28 INTRODUCTION Groundwater is regarded to be one of the most reliable water supply sources for meeting the demands of water for various sectors in India including manufacturing, industries, agriculture, mining, and municipal water supply. It can also be regarded as a vital source in terms of clean drinking water in the country. However, on current trends, it is estimated that 60 percent of groundwater sources will be in a critical state of degradation within the next twenty years and thus have serious implications for the sustainability of agriculture, long-term food security, livelihoods, and economic growth. As groundwater resources are more intensively used, there is an increasing need for monitoring and modeling of groundwater systems. One of the most important and critical hydrological variables is the groundwater level, which is therefore monitored and predicted frequently at different locations and at frequent time intervals. Accurate prediction of the groundwater level helps a water engineer in developing better strategies to reduce the effects of various factors leading to the progressive decline of groundwater levels and thus it assists in the better sustainable management of groundwater sources. Thus, to assess the factors governing the rapid decline of groundwater levels, modeling groundwater resources can help a water engineer in achieving the objective in a better way. Almost all the groundwater flow and available transport models solve the relevant partial differential equation by the finite element method. These modeling methods are very much data and labour intensive and costly. In the recent past, intelligent technique based ANN models is being applied intensively to gain insight into the hydrological processes due to their better performance over the traditional modeling techniques such as empirical models, statistical models (autoregressive, autoregressive moving average models), and as an alternative to the physical-based models. ANNs also treated as Universal approximators are an alternative modeling and simulation tool, greatly suited to dynamic non-linear system modeling. Another attractive featureof ANNs is that they often do not require explicit characterization and quantification of physical assumptions like traditional physical-based numerical models. ANN models also include some drawbacks when handled with a non-stationary signal of a hydrologic process that involves seasonalities that vary from a single day to several decades. J. Indian Water Resour. Soc., Vol. 41, No. 1, Jan., 2021 APPLICATION OF WANN MODEL FOR GROUNDWATER LEVEL FORECASTING IN UR RIVER WATERSHED IN TIKAMGARH DISTRICT, INDIA Ankur Kumar 1 , Ravindra Vitthal Kale 2 , Govind Pandey 3 and V. C. Goyal 4 ABSTRACT The use of Aquifers as a source of water supply is increasing on a global scale, leading to over-exploitation of available groundwater blocks. Thus, there is an increasing demand for checking the groundwater levels for better and sustainable management of groundwater resources. To acquire knowledge about the factors affecting the entire groundwater system, one should know the important variables and how they vary over time. It is well known that the groundwater head is considered to be one of the most essential hydrological variables and hence, it is monitored and predicted frequently at different locations and at frequent time intervals. Particularly, the groundwater prediction in hard rock areas is a complex task with the use of physically-based models as compared to the data-driven models. Therefore, in this study, an attempt has been made to verify the adequacy as well as the efficacy of the Artificial Neural Network model (ANN) and Wavelet-ANN conjunction (WANN) models in the prediction of groundwater levels in the Ur River watershed in Tikamgarh district of Madhya Pradesh, India. Although the Ur river basin having mainly granite type of aquifer, the obtained results reveal that the WANN and ANN models can be used to predict the groundwater levels in this watershed. The application of the ANN model in the groundwater prediction gives a higher estimate of the RMSE values during calibration and validation as compared to those obtained with the application of the WANN model for each one of the observation wells. Further, the WANN model is capable to provide groundwater level prediction with higher efficiency as reflected by higher R 2 values during calibration and validation as compared to the ANN model which indicates a substantial improvement in the model performance. Therefore, it can be concluded that the WANN model provides a significantly accurate prediction of groundwater levels as compared to the results of the ANN model. Besides, the comparison of the scatter plots of time series during calibration and validation indicates that the values of water level depth estimated by the WANN model are more precise than those estimated by the ANN. Thus, this paper reveals the significant features of ANN models for forecasting groundwater levels in hard rock aquifer and their performance enhancement with wavelet theory. Keywords: Modeling, Groundwater level depth, Artificial Neural Networks, Feedforward neural networks, Discrete Wavelet Transform, Mother Wavelet, Decomposition level, Hargreaves Temperature Model. 1. Assistant Professor, CE Dept, BIT, Gorakhpur, India Email : ankur495@bit.ac.in 2. Scientist ‘D’, WHRC, NIH, Jammu, India Email: ravikale2610@gmail.com 3. Professor, CE Dept, MMMUT, Gorakhpur, India Email: pandey_govind@rediffmail.com 4. Scientist ‘F, NIH, Roorkee Email : vcgoyal@yahoo.com Manuscript No. 1549 Received 8 January 2021; Accepted 10 May 2021