Abstract:
Droughts are continuous periods of precipitation that are much below average can
cause water shortages, decreased soil moisture, and a number of negative impact on
ecosystems, agriculture, humans and wildlife. RDI are typically used for monthly and
annually data processing to simultaneously calculate precipitation and potential evap otranspiration on an appropriate time scale. Study improves the methods for choosing
predictors and create an innovative model to forecast drought in Punjab, Pakistan.
This study analyzed precipitation and temperature data from 9 meterological site us ing RDI. This study aim is to compare four well-known models: Autoregressive Inte grating Moving Average (ARIMA), Support Vector regression (SVR), Neural Network
Auto Regressive (NNAR) and Dynamic Linear Model (DLM) in order to assess and
compare the effectiveness of various forecasting model in predicting RDI. The training
(January 1961 to November 2000) to evaluate its performance and for testing data
(December 2000 to December 2016) to estimate their performance. ME, RMSE, MPE
and MAPE for performance measure. And for model selection we use AIC and BIC.
The selection of best model is based on minimum value of RMSE, ME, MPE, MAPE,
AIC and BIC. The evaluation results suggest that NNAR outperforms the other three
models. The study found that the NNAR model demonstrated the highest accuracy
and prediction for RDI forecasting. The NNAR model performed better than other
models capturing RDI changes with unique accuracy. These results demonstrate the
effectiveness of the NNAR model in predicting drought conditions and to evaluate the
model selection for RDI prediction. These finding help improve planning and manage ment efforts for droughts, enabling efficient use of resources and decision-making. The
results of this study provide proof for the decision, particularly droughts, as well as for
creating drought mitigation efforts and applying plan into action to lessen the impact
of drought in Punjab.
