• English 
  • Spanish 

Snow Trends in Northern Spain. Analysis and simulation with statistical downscaling models

In this study we consider a database of binary snow observations (occurrence or no occurrence) in a set of stations from the Spanish National Meteorological Institute (INM) over Northern Spain. A previous quality analysis has been performed based only on the length of missing periods and the stations with less than 10% of missing data in the period 1957-2002 were selected for the study. In the first part of the work a trend analysis is performed regarding the annual number of days in which snow occurred, obtaining a significant decreasing trend in the last 15 years.

Secondly, the connection of snow with large-scale fields simulated by a General Circulation Model (GCM) is analyzed using statistical downscaling methods (in particular analog and weather typing techniques). On the one hand, we validate the probabilistic predictions using Relative Operating Characteristics (ROC) curves. However, due to the low climatological annual frequency of snow occurrence (around 6% in average), appropriate thresholds for the predicted probability have been calculated for each station to obtain deterministic predictions which can be validated with 2x2 contingency tables (for instance, typical values for HIR are around 60% whereas FARs are around 2%). Using this technique the annual number of days with snow is correctly simulated for the different stations, reproducing the observed trends. These results are based on cross-validation studies, in which a window around each of the predicted dates is removed in the analysis.

In order to check the possible extrapolation of statistical downscaling to climate change studies, the following experiment was conducted: the last 15 years of data –in which the decreasing trend was observed- were tried to simulate using the remaining 30 years as the training period to develop the statistical model. Some promising results are reported.