The calendar-window discovery
Daily satellite-gauge correlation is convention-dependent. IMERG-L is dated by the UTC calendar day; the BMKG gauges report on the local-day morning-observation window. So the two references peak at different aggregation-window offsets. Drag the dial and watch the peaks separate - the low daily correlation is a clock problem, not a rainfall problem.
vs BMKG (validation)
vs CPC-UNI (UTC label)
vs CPC-UNI (relabelled +1 day)
What the dial shows. At h = 0 (the UTC day) IMERG-L agrees best with CPC-UNI but poorly with BMKG; at h = −23 h (the local day) the picture flips, and BMKG recovers to about 0.57. No single calendar label maximises agreement with both references at once, so a corrected product is best distributed on an explicit local-observation day with all references harmonised to it. The residual after harmonisation is the genuine retrieval-timing limit - it cannot be relabelled away.
Daily vs monthly
The window fix matters only at the daily scale. Aggregate the same product to calendar months and the shift vanishes - a one-day window change just moves a day or two across each month boundary. Pooled over the GPM era (n =