Understanding the relationship between Visual Environment (VE) and residents’ happiness is crucial for designing psychologically supportive urban environments. However, prior research has primarily focused on the impact of visual semantic elements (e.g. trees, buildings), often overlooking geometric features (e.g. area and perimeter of the visual field) that are also perceived at eye-level and profoundly shape emotional experiences. Moreover, the nonlinear impact of VE factors on residents’ happiness remains largely unexplored (e.g. more visual greenery is not always better). To address these gaps, this study proposes a novel framework that integrates multi-source geospatial data to investigate the nonlinear relationship between integrated VE and residents’ happiness in Wuhan, China. Both semantic and geometric features of VE are systematically measured by combining street view images and building footprints through semantic segmentation and isovist analysis. Residents’ happiness is quantified by extracting emotional information from social media data utilizing the Bidirectional Encoder Representations from Transformers (BERT) pre-trained model. Generalized Additive Mixed Model (GAMM) is then employed to assess the VE’s impact on happiness. Results reveal that certain VE factors exhibit significant nonlinear relationships with happiness (i.e. inverted U- or S-shaped curves), displaying varying threshold effects. For instance, the green view index has an optimal level at 45%, with a dose-response effect emerging above 20%. Similar optimal thresholds are observed for isovist area (80,000 m²), and drift magnitude (40 m). Additionally, factors like wall elements show varied impacts between weekdays and weekends. This study broadens the scope of VE-happiness research by introducing the visual geometric dimensions, highlighting the importance of considering optimal VE ranges to inform evidence-based strategies for designing livable and more sustainable cities.