Integrating WSN, citizen science, and remote sensing to power a data-driven decision support system for resilient smallholder agriculture

Low production and over-exploitation of natural resources often characterize smallholder farms. Extreme and unpredictable weather conditions, soil and land degradation, and lack of farm inputs are some of the underlying factors for this. The recent pandemic (COVID-19) and, in quite a few places, political instability have also exacerbated the poor production of these farms. Farming is a complex activity that relies on multiple intertwined factors, and understanding these at the base level is essential to improve productivity. This calls for extensive resources so a reliable and (near) real-time farm monitoring system can be constructed, which is beyond the capability of most smallholder communities.
This thesis addresses the pressing challenges of food insecurity and limited agricultural data in smallholder farming systems, with a focus on northern Ethiopia. It demonstrates the feasibility of integrating inclusive technology and advanced data analytics to support data-driven agricultural decision-making at the farm level. The work began by establishing a modular, spatiotemporal data acquisition infrastructure
that uses a wireless sensor network, citizen-generated data, and remote sensing data.
Through appropriate preprocessing and spatiotemporal resampling, these heterogeneous data sources were downscaled to farm resolution and allowed to build an Agriculture data cube (ADC) that enables granular insights. This ADC served as the foundation for a knowledge base and inference models. Specifically, a data-driven Decision Support System (DSS) was implemented that integrates fuzzy logic, machine learning, and association rule mining to infer farmlevel information such as seasonal yield and farm health. The study highlighted the importance of farm management, topography, and soil properties in predicting seasonal productivity. These models were evaluated against real and synthetic data and validated using expert input.
A core innovation of the work lies in the construction of a reliable, non-intrusive, and affordable farm-level monitoring system through low-cost IoT-WSN devices and a participatory citizen science approach. This design promotes local engagement, which particularly involves youth and women, and supports continuous data col-lection with minimal technical barriers. These efforts allow farmers to understand the link between their management practice and expected yield, and fosters proactive, evidence-based decision-making.
Although the lack of widespread farm-level validation data posed a limitation, comparisons with existing datasets suggest reasonable
accuracy. The data-driven knowledge base developed in this thesis helps to address the shortage of field experts and offers scalable support for smallholder agriculture. With further refinement, particularly by incorporating more diverse training data and broader expert validation, this framework can be adapted to other regions and scaled to support decisions at zonal, regional, or national levels.