Citizen science leveraging social media-based platforms offers a powerful tool for large-scale biodiversity monitoring and public engagement. However, inherent biases related to observer behavior affect the data patterns. Understanding the drivers of observation hotspots – areas with high data density – is vital for data interpretation and project design optimization. This study investigated the factors forming citizen science biodiversity observation hotspots in the urban forests of Tsukuba Science City, Japan, hypothesizing that human accessibility factors are as important as ecological factors. We analyzed 17,174 filtered wildlife observations from the citizen science platform (2019-2024) across 54 km^2 grid squares associated with Densely Inhabited Districts. We classified forest land cover into three accessibility types (Public Wayside, Public Inland, Remote), based on road proximity and public access status. We compared MaxEnt models for the five taxonomic groups using (1) basic land cover categories and (2) land cover with refined forest accessibility categories. Model performance was evaluated using the AUC. These results strongly support our hypotheses. Models incorporating forest accessibility (Model 2) consistently outperformed basic land cover models (Model 1). In Model 2, human accessibility factors were major contributors to predicting hotspots and specific land cover factors connected to species distributions. These findings highlight the critical role of biases that arise from social variables, defined by accessibility, in driving citizen science data patterns. Accounting for accessibility is essential for interpreting citizen science data, designing effective engagement strategies, and planning biodiversity-friendly, accessible urban green spaces.