Winter Precipitation Forecast in the European and Mediterranean Regions Using Cluster Analysis. Issue 24 (17th December 2017)

• Record Type: Journal Article
• Authors: Totz, Sonja; Tziperman, Eli; Coumou, Dim; Pfeiffer, Karl; Cohen, Judah
• Is Part Of: Geophysical research letters. Volume 44:Issue 24(2017)
• Page Start: 12, 418
• Page End: 12, 426
• Publication Date: 2017-12-17
• Subjects: precipitation anomaly -- seasonal forecast -- cluster analysis; Geophysics -- Periodicals; Planets -- Periodicals; Lunar geology -- Periodicals; 550
• DOI: 10.1002/2017GL075674
• Languages: English
• ISSNs: 0094-8276

Abstract

Abstract: The European climate is changing under global warming, and especially the Mediterranean region has been identified as a hot spot for climate change with climate models projecting a reduction in winter rainfall and a very pronounced increase in summertime heat waves. These trends are already detectable over the historic period. Hence, it is beneficial to forecast seasonal droughts well in advance so that water managers and stakeholders can prepare to mitigate deleterious impacts. We developed a new cluster‐based empirical forecast method to predict precipitation anomalies in winter. This algorithm considers not only the strength but also the pattern of the precursors. We compare our algorithm with dynamic forecast models and a canonical correlation analysis‐based prediction method demonstrating that our prediction method performs better in terms of time and pattern correlation in the Mediterranean and European regions. Plain Language Summary: We have applied a new forecasting technique to the problem of seasonal prediction that involves machine learning. By recognizing related and reoccurring patterns in both the predictors and the predictands our new technique shows improved accuracy in predicting winter precipitation in the European and Mediterranean regions. Our demonstrated technique outperforms both statistical and dynamical models over comparable historical periods. Key Points: We introduce a new scheme to forecast winter European and Mediterranean precipitation using cluster analysis and autumn precursors Clustering analysis identified three dominant patterns of precipitation anomalies across Europe and the Mediterranean Comparison of forecasts with NMME model forecasts and CCA‐based forecasts reveals that the clustering forecasts have the highest accuracy