Intercropping breeding requires strategic adjustments to existing monoculture breeding pipelines rather than entirely new systems, as genetic variation for intercrop performance exists but selecting based on monoculture alone misses 50-60% of potential genetic gain; general mixing ability (additive effects) drives intercrop interactions while specific mixing ability is often not significant, and the key challenge is addressing yield scale asymmetry when modeling productivity metrics across different crop species.
Install our extension to search inside any video instantly.
Dr. Gideon Oyebode: Breeding Crops for Intercropping: Quantitative Genetic Insights
Added:[ Music ] >> Good morning, good afternoon.
I hope you guys can hear me well.
Thank you for participating in this new series.
The International Plant Breeding Seminar, the 11th Series.
So we've been bringing this to you guys for more than five years now.
And this particular series is focused on African Plant Breeders.
The calendar is there, and I think if you register, you should have it already in your calendar, okay?
So we start today and we finish on May 7th.
Again, this seminar is hosted by North Carolina State University here from Raleigh.
We have the pleasure of opening up this series with Gideon Oyebode, who is graduating in August 2026 at Auburn University with a PhD in Plant Breeding and a Minor in Statistics.
His research has focused on quantitative genetic frameworks for breeding crops in intercropping systems, both in tropical and temperate agroecosystems.
Integrating genomics, high-throughput phenomics data in order to develop predictive models that improve selection accuracy in species that are mixed in cropping.
Okay? Previously to his career as a student here in Auburn, he has worked a decade in cassava and cowpea breeding at IITA in Nigeria.
Gideon, it's a pleasure to have you with us.
I'm going to stop sharing here and give you the floor for your presentation.
>> All right.
Trying to get set up here.
All right.
Thank you, Dr. Carlos.
Can you all see my screen really good?
>> Yes, just put it on presentation mode and you're set to go.
>> Okay. All right.
Okay. I think it's fine now, right?
All right.
Can you all hear me well and see my screen?
>> The closed caption is showing again if you want to turn it off.
>> Okay.
>> Yeah, we see your screen and we hear you well.
>> All right.
Perfect. Once again, thank you, Dr. Carlos, for organizing this meeting and having my team come to share what we've been learning from my experience.
In working with the genetics that happens within intercropping trial.
It has been an exciting experience.
And today I will be sharing our learnings from tropical intercropping systems with cassava and cowpea.
And also do a lightning talk through what we are learning from clover and oats intercrop breeding here in the U.S. So to start on a little bit pessimistic note, we have not been able to meet the, you know, heroic goal of ending hunger.
And as we can see currently, there are still over 316 million people in Africa that still are in the range of severe food insecurity.
However, despite these challenges, through the effort of basically plant breeding and then works from agronomy and other sciences.
We've been able to increase the amount of food we produce, looking at the record of major staples.
However, these have come, you know, at a cost to the environment.
And activities related to agriculture is responsible for most of, you know, the environmental pollution we see.
For instance, for eutrophication, we account for over 78% of it.
So this, therefore, means we, as plant breeders, have to redefine how we make selection for our crops.
Historically, and even currently, plant breeding is more in the lower quadrant, where it's corporate based and trade based.
And that is why we select mostly for monocultures.
But considering the challenges that we face, we realize that we have to make a transition towards a kind of systems breeding.
Where we consider, you know, the community in which we are breeding for and also the ecosystem services.
And then it's issues like this that motivates us to tend to study intercropping as an alternative way.
To kind of resolve some of these challenges that are related to how agriculture is being done.
And our major focus in this effort is intercropping, which is the cultivation of two or more crops with overlap in space or time.
And then if done well, it can be a sustainable solution to agricultural intensification.
Because from my first few slides, the conclusion from it is that we have to produce more food for a growing human population in a sustainable way.
And agriculture being done in intercropping has been known to be productive and stable.
Performance from crop in intercrop is also known to be of high quality.
And then it's well recorded that pest and diseases are somewhat well regulated in intercropping system.
And then when you look at the last three point here, intercropping provides better ecosystem services when you compare it to monocultural counterparts.
Now, why does intercropping works?
Intercropping works because of ecological reasons such as overyielding.
So in my talk, I would like to provide some basic definitions as I move on.
Because I'm aware this might not be a popular field of research to all breeders.
So overyielding in intercrop is when the intercrop performs better than the average component in a species mixture.
And then a way better version of it is what is known as transgressive overyielding.
That is when we select crops to perform better when they are intercropped than when they are grown in monoculture.
It might sound difficult, but it's possible.
And the factors that drive monoculture success or failure is being summarized by ecologists into the so-called four C's.
Where you have compensation, complementarity and cooperation as factors that promote performance in intercropping.
However, we have the fourth one on top, which is competition.
Which is the main detriment to productivity in intercropping systems.
