气候变化领域集成服务门户

Autoregressive integrated moving average (ARIMA) is a data mining technique that is generally used for time series analysis and future forecasting. Climate change forecasting is essential for preventing the world from unexpected natural hazards like floods, frost, forest fires and droughts. It is a challenging task to forecast weather data accurately. In this paper, the ARIMA based weather forecasting tool has been developed by implementing the ARIMA algorithm in R. Sixty-five years of daily meteorological data (1951-2015) was procured from the Indian Meteorological Department. The data were then divided into three datasets- (i)1951 to 1975 was used as the training set for analysis and forecasting, (ii)1975 to 1995 was used as monitoring set and (iii)1995 to 2015 data was used as validating set. As the ARIMA model works only on stationary data, therefore the data should be trend and seasonality free. Hence as the first step of R analysis, the acquired data sets were checked for trend and seasonality. For removing the identified trend and seasonality, the data sets were transformed and the removal of irregularities was done using the Simple Moving Average (SMA) filter and Exponential Moving Average (EMA) filter. ARIMA is based on method ARIMA (p,d,q) where p is a value of partial autocorrelation, d is lagged difference between current and previous values and q is a value from autocorrelation. In the present study, we worked on ARIMA (2,0,2) for rainfall data and ARIMA (2,1,3) for temperature data. As a result, it estimated the future values for the next fifteen years. The root means square error values were 0.0948 and 0.085 for rainfall data and temperature data respectively which show that the algorithm worked accurately. The resulted data can be further utilized for the management of solar cell station, agriculture, natural resources and tourism.