Accurate ammonia (NH₃) emission inventories are critical for PM_2.5 mitigation, yet bottom-up estimates remain uncertain, particularly for sector-resolved estimates in humid subtropical regions such as Guangdong, where urban-industrial emissions in the Pearl River Delta (PRD) coexist with dispersed agricultural sources in surrounding non-PRD areas. Here we integrate a ground-based NH₃ network, Fengyun-4B (FY-4B) geostationary NH₃ retrievals, a localized 3 km × 3 km prior inventory, and stretched-grid GEOS-Chem High Performance (GCHP) simulations at 0.2° × 0.2° resolution to inversely constrain monthly agricultural and non-agricultural NH₃ emissions in the PRD and non-PRD Guangdong in 2023. The posterior simulation improved agreement with observations, reducing NRMSE from 55.3 % to 48.4 % and changing NMB from −9.0 % to 4.6 %, with further support from independent NH₃, NH₄⁺, and deposition measurements. Provincial anthropogenic NH₃ emissions decreased from 477.6 to 441.5 kt yr^−1: agricultural emissions declined from 437.2 to 362.5 kt yr^−1, mainly through warm-season reductions in non-PRD areas, whereas non-agricultural emissions increased from 40.4 to 79.0 kt yr^−1, especially during the cool season. Hypothetically removing agricultural (non-agricultural) NH₃ emissions in Guangdong provided provincial PM_2.5 reductions of 3.7 ± 1.5 (0.7 ± 0.4) μg m^−3 and avoided premature deaths of 3251 (1064), highlighting the need to combine dispersed agricultural source controls and targeted non-agricultural source controls over populated PRD.