Prediction of winter rainfall over north-western india using artificial neural network approach

Abstract

India is the second-most populous country located in South Asia, and Agriculture is the backbone of the Indian economy. Agriculture in India depends primarily on rainfall, as irrigation is not well established in the country. Hence, predicting rainfall becomes essential to have an idea and plan the agriculture activities accordingly. Rainfall prediction is also helpful for disaster management (floods and drought conditions). It also helps minimize the risk of human life, flora and fauna, and infrastructure in the case of extreme events. Northwest India is the breadbasket of India, and our study focuses on the winter rainfall of this region. The Himalayas play an essential role in bringing monsoon in the India’s weather and climate, and they mainly affect the North-West Indian region as they are surrounding this region. During the winter months (i.e., December, January, February), the strong upper- tropospheric westerly winds play an essential role in bringing moisture and thus precipitation (rainfall, snowfall). The western disturbances (WD) are the cause of these westerly winds. These are the cyclonic circulations or the extra-tropical air troughs in the mid-latitude originating from the Mediterranean Sea, the Caspian sea, the Black sea that tends to move across the Himalayas from west to east, causing precipitation in northern India, Pakistan, and Afghanistan. Rainfall prediction is one of the most challenging and vital tasks worldwide, and achieving good accuracy is crucial. Prediction of northwest Indian winter monsoon is a more challenging and tedious task due to the heterogeneous topography of the region and the Himalayas surrounding the region, making the prediction more uncertain(due to unavailability of some data). Prediction of Rainfall can be made using various statistical methods, Numerical and Machine Learning methods. We used the Machine learning approach to predict the rainfall and incorporated the ANN approach to better accuracy. This approach is chosen over the others as it can deal with nonlinear inputs and has a better graphical and predictive analysis approach. It also deals with unstructured data, can handle large datasets, and fix the outliers better than any other model. Our study uses European Centre for Medium-Range Weather Forecasts Reanalysis version 5 monthly averaged data sets over the period 1950-2018 and uses the ANN approach using python programming. Local scale weather parameters such as Mean sea level pressure, Humidity, Zonal wind, Meridional or Vwind, Vertical or W wind, Total cloud cover. etc., and remote teleconnections like El Niño–Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), etc., were considered as predictor variables to achieve the task.