Groundwater Level Fluctuation and Prediction in The Jakarta Groundwater Basin Using Non Linear Autoregressive Exogenous (NARX)
Dani Ruswandi (a*), Selly Feranie (a), Gumilar Utamas Nugraha (b)

a) Physics Study Program, Indonesian University of Education
Jl. Dr. Setiabudhi No. 229 Bandung 40154, Indonesia
b) Research Center For Geotechnology, Indonesian Institute of Sciences
Jl. Cisitu, Sangkuriang, Bandung 40135, Indonesia

Abstract

Groundwater in the Jakarta Groundwater Basin is the focus of Jakarta residents in meeting their clean water needs, therefore it is necessary to monitor the quantity of groundwater in the Jakarta Groundwater Basin. Accurate groundwater level prediction provides important information about groundwater quantitatively and provides an overview of the condition of the aquifer. One of the models that can be used to predict groundwater level is Neural Network Non Linear Autoregressive Exogenous (NARX). In this NARX model, there are several modeling parameters that must be optimized to produce accurate model predictions. These parameters include: Exogenous input, time delay and optimization algorithm. The results of the prediction of the output of the model will be tested for accuracy using the value of Root Mean Square Error (RMSE) and Coefficient of Determination (R2). The hydrological factors that will be used as exogenous input variables are potential evaporation, precipitation and humidity. The groundwater level that will be predicted is a monitoring well located at Citra Grand Cibubur, Jalan Alternative Cibubur Km. 4 Bekasi City, West Java Province, located at 6.38 South Latitude and 106.92 East Longitude. The modeling results show that the Bayesian Regularization (BR) optimization algorithm has the most optimal results with the time delay value being dominated by a value of 50, only the exogenous humidity input has a time delay of 100. The most optimal prediction results for each NARX model are: potential evaporation exogenous input with RMSE = 1.02 x 10-6 and R2 = 0.999632- input precipitation RMSE = 2.79 x 10-5 and R2 = 0.999426 - Input humidity with RMSE = 2.46 x 10-5 and R2 = 0.99952 - input humidity RMSE = 6.74 x 10-7 and R2 = 0.999726. From the results of forecasting the depth of the groundwater level of the 4 models have a value different depths but, all prediction models experiencing a trend graph down or to say the depth of ground water getting inside

Keywords: Groundwater Level, Humidity, NARX, Potential Evaporation, Precipitation, Prediction

Topic: Physics