MkulimaScore Geointelligence

Governance & Provenance Charter · MS-GEO-EVID-2026

Know the evidence. Respect its scale.

Geospatial intelligence is only as trustworthy as its provenance discipline. MkulimaScore GEO separates farmer-reported claims, surveyed field boundaries, calibrated Earth observation, and model-derived risk indicators.

Core Principle
No Data ≠ Adverse Condition
Entity Separation
Farmers ≠ Farms ≠ Plots ≠ Activities
Access Control
Role-Gated Section 07 Risk Dossier

01 — Evidentiary Taxonomy

Four distinct classes of agricultural information.

Every data field in a MkulimaScore GEO farm dossier is tagged with its origin class so analysts never confuse a farmer declaration with an independent satellite observation.

CLASS 01 · GROUND ENUMERATION

Field-Enumerated & Farmer-Reported Records

Source: Pre-loan agronomic surveys, cooperative registries, and extension officer logs

Analytical Contribution
Captures primary and secondary crop enterprises, planting dates, input application, and management practices invisible from orbit.
Operational Constraint
Subject to recall bias or optimistic acreage reporting unless cross-checked against surveyed polygon geometry.

CLASS 02 · SURVEYED GEOMETRY

Field-Walked & Delineated Parcel Boundaries

Source: GNSS perimeter walks and high-resolution orthophoto boundary delineation

Analytical Contribution
Establishes exact parcel area (hectares/acres), perimeter, centroid coordinates, and spatial intersection with agro-ecological zones.
Operational Constraint
A verified farm boundary confirms agricultural land use and location; it does not constitute statutory land title or legal ownership.

CLASS 03 · EARTH OBSERVATION

Calibrated Satellite & Climate Observations

Source: Sentinel-2 L2A (10 m), MODIS (250 m), CHIRPS v2.0 (5.5 km), NASA SRTM (30 m)

Analytical Contribution
Objective, repeatable, multi-year observations of canopy vigour (NDVI/EVI), leaf moisture (NDMI), rainfall anomalies, and terrain slope.
Operational Constraint
Constrained by cloud cover in highland wet seasons, mixed boundary pixels on sub-0.15 ha plots, and sensor pixel resolution.

CLASS 04 · ANALYTICAL SYNTHESIS

Model-Derived Risk & Stability Indicators

Source: MkulimaScore GEO 7-Section Dossier & Role-Gated Risk Engine

Analytical Contribution
Combines multi-year spectral stability, rainfall departure, slope suitability, and field completeness into structured underwriting scores.
Operational Constraint
Decision-support evidence for authorised credit and risk officers; never an automated or unappealable lending verdict.

02 — Scale of Validity

Use spatial evidence strictly at the resolution it supports.

A regional rainfall grid, a catchment vegetation index, and a 10 m parcel polygon answer different institutional questions.

CHIRPS v2.0 Rainfall (0.05° / ~5.5 km)

Regional & County Scale (5–50 km)

Supported Interpretation

Monitoring seasonal onset delays, cumulative 90-day rainfall deficits, and county-wide drought exposure across a lending portfolio.

Prohibited Extrapolation

Attributing a 5.5 km grid cell value as an exact on-farm rain gauge measurement for an individual smallholder plot.

MODIS MOD13Q1 (250 m · 16-day) & Ward Aggregations

Catchment & Ward Scale (250 m – 2 km)

Supported Interpretation

Comparing ward-level phenological trajectories against 10-year historical baselines and identifying correlated catchment stress.

Prohibited Extrapolation

Inferring individual field management quality where a 250 m pixel blends multiple smallholder plots, footpaths, and homesteads.

Sentinel-2 L2A (10 m / 20 m) & SRTM DEM (30 m)

Parcel & Field Scale (10 m – 30 m)

Supported Interpretation

Inspecting parcel-specific NDVI, NDMI, EVI, SAVI, and terrain slope on surveyed polygons with interior area ≥ 0.15 ha.

Prohibited Extrapolation

Treating single-date spectral reflectance as proof of crop variety or yield volume without ground calibration.

03 — Epistemic Discipline

No data is never scored as an adverse farm condition.

In tropical highland agriculture, cloud cover and small parcel geometry frequently constrain optical satellite acquisitions. MkulimaScore GEO explicitly distinguishes between missing observations, seasonal fallow periods, and genuine vegetation stress.

Entity Accounting Rule: Farmers ≠ Farms ≠ Plots

A single smallholder household may operate multiple non-contiguous farm parcels across different elevation zones or cultivate multiple seasonal enterprises on a single parcel. All cohort registers report exact denominators for each entity type.

VERIFIED OBSERVATION

Confidence: High

Closed polygon geometry with ≥80% cloud-free Sentinel-2 pixel coverage during the active phenological window and corroborated ground enterprise record.

CLOUD-CONSTRAINED / PROVISIONAL

Confidence: Moderate

Valid parcel geometry exists, but persistent highland cloud cover reduced optical revisit frequency; relies on interpolated seasonal composite or catchment context.

OUT OF PHENOLOGICAL WINDOW

Seasonal Context

Low NDVI during post-harvest fallow or land preparation (e.g., January–February in unimodal cereal zones) reflects normal crop calendar stage, not crop failure.

NO USABLE OBSERVATION

Epistemic Null

No valid satellite pass or ground record is available for the requested period. Treated strictly as missing evidence — never scored as an adverse farm condition.

04 — Institutional Guardrails

What MkulimaScore GEO will not do.

  • No Automated Credit Rejection from Map Pixels

    Satellite vegetation anomalies trigger structured agronomic review or field verification; they are never used to automatically deny credit without human underwriting context.

  • No Cross-Enterprise Signal Misapplication

    Canopy reflectance thresholds calibrated for perennial Arabica coffee are never applied unmodified to annual tubers, zero-grazing dairy units, or aquaculture ponds.

  • No Unlabelled Synthetic or Stale Data

    Every observation in the workspace displays its exact acquisition date, sensor source, and operational vs illustrative status.

MkulimaScore GEO · Institutional Engagement

Verify provenance across the 1,147-farm Kenyan cohort.

Select any farm polygon in the interactive workspace to inspect its Section 06 Agricultural Activity and provenance badges.