Making sense of imperfect data

Namibia Economic, Energy, Climate Risk & Green Hydrogen Observatory

A four-page Power BI report on public debt, inflation, electricity supply, drought risk and the green hydrogen investment pipeline. A Python pipeline builds a star-schema warehouse, checks every run against a data-quality gate that fails closed, and refreshes source data on the first of each month. Every curated figure was audited against primary sources before publication, and several of my own numbers changed as a result.

Namibia Station Monitor

An automated data-quality audit of 56 sources in a global public archive, scoring each one against a documented Data Quality Score (DQS v1.0) and republishing on the 1st of every month without supervision. Surfaced a concrete archive defect: one station present in ISD metadata since 1949 returns HTTP 404 in GHCND. Released as a citable dataset — Igulu (2026), DOI 10.5281/zenodo.21229782.

Modelling Wind Speed Extremes Using Extreme Value Theory: A Case Study for Namibia

Identified that quality control, not model choice, was the dominant source of error in national extreme-wind estimates. Removing 70 spurious records, 0.24% of observations, cut 100-year return levels by 35 to 45%. Produced the first station-level GPD return-level estimates with profile-likelihood uncertainty bounds for six Namibian stations.

doi.org/10.3390/wind6030040 ↗

SACRF — Sparse-Adaptive Climate Risk Framework

A working framework for carrying observational uncertainty through to risk estimates when the underlying record is incomplete. It builds on the extreme-value work above, and asks a question that generalises well beyond climate: how should a model behave when you can measure how bad its inputs are?

Current focus: methodological development, empirical validation, reproducible implementation.

Namibian Climate Risk Sandbox

An open, reproducible pipeline running from raw station records and CMIP6 projections through to a scenario-adjusted valuation for a real asset. A worked example of the full chain: ingest, bias-correct, model the tail, quantify uncertainty, convert to a number a decision-maker can use. Audit and bias-correction phases complete; extreme value and financial layers in build.