And as plant breeders focusing in this area of intercropping, our main goal is to see how we can use selection to reduce or eliminate the effect of competition.
Now, how do you breed for intercropping?
There are generally basic concepts that are used in intercrop breeding, and it's very synonymous to what's being done in like maize breeding.
Major ones that we use are, for instance, the general mixing ability and the specific mixing ability.
Which have similar definitions to what like, for instance, corn breeders use.
So here, general mixing ability, you know, describes the overall performance of a clover across many combinations.
So it's an indication of additive effects of a clone and how it performs overall when you mix it with a lot of different varieties.
And then the flip side of it is the specific mixing ability where performance is superior with a unique set of companions.
So here I'm using clover and oats as my illustration, which is one of the case studies I'll be showing down the line.
So modeling GMA and SMA, you want to fit a model where, on your Y vector axis, you need the plot productivity metrics.
And in the case of clover and oats, the metrics is the biomass.
So you need the biomass of clovers and oats from a plot and then you plug that into the model.
So the clover main effect is your GMA and the oats main effect is your oat GMA.
And then the interaction between the clovers and oats is your SMA.
Now, there are also very important variances such as the producer effect, also known as the direct genetic effect, depending on the literature you are reading.
So this measures the ability of a clover or a cassava to influence its own yield when placed in an intercrop.
And to model this, you just need the yield of your focal crop in your Y vector for your model.
A flip side of it is the associated effect, which measures the impact of clover on the companion crop.
So this begins to describe what we are trying to achieve when breeding for intercropping.
We are not just selecting for single crop performance.
We are looking at how each of the component species is performing in itself and then how it's affecting its neighbor, you know, that it is being grown together with.
So when we try to optimize these systems of intercropping, we can do it by either selecting for materials with superior general mixing ability.
Or we select for materials with superior producer-associate effect.
There won't be much of models like this, I promise.
So now intercropping is interesting, but it has its challenges.
As we all know, it's lagged so well behind monoculture breeding for reasons I explained in my third slides.
Most history of plant breeding has been trade based and corporate based.
So most of the models that exist, most of the machineries, most of the market, even funding is channeled towards monoculture breeding.
And then we can further dissect the challenges of, you know, breeding and intercropping into biological and genetic complexities, as I showed in the left.
And then challenges that are operational that I showed on the right.
So operational challenges, a huge one is, for instance, you don't have a combined harvester.
That is designed to harvest maize and cowpea or maize and soybean intercrop.
So almost every machine is designed for a monoculture system.
But most importantly for us as plant breeders and biologists or agronomists, the biological limitations are most important for us.
And the most obvious one is that in intercropping, there is a massive combinatorial problem.
So, for instance, plant breeders like to select early on in their selection breeding scheme where there is enough genetic variability.
So consider a system where we want to test 200 clovers with 200 oats.
It means we have 40,000 possible combinations.
It is impossible to test all these combinations, especially when you also consider that you might want to replicate to have a better estimate of error.
Other challenges is, as I showed earlier on, in order to have estimates to make ranking and selection, which you need to run special kind of trials that gives you the data set to fit special kind of models like the GMA and the producer and associate effect model.
And also beyond dealing with the G and the G by E, in intercrop breeding, you also have to deal with G by G. You also have to deal with G by E by management.
And then we looked at applying genomic and phenomic predictions to help us in selection.
And then it's an obvious reason that most of the pipeline and models for genomic and phenomic predictions are developed for monoculture settings.
And these challenges tend to be cyclic.
They affect all around what you try to do in breeding for intercropping.
Now, let's narrow down to what happens in intercropping in Africa.
So there is a paradox.
So in reality, the number of farmers that practice intercropping is about 70 to 83%, depending on the source you are referencing.
And then the literature always shows a positive land equivalent ratio.
And this is encouraging because it shows that it improves resilience under stress.
However, there is a contradiction because despite the huge success reported at systems level.
When you look at what is happening biologically, you realize that cassava yield is being penalized.
Or if it is a cassava-maize intercrop, maize yield is being penalized.
So it means this high system productivity that we read in most journals doesn't equate to high crop performance.
To further illustrate that, one of the case studies I'll be showing us is our study with cassava and cowpea intercropping.
Where we selected 120 clones from the NextGen cassava breed base.
And what the panel at the bottom shows is that you can see that almost every one of the 120 cassava clones that we tested showed a land equivalent ratio of greater than one.
But just 8% of those 120 cassava clones are actually showing some form of transgressive yielding or overyielding.
And then when you look at the panel up here is a correlation of BLUPs of the clones from intercrop and then the BLUPs from monoculture.
We can also see here that just a few lines over the threshold where cassava yield is not being dropped.
While all those in yellow are having at least a yield penalty of like 5.9kg.
