A new forecasting model for groundwater quality based on short time series monitoring data

Abstract

Groundwater is an important part of regional water resource, rapid urban development often witness deterioration of regional groundwater quality. This paper proposed a missing-aware-weighted hidden markov model (MWMO-HMM) combining expectation maximization algorithm (EM) with a weighted multi-order HMM to build groundwater quality prediction model with incomplete short-term observations. The proposed model was used to predict hydrogen ion concentration (PH) and chemical oxygen demand (COD) of groundwater in five representative cities. The Nash-Sutcliffe model efficiency coefficients of MWMO-HMM prediction results are respectively 61.51% and 98.06%. Compared with prediction results achieved by auto-regressive and moving average model (ARMA) and gray model (GM), the results show that MWMO-HMM is superior to ARMA and GM, ARMA and GM demonstrate an unstable performance of forecasting. In addition, missing value has a greater effect on ARMA than GM. Furthermore, the integral observations filled with EM algorithm indicates that COD concentration of karst groundwater in Guizhou is affected to some extent by the surface precipitation. The proposed model can predict groundwater quality effectively and meet the management requirements in groundwater prediction based on disintegrated small sample datasets. It would assist decision makers to enhance the decision making for future sustainable development.