Knowledge Base

A mixed methods approach was employed, combining random forest and convolutional neural network models trained on Sentinel-2 satellite imagery over three growing seasons (2022-2024). Ground-truth yield data was collected from 240 farms across two states, with stratified sampling to ensure representation of both irrigated and rain-fed agricultural systems...

The CNN model achieved an R² score of 0.89 on the test set, outperforming the random forest baseline (R² = 0.76) by 17%. Prediction accuracy was highest for maize (RMSE = 0.34 t/ha) and lowest for cassava (RMSE = 0.61 t/ha). Feature importance analysis revealed that NDVI temporal profiles and soil moisture indices were the strongest predictors of yield variance...

Prior studies on mobile banking adoption in developing economies have primarily relied on the Technology Acceptance Model (TAM) and its extensions. However, recent work by Aker & Mbiti (2023) argues that TAM insufficiently captures the role of social networks and trust intermediaries in rural financial decision-making. This gap motivates our mixed-methods approach combining survey data with ethnographic fieldwork...

A genetic algorithm optimization framework was developed to determine optimal solar panel orientation and tilt angles, accounting for urban shading effects from surrounding buildings. 3D building models were generated from LiDAR point cloud data for three study areas, and solar irradiance simulations were run at 15-minute intervals across a full calendar year...

This thesis presents a transfer learning approach using ResNet-50 pre-trained on ImageNet for classifying 14 crop diseases from leaf photographs. The model achieves 96.3% accuracy on a dataset of 54,000 images collected from smallholder farms, and is deployable on mobile devices for real-time field diagnosis...

Our findings demonstrate that low-cost biosand filters combined with solar disinfection (SODIS) reduced E. coli contamination by 99.2% in field trials. This two-stage treatment approach is both affordable and maintainable at the household level, making it a viable solution for communities without access to centralised water treatment infrastructure...