To further summarize that, we can see that just a few clones of the 120 are showing real transgressive overyielding.
While a whole majority, 89% of what we tested, are having a mean yield penalty of 19.3%.
So what this is telling us is that we have to look beyond the system and then look on how to select to optimize crops to perform better in intercropping.
Now this clearly shows us that of course there is a breeding mismatch.
We are selecting in monoculture and then delivering these varieties for farmers that grow them mostly in intercropping.
And then the other one is a scientific gap in the literature.
To date, tropical intercropping has not yet been genetically characterized.
So you hardly find papers on general and producer mixing ability effects coming from the tropics.
Only a few exceptions with a few papers coming from China.
So the core question that my dissertation now attempted to answer with this project is, considering that we've been selecting in monoculture for the past many decades.
Does this clone selected in monoculture still retain some genetic signals that we can use to select for intercropping?
Another way to put this question is, how much more of change do we need to impose on our current monoculture breeding program to make it make selections for intercropping?
Now I'll dive deep now into the first case studies that I'll be sharing with you.
So this is a story of how we tested this hypothesis in cassava-cowpea intercropping.
On the right panel here is a very beautiful decision chart that was developed in a review paper by Virginia Moore and other co-authors, my advisor included.
And the summary of what they are trying to say is that if a breeder decides to do intercropping breeding, it is more of like a system approach.
Where you start by asking yourself basic and fundamental question and then you move down the line based on whether the answer is yes or no.
As we move down the line in the presentation, we would see that to a greater extent, this is like an oversimplification of the whole system.
But it provides a good way to pull yourself together and then help you to define basic questions that needed to be answered.
So I'm not going to provide answers to all the quadrants.
But we tried to focus to make sure we provide answers to the top left and the bottom left quadrant.
Where we basically asking first, what is the cropping system goals for intercropping in Africa?
And then to understand, what are the constraints of those intercropping systems?
And then as the paper suggests, if it's something that breeding can't solve, you pursue breeding.
If breeding cannot solve it, you hand over the problems to the agronomist.
So enough on that, we can spend a whole time discussing this.
But basically on the left, the questions that we tried to answer.
So a quick rundown on what we conclude to be like, the intercropping goals in Africa.
So first and most important is that African farmers practice intercropping because they want all crops to be productive.
And then they choose what to intercrop, basically, based on food preference.
And as we can see from this review by Adam et al.
Which is actually a meta-analysis, they did an awesome job, I recommend it as a read.
And we can see that most of our intercropping is done with some form of legumes.
Cowpea being the top species being intercropped with, a little with maize.
But the other we have here, a legume with a little bit of intercropping with sorghum.
And they also discovered that cassava plus legume intercropping is the most promising for Africa.
Where we expect temperatures to get warmer and, like, the environments to get drier.
Cassava-legume intercrops tends to show more resilience based on their meta-analysis study.
So this begins to give us a sense of how we should direct our selection when we begin to make models with the data that we get.
So in order to do this, Marnin and I designed the experiment and then we pitched it to Dr. Ismail Rabbi, my former boss while working with NextGen, a huge thanks to him.
And then we were able to set up this little, big, small pilot study.
Where we tested 120 cassava clones from the NextGen breed base.
We can talk more on that later on.
And then we used two cowpea lines as testers because that is one of the options that you can use to reduce the cost of running an intercropping trial.
As I said earlier on, there is a huge combinatorial problem and then you have to decide the best way you can use to manage this.
A little bit on how we set it up is shown in this sketch over here.
We used the kind of additive way of setting up intercrop and then we planted both on the same date to have a maximum intercropping competition on the cassava clones.
So I'm going to show two tables from this.
This table here answers two questions.
The first question it answers is, does cassava clones show exploitable genetic variation for performance in intercropping system?
Because this is important as plant breeders, selection only works if it's based on some kind of genetic variance.
The second question it answers is, do cassava genotypes interact with cropping systems and management?
And we can see clearly here, I will run you through the busy table.
So this area of the table shows performance and metrics for monoculture.
While here we see performance and metrics for intercropping.
And obvious answer to question one, is we see clear evidence that there is indeed a substantial and stable amount of genetic variation for cassava.
In both monoculture and intercropping.
And then it's stable across years and then also stable across, you know, when we combine our dataset, which is good news for us.
And then we see the same trend when we look at fresh fruit weight, harvest index, and then root dry matter content.
Another interesting result here is that heritabilities from intercropping system were quite not too far from what you expect from monoculture.
Which puts us in the range of what is being reported in the literature.
Around 0.5 for instance, for harvest index where we went up as high as 0.75 -- or that is for dry matter content rather.
So another interesting thing we discovered and mind you, this is a single location, two-year study.
It is a pilot, but it provides us a lot of information we can start thinking with.
So we observed that there was no significant genotype by management effect for any of the traits I'm showing across years.
