Using large-sample high-frequency records to optimise water quality sampling

River water quality shapes both ecosystem health and human well-being. However, rapid fluctuations in a river's water quality can emerge between sampling intervals and escape detection. These brief events are often linked to intense rainfall or post-drought flushing. Consequently, low-frequency grab sampling can result in an incomplete representation of water quality in a river. By contrast, high-frequency monitoring at hourly or finer intervals can reveal these previously hidden water quality dynamics. What remains unresolved is how frequently we must sample to detect short-lived water-quality events without exceeding realistic monitoring effort. To address this, we systematically examine the effect of sampling frequency on accurately capturing riverine water-quality dynamics, with particular focus on extreme values. We use a novel, large-sample, Germany-wide dataset of multi-year hourly river water quality records from 72 catchments, covering more than 70 per cent of Germany's land area. We focus on eight primary parameters, including conductivity, dissolved oxygen, ammonia, nitrate, pH, phosphate, turbidity, and water temperature. Each time series is sub-sampled along a continuous range of interval lengths (hourly to annually). We analyse different measurement objectives through skewness, water quality duration curves, and weighted regression on time, discharge and season (WRTDS). For each interval, we compute skewness to diagnose extreme values' behaviour, as well as annual duration curves of water quality. WRTDS was applied to determine whether a relatively simple model can overcome sampling-interval-induced inaccuracy. Our results show that low-frequency intervals of one week or longer are consistently associated with a considerable loss of information, most substantially for dissolved oxygen, pH and conductivity. This loss was pronounced for extreme values, while the mean and median were less affected. WRTDS did not substantially combat the information loss associated with increasing sampling intervals. Estimates of skewness and coefficient of variation worsened, while median values showed only minor improvements. We conclude that sampling frequency must align with the monitoring objective and be explicitly incorporated into the interpretation of the results. Coarse sampling can approximate central tendencies, but not extremes or variability. These findings underscore the need for tailored sampling strategies to optimise water quality monitoring and ensure that critical fluctuations are not overlooked.