Home Consulting Derisking Agricultural Loans to Smallholder Farmers using Satellite Imagery(Sentinel_1) and Machine Learning Algorithms

Derisking Agricultural Loans to Smallholder Farmers using Satellite Imagery(Sentinel_1) and Machine Learning Algorithms

by Shola Ogunniyi
farmsky

Africa spends $35 Billion and Nigeria about $5 Billion annually on food importation to cushion its production deficit, even though we have a great weather condition and large expanse of arable land that support farming, we still have a very large army of poor small holder farmers. These farmers complain majorly about access to affordable finance and market price volatility as reasons for the shortfall and poor economic status. They are also faced with frequent drought, unpredictable rainfall, degraded soil, pest and diseases and all of these in no small measure has made lending to farmers a huge challenge.

It’s also unfortunate that banks find it very risky to lend to farmers because there are no proper tools for monitoring and evaluation of loan compliance. They need to visit the farms periodically to access activities on the farm, a very huge expense on the cost of providing credit to the farmers. The reluctance to adopt to ICT adoption in their operations makes lending to them harder, not to mention their old farming methods and ideologies.  

African Development Banks (AfDB), reports that less than 3 % of the total loans in Africa goes to the farming sector which accounts for over 70% of employments and more than 40% of the overall GDP of Africa. It’s obvious farmers are neglected in the share of researches to optimize growth.  

Governments across Africa are making policies to encourage the farmers especially the small holder farmers. In Nigeria for example, the federal government is insisting that lending to deposit ratio must be 65-35%, this has put more pressure on banks to lend out money or not accept deposit. But unfortunately, banks prefer to lend to manufacturing and retail sector because of faster turn around and predictable revenue.

The Nigerian Incentive-Based Risk Sharing System for Agricultural Lending (NIRSAL) is pushing the limits, they gave close to 30Billion Naira to over 100,000 farmers in 2018, but we must be quick to mention their pains of monitoring and evaluation of loan compliance as It’s humanly impossible to visit all these farms during cultivation and post cultivation operations, this inability to actively monitor the operations on the farms seriously affects ensuring the farmers use the funds in compliance to the purpose of the loans.

Several studies has been conducted on the use of remote sensing and geographic information science/technologies in monitoring farmland activities, most of them were focused on using Sentinel 2 and Landsat 1-8 satellite imaging technologies which works with optical sensors. Optical datasets derived from optical sensors are very efficient for monitoring farm activities in areas with less cloud cover or non tropics but not ideal for regions in the tropics or areas with high precipitation. The limitation of optical satellite in capturing data during peak season necessitates the need to proffer another solution to monitor changes in vegetation in times optical sensors cannot capture datasets. This research would be focused on the less investigated option of using sentinel 1 datasets (it is not susceptible to issues like cloud, darkness and fog which hampers the reliability and availability of optical satellite datasets).

Satellite images are getting cheaper with better resolutions and shorter revisit period and no cloud cover issues. Consequently, the aim of this study is to build a model that would combine microware satellite data (Sentinel 1) and machine learning algorithms to detect changes in vegetation in line to the expected farming operations throughout the planting season for lending institutions to use as a decision support system for loan performance monitoring and evaluation. The model combines Interferometry Synthetic Aperture Radar (INSAR) a type of satellite image that is not inhibit by elements of weather planetary wide.

The objectives would be to;

  1. Extract the farm fields to monitor
  2. Classify the images to identify the crops being grown on the fields
  3. Determine the NDVI for the farmland
  4. Examine the farm compliance status from the NDVI

In the next article we would share the step by step procedures we used and explain the results.

You may also like

Leave a Comment