And then when we even combined the dataset together.
So what this means is that the relative ranking and performance of cassava in monoculture versus in intercrop is relatively stable.
However, we begin to see some significant genotype by year interaction.
And then when we go to the bigger model interaction, which is where we looked at genotype by management by year interaction, it wasn't significant.
And so this reflects on the stability or reliability of what we see about the genotype by management interaction.
That the year effect is not actually changing how cassava behaves when you put it in an intercrop or in a monoculture.
So the second table is very interesting because it helps us to answer the question, can monoculture performance reliably predict intercrop performance?
Or does genotype ranking limit selection efficiency?
So, and here we see that the genetic correlation was quite high.
In 2023, when we ran the first analysis, we were like, "Oh no, this is too good to be true."
So we went deeper to like further test, you know, the significance of this value to make sure that they are not statistical artifacts.
So we did likelihood ratio tests and then we ran some Benjamini-Hochberg adjustment just to verify that these values are not, like, inflated.
And on this column, we show the BLUP correlations.
So where we did the correlation of the monoculture BLUP with the intercrop BLUP.
And then that kind of dropped a bit, which is expected.
So while the genetic correlation is explaining the genetic interplay between monoculture and intercropping.
The BLUP correlation is kind of telling you how rankings are happening after we've removed the row column special effect and environmental noise.
So interesting thing about this table also is what we get as selection efficiency at different selection intensities.
Now, earlier on, I made a comment that the decision chart by Virginia Moore kind of gives an oversimplification of what you expect when you dive into intercrop breeding.
And then studying the literature and having this result, I had to recheck again.
Because there is a consensus in the literature that the higher the genetic correlation between the monoculture and the intercrop, you expect to have a higher selection efficiency.
But there are papers by Zimmermann that shows that is not always the case.
In fact, when you look at actual selection efficiency, that is when you select crops in monoculture.
And then try to predict their performance in intercropping.
At best we can get, for instance, for root weight is 44% at 10 selection intensity.
And when you go to 33% selection intensity, it kind of dropped a bit, but it's stable.
Overall, what this is telling us is that we need to look beyond genetic correlation as a metric.
To help us decide and determine whether we want to use monoculture to make predictions for intercropping.
However, these results are interesting.
It's a pilot study and they begin to inspire us to dig deeper and then test with more species and expand to more environment and see what happens to these metrics.
So to further give you a more practical explanation of what we are talking about, thanks so much to Dr. Carlos.
He told me not all my audience will be plant breeders or statisticians.
So in reality, this is what we are talking about.
So the blue column here, each dot is actually each of the 120 clones we tested.
And then for the red dots here, the 120 clones tested in intercrop.
So what we are saying is, you see, we have some clones that they are ranking in monoculture and in intercrop, it's fairly stable, okay?
In monoculture, they were among the top five and then in intercropping, they were among the top five.
And we have some here in the middle that maintains that same relative stable ranking.
And also in ranking the top worst lines, we see that it's also somewhat stable.
However, there are extreme case, which are the lines in red.
Where like, for instance, a line that ranks poorly in monoculture might be ranking high in intercrop.
So one thing this plot figure doesn't shows is direction, which one is higher.
But it kind of gives you the general overview of what we are talking about.
And saying whether rankings remains constant in monoculture versus intercropping.
Because that affects the quality of the decisions we make.
Now, the other thing we tested was, if we can use a few testers to screen cassava clones for intercropping.
Because as I said, there is this huge combinatorial problem in breeding for intercrops.
If you say, okay, you want to test 120 cassava clones with 20 cowpea lines, the number of combinations you have to test explodes massively.
So what we did here was to see if my hypothesis of using two testers that shows constructing competitive ability and feature.
Because studying papers from ecology, we learn that for legumes, part of the features that makes them competitive in intercropping is one, their genetic architecture.
And then secondly, their maturity dates, like how early they flower and all that.
And then these two lines that we selected were one, constructing in terms of plant architecture.
And then secondly, we made sure that they were lines that are popularly grown in the region where we were conducting this experiment to make sure that it's farmer relevant.
And then our result is a bit mimicking the target environments where this kind of innovations are being deployed.
And then right here, we see that the correlation between Oloyin, one of the cowpeas and ITK150-12 is approximately two.
Which is not bad because in essence, it tells you that they are selecting in different directions.
They are giving you different information.
And then when we narrow down to panel B, we see that the amount of stress they impose on cassava also varies.
While 150-12 imposes a penalty of about 16%, Oloyin is imposing about 20%.
So we begin to see that there is actually an effect of the testers we use.
So panel B shows the Jaccard index.
So the Jaccard index is an index that helps you to compare selections from two different models.
So here, consensus shows how these two testers agree in what they select.
