of the leaf number.
[00:41:12.750]These are pretty high R squared values.
[00:41:14.697]And this is across two different years, so 2020, 2021.
[00:41:19.200]That was encouraging
[00:41:20.100]because it's a really surprisingly painful trait to measure,
[00:41:25.080]because you're not just counting leaves one time,
[00:41:26.940]you actually have to label the leaves as they come out,
[00:41:29.820]because then the lower ones will fall off, right?
[00:41:31.650]So Sharpies on leaves.
[00:41:35.970]A fun thing that happened by chance
[00:41:41.163]was that in Michigan we had a couple of field seasons
[00:41:44.775]that were really, really different from each other.
[00:41:47.850]And for those of you familiar
[00:41:49.350]with the Genomes to Fields collaborative,
[00:41:52.537]every two years the germplasm changes.
[00:41:55.080]But this was a set of two years
[00:41:56.583]where it was the same varieties
[00:41:58.167]replicated in our same location,
[00:42:00.240]but it was just very different conditions.
[00:42:03.960]And I'm showing here two different traits.
[00:42:06.270]So this is plant height,
[00:42:07.950]and for plant height we had very tall plants in 2020.
[00:42:10.920]They were shorter in 2021.
[00:42:12.270]And then for yield,
[00:42:13.680]we had lower yield here and higher yield here.
[00:42:17.340]Now, it's kind of conventional wisdom
[00:42:20.670]that when you have taller plants you have higher yield.
[00:42:23.396]We saw the opposite, but there was good reason for it.
[00:42:27.603](chuckles) Good reason for it.
[00:42:29.490]So this is what was happening
[00:42:31.020]with these two growing seasons.
[00:42:32.610]2020 is in red, 2021 is in blue.
[00:42:35.804]In 2020, we had an initially warmer season with some rain,
[00:42:41.070]so things germinated quickly and progressed much faster.
[00:42:44.130]But then at the end of the season
[00:42:45.780]it dropped down and got cold
[00:42:46.920]and didn't rain for a long time.
[00:42:48.300]So the plants thought they were off to a great start,
[00:42:50.310]and then it got worse.
[00:42:51.450]For 2021, it was a cooler, slower, drier start,
[00:42:55.350]so the germination happened later,
[00:42:57.180]things progressed through phenology a little bit later,
[00:43:00.990]but then it was a very long extension
[00:43:03.000]at the end of the season
[00:43:03.840]where it didn't get cold for quite a while,
[00:43:05.670]so you had a lot more growing degree days
[00:43:06.990]accumulating at the end.
[00:43:08.460]What this looks like if you look at the NDVI
[00:43:10.464]is that you see 2020 in green here.
[00:43:13.451]It raises up fast, but then it levels off fast.
[00:43:16.469]2021, a little bit slower,
[00:43:18.570]but then it's this long extended growing season.
[00:43:22.290]So Zhongjie decided that maybe what he could do
[00:43:29.340]was to look at the area under the curve
[00:43:31.848]from the timepoint of silking, flowering time,
[00:43:35.010]he used silking as the cutoff there,
[00:43:38.003]up until where it dropped off, right,
[00:43:40.890]and then used that as a trait.
[00:43:42.750]It turned out
[00:43:43.583]you don't actually have to use the area under the curve.
[00:43:45.030]You can just use the growing degree days.
[00:43:47.010]That's all that matters.
[00:43:47.956]And he also adjusted this at one point
[00:43:50.640]so that we weren't dependent on the value itself
[00:43:52.770]but rather just a percent drop.
[00:43:55.710]But he made some, this is literally just a scatterplot.
[00:44:01.860]There's no fancy modeling. It is a scatterplot of this.
[00:44:04.440]He calls it NDVI-adjusted GDD versus yield.
[00:44:08.370]And so within a year
[00:44:10.230]it may or may not correlate really well,
[00:44:12.360]and across years it correlates really well.
[00:44:13.740]So you get a really pretty accurate sense
[00:44:15.690]of the environmental mean based on this trait.
[00:44:18.987]And I said, that's pretty cool,
[00:44:20.793]but, you know, you're looking at two years
[00:44:24.150]with the exact same genotypes in the exact same location.
