Post Date: 24 June 2026

Integrated Monitoring and Source Characterization of Ambient Methane at a Coastal Urban Site in Hong Kong

Abstract:

Urban methane assessment in coastal environments requires explicit consideration of observational support, spatial representativeness, atmospheric transport, and the strength of source evidence. This thesis develops an integrated CH₄ observation and modelling framework at the HKUST coastal supersite in Hong Kong by combining retrievals from ground-based direct-sun Fourier transform infrared spectroscopy, Sentinel-5P/TROPOMI XCH₄ products, fixed-time near-surface CH₄ measurements, meteorological transport diagnostics, source-inventory data, and multimodal Earth-observation (EO) descriptors. The winter EM27/SUN campaign provided the ground-based column-observation component of the study. Across 24 archived campaign days, 45,312 raw interferograms yielded 3,748 valid XCH₄ retrievals on 20 days. Seventeen campaign days coincided with Sentinel-5P overpasses, but only 13 contained at least one valid TROPOMI pixel within 50 km of the site. At the selected 50 km collocation radius, strict overpass-centred daily pairing retained nine matched days and yielded a mean bias of −3.3 ppb, a root-mean-square error (RMSE) of 23.7 ppb, and a correlation coefficient of r = 0.77. These results highlight the extent to which limited sampling support and spatial representativeness constrain satellite–ground column comparisons under coastal winter conditions.

A leakage-controlled benchmark with strict temporal separation was subsequently constructed from 2,224 model-ready near-surface CH₄ observation slots collected at fixed local times between January 2024 and March 2026. On the independent test block, the final transport-aware model incorporating selected EO context achieved an RMSE of 29.00 ppbv, R² = 0.658, an area under the precision–recall curve (AUPRC) of 0.974, and an F1 score of 0.920. Independent evaluation using 28 whole-air canister samples collected after the benchmark definition and candidate-prioritization surface had been frozen showed that four of the five highest-ranked samples exceeded the predefined campaign-relative field-enhancement threshold, corresponding to Precision@5 = 0.80 and Recall@5 = 0.57.

Overall, this thesis demonstrates that coastal urban CH₄ assessment can be strengthened by integrating traceable XCH₄ column observations, representativeness-aware satellite comparison, transport-aware modelling, and independent field evidence. The resulting framework supports defensible candidate-source prioritization while keeping source interpretation commensurate with the strength and limitations of the available observational evidence.