Surface reflectance calculation and predictive models of biophysical parameters of maize crop from RG-NIR sensor on board a UAV

Unmanned aerial vehicles (UAVs) present themselves as an alternative to overcome the limitations of satellite sensors in monitoring agricultural crops, motivating many studies with UAVs. They can carry sensors, which need studies for better understanding. The present study aimed to vicariously calibrate a Red-Green-Near infrared (RG-NIR) low-cost sensor on board a UAV, and to develop predictive models of biophysical parameters for a maize crop. To achieve this purpose, 15 sets of images were captured over 61 days after emergence (DAE) of the maize crop plantation. Each set of images was mosaicked and had their digital numbers (DN) converted to reflectance. After calibration, normalized difference vegetation index (NDVI) and cumulative NDVI (cNDVI) were calculated to serve as an independent variable in the models for estimating crop parameters. In the field, 54 plants were collected and evaluated for height, leaf area and dry biomass. It was observed that the NIR band had an influence on the red band, but this influence was attenuated with the empirical line calibration. NDVI was able to detect seasonal and spatial variations in maize. The NDVI model obtained on the collection day to estimate the total dry above ground biomass had better results, generating RMSE of 68.68 g m(-2) and R-2 of 0.81, in comparison with cNDVI. For productivity, the result was satisfactory with cNDVI, showing RMSE of 134.00 g m(-2) and R-2 of 0.63. Calibration of the sensor was shown to be important to attenuate influence between bands.

Automated summary

This study involved the calibration and development of predictive models for maize crops using a low-cost sensor mounted on a UAV. The results showed that the NDVI model performed better in estimating total dry above ground biomass compared to cumulative NDVI, and the importance of calibration to reduce inter-band influence.