Forecasting surface water-level fluctuations of a small glacial lake in Poland using a wavelet-based artificial intelligence method

Adam Piasecki,Jakub Jurasz,Jan Franklin Adamowski

doi:10.1007/s11600-018-0183-5

Adam Piasecki, Jakub Jurasz + Show 1 more

Open Access

https://doi.org/10.1007/s11600-018-0183-5

Copy DOI

Abstract

Lake waters are a significant source of drinking water and contribute to the local economy (e.g. enabling irrigation, offering opportunities for tourism, waterways for transport, and meeting utility water demands); therefore, the ability to accurately forecast lake water levels is important. However, given the significant lack of research with respect to forecasting water levels in small lakes (i.e. 0.05 km2 < area < 10 km2), the present study sought to address this knowledge gap by testing a pair of hypotheses: (1) it is possible to forecast water levels in small surface lakes using artificial neural networks (ANN), and (2) better water-level forecasts will be obtained when the wavelet transform (WT) is used as an input data pre-processing tool. Based on an analysis of a case study in Lake Biskupinskie (1.16 km2) in Poland and based on a range of model performance statistics (e.g. mean absolute error, root mean square error, mean squared error, coefficient of determination, mean absolute percentage error), both hypotheses were confirmed for monthly forecasting of lake water levels. ANNs provided good forecasting results, and WT pre-processing of input data led to even better forecasts. Additionally, it was found that meteorological variables did not have a significant impact in forecasting water-level fluctuations. In light of the results and the limited scope of the present study, proposed future research directions and problems to be resolved are discussed.

Highlights

Lake waters are a significant source of drinking water and contribute to the local economy; the ability to accurately forecast lake water levels is important
Based on an analysis of a case study in Lake Biskupinskie (1.16 km2) in Poland and based on a range of model performance statistics, both hypotheses were confirmed for monthly forecasting of lake water levels
Based on the methods of WLF forecasting developed for large lakes, forecasting applications in this particular study were developed by means of artificial neural networks (ANNs) and wavelet transforms (WTs)

Summary

Introduction

Lake waters are a significant source of drinking water and contribute to the local economy (e.g. enabling irrigation, offering opportunities for tourism, waterways for transport, and meeting utility water demands); the ability to accurately forecast lake water levels is important. While small lakes (i.e. 0.05 km2\ area \ 10 km2; Silvestri 2010) are of lesser global significance, they are often very important to local ecosystems and economies This constitutes a valid reason to attempt to forecast WLF in the smaller and far more numerous small lakes which dominate the landscape of high-latitude regions where the effects of climate change are evident (Kundzewicz 2011). Susceptible to natural or anthropogenic changes, these smaller lakes constitute an essential element of local ecosystems and harbour habitats which support numerous plant and animal species As a result, these lakes represent an ideal indicator of the mainly climate-driven changes that may occur in a given region’s water resources. Given the lack of studies that have explored artificial intelligence and WT methods for WLF forecasting in small lakes, the present study aimed to evaluate, for the first time, whether: (1) small lake water levels could be accurately forecast using ANNs, (2) ANNs fed with wavelet-transformed data would yield better results in terms of forecast accuracy, and (3) certain input data had a greater impact on the accuracy of monthly WLF forecasting

Objectives

Methods

Results

Discussion

Conclusion