How does temporal variation affect the value of stream water as a medium for regional geochemical survey?

•We model time series data on stream water analytes from monitored streams.•We model spatial data from a stream survey in the same region as the monitor sites.•Temporally correlated within-stream variation appears independent between streams.•We examine the implications for streamwater as a medium for spatial survey.

Stream water is a key medium for regional geochemical survey for mineral exploration and environmental protection. However, stream waters are transient, and measurements are susceptible to various sources of temporal variation. In a regional geochemical survey stream water data comprise ‘snapshots’ of the state of the medium at a sample time. For this reason the British Geological Survey (BGS) has included monitoring streams in its regional geochemical baseline surveys (G-BASE) at which daily stream water samples are collected, over variable time intervals, to supplement the spatial data collected in once-off sampling events.In this study we present results from spatio-temporal analysis of spatial stream water surveys and the associated monitoring stream data. We show that the variability of monitoring stream data from the G-BASE surveys has a temporally correlated component which can be treated as independent between streams, and therefore as a component of the nugget (spatially uncorrelated variance) of the spatial variograms of stream water survey data. For the variables examined this component was small relative to the spatial variability, which indicates that the value of stream water data to provide spatial geochemical information is not compromised by temporal variability. However, these conclusions are conditioned on the particular data set which was collected only in the summer months, specifically to limit temporal variability. Temporal variation in stream water analyses may be less tractable in wetter conditions. We show how the spatial data from stream water surveys can be mapped by ordinary kriging, with the predictions interpreted as an estimate of the temporal (summer months) mean, and the kriging variance reflecting the partition of the nugget variance of the spatial variogram between spatial and temporal components.