Analyzing Seasonal and Interannual Changes in the West African Monsoon with Machine Learning Approaches: Self-Organizing Mapping
Abstract
Here we present findings on recent interannual and seasonal-scale changes in the West African monsoon system. Using self-organizing mapping (SOM) for inductive pattern discovery, we examine the distribution and characteristics of rainfall over space and time at a finer level of detail than is possible using single or multi-variate spatially constrained indices. Unlike more traditional empirical orthogonal function analysis, SOM produces physically representative patterns that we use to investigate precipitation patterns in this region: spatially, annually, and intra-annually. Existing literature demonstrates an increase in recent years in the variability of annual precipitation in West Africa; projections indicate a possible shift in precipitation toward later in the season. Through self-organizing mapping, we examine the role of the myriad underlying spatio-temporal precipitation patterns in the overall changes in variability and timing of precipitation. We distinguish, at inter- and intra-annual time scales, the contributions of: changes in the timing of shifts between precipitation patterns; changes in the precipitation patterns themselves; large scale precipitation patterns associated with the monsoon; and small scale patterns associated with mesoscale systems. We find that observed changes in rainfall dynamics, regionally and sub-regionally, are associated with both changes in the timing of shifts between rainfall patterns and changes in the rainfall patterns themselves. The spatial-temporal analysis of precipitation dynamics in West Africa points toward monsoon timing as the driving factor behind broader precipitation trends observed in the region. We compare SOM patterns derived from observational data against those derived from model data to evaluate model performance. Comparison of observation- and model-derived SOM patterns is a novel spatially and temporally detailed metric. Examining precipitation in West Africa with the higher-dimensional, more flexible method of self-organizing mapping provides a new tool for disentangling the underlying behavior responsible for precipitation changes in this region, indicating the systems that will dominate changes in precipitation over West Africa as climate change progresses.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2016
- Bibcode:
- 2016AGUFM.A51C0040V
- Keywords:
-
- 3305 Climate change and variability;
- ATMOSPHERIC PROCESSESDE: 3319 General circulation;
- ATMOSPHERIC PROCESSESDE: 3354 Precipitation;
- ATMOSPHERIC PROCESSESDE: 1620 Climate dynamics;
- GLOBAL CHANGE