MBC Credit Scoring v2.0 utilizes the IBM/NASA Prithvi-EO-2.0 satellite model combined with SHAP explainability to refine how we analyze agricultural data in Mozambique. This research direction focuses on how advanced data science can eventually support better decision-making for smallholder farmers.
The Problem with v1.0
MBC's current approach to data analysis uses 6 factors: NDVI satellite data, mobile money history, farm size, crop type, province risk, and repayment history. It provides significant insights, but it has three critical limitations:
- NDVI only — we capture vegetation health but miss biomass, nutrient stress, and 90-day growth trends.
- Black box — when data indicates a challenge, we struggle to explain the underlying factors. Transparency remains a core goal.
- No temporal depth — a single NDVI snapshot does not capture complex seasonal patterns.
Enter Prithvi-EO-2.0
IBM and NASA collaborated to build Prithvi-EO-2.0, a foundation model trained on Harmonized Landsat-Sentinel-2 data. It is open-source on HuggingFace, processes multispectral satellite imagery, and has been proven in production for 29 million farmers through the Agri-Access framework.
Why Prithvi matters for Mozambique
Mozambique shares a similar agricultural profile with other regions featuring smallholder farms, tropical climates, and data scarcity. The model captures:
- NDVI — vegetation health (our current input).
- NDRE — nitrogen stress detection (new).
- Biomass estimation — crop yield proxy (new).
- 90-day temporal trend — growth trajectory (new).
That is 4x more data points per farmer, from the same Sentinel-2 satellite passes we already access.
SHAP Explainability: Transparency in Data
SHAP (SHapley Additive exPlanations) assigns contribution values to each input feature. Instead of a simple numerical score, we get actionable insights:
Credit Analysis: 420 (Grade C) Reason: NDVI below provincial average (-45 points) Limited M-Pesa history (-30 points) Farm size below 1ha threshold (-25 points) Crop type (cassava) lower yield variance (+15 points)
This framework allows for clear communication regarding agricultural analysis. When questions arise about why certain patterns appear, we have a data-driven answer.
The Nature of the Harvest
Agriculture as the engine of wealth: ten great books on farming, land, seed, water, risk and innovation, and who funds agriculture in Mozambique.
PDF ebook, instant download. $5 of every copy buys seed for a farmer.
The Prototype
We built a Python prototype analyzing 5 sample farmers:
| Farmer | Province | v1.0 Score | v2.0 Score | Key Factor |
|---|---|---|---|---|
| João Macuácua | Gaza | 580 | 612 | NDVI trend positive |
| Ana Sibany | Maputo | 520 | 498 | Nitrogen stress detected |
| Carlos Mondlane | Sofala | 610 | 645 | Biomass above average |
| Fatima Cossa | Zambézia | 480 | 475 | Limited mobile history |
| Domingos Tamele | Nampula | 550 | 568 | 90-day growth strong |
Inference time: 0.022 seconds per farmer. This demonstrates the efficiency of the research.
Smallholder Farmers and the Season
For a smallholder farmer in Zambezia working with the COJAZ cooperative, these technologies represent potential future improvements in yield monitoring. Accurate data helps track the health of their crops throughout the growing season, allowing for better-informed decisions regarding the application of fertilizers and the timing of the harvest.
WhatsApp Voice Explanations
The final piece: when analysis is calculated, an automated WhatsApp voice message in Portuguese can explain the result to the farmer:
"Olá João. O seu score é 612, grau B. A sua plantação está saudável, mas o seu histórico M-Pesa é limitado."
This turns complex data into an actionable path for the farmer.
What's Next
- Deploy Prithvi on MBC infrastructure — the model runs on a single GPU.
- Retrain on Mozambican crop data — 52 verified farmers is our training set; 200+ is the target.
- Develop the explainability framework to support transparency.
- Open-source the SHAP integration — building trust through radical transparency.
Frequently asked questions
What is NDVI? NDVI stands for Normalized Difference Vegetation Index, a simple graphical indicator used to analyze remote sensing measurements to assess whether the target being observed contains live green vegetation.
Why is satellite data useful for farmers? Satellite data allows for the monitoring of crop health over large areas without needing to visit every plot, providing insights into nutrient stress and growth trends.
How does SHAP improve data interpretation? SHAP provides a way to attribute the output of a model to its input features, making the decision-making process of an algorithm transparent and understandable.
Conclusion
Prithvi-EO-2.0 is free, open-source, and offers 4x more data points with advanced explainability. The technology is real, and the prototype is ready for further testing. For more information on how we verify field data, visit our Receipt Season campaign. You can also learn more about how sponsoring a farmer works or meet the farmers we currently work with.
Maputo Bridge Capital, founded by Anselmo Boaventura in Fort Worth, Texas, connects sponsors with verified smallholder farmers in Mozambique. A sponsor gives from $25, which goes directly to the farmer's local seed and fertilizer shop, with 92% supporting the farmer and 8% covering verification. Sponsorships are gifts, not investments, and are not tax-deductible. To get involved, sponsor a farmer here.