And using a strict selection intensity of 10%, which is about 12 clones, they could select about seven clones to be the top yielding clone in intercropping.
And then when we look at the heritability, which is a reflection of genetic signal we get from each of those models.
We see that there is quite some differences as well.
So using testers to screen cassava for intercropping might be a viable idea, which is one of the interesting findings we are excited about in this project.
Now, the next thing I'll quickly run you through is that we discovered that there is a problem of yield scale asymmetry.
That can bias and alter genotype ranking when we fit our general mixing ability model.
Again, to refresh your memory, this is what the general mixing ability model looks like.
So on the Y vector, you need a plot productivity matrix where you have to add together the yield of cassava and cowpea.
Now, here's the problem.
So on panel A here, we show the naive unit distribution in their raw form.
So cassava was recorded in kg, and then the range is from five to 58kg.
Cowpea was recorded in grams, the range was 100 to 899 grams, and then -- I'll rush through this.
So we begin to see the impact on the variance decomposition on panel C.
So if we use the naive raw form, we see that what we'll be getting from our model will just be purely the variance from cowpea for obvious reason.
Because the magnitude is way higher than that of cassava.
We don't see cassava variance, and then we don't see the covariance between cowpea and cassava, which is actually what makes this an intercropping model.
Now, Wright et al suggested in his paper, okay, what you can do is put them on the same unit.
So when we flip cassava, I mean, cowpea yield in grams to kg, you just switch the problem.
This time around now, you are getting just purely cassava variance, and then you are losing the variance from cowpea.
So my good friend, thank God, works in the seed systems back home in Nigeria.
And he was like, "Hey, what if you use the Nairametrics?"
And then we got the real fine price value for each of these crops, and then we converted the yield in grams and in kg from each of the species to the Naira equivalent.
And then we put that now into our model, and we see that we begin to see some signals.
And then we begin to see, especially, some covariance.
But still yet, it's still tilted in the favor of cassava.
To cut the story short, I discovered that using a Z-score where we apply a kind of standardized Z-score to both yield metrics.
Gave a more balanced and a more reasonable, you know, decomposition of the variances.
So using a Z-score, we have about 26% of covariance reported, and then you see the variances between cassava and cowpea is being balanced out somehow.
And then also when you look at the correlations coming from the Z-score.
It's more reasonable compared to the raw form that kind of cancel out the effects of cassava.
So down the line, I will be showing you results from our GMA and our producer-associate effect based on Z-score.
Also, I'm showing you this just to further portray that this is not a mathematical artifact.
It is a real problem in the fact that the productivity metrics you choose to use affects your decision, okay?
So here on the X-axis are the six productivity metrics we tested.
And you can see that the way they ranked the clones for performance, it's different.
There is a little bit of agreement between Naira and Z-score, you know, when you take the first 10 to 15.
But as you go down the line, they begin to disagree.
So how we choose to model this is very important.
And then this was a huge finding for us because in all of the intercropping breeding papers from the temperate regions, they did not report on this kind of problem.
Because they did not have to deal with it, which gives a good appreciation of why we need to conduct these experiments using our crops in our environments.
To make sure that the results we are getting can be useful and applicable to what we are trying to do.
So here I'm showing the producer and associate effect joint modeling quadrant.
And this summarizes how we can begin to make selection considering a whole system's level.
Using this, it shows us that the 120 clones we tested fell into different quadrants.
However, based on what we want for an intercropping clone is a clone that is able to improve its own yield in intercropping.
And also have a positive effect on the companion crop that it's being grown with.
And across traits that we studied, we discovered that there is a good amount of clones that falls into this quadrant.
Now, with these two information, we can now begin to position cassava into the existing intercrop breeding framework.
So this brings cassava to a large extent at par in terms of intercrop breeding.
And Dubey et al did an awesome study where they did a simulation to determine when is the best time to make a switch from monoculture selection to intercrop selection.
Because that is what we envisage.
Dreaming or hoping that we have a brand new intercropping breeding scheme.
It can be far fetched but the idea is in the current monoculture breeding scheme, how do we introduce intercropping?
And based on their simulation studies, you can choose to introduce it at the elite yield stage, at the advanced yield stage, or the preliminary yield trial stage.
And based on the metrics of high genetic correlations that we have, we can place cassava at the top right column where genetic correlation is high.
And then because our heritability is around 0.5 on the average for most traits.
It is advisable to kind of deploy selections for, you know, intercropping probably at best, like, the advanced yield stage.
So it means that you can make selections in monoculture for instance in the cassava breeding scheme, the clonal evaluation monoculture, the PYT monoculture.
But advanced yield trials where in reality you actually have fewer clones, you can now start making a switch to test those clones in intercropping.
So this is an exciting finding for us.
And so to summarize this, running a bit short of time.