[00:44:27.990]Yes, it was different environments
[00:44:29.340]but, you know, it was the same data types,
[00:44:32.880]the same drone was used, same sensor, everything.
[00:44:36.450]So he added another year of the Michigan data
[00:44:38.520]from the prior year, from 2019,
[00:44:40.426]so this would add a different set of genotypes.
[00:44:45.090]And it was still, you know, fairly accurate
[00:44:48.480]in its prediction.
[00:44:49.560]So this is where he was showing the area approach
[00:44:53.100]versus just looking at the GDD,
[00:44:54.507]and you don't actually see much of any drop
[00:44:57.450]in the predictive ability.
[00:45:01.080]So then again, I mean, okay, yeah, this is cool,
[00:45:03.000]but it's still just one location.
[00:45:04.200]So we contacted some friends at Wisconsin,
[00:45:06.810]so thank you to Jose Varela and Natalia de Leon
[00:45:09.780]and Shawn Kaeppler, Nathan Miller, Edgar Spalding.
[00:45:11.730]They let us use some of their data.
[00:45:12.870]They had used a different type of drone,
[00:45:14.610]a different type of sensor.
[00:45:15.900]They had processed it completely differently,
[00:45:18.090]but we were able to use the data and kind of adjust it
[00:45:21.030]based on canopy cover,
[00:45:22.075]and then it plotted beautifully onto this graph as well.
[00:45:26.220]So that's kind of fun.
[00:45:27.960]The other thing that he did
[00:45:28.860]was he used a genome-wide mapping approach
[00:45:33.450]on both the NDVI as well as some of the ground traits
[00:45:37.230]across different years.
[00:45:38.489]So this is an example of stand count being identifying
[00:45:42.900]some of the same genes that we see for NDVI.
[00:45:45.750]And then same story later in the season for root lodging,
[00:45:50.280]we see the same thing.
[00:45:51.840]So I think I'm gonna kind of wrap it up at this point,
[00:45:55.950]'cause it's getting a little late,
[00:45:58.440]but we're excited about the idea
[00:46:01.770]of being able to use some of these complex data types
[00:46:04.418]to hopefully improve our predictions
[00:46:06.417]into different environments.
[00:46:08.280]But I want to tease a preprint from James's group
[00:46:12.191]that is using gene expression data from Nebraska
[00:46:18.377]and applying it to both Nebraska and Michigan data types,
[00:46:21.390]which is pretty cool.
[00:46:22.770]So I'd encourage you to check out their preprint.
[00:46:26.742]And then to end this,
[00:46:29.010]I wanna give you a little bit of a teaser I guess,
[00:46:31.980]or maybe it's more of an advertisement.
[00:46:33.633]I teach a few different courses at Michigan State,
[00:46:36.390]but one of the ones that I've enjoyed teaching a lot
[00:46:38.424]is this new course
[00:46:39.810]that we call Frontiers in Computational and Plant Sciences.
[00:46:43.590]And this is a second semester course
[00:46:45.913]of a certificate program we created called,
[00:46:50.467]I think, it's just called a certificate
[00:46:51.870]in computational plant sciences.
[00:46:53.670]So the first course teaches students the basics
[00:46:58.200]of both coding and plants,
[00:47:01.500]because we have students coming in from computer science
[00:47:03.300]and students coming in from plant science,
[00:47:04.860]and so we need to teach them kind of both things.
[00:47:06.900]If you're interested in learning some intro Python
[00:47:09.900]and intro to thinking about plants,
[00:47:12.396]the first semester professors,
[00:47:15.030]Dan Chitwood and Bob VanBuren,
[00:47:16.890]have created a really nice curriculum
[00:47:18.930]called Plants and Python.
[00:47:20.610]You can get that online. Search for it.
[00:47:22.530]It's freely available.
[00:47:23.670]It's available in English and Spanish. So yeah, cool.
[00:47:28.050]I know. They've done a much better job than I have.
[00:47:29.970]So my course is the second semester,
[00:47:32.190]and what we focus on is data analysis.
[00:47:34.830]I throw data sets at students,
[00:47:36.327]and then teach them how to ask questions and get answers.
[00:47:40.950]And one of the fun things that we've been doing
[00:47:42.570]is partnering with industry and other experts
[00:47:45.862]to think about these problems.
