Abstract
{ "background": "Crop monitoring in semi-arid regions remains challenging due to sparse ground data and variable agro-climatic conditions. Existing remote sensing methods often lack integration with in-situ agronomic diagnostics, limiting their operational utility for smallholder systems.", "purpose and objectives": "This paper develops and validates an integrated framework combining satellite-derived indices with diagnostic agronomic modelling to improve the accuracy and actionable insight of crop monitoring for staple cereals in a North African context.", "methodology": "We fused Sentinel-2 time-series data with systematically collected field-level agronomic data on crop status and management. A hierarchical Bayesian model was employed to estimate crop performance, formalised as $y{it} \\sim \\mathcal{N}(\\alphai + \\beta X{it}, \\sigma^2)$, where $y{it}$ is the observed yield for field $i$ at time $t$, $\\alphai$ is a field-specific random intercept, and $X{it}$ is a vector of vegetation indices and agronomic covariates. Model inference used Markov Chain Monte Carlo sampling.", "findings": "The integrated model explained over 70% of the variance in final yield, a 25-percentage-point improvement over a remote-sensing-only baseline. The posterior distribution for the key agronomic covariate (nutrient status index) indicated a positive effect, with a 95% credible interval of [0.14, 0.27] on the standardised yield scale.", "conclusion": "Integrating diagnostic agronomic variables with earth observation data significantly enhances the explanatory power and practical relevance of crop monitoring models, moving beyond phenological detection towards identifying causal factors of performance gaps.", "recommendations": "Agricultural extension services should adopt integrated monitoring protocols that pair satellite analytics with targeted ground diagnostics. Further research should focus on scaling the framework through participatory data collection and digital platforms.", "key words": "precision agriculture, Sentinel-2, hierarchical modelling, smallholder farmers, yield gap, Bayesian inference", "contribution statement