Combining Field Survey and Open Geospatial Data for Transport Infrastructure Decision Support in Data-Scarce Regions

Transport decision support in regions without a road inventory is built on open geospatial layers whose reliability is usually unknown. We ask what that costs, and what a field survey is worth against it. Reading the error budget of thirteen prior studies out of their own result files rather than their prose, 23 of 42 paper-layer pairs carry no measured error at all, and every one of the thirteen studies has at least one. Where error has been measured it is poor: an agreement coefficient of 0.041 between two readings of a volunteered surface layer, and a flood detector at   that returns nothing under canopy. Taking a rural rehabilitation decision that funds the most urgent decile of 7,068 links, and drawing 1,000 truth ensembles from those measured kernels, 325 of the 707 funded links, 46 percent and 594 km, are links a perfectly informed agency would not have funded. Two results follow. Propagating the kernels the program has already measured, with no fieldwork whatever, recovers 34.2 percent of the value of perfect information; it is the largest single return in the study and it is free. And survey effort should go to market and health access before surface condition, and not to served population at any budget, whose entire value at a complete census is no larger than the Monte Carlo noise of the calculation. A hundred visits chosen by decision uncertainty are worth roughly 629 chosen at random. In a worked case on 800 links, a survey costing between GBP 3,902 and 12,697 cuts misdirected funding from 55 links in 80 to 32. The layer ranking holds in 26 of 30 sensitivity variants, and the four that disagree all move the decision rule rather than the data.