UAV remote sensing was carried out on 15 February, 14 March, and 12 April 2019, when the target wheat plants were in various stages of growth (tillering, early growth, and lager growth stage). A commercial-grade UAV (DJI Inspire 1, Shenzhen, China) equipped with a multispectral camera was used. The UAV was flown automatically over the field (Figure 3A) at an altitude of 30 m under the control of a commercially available flight application (Litchi, VC Technology Ltd., London, England). Two cameras—a Zenmuse X5 (DJI, Shenzhen, China) and a Micasense RedEdge (Micasense, Seattle, WA, United States)—were mounted on the UAV to ensure that RGB and multispectral images were captured during the same flight (Figure 3B). Two sets of images of a calibrated reflectance panel placed at about 1 m height were also captured immediately before and after each flight to improve the accuracy of the reflectance data for multi-spectral images. Also, acrylic plates were placed at the four corners of the field and three locations within the field as ground control points (GCPs), and were measured using the Hemisphere RTK differential GNSS device to improve the geolocation accuracy.

Representation of the image capture and data processing in the GAUSS system. (A) UAV overflight to capture image data; (B) types of image data collected; (C) results of image processing.

The captured images were processed using commercial photogrammetry software (Pix4Dmapper Pro, Pix4D, Lausanne, Switzerland). The pixel-by-pixel values over the entire wheat field were determined by using orthomosaic and digital surface modeling (DSM) that are generated from RGB images to calculate the vegetation cover area and plant height, respectively, and reflectance maps were generated from the multi-spectral images to calculate the NDVI (Figure 3C).

To determine which of the image indices best predicted actual wheat yield, the model selection based on the Akaike’s information criteria (AIC) (Akaike, 1974) for results of generalized linear models (GLMs) was used. Accurate geolocation information for both the UAV imagery and the manual sampling points made it possible to map the manually sampled data (actual ground data) into the UAV imagery indices. In the GLM analysis, the dry weight of the harvested wheat ears from each manual sampling location was the response variable. The means of vegetation cover area, plant height, and NDVI at each of the sampling locations, estimated from the UAV images recorded in February, March, and April, were the explanatory variables. Plant height from February was excluded from the analysis owing to low plant height (<10 cm) and consequent low estimation accuracy. The error distribution was Gaussian with an identity link function. The statistical model with the lowest AIC score was selected as the best model (f) for estimating the values of the manually sampled wheat yield data (yp) at a sampling point p from the UAV images, where:

and y^p is the estimate of yp, xi,p represents i-th index (vegetation cover area, plant height, or NDVI in this study) at a sampling point p derived from the UAV images, m is the total number of indices, and ϵ is measurement error, respectively. The best model was then used in all subsequent steps.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.