Risk dataset methodology
Formulas and rescale tables that drive every calculation in this workbook. See DIH-Operational-SOP-v1.docx §4.5 for the full narrative.
Hazard(i) × Exposure(i) × Vulnerability(i) = Risk(i)
All three components lie in [0, 1] so the product also lies in [0, 1]. A multiplicative form is used so that risk collapses to zero whenever any single component is absent — no hazard, no exposure, or no vulnerability means no active risk.
Hazard: validated D-class → H value
Each Inkhundla's validated D-class is adjusted to a hazard value H in increments of 0.20 across the six-class NDMC scale, where None corresponds to 0 and D4 corresponds to 1.
The D-class is derived from the CDI percentile displayed in the table. Since the steps are uniform, the hazard scales linearly.
| D-class | H value | Description | CDI percentile |
|---|---|---|---|
| Normal | 0.00 | No drought signal | > 30.01 |
| D0 | 0.20 | Abnormally dry | 20.01 – 30.00 |
| D1 | 0.40 | Moderate drought | 10.01 – 20.00 |
| D2 | 0.60 | Severe drought | 5.01 – 10.00 |
| D3 | 0.80 | Extreme drought | 2.01 – 5.00 |
| D4 | 1.00 | Exceptional drought | 0.00 – 2.00 |
Vulnerability: IPC phase → V value
IPC phase is rescaled to a vulnerability value V. The mapping is non-linear because IPC severity is non-linear.
Phase 1 is deliberately non-zero, so a severe hazard in a currently food-secure Inkhundla is not silenced by the multiplication.
| IPC phase | V value | Description |
|---|---|---|
| 1 | 0.10 | Minimal / None |
| 2 | 0.30 | Stressed |
| 3 | 0.60 | Crisis |
| 4 | 0.85 | Emergency |
| 5 | 1.00 | Famine |
Exposure composition
Exposure(i) is the arithmetic mean of four sub-indicators, each log-transformed and then min-max normalised across the 59 Tinkhundla within the current cycle.
The log step (log1p) is applied before scaling because these sub-indicators are heavily right-skewed — water demand alone spans roughly 6,600 to 414 million, a 62,000-fold range. Under plain min-max a single Inkhundla takes the value 1.00 and the median collapses to 0.004, which would make a real-but-low reading score lower than no reading at all.
The mean is taken over whichever sub-indicators are present. Cattle count has no source yet and is absent everywhere; water demand covers 45 of the 59 Tinkhundla, so the remaining 14 average two sub-indicators rather than three.
| Sub-indicator | Weight (all 4) | Coverage | Normalisation method |
|---|---|---|---|
| Land use — DVI-agri | 0.25 | 59 / 59 | Dynamic World reclass → agri-mask → zonal mean → log1p → min-max |
| Water demand | 0.25 | 45 / 59 | log1p → min-max of Inkhundla water demand (DWA / JRBA) |
| Population | 0.25 | 59 / 59 | log1p → min-max of Inkhundla population count |
| Cattle count | 0.25 | 0 / 59 | log1p → min-max of Inkhundla cattle count — no source yet |
Risk-class bands
Cut-offs used for risk classification and to match the Recommended Mitigation Actions.
| Threshold ≥ | Class | Operational implication |
|---|---|---|
| 0.5 | Very High | Immediate response; priority resource allocation |
| 0.3 | High | Response planning; activate contingency arrangements |
| 0.15 | Moderate | Monitor and prepare; early-warning triggers active |
| 0 | Low | Routine monitoring only |
Land-cover DVI weights
Per-pixel weights applied before the agricultural mask and zonal aggregation. These feed DVI-agri, the land-use sub-indicator of exposure.
Kept here for reference. DVI-agri values arrive at the workbook already computed per Inkhundla.
| Land-cover class | DVI weight | Note |
|---|---|---|
| Cropland | 0.90 | Rain-fed crops — highest drought sensitivity |
| Grassland | 0.75 | Shallow-rooted grazing |
| Shrubland | 0.55 | Deeper roots — partial resilience |
| Trees / Forest | 0.30 | Established canopy and roots |
| Water / Built / Other | 0.05 | Not directly drought-sensitive |
Notes on applying this methodology
These reference tables drive every calculation in the risk workbook. Three points matter when reading the outputs.
Exposure runs on the sub-indicators available per Inkhundla — cattle count has no source, and water demand covers 45 of 59.
Class counts vary between cycles, because the bands are fixed cut-offs rather than fixed proportions.
Any component at zero drives risk to zero, which is deliberate in the multiplicative form.
ABOUT EDM
The Eswatini Drought Intelligence Hub (DIH) is a vital initiative created through the joint efforts of the National Drought Management Center, the Ministry of Agriculture, the Ministry of tourism & environmental affairs, the Ministry of tinkhundla administration and the Eswatini Meteorological Service. This collaborative platform aims to provide timely and accurate information on drought conditions, helping to mitigate the impacts on agriculture and water resources.
