Unmanned aerial system assisted framework for the selection of high yielding cotton genotypes
Introduction
Currently, cotton (Gossypium sp.) is the world's most important natural textile fiber crop, accounting for around 35 percent of total world fiber use. Also, it is one of the biggest crops in the food oil industry (USDA, http://www.ers.usda.gov/topics/crops/cotton-wool.aspx). Plant breeding is one of civilization’s oldest activities and started with the need of improving plants for human benefit. Today, the world’s food, feed, and fiber production relies on breeding programs to create genetic types that excel in productivity and meet population needs. To that end, molecular plant breeding techniques and bioinformatics have significantly evolved in the past few decades (Cabrera-Bosquet et al., 2012, Hulse-Kemp et al., 2014, Mayes et al., 2005, Yu et al., 2014) to develop cotton genotypes with higher potential yield. The advancement of high throughput genotyping platforms and the access to fast and relatively inexpensive genomic information revolutionized cotton breeding programs. The ability to work efficiently in a genomic level enabled fast advances in innumerous sectors of genetic research. In plant breeding, the use of genetic maps and molecular markers in marker assisted selection to meticulously target the gene/trait of interest greatly expedites the breeding process. Phenotyping, on the other hand, has not developed at the same pace, despite its important role in plant breeding research (Araus and Cairns, 2014). Lack of high-throughput phenotyping method is known to be the main bottleneck for increased breeding efficiency, and current phenotyping methods still remain relatively low throughput, laborious, and time-consuming. Efficient, rapid, and reliable high throughput phenotyping platforms are much needed to decrypt relationship between the full suite of genetic and environmental factors and phenotypic variations at plant- and population-levels (Cobb et al., 2013, Furbank and Tester, 2011).
Traditionally, a plant breeding program focuses on several major steps. First, breeders establish the goal(s) of his program with a defined objective to focus their attention in gathering and/or creating a breeding population (genetic variability). After a base population has been established, one needs to choose one, or a combination of several available breeding strategies to deal with the segregating populations (e.g. pedigree, modified bulk, backcross, etc). Finally, based on the goal(s) of the breeding program, stable non-segregating superior genotypes are identified to be commercially released to the public. Typically, multiple years of intense screening and selection are required to develop and bring new elite genetic types or “cultivars” to market. As a result of the expansion of breeding programs and an increase of demand by the world’s population, breeders are always searching for new methods to increase efficiency, and convenient techniques to facilitate the breeding process.
Although traditional satellite remote sensing technologies have been utilized to address issues in effectively selecting high yield genotypes, the lack of spatial resolution and limited temporal resolution due to their orbital design were mentioned as key hurdles in enabling breeding selection framework using the remote sensing technologies (Berni et al., 2009). Recent advances in sensor technologies now made it possible to integrate small and lightweight sensors into Unmanned Aerial System (UAS). In the early stage of UAS technology for agriculture applications, most research focused on developing UAS data processing scheme and accessing data accuracy. Honkavaara et al. (2013) investigated the processing and uses of UAS image data in precision agriculture. They developed a processing scheme from raw images up to georeferenced reflectance image and digital surface models. The results confirmed that UAS technology has a great potential in precision agriculture and indicated many possible future researches. Gómez-Candón et al. (2014) studied geometric accuracy and crop line alignment while taking into account three different flight altitudes and a different number of GCPs. The results concluded that UAS data collection at a range between 30 and 100 m altitude and using a moderate number of GCPs can result in ultra-high resolution orthomosaic images. Rokhmana (2015) investigated the potential of UAS-based remote sensing for precision agriculture. The system used in the study has the ability to produce imagery with spatial resolution of <10 cm and the average geometric accuracy can be obtained up to 3 pixels, while the production time can be reached more than 500 ha a day. Primicerio et al. (2012) tested autonomous UAS flight for multispectral UAS image acquisition and the acquired images showed good agreement with ground-based spectrometer data.
In addition to the previous literatures, more detailed precision agriculture studies using UASs have been conducted to extract phenotypic features from the UAS data. Gevaert et al. (2015) combined multispectral satellite imagery with hyperspectral UAS imagery for precision agriculture applications for the first time. They constructed spectral-temporal response surfaces for crop phenology monitoring. Gago et al. (2015) compared the performance of different types of UASs with ground-truth plant data in terms of crop water stress. The results showed that thermal indices have a great potential to determine water stress heterogeneity. López-Granados et al. (2016) generated georeferenced weed seedling infestation maps by analyzing of visible and near-infrared UAS images. Based on the classification map, they investigated variability of herbicide treatments against weed coverage ratio. Bendig et al. (2015) combined selected vegetation indices and plant height information to estimate crop biomass. The correlations between vegetation indices and dry biomass were analyzed and then multiple regression models using selected vegetation indices and plant height were developed.
