MAMINDA · SYNTHETIC PROTOTYPE

KCS Fusion Lab

Sentinel-2 NDVI number + crop-health CV + calibrated-sensor input fields

Maize, potatoes and tomatoes. Explore simulated yield and repayment-capacity scenarios.

The yield models are trained on synthetic data. Their input-to-yield relationships are assumptions. Real-farm accuracy, probe calibration and repayment probability have not been established. This is a software prototype.

Loading fusion models…

Crop and observation context

CropDays after planting

Use observations from the same crop field and scoring period. No future observations should enter a forecast.

1. Numeric NDVI from Sentinel-2

NDVI index scoreEnter the actual index, such as 0.65—not a colour, screenshot or percentage.NDVI input sourceSatellite acquisition dateRequired for a user-supplied Sentinel-2 value.Field / extraction reference (optional)

Use a cloud-screened field-level index. This demo accepts the number; it does not download Sentinel imagery.

2. Crop-health computer vision

CV input

Assumed healthy-class score within the declared cropThis is a simulated classifier score, not disease severity.

3. Sensor input values

Sensor input sourceCalibration record referenceThis records your reference; the app cannot verify it.Soil moisture — volumetric water content (%)Soil temperature (°C)Soil pHSoil EC (mS/cm)1 mS/cm = 1,000 µS/cm.

Use calibrated means for the same preceding 14-day window. This app does not calibrate hardware. A probe reading labelled RH% is not automatically volumetric soil moisture. Unverified NPK readings are excluded.

4. Optional repayment-capacity scenario

Illustrative amounts below use one currency and the same repayment date. These are user assumptions, not live prices.

Cropped area (ha)Share of harvest sold (%)Selling price per tonneCash costs still payable before repaymentExclude this loan repayment and costs already paid; do not count financed inputs twice.Household commitments and other debts dueOther cash available by the due dateTotal repayment due, including interest

SIMULATION-TRAINED OUTPUT

Yield scenario

Lower scenario

Central estimate

Upper scenario

Yield units: t/ha. The range describes the assumed simulator, not validated uncertainty on real farms.

Scenario	Cash available	Repayment coverage	Surplus / shortfall

No validated KCS score or repayment probability is produced. Coverage of 1.00× means scenario cash equals the scheduled repayment. It is not a loan recommendation.

Model provenance and next validation

36,000 synthetic farm-seasons were generated using documented assumed crop-growth, soil-stress and health relationships. Each crop has separate full and no-CV XGBoost models. Simulator test results are not field accuracy results. Source data and labels are entirely synthetic, even when a real photo is used at inference.

Real crop photos are analysed by the existing, separately trained 20-class YOLO26n classifier. Its healthy-class score is a proxy, not a measured percentage of healthy plants. That score's relationship to yield remains assumed in this prototype.

Next validation requires matched field NDVI, frozen CV outputs, documented sensor calibration and measured harvests. Repayment modelling additionally needs actual loan outcomes.

Fusion model, generator assumptions and synthetic-only metrics · Earlier regional baselines, retained separately