Abstract:
The Senanayake Samudraya Reservoir in
Ampara District of Sri Lanka is a crucial piece of water
management infrastructure. It supports various sectors,
including agriculture, industry, and domestic water needs.
The country experiences variable rainfall patterns
throughout the year due to its tropical climate and diverse
topography. Catchment area of the reservoir is 994.22 sq.km
and command area is 120,254 ac (488.66 sq.km). Data for the
reservoir was collected from irrigation department, Ampara
and 26-year (1990-2015) rainfall and temperature data
collected from Department of Meteorology and Irrigation
department. This study developed a comprehensive water
level forecasting system for the Reservoir to improve water
resources management system. Both Hydrologic
Engineering Centre-Hydrologic Modelling System (HECHMS) and Soil Conservation Service-Curve number (SCSCN) methods were used to simulate the surface inflow and
compared, with SCS-CN proving more reliable for this
study. Then, the base flow of the reservoir was determined.
The Long Short-Term Memory (LSTM) recurrent neural
network (RNN) was employed to forecast water level
fluctuations. The model performance was evaluated based
on the computed statistical parameters. The performance of
model is very good with Coefficient of Determination (R2) =
0.998, Mean Square Error (MSE)= 0.0357 and Root Mean
Squared Error (RMSE)= 0.189. Finally, it can be concluded
that the model can be used effectively in forecasting water
level fluctuations, providing valuable insights for reservoir
management.
