Seed Sorting: From Harvest to Genebank

11 June 2019

Keeping up with the demand for rice seeds: The automation of seed sorting at the International Rice Genebank

To an untrained eye, all sheep in a paddock look alike. But shepherds can call all the sheep in their flock by name and recognize the distinguishing features of each animal. To most of us, the seeds in a bag of freshly harvested rice might look alike. However, a trained genebank technician can immediately spot differences and quickly separate seeds worth conserving from those destined for the bin.

At the International Rice Research Institute (IRRI), an automatic seed sorter is now making the task of sorting seeds more efficient.

The International Rice Genebank at the International Rice Research Institute (IRRI) processes a staggering number of seeds. When the genebank receives seeds from the field, they are sorted to select healthy and mature ones. “To keep the rice for as long as possible in storage we need high quality seed, so every single seed that is harvested is inspected,” said Ruaraidh Sackville Hamilton, former Head of the International Rice Genebank. Traditionally, this colossal task is performed by workers who rapidly pick through the seeds by hand.

Rising Demand for Rice Germplasm

The genebank distributes up to 40,000 rice samples a year, a huge increase from only a few years ago. The demand has been partly driven by advances in sequencing technologies, which enabled the first rice genome to be sequenced in 2005, and, following this, the 3,000 Rice Genomes Project launched in 2012.

“Our genebank holds seeds of all the varieties included in the 3,000 Rice Genomes Project, and researchers have been increasingly eager to use them as they find out more about the genetics behind the traits they are looking for,” said Venuprasad Ramaiah, the current head of the genebank.

That means the genebank must multiply and process more seeds, and it must do it quickly and efficiently. That presented a challenge to IRRI genebank staff.

“There were times when the harvests in the current season were ready to be sorted, but some of the harvests from the previous season were not yet processed,” said Flora de Guzman, who oversees genebank operations. The delay in processing not only means a longer time before the seed is distributed but also possibly shorter longevity in storage.

“One of the factors that affects the amount of time you can keep a seed healthy in storage is the time it takes to process the harvest for storage,” said de Guzman. The longer the processing time, the longer the seeds are exposed to conditions that are detrimental to seed quality and longevity. In fact, one of the research activities under the CGIAR Genebank Platform examined seed longevity of all the mandate crops and optimized protocols for seed processing, testing and storage so that the genebanks can increase how long they can keep seeds in storage, as well as the intervals between viability test.

Toward Greater Efficiency in Seed Sorting

One of the scientists at IRRI thought it was worth exploring to see if introducing more automation into seed sorting would help remove bottlenecks. The use of automated seed sorters is well established among large seed companies. But a genebank sorts smaller samples of more diverse types of seed than do seed companies and existing sorters would not be able to deal with the lack of uniformity.

With support for upgrading under the Genebanks CGIAR Research Program (the precursor of the Platform), IRRI was able to consider investing in a machine specifically customized to deal with diverse rice samples. After some background research, a Dutch company, SeQso, was chosen to design a bespoke automatic seed sorter for the rice genebank.

Donald Villanueva, an IRRI staff member who participated in the Genebank Platform Impact Fellowship Program in 2018, explained how it works. “Staff first sort a set of reference seeds to show the machine which seeds are good to keep. The machine converts this into parameters of seed size, shape and color. Then the machine is fed with the whole sample and proceeds to sort them. It does not sort perfectly, so staff re-examine the smaller sets of accepted seeds to remove any erroneously accepted bad ones.”

Increased Throughput with Less Manual Labor

The automatic seed sorter has streamlined seed sorting and increased throughput. It can accommodate eight sets of samples in one run, which lasts for 8-12 hours, depending on the amount of seed. The genebank staff usually prepare two runs: one batch for the day and the other for overnight, handling up to a total of 1.1 million seeds a day. The machine can operate for 24 hours a day, whereas manual seed selection only occurs during office hours, unless there is overtime.

“We estimate the automatic seed sorter can lead to a 30-40% reduction in the staff time needed for seed sorting,” said Donald. This estimate includes the labor incurred in the manual re-examination of seed sorted by the machine. “If the machine could work faultlessly without post-operation quality control, that reduction would increase to 60-80%.”

“The harvest takes less time to get to storage and we have reduced the need for staff overtime,” said Venuprasad.

Limits To Automation

This is only the beginning of improvements in seed sorting; there are still challenges to address. For example, the seed sorter cannot handle all seed types. Some varieties have long awns, the bristly extensions of the spikelet originally evolved to attach the seed to animal fur in the wild species. Unfortunately, these get stuck in the seed sorter tubes. What is more, some features of certain seeds are difficult for the machine to recognize, including seed coat color.

“This machine does not entirely replace manual seed sorting,” said Imelda Boncajes, a research technician at the IRRI genebank. “There are some characteristics of the seeds which cannot be read by the machine. Thus, some batches of seeds still have to be sorted manually.”

There are also cases where the machine erroneously accepts bad seeds and rejects good ones. The sorter can only scan the part of the seed facing upwards, so off-types are selected, while the misalignment of the seeds on the conveyor belt can cause good seeds to be discarded. Currently, the machine uses very simplistic measures of size, length, breadth and average color.

Yet seeds differ in all sorts of ways – for example the degree of curvature at one end, little spots of color or ridges of different sizes. These are all significant. Detecting color is also a challenge with the sorter IRRI is currently using. More expensive cameras are available, however, that can provide higher resolution images and a quantitative measure of color that comes closer to what the human eye can see.

Software development also has the potential to make the machine better at distinguishing some small differences either through more sophisticated programming or machine learning, where the computer works out the parameters for itself.

Categories: Genebank Platform, Genebanks, Rice