Large-Scale Circulation Anomalies Conducive to Extreme Precipitation Events and Derivation of Daily Rainfall in Northeastern Mexico and Southeastern Texas

Abstract

The severe impacts of climate variability and climate hazards on society reveal the increasing need for improving regional- and local-scale climate diagnosis. A new downscaling approach for climate diagnosis is presented here. It is based on artificial neural network (ANN) techniques that derive relationships from the large- and local-scale atmospheric controls to the local winter climate. This study documents the large-scale conditions associated with extreme precipitation events in northeastern Mexico and southeastern Texas during the 1985?93 period, and demonstrates the ability of ANN to simulate realistic relationships between circulation?humidity fields and daily precipitation at local scale. The diagnostic model employs a neural network that preclassifies the winter circulation and humidity fields into different patterns. The results from this neural network classification approach, known as a self-organizing map (SOM), indicate that negative (positive) anomalies of winter precipitation over the study area are associated with 1) a weaker (stronger) Aleutian low, 2) a stronger (weaker) North Pacific high, 3) a negative (positive) phase of the Pacific?North American pattern, and 4) cold (warm) ENSO events. The atmospheric patterns classified with the SOM technique are then used as input to another neural network (feed-forward ANN) that captures over 60% of the daily rainfall variance over the region. This further reveals that the SOM preclassification of days with similar atmospheric conditions succeeded in emphasizing the differences of the atmospheric variance that are conducive to extreme precipitation. This resulted in a downscaling model that is highly sensitive to local- and large-scale weather anomalies associated with ENSO warm events and cold air outbreaks.