A Comparison of UAV- and TLS-derived Plant Height for Crop Monitoring: Using Polygon Grids for the Analysis of Crop Surface Models (CSMs)

Multi-temporal crop surface models (CSMs) are a reliable method for agricultural crop monitoring. They provide 3-dimensional representations of crop canopies, preferably available as a multi-temporal dataset. From the CSMs the spatial distribution of plant height can be derived. The data for the CSMs are captured by remote sensing methods including terrestrial laser scanning (TLS) and imagery from unmanned aerial vehicles (UAVs) combined with computer vision techniques. Previous studies underlined the suitability of both methods. However, it remained an open question if both methods provide actually comparable information. We assume that the differing viewing angles of both sensors influence the resulting CSM and that the UAV-based CSMs contain crop density information due to the nadir sensor position. Therefore, we expect a lower mean plant height and higher variation in the UAV-based CSM. The correlation between plant heights from both methods was analyzed and complemented by using polygon grids for spatial analysis. The polygon grids provide descriptive statistics for each raster cell by zonal statistics to investigate the data's potential as a density measure. Through this analysis it is possible to maximize the extraction of spatial information for larger grid cells though it is not comparable to standard resampling methods. We analyzed CSMs at early, middle, and late growth stages from a barley experiment field and found a high correlation (R-2 = 0.91) in plant height derived from both methods. The UAV-derived plant height was generally lower than the TLS-derived plant height at all growth stages. However, contrary to the expectations the coefficient of variation was higher in the TLS data set.