6 — Data
Thesis · APA section — sources, access, harmonization, variables
Thesis · Section 6 of 7 · Paper landing
6.1 Source datasets
| Dataset | Variable | Resolution | Provider | Reference |
|---|---|---|---|---|
| HMIS Typhoid Cases | Monthly OPD typhoid case counts | District-month | MoHP / DoHS, Nepal | (Ministry of Health and Population, Government of Nepal, 2023) |
| DRR Flood Events | Flood event counts | District-month event records | NDRRMA / MoHA, Nepal | (NPDRR Nepal, 2025) |
| CHIRPS v2.0 | Precipitation (mm) | 0.05° → district mean | Climate Hazards Group, UC Santa Barbara | (Funk et al., 2015) |
| ERA5-Land | 2 m air temperature, humidity | 0.1° → district mean | ECMWF · Copernicus C3S | (Muñoz-Sabater et al., 2021) |
| District shapefiles | Administrative boundaries | District polygons (2021 census) | CBS, Nepal | (Central Bureau of Statistics, 2021) |
6.2 Dataset descriptions and credibility
6.2.1 HMIS Typhoid Incidence
The HMIS is the official national disease surveillance system of Nepal, operating under MoHP statutory authority since the 1990s. Used in national health planning, WHO country reports, and published academic research as the primary reference for Nepal’s communicable disease burden. After harmonization, the HMIS panel comprised 9,790 district-month records, of which 7,327 had non-zero, non-missing case data for analysis.
6.2.2 DRR Flood Events
The DRR Portal operates under the Disaster Risk Reduction and Management Act 2017 and represents the official disaster event registration system. The 2015–2023 dataset contained 873 flood-type event records, with 86 % during June–September.
6.2.3 CHIRPS Precipitation
CHIRPS (Funk et al., 2015) combines satellite-derived Cold Cloud Duration with quality-controlled rain-gauge measurements at 0.05° (~5 km) spatial resolution. Validation studies for Nepal report Pearson r = 0.75–0.92 against gauge measurements for monthly totals.
6.2.4 ERA5-Land Temperature and Humidity
ERA5-Land (Muñoz-Sabater et al., 2021) provides land-surface data at 0.1° (~9 km) spatial resolution and hourly temporal resolution from January 1950, with 4D-Var data assimilation. Mean absolute errors below 1.5 °C for monthly mean temperature over South Asia.
6.2.5 CBS District Shapefiles
District polygon boundaries (2021 Census; 77 districts across 7 provinces) for area-weighted spatial aggregation of CHIRPS and ERA5-Land grids (Central Bureau of Statistics, 2021).
6.3 Access conditions
| Dataset | URL | Access |
|---|---|---|
| HMIS | hmis.mohp.gov.np | Restricted (MoHP authorization) |
| DRR Portal | drrportal.gov.np | Open (public) |
| CHIRPS | chc.ucsb.edu/data/chirps | Open (no registration) |
| ERA5-Land | cds.climate.copernicus.eu | Free (Copernicus CDS account) |
| CBS shapefiles | cbs.gov.np | Open (public) |
6.4 Raw and engineered variable definitions
6.4.1 Raw variables
| Variable | Definition | Unit | Range | Notes |
|---|---|---|---|---|
cases |
Monthly OPD enteric-fever cases per district | count | 1 – 2,772 | Primary outcome |
precip |
District-mean precipitation (CHIRPS) | mm | 0 – 1,100 | Area-weighted |
temp_mean |
District-mean 2 m air temperature | °C | −18 to +31.5 | Mean of daily |
humidity |
District-mean 2 m relative humidity | % | 25.7 – 95.7 | Derived from specific humidity |
flood_events |
Count of flood events recorded | count | 0 – 8 | DRR Portal aggregation |
6.4.2 Engineered features (used in ML models)
| Feature | Derivation |
|---|---|
log_cases |
log(cases + 1) — primary ML outcome |
cases_lag1 |
cases shifted 1 month within district |
precip_lag1 |
CHIRPS precipitation shifted 1 month |
temp_mean_lag1 |
ERA5-Land temperature shifted 1 month |
humidity_lag1 |
ERA5-Land humidity shifted 1 month |
flood_lag1 |
DRR flood count shifted 1 month |
precip_roll3 |
Rolling mean precipitation over t, t−1, t−2 |
temp_roll3 |
Rolling mean temperature over t, t−1, t−2 |
monsoon |
1 if month ∈ {Jun, Jul, Aug, Sep}; else 0 |
month_sin, month_cos |
sin/cos(2π × month / 12) — cyclical encoding |
All lag and rolling features computed within each district separately to prevent cross-district data leakage.
6.5 Harmonization pipeline
- Calendar conversion — Bikram Sambat → Gregorian via 12-month mapping table (max alignment error ~15 days).
- District-code reconciliation — CBS 3-digit codes as primary spatial merge key; 12 name-spelling variants resolved.
- Numeric cleaning — Comma-formatted HMIS counts parsed; 847 records corrected.
- Spatial aggregation — CHIRPS and ERA5-Land grids aggregated to district polygons by area-weighted zonal statistics.
- Missing-data handling — 2,463 district-months excluded; no outcome imputation (~25 % missing, consistent with documented HMIS completeness; (Bhatt et al., 2023)).
After harmonization, N = 7,327 valid district-month observations remain across 77 districts and 108 months.
6.6 Data quality and limitations
- HMIS under-reporting is likely most severe in Karnali and Sudurpashchim Provinces where facility density is lowest.
- Diagnostic misclassification — HMIS data based primarily on clinical diagnosis; may include malaria, dengue, leptospirosis without lab confirmation (Parry et al., 2002).
- Spatial aggregation of CHIRPS and ERA5-Land smooths over intra-district topographic and climatic heterogeneity.
- Flood-event underregistration — small-scale localized events may fall below formal-registration thresholds.
- Ecological fallacy — findings describe population-level associations and cannot infer individual-level risk.
6.7 Reproducibility
cd thesis
python3 main.pyReads CSV source files from thesis/data/; supporting modules in thesis/src/. Outputs to thesis/output/figures/, thesis/output/tables/, thesis/output/analysis/, and thesis/output/models/.
Python software dependencies
| Library | Version | Purpose |
|---|---|---|
| pandas | 1.5.3 | Data manipulation; calendar conversion; aggregation |
| numpy | 1.23.5 | Numerical computation |
| scikit-learn | 1.3.0 | RF and MLP training; CV; StandardScaler |
| xgboost | 1.7.6 | XGBoost training; Bayesian HPO |
| scipy | 1.10.1 | Ensemble weight optimization (SLSQP) |
| xarray | 2023.1 | NetCDF ERA5-Land extraction |
| rasterio | 1.3.6 | GeoTIFF CHIRPS raster processing |
| geopandas | 0.13.0 | Spatial aggregation to district polygons |
| matplotlib | 3.7.1 | Figure generation |
| seaborn | 0.12.2 | Correlation heatmap; boxplots |
Python 3.10 on standard laptop hardware. No GPU required. Pipeline runtime: ~2 minutes on CPU.
← Back: 5 — Discussion · → Next: 7 — Sources