[00:47:47.730]So I'm showing you some of the phenomics ones.
[00:47:49.380]We do all kinds of computational projects,
[00:47:52.050]including bioinformatics,
[00:47:53.010]but this is more where my expertise is.
[00:47:56.160]So this was an example of a Syngenta sweet corn breeder
[00:47:58.650]who had images of ears, and one of her questions
[00:48:01.170]was whether we could calculate the blank fill.
[00:48:04.950]So blank fill is where you have an ear like this
[00:48:07.440]where there's places that you don't have kernels,
[00:48:10.170]you just have the cob.
[00:48:11.340]And so the students were able
[00:48:13.200]to create two different thresholds
[00:48:14.910]where they figured out, okay, here's the area of the ears,
[00:48:18.150]here's the area of the kernels,
[00:48:19.380]you subtract the two, you get the blank fill.
[00:48:22.650]This is grad students.
[00:48:23.786]This looks like a fairly straightforward, easy project,
[00:48:26.773]but this is one of three modules we had that semester,
[00:48:30.930]so a third of a semester,
[00:48:32.010]and none of these students
[00:48:33.060]had ever done any image analysis before.
[00:48:35.130]So it's really pretty great
[00:48:37.170]that they're able to, you know, work that quickly
[00:48:39.090]to get some pretty impactful progress.
[00:48:42.660]I wanna give one more example, because it's fun to me,
[00:48:47.370]but also I think educational for all of us here.
[00:48:52.350]This is an example from a Weaver Popcorn field trial,
[00:48:54.930]where we had driven a robot through and taken LiDAR data
[00:48:58.230]and then looked at different plant traits,
[00:49:00.330]and then this is potatoes looking at late blight disease.
[00:49:03.990]So what could we calculate, you know,
[00:49:06.150]which ones have disease and which ones don't
[00:49:07.830]based on aerial imagery.
[00:49:09.480]But the reason I like to bring up these two examples
[00:49:11.619]is because the first time the students worked with the data,
[00:49:14.998]they came back and said,
[00:49:17.047]"Okay, here's what we're seeing. What is happening?"
[00:49:20.940]And so in this case they're like, "Where is the corn?
[00:49:23.790]We know there's supposed to be corn in these plots,
[00:49:25.410]but this just doesn't make any sense."
[00:49:27.111]But what was happening is this robot with a laser sensor,
[00:49:30.930]anytime the laser was making it
[00:49:32.340]all the way through the canopy and hitting the sky,
[00:49:35.160]you got these returns that were basically infinity.
[00:49:37.680]So as soon as they filtered all those out,
[00:49:39.686]this is where all the corn was, right?
[00:49:41.790]That's where the canopy is,
[00:49:43.140]all in that blob down at the bottom.
[00:49:44.910]Similarly, here.
[00:49:46.620]They showed me this graph and they said,
[00:49:47.737]"Well, we don't know.
[00:49:48.660]Like these different plots have different values,
[00:49:50.310]but we don't know what's going on."
[00:49:51.750]And I said, "That's not vegetative tissue. You have soil."
[00:49:55.770]And I don't know why you have soil,
[00:49:57.600]because I was the one that had created the shape files
[00:49:59.580]and pulled out the data, so I was very concerned.
[00:50:01.809]But that was how we discovered that our drone imagery
[00:50:05.968]between the RGB/multispectral and the hyperspectral data
[00:50:11.220]had about a one meter diagonal shift.
[00:50:13.590]So wherever we thought we were extracting plant data,
[00:50:15.960]we were actually just pulling out soil part of the time.
[00:50:20.790]That's been a real challenge to work with.
[00:50:22.590]But this to me is a really useful process
[00:50:27.120]for graduate students to work with real crazy data and learn
[00:50:32.490]how do you go about troubleshooting and figuring out
[00:50:34.020]some of these like complex questions and problems.
[00:50:36.930]So I'm gonna leave it at that.
[00:50:38.580]I have a quick slide of funding acknowledgements to give,
[00:50:41.783]but then I would be happy to take any questions.
[00:50:44.910]And again, thank you so much for inviting me.
[00:50:49.260]All right, questions for Professor Thompson.
