ContextThe main purpose of statistical downscaling methods is to model the relationship between large-scale atmospheric circulation and climatic variables on a regional and subregional scale. Downscaling is an important area of research as it bridges the gap between predictions of future circulation generated by General Circulation Models (GCMs) and the effects of climate change on smaller areas. In this study we present the first results of a statistical downscaling model, using a neural network-based approach by means of multi-layer perceptron networks. As predictands, various indices associated to temperature and precipitation extremes in Romania are used over the entire country (for temperature extremes) and on selected homogenous areas (for precipitation extremes). Several large-scale predictors are tested, in order to select the optimum statistical model for each predictand. Predictors are considered separately or in various combinations.

Statistical downscaling of summer extremes in Romania using artificial neural networksMarius-Victor Birsan & Alexandru Dumitrescu

Meteo Romania (National Meteorological Administration), Department of Climatology, Bucharest, RomaniaE-mail: marius.birsan@gmx.com

Acknowledgements. We thank Dr. Aristita Busuioc for her help on selected large-scale predictors. This work has been realised within the research project Changes in climate extremes and associated impact in hydrological events in Romania (CLIMHYDEX), code PN II-ID-2011-2-0073, financed by the Romanian Executive Agency for Higher Education Research, Development and Innovation Funding (UEFISCDI).

MethodsArtificial Neural Networks (ANN) are mathematical models inspired by the structure and operation of biological neural networks. ANN is a densely interconnected network of independent adaptive processing units called neurons. Neurons in an ANN are arranged in a layered structure. Neurons are connected to those in the adjacent layers. Each input connection to a neuron has an associated adaptive weight to model the synaptic learning. There are no connections between the neurons within a layer.

A multi-layer perceptron (MLP) ANN consists of an input layer, an output layer and one or more hidden layers. The input layer is processed forward through the hidden layer(s) to calculate the output at the output layers. This one-directional forward process is called feed-forward method.

Large-scale predictors• Principal Components PC1, PC2, PC3 of temperature at 850 hPa (domain: 20-30E, 40-50N);• Principal Components PC1, PC2, PC3 of sea-level pressure (domain: 5-45E, 30-55N);• Principal Components PC1, PC2, PC3 of specific humidity at 700 hPa (domain: 20-30E, 40-50N).

Indices of extremesFrtmax90 – frequency of very warm days (Tmax ≥ 90-th percentile computed for the reference period

1971-2000); Dtmax90 – maximum duration of very warm days (consecutive days with Tmax 90-th percentile); ≥Frtmin90 – frequency of very warm nights (Tmin 90-th percentile); ≥Dtmin90 – maximum duration of very warm nights (consecutive days with Tmin 90-th percentile); ≥Frpp90 – frequency of days with exceding precipitation (daily cummulated precipitation 90-th ≥

percentile computed for the reference period 1971-2000); Dmaxpp0 – maximum duration without precipitation (consecutive days with precipitation < 0.1 mm).

Preliminary resultsThe period 1962-1990 was used for training, the validation being done for 1991-2010. The MLP-ANN performed well for all temperature-related indices.

Input Hidden layer(s) Output

Correlation coefficients between summer FRTMAX90 index and: trained data series,1962-1990 (left); validated data series, 1991-2010 (right).

Regarding precipitation-related indices, the model managed to capture rather well the dynamics, but the extremes were generally underestimated.

Romania is the largest country in Eastern Europe, with an area of 238,391 km². The terrain is fairly equally distributed between mountainous (Carpathians), hilly and lowland territories. Elevation varies between zero and 2544 m.a.s.l. It has a transitional climate between temperate and continental with four distinct seasons, and with various climate influences, reflected by the hydrological regime: oceanic (western part) Mediterranean (South-West), Baltic (North), semi-arid (East), and Pontic (South-East).