Spatiotemporal Graph Neural Network Yield Forecasting for Smallholder Agriculture Under Rain-Gauge Network Attrition: A Computer Robustness Evaluation

Rain-gauge networks across sub-Saharan Africa are thinning at precisely the moment when smallholder yield forecasting systems have begun to depend on them. the present analysis asks a narrow question with broad consequences for the design of agricultural machine-learning pipelines: how much forecasting skill does a spatiotemporal graph neural network retain when the gauge graph it consumes is progressively degraded? Because no observational benchmark of sufficient coverage exists, the study is framed as a computer robustness evaluation rather than an empirical forecast. A formal definition of the gauge-attrition operator is given, a message-passing architecture is specified over a spatial adjacency graph, and a replication-based evaluation is designed around a before-after-control-impact estimator with a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64). Power was estimated across 2000 design cells spanning cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The design attains 80% power at 54 clusters for a 5-percentage-point effect. The contribution is a reusable robustness protocol and a set of design rules that connect gauge density to detectable forecast degradation, so that yield-forecasting systems can be stress-tested before deployment rather than after failure.