[00:50:54.360]No one has anything immediately.
[00:50:55.920]So one question I had,
[00:50:57.055]I've seen that figure with all the lines before.
[00:50:59.400]I don't think I've ever actually heard the explanation
[00:51:02.070]for what was the reasoning behind that.
[00:51:04.770]Are there specific traits that would be really important
[00:51:07.890]for parameterizing crop growth models
[00:51:09.510]that are still really hard to pull out of phenotyping?
[00:51:12.450]I mean, what's the sort of last frontier
[00:51:15.180]in terms of what we need to be measuring that we can't yet?
[00:51:18.600]What the modelers would tell you
[00:51:19.910]is they would love to have LAI
[00:51:23.130]and biomass accumulation over time,
[00:51:26.070]as those wouldn't be the parameters, they'd be the outputs
[00:51:31.230]that they'd be measuring metrics against.
[00:51:34.560]So anything that you can do a repeated measure over time
[00:51:37.050]and kind of make sure that you're on track
[00:51:39.090]over the course of the season,
[00:51:40.410]I think would be their wishlist.
[00:51:42.660]The other ones that are really hard for us,
[00:51:44.264]we don't really measure some of the timing.
[00:51:49.230]So flowering time again is kind of an output trait
[00:51:51.570]in most of the models, but it's like timing
[00:51:54.000]between different transitions is a challenge.
[00:51:58.500]And then a lot of the efficiencies,
[00:51:59.640]nitrogen use efficiency, water use efficiency,
[00:52:01.620]radiation use efficiency.
[00:52:02.771]Those would be awesome.
[00:52:04.980]And that is part of what we're working on a little bit
[00:52:08.280]with some of our sort of more physiology based projects.
[00:52:12.120]But those are ways you can estimate things
[00:52:14.460]like radiation use efficiency,
[00:52:15.540]but measuring it is a little bit painful.
[00:52:33.090]That was a really great talk,
[00:52:34.260]and it's super, I don't know,
[00:52:36.450]it makes me think about all these
[00:52:37.801]like in silico farming or something, you know.
[00:52:39.150]You could do all of this stuff,
[00:52:41.010]model how everything would be.
[00:52:42.538]But with that, how far away do you think we are
[00:52:45.840]from like acquiring all of these like, you know,
[00:52:48.150]soil parameters, weather data, everything,
[00:52:50.130]and then just growing in a computer first
[00:52:53.370]to see how it will look,
[00:52:55.106]and then- Indus Share does that,
[00:52:56.280]and they do it in particular years.
[00:52:58.200]Yeah, and so I've seen talks.
[00:52:59.970]They'll put up a figure
[00:53:02.400]that shows like county level yields
[00:53:04.740]of what this hypothetical genotype,
[00:53:06.791]how it would've performed in the drought of 2012 and show...
[00:53:10.500]So then they can test out in silico new sort of gene edits
[00:53:14.520]or like new varieties to create,
[00:53:16.413]and not only use that to make decisions
[00:53:20.040]about what breeding lines to go after,
[00:53:22.385]but also placement, right?
[00:53:24.690]So which variety should they market where?
[00:53:28.984]So yeah, totally.
[00:53:32.760]I think there are a number of companies
[00:53:34.470]that are already doing things like that.
[00:53:36.630]Although, like I said, not quite how we're proposing it,
[00:53:41.190]but adjacent to that,
[00:53:47.370]I can (indistinct)
[00:53:51.750]Thank you. Hi, Addie.
[00:53:53.965]Great talk.
[00:53:54.857]I have one question that it was truly amazing
[00:53:58.153]when we saw the differences between the plant height
[00:54:01.312]and the yield for Michigan in different years,
[00:54:05.820]that we explain the differences in yield
[00:54:08.970]because all the environmental temperatures or rainfall.
[00:54:14.700]So when we are predicting
[00:54:17.580]and we are using the environmental factors,
[00:54:20.500]are we including all these factors
[00:54:23.610]for the lifecycle of the plan,
[00:54:26.138]or are we just like selecting the most important timepoint
[00:54:32.578]and then just input that into the models?
[00:54:37.710]You can do it either way.
