Evaluating different methodological approaches for very high spatial resolution mapping of agricultural areas exploiting UAV data: a case study from Greek agricultural site

The rapid technological developments of recent years have enabled new methods for acquiring aerial photographs and high-spectral-resolution imagery. In this context, unmanned aerial vehicles (UAVs) offer significant potential for high-resolution Land Use/Land Cover (LULC) mapping, allowing clear distinction between natural and human-made features. UAV-based approaches provide high accuracy, faster data acquisition, and cost-effective solutions for detailed LULC analyses. However, there is a fertile ground in evaluating different methodological approaches and testing different algorithms for obtaining robust and transferable results. To this end, the present study aims at comparing two advanced classification techniques for mapping agricultural areas using multispectral UAV data over a typical agricultural site. The area selected for the study consists of crops and agricultural land located near the town of Amygdales, in the regional unit of Grevena. The two techniques are SVM (Support Vector Machines) and MLC (Maximum Likelihood Classification). In overall, results showed that the SVM proved to be more accurate with an overall accuracy of 79.45% compared to 78.91% for MLC, while both methods achieved a Kappa coefficient of 0.72. The statistical significance of the findings was further confirmed from the Mc-Nemar statistical significance results which were also computed. The results evidenced the capability of both methods obtaining LULC maps at very high spatial resolution. All in all, the methodological approach presented herein provides potentially a low-cost solution in mapping agricultural areas at very high spatial resolution which may be also fully transferable and reproducible to other locations too, which offer potentially important pathways to be used in precision agriculture applications. Such information can be of practical value to both farmers and decision-makers in reaching the most appropriate decisions for field management.

Keywords: Precision Agriculture, Mapping, UAVs, Classification, Machine Learning, Support Vector Machine, Maximum Likelihood

Acknowledgement

The participation of George P. Petropoulos study is financial supported by supported by the ACCELERATE MSCA SE program of the European Union’s Horizon research and innovation program under grant agreement No. 101182930.