Estimating city–scale emissions using gridded inventories lacks direct, precise measurements, resulting in significant uncertainty. A Kalman filter integrates diverse, uncertain information sources to deliver a reliable, accurate estimate of the true system state. By leveraging multiple gridded inventories and a Kalman filter fusion method, we developed an optimal city–scale (3 km) FFCO₂ emission product that incorporates quantified uncertainties and connects global–regional–city scales. Our findings reveal the following: (1) Kalman fusion post–reconstruction reduces estimate uncertainties for 2000–2014 and 2015–2021 to ±9.77% and ±11.39%, respectively, outperforming other inventories and improving accuracy to 73% compared to ODIAC and EDGAR (57%, 65%). (2) Long–term trends in the Greater Bay Area (GBA) show an upward trajectory, with a 2.8% rise during the global financial crisis and a −0.19% decline during the COVID-19 pandemic. Spatial analysis uncovers a “core–subcore–periphery” emission pattern. (3) The core city GZ consistently contributes the largest emissions, followed by DG as the second–largest emitter, and HK as the seventh–highest emitter. Factors influencing the center–shift of the pattern include the urban form of cities, population migration, GDP contribution, but not electricity consumption. The reconstructed method and product offer a reliable solution for the lack of directly observed emissions, enhancing decision–making accuracy for policymakers.