Use of Remote Sensing Technologies and UAV to Phenotype Peanuts Varieties at the Tidewater AREC
Phenotyping varieties and inbred lines to identify high yielding individuals, tolerant to specific environmental conditions such as drought, is a common process in breeding. However, these screenings require to collect data on hundreds or thousands of genotypes, several times throughout the growing season. It is often time consuming, labor intensive and even subjective based on the trait characterized. The fast development of remote sensing technologies using unmanned aerial vehicles (UAV) (1) is recognized as beneficial to increase the throughput of data collection, its accuracy and reliability.
At the Tidewater Agricultural Research and Extension Center, our peanut physiology program conducted several experiments in southeastern Virginia using UAV and photogrammetry for peanut phenotyping. For example, one experiment compared 56 recombinant inbred lines (RILs) under either rain shelter-covered or irrigated plots (2). The shelters (3) were pulled over the plots temporarily throughout the season, to allow the establishment of drought. After 45 days under the shelters, the plot assessment clearly demonstrated drought effect on peanut plant growth, i.e. the smaller plant growth for covered plots as compared to plants in irrigated plots (4, 5), and differences among the genotypes.
Aerial imagery can help quantify this information from the analysis of pixel color (6). In this way, the percentage of green pixels representing green leaves (arbitrarily called Green Area or GA) from images collected under both water regimes can provide information throughout the growing season on the amount of biomass and health of the plants (7). The RILs under irrigated conditions (the continuous lines on the graph), showed a steep increase of GA over time to quickly reach 95 – 100 %, i.e. images were almost entirely green pixels, and maintained this value until harvest. However, the RILs which were temporarily under the shelters (the dash lines) showed lower GA values and larger GA variability among the RILs after stress cessation. This information allowed to identify the RILs with tolerance to drought (those with higher GA values, representing more biomass accumulation) as compared to drought susceptible RILs (those with lower GA, representing less biomass accumulation). The same techniques allowed to assess the recovery from drought, with several RILs displaying an important increase of GA several days after the end of the drought stress, while other displaced no further GA increase.
Other aerial indices estimating plant biomass and health can be computed from the drone images, including the normalized difference vegetation index (NDVI) (8). The higher NDVI value (red in the image), more biomass and healthier are the plants. A value closer to 0 (light blue in the image) represents stressed and unhealthy plants, while negative NDVI values (dark blue) represents the bare soil. A similar analysis can be performed using thermal or long wave radiation from the canopy (9). Healthy plants transpire, for which they are cool with less thermal radiation from the canopy (pink-violet colors in the image). Unhealthy plants, in absence of moisture, transpire less, for which they have warmer canopies and increased thermal radiation (yellow in the image).
Plants under drought stress tend to wilt. A common wilting assessment performed in the field is visual rating using a wilting scale from 0 to 5 (10). Wilting evaluation in this way is slow and subjective, therefore not very reliable. In a study using 102 peanut accessions, our team developed a model using proximal (handheld) red-blue-green (RGB) camera and UAV camera to estimate wilting fast and more accurately. Our current model, recently published https://doi.org/10.3389/fpls.2021.658621, successfully differentiated the turgid versus wilted plants with 77% accuracy when using proximal camera, and 88 % accuracy when using UAV camera.
These results demonstrate that remote sensing technologies have improved accuracy, throughput, and spatial and temporal coverage in comparison with classical phenotyping.