Exploring the frequency of hydroclimate extremes on the Niger River using Monte Carlo methods

Flooding and low flow events along the Niger River in West Africa can have a large impact on irrigated agriculture, regional food security and water resources. A statistical simulation framework is applied to explore the future frequencies of threshold-crossing events, focusing here on extreme streamflow values on the Niger River. The statistical simulation framework is based in large part on an earlier paper, (Siebert and Ward, 2011) that focused on low seasonal rainfall totals in the Millennium Villages Project. This methodology decomposes climate variability into global change (GC), which imposes a trend on the mean, multidecadal variability (MDV), which is simulated by an autoregressive process and interannual variability (IV), which is represented by white noise sampled from the normal and skew normal distributions consistent with parameters observed in the historical record. Monte Carlo simulations are undertaken for various combinations of the above components through the early 21st century, and the authors evaluate the extent to which future event frequencies could be estimated in real-time. Preliminary analysis shows that a lag-1 year autocorrelation of 0.6-0.7 is appropriate for several hydrological stations on the Niger River based on the historical record. Regional rainfall indices (both satellite and gauge) will also be explored along similar lines, although the autocorrelation of the rainfall tends to be weaker than for streamflow. Several Monte Carlo-based sensitivity studies are intended to explore the impact of a range of trends, lag-1 autocorrelation values, variance values and skew values on the frequency of extreme events. Simulations to date in (Siebert and Ward, 2011) and in preliminary dissertation research highlight a number of general principles. For example, relatively small changes in the average induce much larger changes in the number of threshold-crossing events. The magnitude of change is also shown to be sensitive to the threshold studied, as well as to site-specific climate features (here, coefficient of variation and skewness). The framework developed permits quantification of how, especially in the near term (of order 30 years), MDV can strongly add to uncertainty about future event frequencies: a finding that may have important implications for optimal climate risk management. The context of this work within the framework of developing dissertation research is to evaluate the sensitivity and viability of index insurance in a changing climate context. Quantifying the frequency of threshold crossing extreme events is crucial to this endeavor. However, the development of quantitative statistical simulation approaches to explore the frequency of extreme events in a changing climate can be expected to have application in a number of climate risk management settings beyond index insurance.