Remote sensing for cassava health and yield
Researchers: Femi Adekoya; Dr Kanthu Joseph Mhango
Cassava mosaic disease is both a plant-health problem and an information problem. By the time visual symptoms are unambiguous, opportunities for surveillance, intervention and realistic yield planning may already be constrained. This project develops a sensor-to-decision framework that connects early crop signals with disease progression and production consequences.
Scientific challenge
Early disease effects can be subtle, spatially heterogeneous and confounded with water, nutrient and canopy-development differences. A classificaion model that works in one experiment/environment is therefore insufficient. Useful inference must distinguish disease-related trajectories from competing stress processes, transfer across sensing scales and remain connected to crop function.
Research architecture
The project combines leaf- and canopy-scale spectroscopy, UAV multispectral imaging, field observations and crop-growth information. High-resolution spectral measurements are used to examine where informative signals arise and whether those signals can be represented by more deployable multispectral sensors. Disease classification models will be linked to photosynthetic performance and yield forecasting models to provide conditional probabilities for disease progression, yield and their interactions. We will therefore produce a method for calibrating mechanistic and data-driven yield models with disease progression information. Radiative-transfer modelling supplies a biophysical bridge between reflectance and crop properties such as canopy structure and pigment status.
Machine-learning yield forecasting models will integrate spectral, spatial and temporal evidence rather than treating each acquisition as an independent snapshot. This dynamic layer will allow sensor-derived crop disease states to update expectations about development and production.
Intended contribution
The aim is an operationally credible workflow for surveillance and forecasting in cassava systems, high-throughput enough for field deployment, interpretable enough to support scientific scrutiny, and modular enough to adapt as sensors and datasets change. The public synopsis intentionally withholds emerging model configurations and results while the research develops.