Spatial Awareness: The potential for satellite imaging data to drive scale in rural finance

This is a repost of blog which originally appeared at Fern Software blog here co-written by Howard Miller & Aparna Phalke of Harvesting

To provide financing to a farmer, a bank or MFI needs to have a good estimate of that farmer’s likely incomes — not just how much they will earn, but when will they earn it and the risks associated with different income streams.

For most farmers, the best predictors of cash flows are directly related to crop production. If we can predict the size and timing of a crop’s yield, and map this on to market prices, we have a good idea of how much, and when, a farmer will be able to make repayments.

But how do we predict agricultural outputs? Traditionally this has been difficult. Static data on agricultural inputs, such as usage of certified seeds and fertiliser, are useful but only tell part of the story. What if we could model yields dynamically, in a way that financial institutions could then use to model risk and loan utilisation?

Thankfully, advancements in satellite imaging data and machine learning are facilitating huge advancements in this space. Data on the soil condition, weather conditions and the cropland management practices is increasingly available, even in extremely remote regions. The temporal raw spectral band data of satellite images can be converted to meaningful spectral indices such as Normalised Difference Vegetation Index (NDVI) and Normalised Difference Water Index (NDWI) which can be used to monitor crop growth, health and status throughout the cropping cycle. The resolution of publicly available data (Landsat data at 30m or Sentinel-2 data at 10m-60m) is getting us close to being able to model yields even on very small holdings of land.

This provides a unique opportunity to drive massive scale in rural finance. Using advanced data analytics and machine learning techniques we can identify the crops planted on different parcels of land without having to visit thousands of farms. We can model the likely value and timing of harvests and build models that predict the likelihood of default for large populations of farmers. We can track progress of crops through the cropping cycle and compare it to historical satellite data to provide an early warning system for any factors that could lead to loan defaults.

Harvesting, an emerging markets focussed FinTech company, uses satellite imaging data as the foundation for AI-driven rural financing platforms for banks and MFIs. This is driven by the belief that this is the only way to effectively drive scale in remote, rural and data-poor markets. Harvesting is building the technology that transforms this information from the sky into better agricultural financing and improved livelihoods for the rural poor.