Agricultural Research Management Information System

Home
Research Summary
Search
About Us
- ARMIS
- Brochure
Contact Us
Report
User Request
Data Input
Help
- Operation Manual
  - PDF
  - Video
- Program Area & Commodity
We have reached 37600 number of research entries at this moment.

- Logout

Research Detail

Home
Research
Detail

Multi-step Ahead Prediction of Groundwater Level Fluctuations Using Coupled Wavelet Transform and Long Short-term Memory Networks (2019-2020)

D. K. Roy
Senior Scientific Officer
Irrigation and Water Management Division, BARI, Gazipur 1701

S. K. Biswas
Senior Scientific Officer
Irrigation and Water Management Division, BARI, Gazipur 1701

K. F. I. Murad
Scientific Officer
Irrigation and Water Management Division, BARI, Gazipur 1701

Abstract/ Executive Summary:

Groundwater level prediction is important for sustainable usage of scarce groundwater reserves of an aquifer to ensure the development of a meaningful groundwater abstraction management strategy. This study evaluated the prediction accuracy and estimation capability of a deep learning algorithm, Long-Short Term Memory (LSTM) network, for multi-step forward forecast of groundwater levels at two observation wells in an aquifer system of the Gazipur Sadar Upazilla, Bangladesh. Model independent partial autocorrelation functions-based feature selection approach was used to recognize appropriate input variables for the prediction models. Root Mean Squared Error (RMSE) criterion was used to calculate the training and test performance of the LSTM models to select the appropriate numbers of hidden layers and hidden neurons within each hidden layer. The prediction accuracy of LSTM network was evaluated using five statistical performance evaluation indices: RMSE, Scatter Index, Maximum Absolute Error, Median Absolute Deviation, and a-20 index. Results revealed that the developed LSTM models were capable of predicting one-, two-, and three-week ahead groundwater levels at the observation wells GT3330001 and GT3330002. In general, the prediction performances of the LSTM models at GT3330001 were better than those at GT3330002. The overall results indicate that the proposed LSTM models could be successfully employed to predict multi-step ahead groundwater levels using previous lagged groundwater levels as inputs. For improving prediction accuracy, wavelet transform based data pre-processing may be adopted.

Keywords: DL-based prediction model, LSTM, Groundwater,

Location: Gazipur Sadar Upazilla, Bangladesh

Start Date:

End Date:

Program Area: Crop-Soil-Water Management

Commodity: Groundwater, Water management

Objectives:

To delve into the potential of a DL-based prediction model, LSTM in predicting multi-step ahead groundwater level in the selected observation wells.

Methodology:

Study area and the data: The study area is situated in the Gazipur Sadar Upazilla having an aerial extent of 446.38 km². It is located between 23.88°N and 24.18°N latitudes and between 90.33°E and 92.50°E longitudes. Pumped groundwater appears to be the prime water resource for household usage and crop irrigation. Excessive abstraction of groundwater from the aquifer has been continuing at an increasing rate every year resulting in a gradual declination of groundwater level. To model future scenarios of groundwater table fluctuations in the selected observation wells, especially to provide multi-step ahead forecast of groundwater levels, previous data on groundwater level fluctuations were used in this research. For this, historical weekly data on groundwater level fluctuations were collected from Bangladesh Water Development Board. Collected data at different observation wells were carefully checked and two observation wells, namely GT3330001, and GT3330002 were selected based on the criterion of least amount of missing entries. The observation well GT83330001 is positioned between 23.93⁰N latitude and 90.42⁰E longitude. The position of the observation well GT3330002 is between 23.96?N latitude and 90.48?E longitude. However, there were some missing values in the groundwater level datasets in the selected observation wells. These missing entries were imputed using the „moving median' approach of data imputation in which a moving median with a specified window length was used to fill missing numeric data. The observation wells GT3330001 and GT3330002 had 2012 (from 07 January 1980 to 17 September 2018) and 1937 (07 January 1980 to 26 December 2016) weekly groundwater level entries after the imputation of missing entries. The starting LSTM block utilize the networks? initial state and the starting time-step of the sequence to calculate the first output and the modified cell state. In order to calculate time step's output and the modified cell state c_t the block employs the networks? present state (c_t-1, h_t-1) and the following time phase of a sequence. There are two types of states in a layer, namely hidden state (also referred to as an output state) and cell state. The purpose of the hidden state is to contain output of an LSTM layer for any particular time step t whereas the cell state stores the evidence acquired from the prior time phases. For every single time phase, an LSTM layer either puts in evidences to or takes away evidences from the cell state. The gates are used as the controlling components of these modifications for any particular LSTM layer. The following four components are employed to regulate the cell and hidden states of an LSTM layer: a. Input gate (i): Control level of cell state update; b. Forget gate (f): Control level of cell state reset (forget); c. Cell candidate (g): Put in information to the cell state; d. Output gate (o): Control level of cell state added to hidden state.

Source of Information: Annual Research Report 2019-2020, Irrigation and Water Management Division, BARI, Gazipur

Article Link (Online Journal):

Funding Source:

1.

Budget:

Total Budget (Tk.) :

Achievement/Findings:

Precise and robust prediction of groundwater levels can be effectively employed in developing a sustainable and efficient management strategy for groundwater resources. This judicial planning will aid in optimal abstraction and usage of groundwater for agricultural, domestic, and industrial purposes. This study aimed at developing a robust prediction tool for one-, two-, and three-week ahead groundwater level fluctuations using LSTM models. The suitable weekly lag times of groundwater levels were used as inputs to the prediction models while the output from the models was the one-, two-, and three-week ahead groundwater levels. The selection of optimal combination of inputs for the models was executed through careful examination of the PACF functions. The performance comparison of the proposed models was performed by using several statistical performance evaluation indices. Results of the present study indicated that LSTM models could be used to predict multi-step ahead groundwater level fluctuations. However, adopting wavelet-based data pre-processing step could be employed to improve the performance of the developed LSTM models. Therefore, this study should be continued to evaluate the effects of data pre-processing using wavelet transforms on the prediction performances of the proposed LSTM models.

Knowledge Management: Report/Proceedings