We realize that intercropping creates a system advantage but, at crop level, there is a yield penalty.
And then when you read intercropping agronomy papers, you need to read beyond the land equivalent ratio.
And then look for yield penalties because we expect that to happen.
And then we discover that genetic variation does exist in intercropping environment.
However, selecting based on monoculture alone you get to miss 50 to 60% of your genetic gain.
And then general mixing ability is present and significant and specific mixing ability is not significant which is a good thing for us.
It simplifies the breeding of intercropping and seed systems for intercropping a lot.
The big conclusion here is that intercropping breeding does not require a new breeding pipeline.
Existing monoculture breeding pipelines can just be strategically adjusted based on the genetic correlation you have and your broad sense of heritability.
To know when you can start making test in intercropping breeding.
And then our results came from, like, lines that were from advanced yield selections.
Which really makes it something we are very excited about.
However, there are limitations to this we can discuss down the line.
So future directions to this is we'll be exploring genomic predictions.
And also, we are open to collaborations, other CG centers or national research centers that would like to answer similar questions.
Now, I'll quickly move to case study two, which is what we've been learning from clover-oat intercropping.
Now, back to the chart again, the first thing you want to do in breeding for intercropping is to ask yourself, what is the breeding goal?
What is the objective of the farmers, of the producer in doing this intercrops?
And here for clover-oats is visually for cover crops.
So cover crops are grown when cash crops are not growing to improve soil, reduce erosion and manage nutrients.
And then sometimes it's been grazed.
But I'll quickly like to introduce you to the soils of Alabama because we are basically a cover crop breeding lab.
So the soils of Alabama are basically ultisols background.
And what this means is that they have a very high affinity to fix aluminum and iron oxides to phosphorus.
And when that happens, when you do a soil test, for instance, it tells you there is a lot of phosphorus in the soil.
But those phosphorus have been bound in a way that it's not available to plants to use.
And also during the winter periods, we get an appreciable amount of rainfall.
So without cover crops, you get to lose a lot of nitrates and nutrients from your fields, which leads to a lot of environment problem.
But when you have cover crops in the field, for instance, it can reduce nitrate leaching by like 30 to 70% during peak winter season.
So this is a common saying among [inaudible] people.
That without winter cover crops, nutrients leave the field.
With cover crops, they stay in the system.
And then this begins to reflect back to how we manage the environment.
So while having some residual nutrients in your field, for instance, it means the amount of synthetic fertilizer you need to apply in the following year would be lower.
So adoption of cover crops in the U.S. have been on the increase, but it comes with new challenges.
Because when farmers are asked of the reason why they don't adopt or why they fail to adopt, for instance, economic returns comes out as a major reason.
Also when farmers are asked, like what are the top three things farmers talk about?
What are the top three questions they ask?
We see that species mix selection and economic returns rank high.
And when farmers were asked -- okay, some were doing intercropping, but they stopped along the line.
We realized that again, lack of economic return happens to be one of the reason why they stopped using intercropping.
Which reflect back again to the chart I showed about how we envision breeding and selection.
So this makes us to start thinking, how do we redesign cover crop breeding to improve farmer adoption by delivering reliable economic returns alongside ecosystem service?
And that is why we came up with the idea of developing a grazable clover-oat cover crop mixtures.
Now, this is what we are trying to do in this project.
I won't spend much time on it for sake of time, but we can come back to it.
But the general idea is to have a spaced plant for clovers, which is our focal crop here.
We select best materials here, and then we split the seeds to do some kind of a single row mixture trial.
Select the best mixture performance from the mixture trials and go back to the seeds, and then use that to establish the next spaced trial.
So it's more of like a pedigree selection.
So quickly here, we see clearly the combinatorial problems in doing intercrop.
So we wanted to test here 177 clover lines in mixtures with 26 oats.
So it means we have 4,600 combinations to test.
So this will be intractable.
And then the solution to this is the so-called incomplete factorial design.
And here we worked with the option D, where you test a subset of clovers with a subset of oats.
So this way, all clovers are tested and all oats are tested.
And then this gives you the ability to run the models that I showed earlier on.
So the final statistics formulas I show you today.
So up here is the baseline for phenomic modeling.
For clover and oats and SMA main interactions, you have identity.
You have nothing to borrow information from.
But as you begin to use, you know, genomic prediction, you get to have the snip information here.
And then you can build a kernel off of this, which helps you to borrow information from neighboring clovers.
And then with this you can make predictions for clovers you did not test.
And most importantly you can make predictions for a combination of clovers and oats that were not tested.
So what does the genetics look like in this setup?
Similar to what we see in the cassava-cowpea genetic GMA, it's present and significant.
While SMA is either absent or not significant.
And then we see that there is a appreciable number of heritability but lower for obvious reasons but higher early growth rates compared to used traits.
