Page 842 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Model Training Supervised machine learning models are trained on soil and weather data
and fine tuning to enable accurate predictions. As data increases, lightweight deep learning
models are fine-tuned for TinyML deployment, allowing real-time, low-power
crop recommendations directly from the sensor station without constant
cloud connectivity.

Testbeds or Pilot Pilot tested by Team GenStorm using Arduino-based TinyML models for
Deployments smart weather monitoring, as documented [1]

Code repositories Code repository containing Arduino implementation, TinyML model files,
dataset, and documented [2]

2 Use Case Description

2�1 Description

Access to real-time, localized soil and weather data is a major challenge for farmers, especially
in rural and underserved regions. Without accurate and timely information, farmers struggle
to make informed decisions about planting, fertilization, irrigation, and harvesting, leading to
lower yields, resource inefficiencies, and environmental degradation. Traditional soil testing
methods are slow, expensive, and often inaccessible, while available weather data typically
lacks the local precision needed for effective farm management.

The main objective of this project is to bridge this gap by developing an AI-enabled soil
analysis and weather station that delivers real-time, actionable data to farmers. Specifically, the
project aims to (1) collect live soil and weather metrics, (2) train predictive models tailored to
local farm conditions, (3) deploy a TinyML-enabled device offering real-time, data-driven crop
recommendations, and (4) make the generated data publicly available to inform policymaking
and support evidence-based agricultural strategies.

The technological approach combines Internet of Things (IoT) sensor networks, AI-based
predictive modeling, and edge computing using TinyML. The soil and weather station, designed
without moving parts for durability, collects critical parameters such as rainfall intensity, soil pH,
temperature, and NPK levels. Data will be streamed in real-time to cloud servers using AWS,
visualized through the Grafina platform, and used to train machine learning models that are
later fine-tuned for lightweight, on-device inference. By using TinyML, the system minimizes
dependency on constant cloud access, ensuring farmers receive instant recommendations
even in remote locations.

The expected impact includes enhanced agricultural productivity through data-driven decision-
making, optimized use of fertilizers and water, improved crop yields, and more sustainable
farming practices. On a broader scale, publicly available data will empower policymakers to
implement targeted interventions, promote farmland restoration, and drive innovation in post-
harvest technologies, ultimately contributing to food security and environmental sustainability
in Ghana and beyond.

Use Case Status: Ongoing

806