Title: Yieldguard Tz: A Systematic Review Of Machine Learning And Remote Sensing Approaches For Avocado Yield Prediction In Smallholder Farming Systems Across Tanzania
Authors: Johnson George Mlelwa,Lawi Augustino Kihumbu,Alfa Edward Chengula,Carlos Ngolongolo,Catherine Peter Swai,Baraka Mwagala,Jeza Tunje
Volume: 10
Issue: 4
Pages: 84-92
Publication Date: 2026/04/28
Abstract:
Avocado farming has emerged as one of Tanzania's most promising horticultural enterprises, yet smallholder farmers across all producing regions face persistent and severe yield variability caused by unpredictable climate patterns, the alternate bearing phenomenon, pest and disease pressure, and the near-total absence of accessible, data-driven decision-support tools. Nationally recorded production grew from approximately 50,000 metric tons in 2020 to 195,000 metric tons by 2023, yet post-harvest losses of 15-30% and income instability remain pervasive because farmers cannot reliably forecast seasonal yields in advance. This review paper analyses the current state of machine learning, satellite remote sensing, and digital agricultural advisory systems applicable to Tanzania's smallholder avocado farming context, synthesizing literature on Sentinel-2 multispectral crop monitoring, Random Forest and XGBoost yield prediction modelling, alternate bearing dynamics, and mobile-based agricultural advisory platforms in East Africa. Drawing from South African avocado alternate bearing studies achieving AUC up to 0.95 using Sentinel-2 vegetation indices and climatic variables, Tanzanian avocado suitability distribution models, national-scale crop mapping methodologies from East African smallholder systems, and existing mobile advisory platforms, the review identifies a critical research gap: no comprehensive, locally adapted system in Tanzania integrates Sentinel-2 remote sensing data, real-time climatic variables, soil information, and farmer-reported inputs into a validated machine learning yield prediction model delivered through an accessible multi-channel platform (mobile app, web dashboard, SMS/USSD) tailored for low-literacy smallholder farmers. The proposed YieldGuard TZ system directly addresses this gap through an ensemble machine learning framework (Random Forest, XGBoost) with SHAP-based explainability integrated into a Flutter mobile application and web dashboard specifically designed for Tanzanian smallholder farmers, targeting prediction accuracy of Rē ? 0.85 and RMSE ? 12 kg/tree.