Benchmarking Evaluation of Sparse Gaussian-Process Temporal Fusion for Computer Agriculture Irrigation Advisory Under Communication Blackouts: An Artificial Intelligence Prediction-Interval Audit for Water-Rights Equity

Irrigation advisory systems that rely on continuous telemetry fail precisely when they are most needed: during communication blackouts, when farmers must decide whether to irrigate without fresh sensor data. The evaluation is a replication-based design study: no observational data are used or claimed. Operating characteristics are characterised under a declared variance structure with between-cluster SD = 6.0, residual SD = 4.5, and intraclass correlation = 0.640, across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. Power curves, minimum detectable effects, coefficient intervals, density estimates, power-cluster scatter and residual diagnostics are reported.52 clusters are required for 80% power at a 5-percentage-point effect, and that interval coverage degrades systematically at low cluster counts, producing equity-relevant asymmetries in advisory reliability across water-rights holders. The contribution is a formal characterisation of the prediction-interval audit as a benchmarking instrument for irrigation AI, with explicit boundary conditions and a derived equity bound.