Rating curve uncertainty and change detection in discharge time series: case study with 44-year historic data from the Nyangores River, Kenya

The intersection of the developing topic of rating curve and discharge series uncertainty with the topic of hydrological change detection (e.g., in response to land cover or climatic change) has not yet been well studied. The work herein explores this intersection, with consideration of a long-term discharge response (1964-2007) for a similar to 650-km(2) headwater basin of the Mara River in west Kenya, starting with stream rating and daily gauge height data. A rating model was calibrated using Bayesian methods to quantify uncertainty intervals in model parameters and predictions. There was an unknown balance of random and systemic error in rating data scatter (a scenario not likely unique to this basin), which led to an unknown balance of noise and information in the calibrated statistical error model. This had implications on testing for hydrological change. Overall, indications were that shifts in basin's discharge response were rather subtle over the 44-year period. A null hypothesis for change using flow duration curves (FDCs) from four different 8-year data intervals could be either accepted or rejected over much of the net flow domain depending on different applications of the statistical error model (each with precedence in the literature). The only unambiguous indication of change in FDC comparisons appeared to be a reduction in lowest baseflow in recent years (flows with >98% exceedance probability). We defined a subjective uncertainty interval based on an intermediate balance of random and systematic error in the rating model that suggested a possibility of more prevalent impacts. These results have relevance to management in the Mara basin and to future studies that might establish linkages to historic land use and climatic factors. The concern about uncertain uncertainty intervals (uncertainty(2)) extends beyond the Mara and is relevant to testing change where non-random rating errors may be important and subtle responses are investigated. Copyright (c) 2013 John Wiley & Sons, Ltd.