Combining UAS and Sentinel-2 Data to Estimate Canopy Parameters of a Cotton Crop Using Machine Learning

Abstract

Immense growth is witnessed in the application of unmanned aircraft systems (UAS) for precision agriculture in recent years. Though UAS can provide precise high spatial resolution data, large aerial coverage is still practically not feasible due to limited battery, flight time and large size of the data. Contrarily, satellite images cover larger area but provide coarser spatial resolution data compared to UAS imagery. The objective of this research was to combine UAS's ability to provide precise high spatial resolution data with the vast areal coverage provided by satellite data to estimate canopy parameters. Experimental field was divided into 10m x 10m grids to match Sentinel-2 pixel size. High resolution UAS data was collected covering the entire experimental field over which grid-wise canopy parameters were computed such as canopy height and canopy volume. Using artificial neural networks (ANN) based regression model, a relationship was developed between Sentinel-2 pixel values and UAS based canopy parameters with R2 greater than 0.9. ANN model results were also compared with support vector machine (SVM) regression model results.