Comparing British streamflow scenarios with recent observations: a tool to monitor climate impacts and adaptation?

Climate change impact projections are being used by water managers for long-term water resources planning. The UK was one of the first countries to ‘mainstream' climate change into water resources planning. Some of the earliest climate change impact studies on river flows in the UK go back to 1989. Almost ten years later, in 1998, water companies started including climate change flow factors into long-term water resources planning. There have been five generations of UK national climate change scenarios since 1991 and an equal amount of hydrological scenarios. This research project compares observed changes in British streamflow with those projected from climate-hydrological scenarios for 15 British catchments for monthly, seasonal and annual flows.

Recognising that direct trend comparisons are not helpful over short hydrometric records, and that this is a real problem for learning; a methodology employing the idea of a future scenario space was developed to enable learning through monitoring and falsification of scenario assumptions. Model catchments used in all 4 previous guidance reports (by UKWIR) were first filtered to find those with adequate hydrometric records. Streamflow data and streamflow scenario data for the catchments were then extracted and adjusted to enable comparison at monthly, seasonal and annual scales. Scenario uncertainty spaces: the graphical space in which future trends in a variable should reside if observations are to provide 'confirmation' of assumptions used in adaptation planning – were constructed for each catchment at each time period using the UKWIR scenarios, and the scenarios for natural multi-decadal variability provided by scenarios Anomaly A and Anomaly B in UKWIR02. Comparisons between the spaces and observations were analysed using, three key descriptive statistics for the period 1976-1996: (1) the number of years with data points lying outside the uncertainty space over the period – ‘exceedences'; (2) the sum of the distances (positive and negative) spanned between exceedence points and the uncertainty space at each year throughout the period – ‘distance'; (3) and the maximum range spanned by the scenarios at 2026 –‘scenario range'. Exceedance is a simple measure to show where and how often observed streamflow trends have not been within the limits of given scenario expectations, the distance statistic compliments exceedances by giving an indication of the extent to which scenarios have failed to capture observed trends throughout the period. Scenario range describes the precision of the combined scenario sets and representation of natural multidecadal variability; and also acts as a proxy for the extent of uncertainty about future trends in streamflow for the 2020s.

Overall in our sample of UK catchments for the period 1976-1996, key findings are that just 18.5% of cells (where each cell represents a graph) show no exceedences at all, and that 24.5% of cells were found to have at least 50% of the observed series (n>=10) outside of the bounds for natural variability and climate change. Positive exceedences were more than twice as frequent than negative exceedences (43:16) with that largest distances exclusively occurring in the positive direction for catchments in the south east in autumn, and in the north during winter. This finding would seem to fit with a generally warmer, wetter UK suggested in many climate scenarios but does not reflect the seasonal and regional differences predicted by climate scenarios.

Causes for exceedences vary on a regional and temporal scale. The overarching observation is that the representation of natural multidecadal variability by scenarios Anomaly A and Anomaly B appears to be too small when compared to catchments with longer observational or backcast streamflow records. Case study catchments Eden, Medway and Thames; where a more details historical analysis of multi-decadal variability was performed to improve the scenario spaces' representations, resulted in greatly reduced exceedences and larger uncertainty bounds for natural multi-decadal variability. One key exception among these three catchments was summer in Eden where recent low streamflow trends and exceedences observed during the 1980's appear to be at the limit of the newly defined scenario space. This indicates that a more in depth analysis may be required for both historic variability, and streamflow scenarios. Historical records for both Medway and Thames indicate that recent declining trends are well within the bounds of natural multidecadal variability. The methodology developed and presented in this study provides a means of monitoring adaptation decisions based on scenarios in such as way as they may be falsified at some point in the future by ongoing observations (monitoring). In this way, despite the inability to attribute observed trends and make comparisons with climate scenarios in this way, we can learn about the decisions made based upon scenarios through falsification of the assumptions used to define scenarios spaces. This method allows learning despite uncertainty, and provides a rational for the monitoring of scenario performance over time and thus the evaluation of adaptation decisions made.