Development of high-throughput phenotyping models for the estimation of quantitative traits in soybean breeding collections

[P. D. Ranđelović]

123 str.

Biotehničke nauke - Genetika i oplemenjivanje / Biotechnical Science - Genetics and breeding

U okviru oplemenjivačkog procesa koji za krajnji rezultat ima stvaranje nove sorte, prikupljanje podataka o fenotipskim karakteristikama biljnog materijala predstavlja esencijalan ali i prilično naporan posao. S obzirom da oplemenjivači soje svake godine rade na materijalu koji se sastoji od stotina i hiljada genotipova roditelja i hibridnog potomstva od F1 do F6-7 generacije, a da je raspoložive radne snage sve manje, postaje neophodno uvoditi savremene alate i tehnike za efikasniju i precizniju determinaciju kvantitativnih osobina. Imajući u vidu potencijal daljinske detekcije i fotogrametrijske obrade snimaka u oplemenjivanju biljaka, cilj ove doktorske disertacije bio je razvoj modela za visokopropusnu fenotipizaciju kvantitativnih osobina soje. Istraživanje je realizovano u četvorogodišnjim poljskim ogledima, u dve faze, pri čemu je u prvoj kreiran model za predikciju broja biljaka/m2 dok su modeli za procenu pokrovnosti, visine, biomase i prinosa soje razvijeni tokom druge faze istraživanja na divergentnoj germplazmi koja je gajena na dva tipa zemljišta, karbonantnom černozemu i peskovitom lesu (simulacija suše). Kalibrisanje modela je obavljeno na osnovu podataka izdvojenih sa digitalnih fotografija (RGB, multispektral) biljnog materijala i stvarno izmerenih vrednosti osobina za koje se modeli prave. Pored toga, analizirana je veza između podataka visokopropusne fenotipizacije i morfoloških i produktivnih svojstava soje, te mogućnost korišćenja rezultata predikcionih modela za selekciju superiornih genotipova. Kreirani modeli su pokazali visok nivo preciznosti što je potvrđeno R2 vrednostima koje su se u zavisnosti od osobine kretale od 0,76 do 0,94. Podaci dobijeni pomoću visokopropusne fenotipizacije pokazali su određen stepen povezanosti sa pojedinim morfološkim i proizvodnim karakteristikama soje (r = ±0,5) što je od posebnog značaja za ranu karakterizaciju oplemenjivačkog materijala. Konačno, rezultati predikcionih modela u kombinaciji sa podacima koji su dobijeni klasičnom fenotipizacijom, omogućili su odabir najboljih genotipova u različitim uslovima gajenja koji kao takvi mogu biti iskorišćeni u budućim oplemenjivačkim programima.

In the plant breeding process, which results in the development of new varieties, phenotyping represents an essential yet demanding task. Each year, soybean breeders work with material that consists of hundreds or even thousands of genotypes from F1 to F6-7 generations. This fact, alongside the current decrease in available manpower, demands the implementation of modern tools and techniques that will enable a more efficient and precise determination of quantitative traits. Knowing the potential of remote sensing and image photogrammetry in plant breeding, this doctoral dissertation aimed to develop models for high-throughput phenotyping of quantitative soybean traits. The research was carried out in four-year field trials in two phases. The model for the prediction of soybean plant density (plants/m2) was created in the first phase, while models for canopy coverage, height, biomass and yield were developed during the second phase of the study on divergent germplasm grown on two soil types, carbonate chernozem, and sandy loess (drought simulation). The models were calibrated based on the data extracted from digital images (RGB, multispectral) of soybean plots and actual values of the traits for which the models were created. In addition, the relationship between high-throughput phenotyping data and morphological and agronomic soybean characteristics was analyzed. Also, the results obtained with the prediction models were used to select superior genotypes. The created models showed a high level of precision, which was confirmed by R2 values that ranged from 0,76 to 0,94, depending on the trait. The high-throughput phenotyping data was correlated with some morphological and productive soybean traits (r = ±0,5), which is very important for the early characterization of breeding material. Based on the data obtained with the prediction models and traditional phenotyping, the genotypes with the best performance in different growing conditions were highlighted as a great source of superior traits for future breeding programs.

visokopropusna fenotipizacija, soja, oplemenjivanje, daljinska detekcija, fotogrametrija, broj biljaka/m2, pokrovnost, visina, biomasa, prinos

high-throughput phenotyping, soybean, breeding, remote sensing, photogrammetry, plant density, canopy coverage, height, biomass, yield

631.527:633.34(043.3)

Tekst