Vasyl Cherlinka

Hyperspectral (HS) and LiDAR sensing provide complementary information for vineyard monitoring. HS imagery captures detailed spectral signals related to canopy physiology and biochemistry but is often affected by contamination from inter-row soil and weeds. LiDAR offers precise measurements of canopy structure yet lacks biochemical sensitivity. Their integration has the potential to overcome these limitations, and remotely piloted (unpiloted) aerial vehicles (UAVs) provide the flexible platform needed to collect both datasets at very high resolution over vineyards. In this study, UAV-based hyperspectral and laser scanning data were collected in the Slovak part of the Tokaj wine region to evaluate their combined potential for distinguishing vine from non-vine areas, producing dense point clouds with more than 600 points per square metre and hyperspectral imagery consisting of 172 bands at 0.1 m spatial resolution. Four datasets were prepared: hyperspectral imagery alone, hyperspectral imagery combined with canopy height, simulated natural colour imagery alone, and simulated natural colour imagery combined with canopy height. All datasets were transformed using principal component analysis, and the resulting features were classified with a supervised maximum likelihood classifier. Accuracy was evaluated using 1,000 field-validated reference points. The classification based only on the hyperspectral data reached 89% overall accuracy but performed poorly for vine detection, with a producer’s accuracy of 48.9% and an F1-score of 0.61. When canopy height information was included, performance improved to 96% overall accuracy, a Kappa coefficient of 0.85, and an F1-score of 0.88. Simulated natural-colour imagery combined with canopy height achieved intermediate results, with 93% overall accuracy and an F1-score of 0.79. These findings confirm that integrating spectral and structural information enhances vineyard mapping and provides a reliable basis for precision viticulture applications.

Soil monitoring programs in many countries often do not measure such an indicator as soil bulk density. At the same time, modern problems of assessing the condition of the soil cover require its use in soil management of agricultural landscapes. We see the solution to this problem in geospatial modeling. We have developed a technique for simulating soil bulk density, the algorithm of which consists of five stages: the creation of model maps of agro-industrial soil groups, modeling of a clay content map, construction of a map of humus content, selection of empirical dependences of soil bulk density on humus and clay content with evaluation reliability of these dependencies and their implementation in GRASS GIS and, finally, cartographic modeling of the spatial continuous distribution of soil bulk density. At the same time, parameters for improving the quality of methods modeling and mapping the soil bulk density were established. First of all, this is an increase in the sampling dataset of density, humus, and clay content to improve the reliability of the selected regression formula. It is also an improvement in the quality of spatial modeling of the humus content using geostatistical methods or modeling. To improve the clay content map, it is necessary to increase the resolution of DEM and predictors, to choose more accurate sources of DEM and prediction algorithms. The application of the presented methodology is shown in a part of Chernivtsi region, Ukraine, with the possibility of use on the scale of the entire country. URL: https://www.gcass.science.upjs.sk/