BruceOnyango/east-africa-aq-analytics

Statistical analysis of weather drivers of PM2.5 in four East African cities, via the OpenAQ and Open-Meteo APIs, with an interactive Dash dashboard. # NEMA · Weather & Air Quality in East African Cities A data-analytics consultancy engagement for the **National Environment Management Authority (NEMA)**, quantifying how weather conditions drive fine-particulate pollution (PM2.5) in **Nairobi, Kampala, Kigali, and Addis Ababa**, using two public APIs. The emphasis is a defensible, end-to-end analytics workflow, from API acquisition through data-quality assessment, statistical modelling, and evidence-linked recommendations, surfaced in an interactive dashboard. > **Headline result.** Weather has statistically robust *associations* with PM2.5 > (rain scavenging, wind dispersion) but limited standalone *predictive* skill at the > daily scale. The analysis is careful to distinguish the two. --- ## Table of contents - Data sources - Quickstart - Running the pipeline - The dashboard - Testing & coverage - Project structure - Methodology - Key findings - Reproducibility & data notes --- ## Data sources Two independent APIs are merged into a single **city–day panel** on `city + date`, with timestamps aligned to each city's local timezone before joining. | Role | API | Endpoint(s) | Variables | |------|-----|-------------|-----------| | Dependent variable | **OpenAQ v3** (`api.openaq.org`) | `GET /v3/locations`, `GET /v3/sensors/{id}/measurements` | `pm25` (µg/m³) | | Independent variables | **Open-Meteo Historical** (`archive-api.open-meteo.com`) | `GET /v1/archive` (`daily` + `hourly`) | temperature, precipitation, wind speed, relative humidity | - **OpenAQ** requires a free API key (register at openaq.org). Requests are split into date chunks with retry/back-off, because large multi-year windows time out server-side. - **Open-Meteo** is keyless. Daily temp/precip/wind come from the `daily` block; humidity is aggregated to daily from the `hourly` block. Study window: **2020–2024**. Unit of analysis: **one city-day**. --- ## Quickstart **Prerequisites:** Python 3.10+ and Git. ### 1. Clone and create a v …