[00:54:39.810]So there's a lot of different ways
[00:54:41.550]to go about incorporating environmental variance,
[00:54:43.500]and you can use environmental covariates
[00:54:45.448]either extracted as particular timepoints.
[00:54:50.460]The challenge there becomes that
[00:54:51.881]if you have a drought from June 6th to June 12th,
[00:54:56.700]it could be a different growth stage
[00:54:58.680]for different varieties.
[00:54:59.513]And so if you're using that cofactor as just a number,
[00:55:03.390]it's not going to reflect that interaction.
[00:55:06.750]So a couple of ways you can go about that.
[00:55:08.910]You can create windows based on flowering time
[00:55:11.640]as a metric of your maturity.
[00:55:13.350]But you can also do what we did in a paper
[00:55:18.000]that came out recently, "Nature Communications."
[00:55:20.940]Gustavo de los Campos is the corresponding author.
[00:55:23.760]I forget who the first author is. I'll think of it later.
[00:55:27.600]But it's a repackaging of all the years so far
[00:55:33.330]of the Genomes to Fields data.
[00:55:34.800]So if you're working with Genomes to Fields data,
[00:55:36.120]I would encourage you to check that out,
[00:55:37.320]because we not only like cleaned up and reprocessed
[00:55:41.790]and repackaged some of the data,
[00:55:42.840]but also extracted environmental covariates
[00:55:45.060]from a crop growth model.
[00:55:46.500]So we ran each location through APSIM,
[00:55:48.540]and then got out the covariates that were important
[00:55:50.820]from that physiological model.
[00:55:52.080]So it's a different way
[00:55:53.160]to go about this same question, for sure.
[00:55:57.690]So yeah, the answer is you could do it any way you want.
[00:56:02.308]How successful it's gonna be is gonna depend, you know?
[00:56:07.380]For us, what we're getting at mostly
[00:56:09.030]is just how long was the grain fill period, right, on that.
[00:56:12.690]That was the main thing there.
[00:56:14.400]But you could argue
[00:56:16.320]that that's a very Michigan-centric issue,
[00:56:18.750]because a lot of other places
[00:56:19.980]don't have such a short growing season.
[00:56:22.260]So maybe you guys have lots of nice warm, sunny days
[00:56:25.130]and it never is gonna be an issue that you don't have
[00:56:27.840]a really, really like full grain fill period.
[00:56:31.830]Thank you. Great answer.
[00:56:33.420]Lopez-Cruz. Thank you. Lopez-Cruz.
[00:56:40.110]Any other questions?
[00:56:44.772]I could picture him but not his name. Sorry.
[00:56:52.410]So this is thinking about
[00:56:55.863]that industry is already doing this predictive modeling.
[00:57:00.720]And obviously industry breeding programs
[00:57:04.020]have a lot of support research groups
[00:57:06.450]that allow that to happen.
[00:57:08.007]How do you think we can implement some of this
[00:57:12.363]in actual, you know, public breeding programs?
[00:57:17.580]Because I think one of the challenges
[00:57:19.359]is processing all that data in the timeline
[00:57:24.705]that you can actually make your selection decisions
[00:57:27.854]and get those back into the ground.
[00:57:32.160]And for me in winter wheat,
[00:57:35.370]my turnaround time is six to eight weeks.
[00:57:39.630]So (chuckles) I'd love, I would absolutely love
[00:57:43.530]to be able to do more of that,
[00:57:45.120]but my turnaround time is so short
[00:57:48.570]that as a public plant breeder,
[00:57:52.868]where it's just my lab running this program,
[00:57:56.100]that's a real challenge.
[00:57:57.510]How do you see...?
[00:57:58.530]You know, is there a way to solve that problem
[00:58:01.560]such that we might be able to catch up a little bit
[00:58:04.890]with industry?
[00:58:08.250]I think my first answer would be
[00:58:14.850]that you can get shockingly far with just RGB data.
[00:58:20.760]So I showed a lot of like LiDAR and multispectral,
[00:58:23.040]and the multispec is not a significant lift
[00:58:26.310]beyond RGB, really,
[00:58:28.260]but you can get really, really pretty far with just that,
[00:58:32.100]and that, once you have the pipelines in place,
[00:58:34.770]can be processed very quickly.
