LAND-MAP sensor-fusion simulation: synthetic landfill survey generator, spatial-validation classifiers, value-of-information economic layer, and sensor-selection and decision-viability maps

A simulation-based methodology paper. No field, laboratory or test-bed data were used; every performance figure is a model prediction from a synthetic, physically motivated but unvalidated model. Submitted to Environmental Monitoring and Assessment.

An architecture for pre-excavation landfill characterisation — a magnetometer, electromagnetic induction, ground-penetrating radar (GPR), thermal imaging and a gas array, fused by spatial machine learning into material and gas-hazard probabilities and coupled to an explicit value-of-information (VOI) economic layer — is evaluated against the specific, falsifiable questions it raises rather than the claims originally made for it. Under a spatially blocked, non-leaky validation protocol, an XGBoost classifier reaches pooled F1 of 0.99 (metal), 0.98 (organic) and 0.87 (construction and demolition debris), against 0.32 (plastic) and 0.24 (glass); random per-cell splitting did not detectably inflate performance relative to spatial blocking (95% CI including zero for all three classifiers tested), a tested null result rather than an assumption.

The paper's central finding is that predictive utility is not decision utility, and that the economically optimal sensor is not a fixed property of the architecture: ground- penetrating radar contributes most to classification accuracy, but only the magnetometer shows a reliably positive incremental value of information at baseline economics — and that ranking itself changes under deliberately different, still economically nameable site compositions (polymer-rich, C&D-rich, glass-rich), shifting the economically preferred sensor to GPR or EMI. A two-dimensional decision-viability map over sensing-degradation and economic-adversity severity shows the resulting VOI-zero boundary is considerably more sensitive to adverse secondary-material markets than to sensor degradation itself. Pooling 40 independent survey realisations, the simulated survey has positive median VOI ($924) and a median break-even survey cost of $1,451. A field validation and falsification plan with pre-specified acceptance criteria, and a Status of Claims and Assumptions table classifying every result by evidentiary status, are given.