Efficient Utilization of Low Altitude Remote Sensing Technology for Crop Phenotype Estimation
Discipline: Biological Sciences
Subcategory: Plant Research
Session: 3
Room: Marriott Balcony B
Roy L. Davis II - Texas A&M University
Co-Author(s): Thomas M. Chappell, Texas A&M University, College Station, TX; Young-Ki Jo, Texas A&M University, College Station, Texas
Low altitude remote sensing (LARS) applications have rapidly increased in agricultural use with the increasing accessibility and decreasing cost of unmanned aerial systems (UAS). Applications of this technology for high-throughput phenotyping and precision agriculture have been explore, but major limitation remain in the efficient use of the technology and the data generated. To realize the potential benefits of LARS, generated data must be used efficiently.
We hypothesize that LARS-sourced data are functionally surrogates for human-sourced measurements, instead of being unprecedented in form and requiring generation of analytical approaches de novo. To prove this concept, we developed an analytical approach for inferring phenotypes of rice using LARS-sourced data that will be explainable to diverse stakeholders.
Trials to determine effect of nitrogen inputs on rice were conducted, using varieties XL753 and CL272. Nitrogen treatments were applied at four prescribed time points during the growing season. Data derived from UAS-captured images were used to analyze the relationships between grain yield, plant height, and nitrogen input. From raw data, 3D point clouds were generated, and then analyzed as spatial and spectral data. We discovered that ‘GPS control points,’ a labor-intensive and costly means of establishing ground level for spatial analysis of 3D images, are not necessary for accurately inferring plant heights. We fit finite mixture models that facilitate calculation of average plant height without requiring the spatial position of the ground to be known: we simply estimated the positions of two latent classes and calculated the difference between them for an optimized spatial extent, to infer height. Thereafter, we analyzed the relationship of inferred height to grain yield, finding that inferred rice height can be used to predict grain yield, describing an appreciable amount of yield variation for the variety XL753 (R2=.724). We also determined that the height/yield relationship is variety-dependent: for the CL272 variety the relationship is weak (R2=.260). Similarly, spectral (red-green-blue) measurements describe yield variation in this system (R2=.738).
Findings offer exciting promise for the use of LARS in areas where precision agriculture is emergent. Because we have determined analytical techniques that negate the use of expensive or unavailable additional technology, ability to conduct high-throughput phenotyping without that technology results. Future research will test applicability of this approach in other crop systems, as well as its generalizability to pest and disease detection.
Reference: Yin, X., McClure, A., Jaja, N., Tyler, D. D., Hayes, R. M. (2011). In-season prediction of corn yield using plant height under major production systems. Agronomy Journal. 103:923-929.
Funder Acknowledgement(s): Funder Acknowledgements: Funding is provided by the USDA NIFA Hatch Project #TEX09705. I thank F. Dou, Ph.D. and T. Wilson Ph.D. at the Texas A&M Research and Extension Center in Beaumont, TX.
Faculty Advisor: Thomas M. Chappell, thomas-chappell@tamu.edu
Role: Of this research, I processed and analyzed the data that was received following the creation of the three-dimensional point clouds. I also revised a previously developed statistical analysis program, developed by my advisor, to better suit the needs of the research at hand. I further analyzed the effects of the nitrogen treatments on grain yield and the height response.