Application of Self-Organizing Maps to Observed and Simulated Daily Precipitation over the Tropical and Southern Pacific Ocean

Self-organizing maps (SOMs) comprise a class of artificial neural networks that aim to organize complex input data through computation of a set of M x N representative maps. Here we use an SOM routine to isolate the spatial patterns inherent in daily austral summer (December-January-February or DJF) rainfall over the tropical and southern Pacific Ocean basins from Tropical Rainfall Measuring Mission (TRMM) satellite observations as well from an ensemble of models from Phase 5 of the Coupled Model Intercomparison Project (CMIP5). Computing a 2x2 SOM from all available DJFs from 15 years (1998-2013) of TRMM data yields two maps that may be regarded as Intertropical Convergence Zone (ITCZ)-active, in which precipitation is more intense over the ITCZ region compared to the South Pacific Convergence Zone (SPCZ) region, while the remaining maps are SPCZ-dominant. The latter reflect a spatial translation of the principal SPCZ diagonal consistent with the impacts of the El Niņo/Southern Oscillation (ENSO) or analogous low-frequency modes of variability on the SPCZ as shown in prior studies. Comparing the CMIP5-based SOMs to TRMM reveals some broad similarities in the orientation and extent of large-scale features, as well as spurious features like the spurious Southern Hemisphere ITCZ in the Eastern Pacific, which underscore errors or biases present in the models. Because of the pronounced impact of ENSO, we further consider SOMs computed separately for each of the El Niņo and La Niņa phases. While the overall position of the SPCZ is sensitive to the phase of ENSO, within each phase, similar configurations of the slope of the SPCZ diagonal occur. Thus, while the mean position of the SPCZ may be dominantly controlled by ENSO phase, the distinct orientations within the same ENSO phase point to higher-frequency modulation of SPCZ slope. To investigate these controls further, we construct composites of pressure-level winds and specific humidity from the Climate Forecast System Reanalysis product based on the assignment of days to each SOM. These composites show large-scale features in wind and humidity that are consistent with the rainfall maps and which we interpret in terms of previously described mechanisms of SPCZ variability.