Haonan Zhang

• Developed an end-to-end machine learning pipeline for automated subsea cable detection in Sub-Bottom Profiler (SBP) data, enabling near–real-time outputs for AUV deployment • Trained a YOLOv5 model on Ground Penetrating Radar (GPR) data and successfully generalised to SBP data, demonstrating robust cross-domain hyperbola detection • Applied seismic preprocessing techniques including bandpass filtering, time-gain control (TGC), and automated seabed picking to improve signal quality and reduce false positives • Translated YOLO detections into engineering metrics, estimating cable position and burial depth (via two-way travel time) • Analysed multi-year MBES datasets to generate bathymetric maps and cross-sections, identifying seabed evolution and geohazards (e.g. scour, sediment mobility) • Assessed subsea cable exposure risk by integrating bathymetric changes and seabed morphology across survey campaigns

MSci Project: Integrating Harmonic Analysis and Recursive BiLSTM for Accurate Sea Water Elevation Prediction Developed a hybrid forecasting model combining harmonic tidal theory with recursive BiLSTM, achieving R² up to 0.97 in short-term sea level predictions across four UK coastal sites. Mid-term (3–8 day) forecast accuracy improved by 15–20% over standard BiLSTM through the integration of solar and lunar tidal harmonics as exogenous inputs. A recursive self-correcting multi-step framework enabled 30-day predictions while mitigating error propagation. Processed and cleaned 140,000+ tidal measurements from BODC archives with outlier handling, normalization, and missing data recovery. Sequence-to-sequence LSTM hyperparameters were tuned to optimize temporal input windows and output horizons for robust performance.