Running such a large number of models with daily data at a global scale, and for such long time periods had never been attempted before. For it to be successful it needed a strict framework. It is worth remembering that these models have evolved independently over a number of years and have been subject to different institutional (and national) influences. And, each has its own operational needs. The modelling groups came together to agree the framework, and during this they established links that would be of mutual benefit during and beyond the lifetime of WATCH.
The over-riding requirement of the framework was for all models to accept the same input data – namely the WATCH Forcing Data, and WATCH Driving Data - and also non-climatic data such as land cover and irrigation. This requirement underpinned our ability to perform true comparisons.
To comply, models needed to be adapted. The changes that were necessary included moving from monthly to daily time-steps, and introducing the capability to accept and utilise new data types.
Other protocols were established within the framework, such as a standard agreement on the “land mask” (the grid cells describing the Earth’s surface that were to be used in the study), and the paths that rivers should take over the landscape.
The changes to the models were completed before the WATCH Forcing Data was available, so the inter-comparison was trialled using alternative data sets containing the agreed variables for the periods 1961-1990 and 2071-2100. Later runs used the WATCH Forcing Data.
A number of individual runs were completed. Ones that assumed natural pristine conditions were then compared with runs that included human influence. The final phase used the WATCH Driving Data, and looked at the 21st century, with the focus on the potential impacts of climate change.
The first phase revealed large disagreements between the modelled river flows. Interestingly – and a little unexpectedly – the two different types of models did not fall into two distinct camps of over-estimation and under-estimation. The exception to this was in areas where snow is a major influence on the hydrology.
The main root of the differences lay in the way that each model handled evaporation, with a number of models becoming unstable when evaporation rates changed significantly. This was a concern because changes in evaporation are an anticipated feature of climate change. Until now, the uncertainties attached to rainfall estimates have received great attention, but it was clear from the first results of Water MIP that equal attention needs to be paid to evaporation.