Traditional methods for individual tree-crown (ITC) detection (image classification, segmentation, template matching, etc.) applied to very high-resolution remote sensing imagery have been shown to struggle in disparate landscape types or image resolutions due to scale problems and information complexity. Deep learning promised to overcome these shortcomings due to its superior performance and versatility, proven with reported detection rates of ~90%. However, such models still find their limits in transferability across study areas, because of different tree conditions (e.g., isolated trees vs. compact forests) and/or resolutions of the input data. This study introduces a highly replicable deep learning ensemble design for ITC detection and species classification based on the established single shot detector (SSD) model. The ensemble model design is based on varying the input data for the SSD models, coupled with a voting strategy for the output predictions. Very high-resolution unmanned aerial vehicles (UAV), aerial remote sensing imagery and elevation data are used in different combinations to test the performance of the ensemble models in three study sites with highly contrasting spatial patterns. The results show that ensemble models perform better than any single SSD model, regardless of the local tree conditions or image resolution. The detection performance and the accuracy rates improved by 3–18% with only as few as two participant single models, regardless of the study site. However, when more than two models were included, the performance of the ensemble models only improved slightly and even dropped.

No abstract

No abstract

Due to the COVID-19 pandemic, researchers have had to find different resources in order to continue their research and the use of online information can represent a temporary solution. Our research is mainly focusing on a landscape which offers services and disservices. Recently, numerous studies that rely on landscape disservices have appeared. We associate wildlife-human-interactions (WHI) and human-wildlife-interactions (HWI) as part of landscape disservices. More precisely, in the first category (WHI) we have included the interaction of the wild animals with human and in the second category (HWI) we have created a database with animals attacked or/and killed by human. In order to sustain this analysis, we have selected data from local newspapers and Facebook groups, which supports our hypothesis that online resources could provide valuable data. The study area is represented by the Southern and Eastern Carpathians. The most affected mammals for this type of interactions (HWI) are bears, followed by wild boars and red deer, while WHI has intensified in the last five years. Based on the analysed data we can conclude that the animals who generate the most disservices to humans are bears and wild boars. The solutions we have identified, which also include online sources, for both HWI and WHI are relocation, rescue, capturing of the animals in reservations or, as a last resort, euthanasia. In order to reduce these types of interactions it is important to promote ecological education, development and promoting of certain attitudes and behaviour that have a visible impact upon HWI and WHI.

This paper presents an approach to detecting patterns in a three-dimensional context, emphasizing the role played by the local geometry of the surface model. The core of the associated algorithm is represented by the cosine similarity computed to sub-matrices of regularly gridded digital surface/canopy models. We developed an accompanying software instrument compatible with a GIS environment which allows, as inputs, locations in the surface/canopy model based on field data, pre-defined geometric shapes, or their combination. We exemplified the approach for a study case dealing with the locations of scattered trees and shrubs previously identified in the field in two study sites. We found that the variation in the pairwise similarities between the trees is better explained by the computation of slopes. Furthermore, we considered a pre-defined shape, the Mexican Hat wavelet. Its geometry is controlled by a single number, for which we found ranges of best fit between the shapes and the actual trees. Finally, a suitable combination of parameters made it possible to determine the potential locations of scattered trees. The accuracy of detection was equal to 77.9% and 89.5% in the two study sites considered. Moreover, a visual check based on orthophotomaps confirmed the reliability of the outcomes.