This study evaluates whether existing long-term monitoring data are sufficient to distinguish natural background concentrations from anthropogenic influences on chemical river water quality in Lake Sevan basin. A comprehensive dataset covering physicochemical parameters, nutrients, and trace metals was analyzed for nine major tributaries with different geological settings and land-use characteristics. Multivariate statistical analysis was applied to identify baseline conditions and deviations attributable to human activities. The results indicate that water chemistry is primarily controlled by lithology and hydrological regime, particularly in minimally impacted headwater regions. In contrast, elevated concentrations of nutrients (e.g., nitrate and phosphate) and selected trace elements were associated with agricultural runoff, urban discharge, and localized industrial inputs. Spatial patterns reveal clear gradients of increasing anthropogenic impact downstream and in densely populated sub-basins. The study also demonstrates that, while the current monitoring network is suitable for assessing the overall chemical status of rivers, it is less effective in defining natural background levels and quantifying individual pollution sources due to limited upstream reference conditions. Overall, this approach provides a scientific basis for improved water quality management and policy implementation in the Lake Sevan basin. The findings highlight the importance of integrating long-term monitoring data with statistical tools to support sustainable watershed management in vulnerable catchments.

Reinforcement Learning (RL) has shown strong potential for Unmanned Aerial Vehicle (UAV) research. However, practical deployment of RL for UAV remains limited by the lack of realistic, standardized, and easily transferable (sim-to-real) simulation environments, as well as the difficulty of integrating RL-based methods with established flight-control frameworks. To address these challenges, we introduce \textbf{RLDroneSim}~\cite{ref-rlpid}, a realistic and extensible platform for RL-based UAV experimentation that builds on the integration of ArduPilot Software-in-the-Loop (SITL) with the Gazebo simulator~\cite{ref-ardupilot-gazebo}.RLDroneSim allows RL agents to operate on top of the existing ArduPilot firmware rather than replacing core components, enabling augmentation of current UAV functionality with minimal integration effort. The platform provides a fully Gymnasium-compatible interface~\cite{ref-gymnasium}, supports any Gymnasium-compatible RL library, and enables flexible experimentation across various UAV control tasks while preserving firmware-level realism. Its modular design includes a clearly defined architecture, environment interface, data-flow pipeline, and support for scalable multi-simulation execution.In this paper, we present the architecture of RLDroneSim, describe its main components and environment design, and demonstrate its integration with standard RL libraries such as Stable Baselines3~\cite{ref-stable-baselines3}. The platform is released as open source with extensive documentation~\cite{ref-rldronesim-docs} to support reproducibility, adoption, and future research on RL-based UAV control and sim-to-real transfer.