So right now our genomic prediction accuracy sits about like 1.2.
We are going to take it because it's a starting point and then we are currently thinking on how to optimize the pipeline to have a better prediction.
The last thing I want to share with you is how we are applying our phenomics in intercrop breeding.
And the goal here is -- it is very tedious to do what we call botanical separation.
So for this experiment we had 600 plots and then for each of the plots we have to do what Philip is doing right here on panel A. Where we have to separate the clovers from the oats.
So the idea is, can we use phenomic prediction from proximal and aerial imaging to predict the proportion of clovers and proportion of oats?
And using machine learning models, I looked at the data.
And what it is telling us is that at the early stage of the trial, the proximal imaging gives some advantage over UAV phenotyping which we have at the bottom frame line.
But towards the middle of the trial, you don't see a clear advantage.
And then towards the end of the trial, you need to make a switch.
So this panel here explains it better.
So the green is the FarmNG proximal.
So we see that, like, from when the plant starts to grow to a point where the trial is, like, a hundred and, let's say, 18 days.
Then you need to make a switch using UAVs to predict clover proportion in order to have higher prediction accuracy.
So we are excited about these results and we are looking at ways to optimize the system.
Because currently for these systems, we did not do image-based segmentation.
So things like occlusion of the oats over the clover, it's affecting the accuracies that we see.
So the key learning from this is that intercropping performance just like the cassava system is being dominated by the GMA here also.
And then genetic signal is trait-dependent.
We see higher, you know, heritabilities like early growth traits.
Which is good because you can sort of use those early signals in, like, a multi-trait genomic prediction to improve your accuracy, so that is not bad.
And then we see that phenomics can work and then also genomic prediction is also feasible but there is a room for optimizing.
As I said, these are all pilot study that gives us initial ideas of what we are expecting.
So our conclusion so far -- and I said so far intentionally because we are still learning.
Significant genetic variation for intercropping does exist and we should explore it.
But using monoculture to predict intercropping performance gives you, you know, partial genetic gain.
In all of the systems we've studied the general mixing ability, which is the additive framework, is what drives, you know, the interaction we are seeing.
And then phenomics and genomics can support, you know, intercrop breeding.
But right now we see clearly that it is trait and context-dependent and it requires a lot of optimization.
Then finally intercrop requirement requires targeted testing.
So you don't need to, like, totally overhaul the current monoculture selection schemes that we have.
We just need to make some strategic adjustment at some certain point.
At this point I would like to acknowledge all my donors and funding agencies.
Dr. Marnin D. Wolfe, my advisor, have been awesome.
And a huge thanks once again to Dr. Rabbi and his Cassava Breeding Team.
And huge appreciation to Mr. Peter Iluebbey.
He employed me first time ever as an undergraduate intern in 2011 and which was a huge starting point for my career.
I'm ever grateful.
And all the guys that worked on this project.
Huge thanks to collaborators in UF and also to the Wolfe Lab.
At this point I would like to thank you for being an awesome audience and for listening to our story on understanding the genetics of intercropping.
And I'm excited to hear your questions because ideas don't come fully formed but they get built up and develop as you move on in, like, taking actions on what you know and what you can do.
Which is, like, the exact point we are at right now with this thought of line of doing breeding.
So thank you very much, that is it from me.
>> Thank you so much, Gideon.
Well done, good presentation.
We have already a question here.
Let me go to it, from Dinesh.
He's saying, could you go back to slide 13?
>> Mm-hmm.
>> And can you explain the difference between BLUP correlations and genetic correlations, okay?
And he asked, "Do these BLUP include special components from your bivariate model with autoregressive correction or purely gBLUPs?"
>> Okay. So the genetic correlations, you know, as I put some -- what's it called now?
A little bit of details under the table is from a bivariate mixed model in ASReml-R.
Where we fitted the monoculture traits performance and the intercrop performance using the, you know, [inaudible] function.
And then we did the AR, also a special modeling to remove environmental effects and all that.
So what it does, in essence, is it fits the performance of intercrop and monoculture on the diagonal.
And then on the off diagonal you have the performance of the interaction between intercrop and monoculture.
And then it uses these genetic variances to estimate the correlations that you are seeing.
For the BLUP and Pearson correlations, those come from a simple linear mixed model in ASReml-R.
Of course, with some kind of special modeling.
And an extra detail from this was that we use most especially for root weights, we use the number of stands harvested as a covariate to further adjust for the yield.
So these are pure genetic estimates, as they are expected to be, you know, best linear unbiased predictors in the sense of what they are.
So adjusted for, you know, stand yield, adjusted for special row column effect and all that.
So Pearson and Spearman, I showed that because they answer different questions.
And depending on how, you know, you see these metrics, you can look at the table in either way.
Yeah.
>> Okay. Any other questions from the audience?
