Prediction of Maize Grain Yield before Maturity Using Improved Temporal Height Estimates of Unmanned Aerial Systems

Abstract

UAS captured increased genetic variation compared with manual terminal height. There were small significant differences in ground filtering methods to extract plant structure. Higher resolution did not improve imagery informativeness with regard to plant height. Logistic function provides informative phenotypes for temporal maize growth. Correlation and prediction accuracy of grain yield increased by ∼20% with UAS heights. Weekly unmanned aerial system (UAS) imagery was collected over the College Station, TX, 2017 Genomes to Fields (G2F) hybrid trial, across three environmental stress treatments, using two UAS platforms. The high‐altitude (120‐m) fixed‐wing platform increased the fraction of variation attributed to genetics and had highly repeatable (R > 60%) height estimates, increasing the genetic variance explained (10–40%) over traditional terminal plant height measurement (PHTTRML ∼30%), as well as over the low‐altitude rotary‐wing UAS platform (10–20%). A logistic function reduced the dimensionality (>20 flights) of each UAS dataset to three parameters (inflection point, growth rate, and asymptote) and produced a more robust predictive model than independent flight dates, effectively summarizing (R2 > 0.98) the UAS flight dates. The logistic model overcame the need to use specific flight dates when comparing different environments. The UAS height estimates (r = 0.36–0.48) doubled the correlations to grain yield in this G2F experiment compared with PHTTRML (r = 0.23–0.28). Parameters of the logistical function achieved equivalent correlations (r = 0.30–0.46) to individual flight dates (r = 0.36–0.48), improving grain yield prediction by ∼400% (R2 = 0.25–0.34) over PHTTRML (R2 = 0.06–0.08). Incorporating other UAS‐derived parameters beyond plant height may allow yield to be accurately predicted before maturity, speeding breeding programs. A new public R function to generate ESRI shapefiles for plot research is also described.

UAV Based Imaging Platform for Monitoring Maize Growth Throughout Development

“…Other approaches attempt to create the DTM by identifying and extracting ground points from the pointcloud and interpolating the height of intermediate points to create a surface model of the ground and then taking the difference between the DTM and DEM, but these approaches require processing large 3D point clouds and finding optimal parameters for the interpolation algorithm which can be computationally intensive (Su et al, 2019). Different algorithms for generating a DTM including DBM and three point cloud interpolation methods were evaluated by Anderson II et al (2019) to see which provides higher accuracy when estimating plot heights. They found that all methods had similar, consistent performance in flat, uniform fields like those used in breeding trials.…”

“…Canopy coverage estimates have been gathered for soybeans using digital cameras and these have been found to highly correlate to canopy light interception measurements (Purcell, 2000) and grain yield (Xavier et al, 2017). Previous studies have estimated PH from UAV imagery in sorghum (Chang et al, 2017;Shi et al, 2016;Watanebe et al, 2017), wheat (Madec et al, 2017;Michalski et al, 2018;Holman et al, 2016), cotton (Feng et al, 2018), barley (Bendig et al, 2014 ) and maize (Anderson II et al, 2019;Anthony et al, 2014;Geipel, Link and Claupein (2014);Grenzdörffer, 2014;Su et al, 2019;Varela et al, 2017 ). Although these studies have reported high correlation in PH measurements across various dates to manual measurements, they lack estimates of how daily correlations of imagery-derived PH (PHUAV) and ruler-derived PH (PHR) vary throughout different growth stages and how this compares to the inherent error in PHR measurements.…”

Unoccupied aerial system enabled functional modeling of maize (Zea mays L.) height reveals dynamic expression of loci associated to temporal growth

“…Implementation of HTP systems provides the ability to collect temporal phenotypic measurements on large representative populations within field settings, to understand how individuals interact with their environments [3, 5, 6]. Unoccupied aerial systems (UAS) are especially useful to increase the size of populations and field studies investigated, collecting RGB images, and reconstructing three dimensional representations of field crop trials using structure from motion methodology [6–15]. UAS height estimates of maize have previously been validated using correlations to traditional manual measurements and evidence of equivalent or greater phenotypic variation partitioned to genetic factors [7, 9, 15, 16].…”

“…Unoccupied aerial systems (UAS) are especially useful to increase the size of populations and field studies investigated, collecting RGB images, and reconstructing three dimensional representations of field crop trials using structure from motion methodology [6–15]. UAS height estimates of maize have previously been validated using correlations to traditional manual measurements and evidence of equivalent or greater phenotypic variation partitioned to genetic factors [7, 9, 15, 16]. To our knowledge the majority of reported field based phenotyping of maize with HTP platforms has focused on hybrid trials [6-9, 15, 17-19] but, limited reports have been published on the evaluation of inbred trials [20–22], specifically genetic mapping populations.…”

“…Maize height is important as a physiological and a highly heritable agronomic trait [23–26] commonly collected due to its ease of measurement, agronomic importance, and correlation to hybrid grain yield in some environment and management scenarios [15, 27–32]. Manually measured plant height is commonly collected after reproductive maturity as the distance from the ground to the tip of the tassel, flag leaf, or peduncle.…”