[00:58:36.810]Something that I benefit from a lot
[00:58:39.000]that I think would be a really nice
[00:58:40.883]wishful thinking future situation for people
[00:58:44.100]is that we actually have a fee for service drone group
[00:58:49.140]on campus.
[00:58:49.973]So I work with a remote sensing
[00:58:50.910]geospatial information system group.
[00:58:52.710]They do the flight planning
[00:58:53.790]and the ground control points and the flights.
[00:58:55.680]They preprocess all the data, so they give us orthophotos.
[00:58:58.560]All we have to do is draw the plots and extract the data,
[00:59:00.870]and then we run with it.
[00:59:01.703]So if everybody had a facility like that, that'd be amazing.
[00:59:05.640]I also drive six hours to our field.
[00:59:09.416]Yeah, if you pay for it, yeah, that's a problem.
[00:59:11.310]So it'd have to be worth the trade off, right?
[00:59:14.700]Alternatively you can have a really well-trained tech.
[00:59:16.710]So it's like anything.
[00:59:18.240]It's gonna be a balance between cost
[00:59:20.220]and what benefit you see.
[00:59:22.140]And maybe there is a benefit, maybe there's not.
[00:59:24.693]Maybe you could get pretty far
[00:59:27.900]by just taking some individual images
[00:59:30.450]and not having to process a lot of stuff,
[00:59:32.040]just pulling out some information.
[00:59:33.330]I don't know.
[00:59:35.806]But let's see.
[00:59:41.640]I think the wheat breeding program at Michigan State
[00:59:45.660]uses RGB and multispectral data from drones
[00:59:50.750]and has found it sufficiently useful
[00:59:53.684]that that is part of their program.
[00:59:57.840]It's probably a different situation,
[00:59:59.610]because their farm is about 20 minutes south of campus.
[01:00:03.630]So if you're six hours away, that's a hard,
[01:00:08.580]but you'd also have to calculate
[01:00:10.501]how much time are you spending taking the traits manually
[01:00:13.912]and how much is your human capacity worth,
[01:00:18.090]and do you have sufficient undergrads
[01:00:19.560]to do that work in the first place?
[01:00:21.176]So, you know, everything's a trade off.
[01:00:31.110]I think this will be the last question,
[01:00:33.750]but CROPS is providing pizza after this,
[01:00:36.420]so we'll have many more opportunities (indistinct)
[01:00:39.690]Yes, (indistinct)
[01:00:41.670]I'm just thinking,
[01:00:42.737]'cause you mentioned our LemnaTec phenotyping greenhouse.
[01:00:46.080]So is the type of hyperspectral data
[01:00:48.000]that you collect from like a LemnaTec facility
[01:00:50.970]equivalent to what you collect from a drone?
[01:00:52.560]Or do you have to, like,
[01:00:54.180]is one better than the other, or...?
[01:00:56.940]This is a loaded question, sir.
[01:01:00.510]Yes, one is better than the other,
[01:01:02.430]but it depends on your question.
[01:01:03.780]So for whatever question you're asking,
[01:01:05.880]one of those is definitely a better choice.
[01:01:09.480]But without knowing your question,
[01:01:10.440]I can't tell you which one.
[01:01:11.850]So if you have more basic science questions,
[01:01:17.160]if you're looking at the impact of one gene,
[01:01:19.320]if you're looking at, so like specific questions,
[01:01:22.983]I think the greenhouse is the way it go.
[01:01:25.320]If you are instead looking
[01:01:27.510]at an agricultural production setting
[01:01:30.510]and you're asking questions about, you know,
[01:01:33.115]comparing how things grow in a field
[01:01:35.999]in the context of a plant canopy,
[01:01:39.960]I think that's more relevant in the field.
[01:01:44.070]And that's not to say
[01:01:44.903]that some of those greenhouses don't get close
[01:01:46.740]to replicating field conditions,
[01:01:47.940]but it's always only gonna be close, not all the way there.
[01:01:56.677]All right, now let's thank the speaker.
[01:01:58.459](attendees applauding)

The screen size you are trying to search captions on is too small!

You can always jump over to MediaHub and check it out there.

Comments

0 Comments

Exploring Maize Resilience Through Genetics, Phenomics, and Canopy Architecture

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments