The development of snowmelt runoff models in the Scottish Highlands
Detailed snow surveys were carried out in the Allt a Mharcaidh catchment on the western edge of the Cairngorm mountains during the winters of 1985/86,1986/87 and 1988/89. Snowpack data collected included depth, density, areal extent and water equivalent. From these data it was possible to determine seasonal patterns in snowpack behaviour and relate these to the initial snowpack water equivalent volume and timing of the snow accumulation and ablation. Using meteorological and flow data collected in the Mharcaidh by the Institute of Hydrology as part of the SWAP project simple linear regression relationships were determined. These indicated that the availability of detailed meteorological data did not improve the ability to simulate observed flow and that a successful regression could be established using simple and readily available data. Using this data temperature index models were developed and tested on the Mharcaidh. These showed that the mean daily temperature provided a better index of melt than more complex indices and that simple changes regarding the addition of a freezing level hindered the model performance despite being closer to reality than other assumptions made in the model. This suggested that the degree of complexity in the model has to be similar for all operations to obtain optimum results; having one particularly complex sub-model reduces the performance of the others. Two other types were tested on the Mharcaidh based on the layered structure developed by Martinec (1975) and Anderson's (1968) method using temperature and windspeed as an index to the energy changes at the snowpack boundary during rain-on--snow events. These again show that simple methods using readily available data can produce acceptable results and that increasing the complexity of the model does not produce a similar increase in performance. The three different models were then run on different datasets for different catchments and years. The dependence of Anderson's method on good quality data is highlighted suggesting that it is not as widely applicable as the other models. The level of performance for all models is related to the extent and depth of the snowpack indicating that further improvements may be necessary to the hydrological components of the model rather than the melt sub-model itself. The models were tested in simulated real time conditions on one dataset and, following this, guidelines for use in real time to predict snowmelt runoff are given.