UAS can perform aerial data acquisition missions without manual control. To address the limitations of traditional remote sensing technologies, we propose to utilize an Unmanned Aerial System (UAS) as a data acquisition platform and develop a framework to select high yield cotton genotypes based on measurements from the UAS data. The main objectives of this study is to develop a systematic framework to select high yield cultivars using data acquired from UAS so that the developed framework can be used to empower breeders to improve their cultivar selection process.
Section snippets
Study site
A dryland field trial was established at the Texas A&M AgriLife Research and Extension Center (27° 46.948′N, 97° 33.605′W, elevation 16 m above sea level) at Corpus Christi, TX during the summer of 2015 (Fig. 1). The dryland field trial was approximately 35 m in width and 110 m in height. Seeds of 48 different cotton genotypes were planted on March 27, 2015 at a rate of 13.3 seeds/m in north to south oriented rows. Crop emergence occurred six days later on April 2, 2015. Plots were arranged in
Genotype/Entry selection process
The field trial included breeding lines (32) and commercially available varieties (16), for a total of 48 genotypes. Phenotypic features extracted from the UAS data were summarized by each genotype. Among the summarized features, four variables - mean equivalent boll diameter (MEBD), mean boll perimeter (MBP), mean boll area (MBA), and mean canopy cover (MCC) - were chosen based on preliminary regression analysis for their correlation with final harvested lint yield (Table 2) so that these
Conclusion
In conclusion, here we briefly demonstrate the potential of UAS technology and advanced image analysis techniques to facilitate cotton breeding efforts. By utilizing UAS-derived information regarding open cotton bolls and canopy cover, a four-step selection process for high yielding entries was performed. Ultimately, the selection process increased minimum and average lint yield of the remaining population by 7.4 and 10%, respectively. UAS-selected entries and genotypes matched 80 and 73%,
Acknowledgements
We appreciate Dow AgriSciences for providing materials for field experiments, Rudy Alaniz and Clink Livingston at Texas A&M AgriLife Extension and Kenneth Schaefer at Texas A&M AgriLife Research for their support.
References (17)
- et al.
Field high-throughput phenotyping: the new crop breeding frontier
Trends Plant Sci.
(2014) - et al.
Combining UAV-based plant height from crop surface models, visible, and near infrared vegetation indices for biomass monitoring in barley
Int. J. Appl. Earth Observation Geoinformation
(2015) - et al.
Phenomics - technologies to relieve the phenotyping bottleneck
Trends Plant Sci.
(2011) - et al.
UAVs challenge to assess water stress for sustainable agriculture
Agric. Water Manage.
(2015) The potential of UAV-based remote sensing for supporting precision agriculture in Indonesia
Procedia Environ. Sci.
(2015)- et al.
Sequential operations in digital picture processing
J. ACM
(1966) - Berni, J.A., Zarco-Tejada, P.J., Suárez Barranco, M.D., Fereres Castiel, E., 2009. Thermal and narrow-band...
- et al.
High-throughput phenotyping and genomic selection: the frontiers of crop breeding converge
J. Integr. Plant Biol.
(2012)
Cited by (37)
3D reconstruction and characterization of cotton bolls in situ based on UVA technology
2024, ISPRS Journal of Photogrammetry and Remote SensingDeveloping an image processing pipeline to improve the position accuracy of single UAV images
2023, Computers and Electronics in AgricultureA support vector machine and image processing based approach for counting open cotton bolls and estimating lint yield from UAV imagery
2023, Smart Agricultural TechnologyCitation Excerpt :Selecting a fixed threshold could lead to a high omission or commission error due to removal of lower reflectance cotton boll pixels or inclusion of higher reflectance background pixels, respectively. For example, Jung et al. [11] used a threshold value of 190 for the red band to segment cotton boll pixels from the background, which worked well for that study area, but using this threshold for UAV images in this study led to high omission error. We also used a fixed red/blue band ratio of 1.2 in the candidate recognition process to avoid high omission error, but we recommend evaluating classification accuracy by changing this threshold value.
Combining a cotton ‘Boll Area Index’ with in-season unmanned aerial multispectral and thermal imagery for yield estimation
2023, Field Crops ResearchCitation Excerpt :The resultant cotton boll area from this methodology had a strong positive correlation with cotton yield (R = 0.80). Jung et al. (2018) classified cotton bolls using a threshold value on the red band from DJI Phantom 2 camera. Boll pixels were clustered into individual cotton bolls then mapped and analyzed using patch analysis.
Efficacy of cotton harvest aids applications with unmanned aerial vehicles (UAV) and ground-based field sprayers – A case study comparison
2022, Smart Agricultural TechnologyDownscaling solar-induced chlorophyll fluorescence for field-scale cotton yield estimation by a two-step convolutional neural network
2022, Computers and Electronics in Agriculture