Identify Precipitation Pattern Using Multi-scale Sample Entropy

Abstract:

In an effort to seek new perspectives on identifying precipitation patterns associated with the precipitation time series, this study explored the potential use of the information metrics through Multi-scale Sample Entropy (MSE) analysis. The objectives were to develop MSE analysis in investigating if discernable changes in long term patterns could be identified when the information metrics in the data were studied in terms of how they change with scales. Scales, in the present context, are the intervals of days that sample entropy (SE) is sampled within a time series. For this study we looked into the characteristics of precipitation before and after 1980 for the regions upstream of Yangtze River in southwestern China, based on the daily rain-gauge data collected from 70 gauges since 1951. The results suggest three main patterns of SE with scale, they are: significant decrease, relatively flat and significant increase. These three patterns correspond, respectively, to the downstream, midstream and upstream of the upper Yangtze River region. By the nature of entropy, a significant decrease in SE implies more regularity with scale, which could mean a longer continuous drought or a more evenly distributed continuous precipitation. In this case, our analysis shows that it is attributed to the longer continuous drought. For the case of significant SE increase, it was found to be tied to an increase in the rain frequency. These results appear to show that the MSE analysis could indeed be useful for long term precipitation study.