Hybrid bias correction of daily satellite precipitation over Indonesia

This dashboard presents the findings of a hybrid bias-correction framework for daily satellite precipitation, worked through the Indonesian archipelago as a case study. Satellite rainfall from IMERG-L is corrected toward gauge observations in four stages - Linear Scaling, Empirical Quantile Mapping, a Generalized Pareto tail, and a Convolutional Neural Network refinement (together, LSEQM+DL) - then validated against an independent network of BMKG stations.

The pages let you explore what the correction achieves and where its limits lie: the daily distribution moves onto the gauge, the day-by-day timing does not, and a large part of the timing gap measured against the station network turns out to be a fixable calendar-window artefact rather than a retrieval limit. Every figure is computed from the same processing pipeline.

What the correction fixes - and what it structurally cannot

The corrected product moves to the gauge distribution, but the day-by-day timing does not improve. Daily Pearson r stays near against CPC-UNI, the calibration target, and near against the independent BMKG stations, and neither moves across the three stages. Both use the same construction - a median across the reference's own units within each dekad, then averaged over the 36 dekads - so they are directly comparable.

Part of that daily timing limit is a calendar-window artefact, and it is measurable separately. IMERG-L accumulates on a UTC day while a BMKG gauge day ends at the morning observation, so the two are offset. Re-pairing IMERG-L to the gauge day lifts the pooled daily r against BMKG from to (at h). That is a different measurement from the two above - a single pooled correlation over station-days rather than a spatial median - and it applies only to the BMKG comparison. CPC-UNI dates its totals to the UTC day like IMERG, so it carries no such offset. See the window diagnostic.

Study area & data coverage

180 BMKG stations (7 island groups) over the 0.1° IMERG footprint; hover a station for its details.

How to read this dashboard

A slide deck can carry the introduction and methods; these pages carry the findings. Move through them with the sidebar.