Talk

From Pixels to Plots: Transforming Lives of African Smallholder Farmers through Geospatial Data

giovedì 23 maggio

15:25 - 15:55
StanzaPanino
LinguaInglese
Livello audienceBeginner
Elevator pitch

During this talk, we’ll explore how One Acre Fund uses python, streamlit and Google Earth Engine to support millions of smallholder farmers in Africa. We will cover how to identify individual farmer fields, assess soil conditions, give farming recommendations, and combat deforestation.

Descrizione

One Acre Fund is a non-profit organization helping smallholder farmers in Africa. We provide millions of farmers in 10 countries with practical support like quality seeds, fertilizers, market access and training. Our approach is hands-on, focusing on increasing farm productivity and promoting sustainable practices. Through our work, we aim to break the cycle of poverty for these farmers and contribute to food security in their communities.

During this presentation, I will focus on the geospatial aspect of my job as a Data Scientist. I will share my experiences in utilizing satellite imagery through Google Earth Engine, a free platform that provides petabytes of geospatial up-to-date data and powerful analysis capabilities. The focus will be on how we’ve employed Python, unlocking information about farmers’ fields.

The first part of the talk will showcase the process of accessing satellite information from Google Earth Engine. We’ll explore how to extract valuable insights, including soil composition, rainfall patterns, and temperature variations, all crucial factors in optimizing agronomic decisions.

Then I will show how we worked together with Berkeley and MIT to delineate fields accurately, by using a Residual U-Net (ResUNet), a deep learning architecture tailored for semantic segmentation. This step is key in identifying individual farmers’ plots, which makes future analysis possible.

After that, I will demonstrate how we leverage this geospatial data to recommend optimal planting dates and maize types based on specific soil characteristics, temperature profiles, and rainfall patterns (both historical and real-time). I will show how we use Machine Learning for this and how we avail the results through API endpoints.

The final part of the talk will feature a Streamlit app that visually represents all the geospatial information gathered. This interactive tool not only aids in data exploration but also serves as a model for how developers and data scientists can craft user-friendly interfaces to communicate complex agricultural insights effectively. I will show how our end users in the field use the tool to check for deforestation, which is important for the reliability of their carbon credit projects.

Throughout the presentation, the emphasis will be on open-source python packages like geemap, sklearn, streamlit and pytorch. Attendees will gain practical insights into replicating and customizing these methodologies for their specific applications. In the end, I hope the attendees learned how to use Google Earth Engine, API endpoints and Streamlit to work with geospatial data. And of course, this presentation will give insights into what it is like to be a Data Scientist working for an African NGO.

TagsVisualization, Big Data, Machine-Learning, Deep Learning, GEO and GIS
Participant

Emiel Veersma

Over the last four years, my role as a Data Scientist at One Acre Fund has involved providing support to smallholder farmers across Africa. Based in Kigali, Rwanda, I am tasked with overseeing various Data Science projects and establishing both a Data Science infrastructure and data capabilities across the organisation. The Data Science initiatives encompass Credit Scoring, Fraud Detection, Repayment Predictions, Fertilizer Price Predictions, and Yield Optimization. Our efforts in developing the Data Science infrastructure and capabilities were recognized, and we received the Dataiku Frontrunner Awards in the Best Positive Impact Use Case (Non-Profit) category.