Case Studies

The framework applied to real data: how to read the results, a runnable Bali example, and the full Indonesia evaluation.

The framework is modular and transferable - it is not tied to any one region. These case studies show what it produces when applied to real data, at two very different scales:

Reading the Results

Start here. A short visual glossary of every metric the framework reports - what each one measures, which direction is better, and where the target sits.

Open the glossary

Bali Example

The small, runnable case. 4 BMKG stations, ships with the repo, runs end-to-end in ~15 minutes on a free Colab CPU. The executed notebooks show every cell and figure.

Open the Bali example

Indonesia

The full operational case. Inputs and outputs span 2001-2025 (on Zenodo, too large for the repo); validation is against 171 independent BMKG stations whose gauge record runs 2001-2021. Presented as curated figures and analysis.

Open the Indonesia study

Why two cases

The Bali bundle exists so anyone can run the whole pipeline and reproduce a result in minutes, without downloading anything. It is the proof that the toolchain works end to end.

The Indonesia study is what the framework was built for: a sparse-gauge tropical archipelago where a single global correction parameter would wash out real regional variability. It is also where the framework’s honest limits show up most clearly (see the correlation ceiling).

Both cases use the same src/ code and the same config structure. Swapping region is a matter of pointing the config at different inputs - see Data Preparation and Configuration.

Note

The two cases are not directly comparable on headline numbers. Bali has 4 stations clustered in the south; Indonesia has 171 spread across the archipelago. The Bali figures are useful for checking the toolchain, not for drawing statistically reliable conclusions on their own.

Back to top