←Back

504 – Spatial Modeling on the Sphere and Over Large Domains

Time Series Analysis of New England Weather Data

at Bridgewater State University

Climate change is an issue that has been at the forefront of public concern for the past few decades and understanding it as well as the direction it is heading is of the utmost importance. There has been extensive research in the past that has addressed this issue on a global scale, yet the resulting findings have been unable to resonate with and convince populations of the local implications of such research. The purpose of this study is to analyze weather data of the New England region in the United States of America to (1) explore trends in snowfall, snow depth, precipitation and average temperature, evaluate the significance of such trends and (2) create time series models that will accurately forecast future values that may be of some use in understanding the future of weather in the region. The structure of this study was organized according to each of the six states in New England, which include Rhode Island, Connecticut, Massachusetts, Maine, Vermont and New Hampshire. For the first of the two tasks, simple linear regression models were applied to each metric against time to better visualize the linear trends inherent in the data, which are valuable for explanatory purposes. As for the second task, various time series model types were tested, but ultimately ARIMA and seasonal ARIMA models were fit to the data in an attempt to predict future monthly values for each metric. The assumptions of these models were also checked via diagnostic measures such as the residual autocorrelation function (ACF), partial autocorrelation function (PACF), the Ljung-Box statistics for the residuals, as well as the usual diagnostics for normality and constant variance. Stationarity was also considered in the creation of these models, and as a result suitable models for prediction for certain weather metrics were unable to be acquired since they failed to meet this requirement. In total, 24 linear regression models and 24 time series models were created to better understand the nature and trajectory of climate change in New England.