Diagnosis of insidious data disasters

Everyone taking field observations has a story of data collection gone wrong, and in most cases, the errors in the data are immediately obvious. A more challenging problem occurs when the errors are insidious, i.e., not readily detectable, and the error-laden data appear useful for model testing and development. We present two case studies, one related to the water balance in the snow-fed Tuolumne River, Sierra Nevada, California, combined with modeling using the Distributed Hydrology Soil Vegetation Model (DHSVM); and one related to the energy balance at Snoqualmie Pass, Washington, combined with modeling using the Structure for Unifying Multiple Modeling Alternatives (SUMMA). In the Tuolumne, modeled streamflow in 1 year was more than twice as large as observed; at Snoqualmie, modeled nighttime surface temperatures were biased by about +10°C. Both appeared to be modeling failures, until detective work uncovered observational errors. We conclude with a discussion of what these cases teach us about science in an age of specialized research, when one person collects data, a separate person conducts model simulations, and a computer is charged with data quality assurance.