| ABSTRACT | The Black Sea water budget, crucial to the thermohaline and ecological functioning of the basin, is very sensitive to the variability of precipitation over its relatively large catchment area. Estimates of the Black Sea water budgets of the past are available, as the required river inflows (as well as evaporation and precipitation over the basin) are provided by models and reanalysis products. However, despite the availability of air-sea flux projections for various future climatic scenarios for the region, it is not possible to assess the water budget, as river inflow projections have not yet been estimated. We investigate the possibility of producing Black Sea river inflow projections for the period 2006-2100, based on future scenarios of precipitation over the catchment area. This is accomplished with the aid of an Artificial Neural Network (ANN), reproducing non-linear relationships between rainfall over the catchment area and river inflow, obtained by the HYPE hydrological model (HYdrological Predictions for the Environment), ran for the period 1980-2006. Datasets of past monthly-averaged rainfall and riverine inflow are used as the ANN's input, and the next time-step inflow as the output. The ANN is trained and tested using part of the dataset as the training set and part of it as the testing set. Results show that the ANN reproduces river-inflow with a good correlation to the expected, and moreover it reproduces most of the variability in the dataset, when ran for the whole period using values of the year 1980 as initial conditions. Finally, the ANN is fed with projected precipitation over the period 2006-2100 for various scenarios, for which river inflow estimates into the basin are produced. The potential impact of the riverine contribution variability to the Black Sea functioning are discussed based on our preliminary results. |