Prediction of rainfall in Dachigam catchment and generation of time series autoregressive model

Abstract

The study was conducted with the prime objective to generate a stochastic time series model, capable of predicting rainfall in Dachigam catchment area of Dal lake. It covers an area of 141 sq. km. The rainfall data of the catchment from the year 1993 to 2013 was collected and used for the generation of the model. Autoregressive (AR) models of orders, 1 and 2 were used for annual rainfall series and different parameters were estimated by the general recursive formula. The goodness of fit and adequacy of models were tested by Box-pierce portmanteau test, Akaike Information Criterion and by comparison of historical and simulated graphs. The AIC value of rainfall for AR (1) model was (234.81) which is less then the rainfall AR (2) model (241.06) and satisfies the selection criteria. The mean forecast error is also very less in case of rainfall AR (1) model. On the basis of the statistical test, Akaike Information Criterion the AR (1) model with estimate model parameters can be used efficiently for the future predictions in Dachigam catchment. The graphical representation between historical and generated correlogram has also proved a very close agreement between simulated and observed rainfall values. The coefficient of determination R2 for rainfall AR(1) model is 0.93. The comparison between the measured and simulated rainfall by AR (1) model clearly shows that the generated model can be used efficiently for the prediction of rainfall in Dachigam catchment, which can benefit the farmers and research workers for water harvesting, ground water recharge, flood control and development of their water management strategies.