The K-State soybean breeding program has teamed up with the spectral analysis lab of Dr. Kevin Price, professor of agronomy, to explore ways to increase the efficiency of the soybean breeding line selection process.
The most time-consuming and expensive aspect of our breeding program at K-State is in harvesting the many thousands of early generation lines, weighing the seed, and determining yield. The early screening stage of the breeding process also takes up a lot of acreage.
If we can find a way to separate out 50 percent or more of the very-low-yielding lines without the need to combine harvest and weigh the seed, that would reduce the time and cost of our breeding program considerably.
Spectral analysis is being used in the Agronomy Department at K-State to determine the level of photosynthetic activity of vegetation in many different situations. We decided to work with Dr. Price’s spectral analysis team to try using this new technology in our soybean breeding nursery. The goal was to find out how effective this technology might be in predicting yields, stress tolerance, and disease resistance as a way to eliminate unpromising lines early in the process.
To do this, we used a ground-based spectroradiometer to gather spectral data at various stages of growth, and correlated the results with actual yield data. We have spent the last two years trying to determine exactly what data to collect and how often, and whether any of the spectral regions being measured would have a good correlation to yield. Spectral analysis doesn’t have to be accurate enough to separate lines with a yield difference of just 1 or 2 bushels per acre. If it can separate lines with a yield difference of 5 to 10 bushels, that would be a great help in the preliminary stages of line evaluation.
With financial support from the Kansas Soybean Commission, during the past two years, we have been testing this technology. A ground-based spectroradiometer has been gathering wavelength data in both the visible and infrared spectrums, resulting in thousands of pieces of data on each genotype.
We have a “training” population of different soybean varieties we are using to develop models from the spectral data we have collected to help us predict the phenotype, or performance of a variety. We intentionally selected varieties known to have a large difference in yield potential for this initial testing phase.
The graph above is the result of one model that graduate student Brent Christensen has developed comparing actual yields to yields predicted by the model. On the X axis is estimated yield based only on spectral data using the model. On the Y axis is the actual seed yield at harvest. Each box on the graph represents a Group III variety. Each diamond represents a Group IV variety.
If the model was predicting yields perfectly using the spectral data, all the boxes and diamonds would fall on the line. They don’t, but in most cases the estimated yield is fairly close to the predicted yield.
With this model, and using only the spectral data taken at the seed fill stage to make selections, we would have retained all of the highest yielding varieties by selecting the best half.
If we can repeat the kind of results we have achieved in the training population with experimental varieties from other populations, the precision should be accurate enough to cull out lines having a low yield potential at the earliest stage of evaluation. If we can discard low-yielding lines without having to harvest them and weigh the seed for yields, this will have tremendous value to the breeding program in terms of saving time, space, and money.
We only have two years of data so far. We are expanding our research into this new technology, developing more robust models, using different types of sensors, adding genotypes, and evaluating the methods of measurement. Also, this summer, we will test the use of aerial sensors in addition to the ground-based sensors.
Our goal is to be able to use spectral analysis to achieve a dramatic reduction in the cost of producing a unit gain in yield potential, and the results so far are promising.
This technology is also being evaluated for its ability to detect yield differences in wheat genotypes, in the program of USDA-ARS wheat geneticist Dr. Jesse Poland.