There's a question here from Simeon.
"I was thinking for the phenomics, the UAV use interchangeable with proximal, just to know if ground truth were adopted for the oat and clover system?"
>> Yes, of course.
There will always be a ground truth because, at the end of the day, what you are doing is prediction.
So you have to have the true observation that you benchmark your predictions against.
So our prediction accuracy is that we show there is actually the correlations between our predicted values.
And then the so-called ground truth, which is the actual dry biomass from each of those plots, yeah.
>> Okay. I don't know if it's a question or an observation that I have.
I mean, it's very good what you're doing, what you have done in IITA and what you're doing here.
Particularly from the fact that you are finding that there's good correlation or decent correlation between what you do in monocrop and what you do in intercropping.
But there are there are exceptions, okay?
>> Mm-hmm.
>> So the question is not so much from the economics of the system.
But from the economics of a breeding program.
If you want to implement and you set it up there, when to implement the intercropping evaluation.
Then later you implemented the least amount of resources used because you narrowed down your genetic base, okay?
>> Yes, sir.
>> I think this is very important because if you don't have restriction of resources, of course you can run two parallel programs.
But that's not the case.
The other question I have is were there any evaluations done of those materials within NextGen on farm and see what the perception of farmers is?
Because what you are getting is from the research perspective, which are the most effective combinations.
But farmers may have a different mindset and farmers may want a particular clone because it produces more biomass that they use it as fuel, in the case of cassava.
Just making something up.
Or in the case of cowpeas, they use it as fodder or something.
I mean, there's a lot of aspects that the farmers may think of in an intercrop situation.
And it's not just yield per se or grain or roots in cassava.
Has there been evaluating in like Tricots or another type of on-farm testing?
>> Yes. So there is currently a Tricot project going on with cassava.
And I also believe with cowpea, but that is not my project.
And I don't have an idea of what is happening there, but I know there is a Tricot project going on with those.
However, trying to respond to your question, in my third slide, you know, I showed you, like, our approach to intercrop breeding is a kind of systems approach.
Where we are not just selecting cassava for the sake of high yield, but we are considering the communities in which these hybrids will be deployed into.
And then that is why, for instance, the way we set up the intercropping system, we use the additive system.
Whereby we maintain the same density between monoculture and intercrop, which is what a farmer would do, right?
The smallholder subsistence farmers doesn't have a precision planter that plants, you know -- >> Right.
>> Yeah. So they kind of crowd everything together.
That is what we did.
And then, as I mentioned, for our choice of cowpea lines that we use as testers.
We made sure that there were cowpea varieties that were popular around the Ibadan axis, okay?
So the general idea, you know, as you said, is just to make sure that our selection environment is mimicking the target environment in which these technologies will be deployed.
But hopefully this gets to be a big and major thing in IITA.
And then we hope to get, like, funding.
And then which would enable us to do a kind of deeper, you know, analysis and evaluation of what is going on around intercropping in Africa.
Nice question, thank you.
>> All right.
Well, if you want to -- let me see this, maybe, no.
Okay, want to stop sharing?
>> Okay.
>> I need to share the screen again.
All right.
So thank you so much for participating in this.
Thank you so much, Gideon, for your presentation.
>> Thank you.
Thank you for having me.
>> And thank you to the audience.
And a special thanks to Brandon, who is our IT help here at North Carolina State University.
There's a lot of work that goes into organizing this seminar series in preparation.
So Brandon is an amazing resource for us.
Next week, at the same time, we'll have Mathieu Ayenan from the World Vegetable Center talking about breeding okra, breeding in the World Vegetable Center in Benin.
I think Mathieu is part of the audience.
So we look forward to your presentation and we look forward to the participation of everybody, okay?
Thank you so much again.
I hope everybody has a wonderful weekend ahead and I will see you next week.
Thank you.
>> Thank you so much.
>> All right.
Goodbye.
>> Bye.
Related Videos
I Found 7 Golden Orb Spider In The River !! Spiny Spider, Weaver orb Spider
insect_geography
1K views•2026-06-16
Your nose is more than a breathing tube...
HealthInSeconds_1
2K views•2026-06-16
Why do marmots always look so dramatic
CodeFauna
3K views•2026-06-16
Your Axolotl Is a Salamander That Never Grew Up
dailywildreports
661 views•2026-06-17
King Vulture: The Colorful King of the Rainforest Skies!
NatureChirps-05
185 views•2026-06-18
The Biggest Lies In The Animal Kingdom!
InfiniteFactssofficial
1144K views•2026-06-15
Humpback Whale, Whale Shark, Great White Shark and Mako Shark Giant Ocean Adventure for Kids
EvieWildTales
5K views•2026-06-18
Thunder Mountain in Juneau, Alaska
Raven-Orix
1K views•2026-06-14
