UAV time-series imagery show diversity treatment effects on cabbage growth

The demand for more sustainable farming is driving interest in alternative cropping systems, such as strip intercropping. In such systems, two or more crops are grown simultaneously on the same field, offering advantages such as increased biocontrol of weed, pest, diseases, and increasing productivity in resource-limited ecosystems. However, with strip intercropping, complexity increases and quantitative data to study the competition between plants are still limited due to the current manual process of acquiring data. While individual-plant data would facilitate this study, the manual acquisition is not feasible for large-scale experiments. Alternatively, unmanned aerial vehicles (UAV) equipped with high-resolution camera can cover large areas and estimate individual-plant growth from RGB imagery using automated image-processing methods. This study investigated its applicability to monitor the plant-height development of individual cabbage plants in space and time with sufficient accuracy and to identify the potential differences between strip intercropping treatments. Using RGB imagery and structure-from-motion analysis, a digital surface model (DSM) was created. Individual plant-height was calculated from the DSM by estimating the height of the vegetation and the height of the soil. Comparing the height estimations with ground-truth height measurements showed an overall root mean square error (RMSE) of 4.67 cm, which is in the same range as the 4cm standard deviation between measurements of multiple observers. The UAV-based height estimation of individual plants was used not only to compare the development in a strip intercropping field to that in a monoculture but also to compare with various treatments in the strip intercropping system. The results show that the plants grew faster in intercropping conditions than in monocropping conditions, with a subtle difference between treatments. Our results illustrate that with a UAV-based imaging approach we can go beyond current experimental practice and collect vast amounts of data on individual plants with high spatial and temporal resolution with an accuracy similar to that of manual measurements.