Our earlier pixel-aligned Himawari-9 study (ISRS 2026) found weak aerosol optical depth (AOD)–surface PM₂.₅ correspondence in Thailand’s Chiang Mai Basin. The diurnal cycles diverged: surface PM₂.₅ peaked under a shallow morning boundary layer, whereas AOD peaked in the afternoon. AOD added little to diagnostic models using lagged ground PM₂.₅. We now quantify vertical-state dependence, test retrieval transfer, and assess PM₂.₅ prediction without ground PM₂.₅ predictors. Using 2023–2025 observations, interaction regression tested how the AOD–PM₂.₅ association varied with TROPOMI aerosol-layer height (ALH) minus ERA5 boundary-layer height (BLH), using basin-day PM₂.₅ averages from 13 DustBoy stations and operational Himawari-9 AOD. A learned Himawari-9 Level-1 AOD retrieval was evaluated against three AERONET sites with held-out-site and held-out-year tests. Gradient-boosting PM₂.₅ models were tested at withheld basin stations, in withheld years, and at six external sites. Across 377 basin-days, the AOD–PM₂.₅ association weakened as ALH–BLH increased; substituting AERONET AOD from one in-basin site also yielded a negative interaction (95% CI excluding zero). The index cannot isolate ALH from BLH contributions. Retrieval correlations were 0.896–0.934 in held-out years and 0.731 at held-out sites. Pooled 2023 predictions at withheld stations were strong (r = 0.938; RMSE = 19.02 µg m⁻³), but same-day spatial-anomaly correlation was 0.180. Transfer beyond the basin core was limited. A matched comparison did not establish an RMSE benefit from retrieved AOD beyond Level-1 reflectance, meteorology, and terrain (station-equal ΔRMSE = −0.104 µg m⁻³; 95% CI: −1.245 to +0.815). Nominal 80% intervals covered 67.7% of held-out 2023 observations; 135 of 324 grid cells met model-support criteria on at least one day. Thus, satellite-assisted PM₂.₅ monitoring requires more than diagnosing surface–column mismatch: retrieval transfer, spatial discrimination, uncertainty, and model applicability need separate evaluation.