CART and PSO+KNN algorithms to estimate the impact of water level change on water quality in Poyang Lake, China

Abstract

Rapid urbanization and global warming have caused a sequence of ecological issues in China including degradation of lake water environments which is one of the many consequences. Lakes are an important part of a biological system where a plethora of amphibian plants and animals reside. Other than this, they have a noteworthy impact in providing water for landscape irrigation, for domestic utilization, and most importantly sustaining a healthy ecosystem. Poyang Lake is the largest freshwater lake of China, with its rich water and biological resources for irrigation, water supply, shipping, and regulation of the flow; additionally, this lake can relieve the impact of droughts and floods by storing huge quantities of water and discharging it during shortages. However, the water environment is a standout among the most critical issues in Poyang Lake. This paper proposes two classification algorithms, i.e., classification and regression trees algorithm and particle swarm optimization + k-nearest neighbors algorithm to build up a connection between the water level and the primary water quality parameters of Poyang Lake. Two models have been trained with 8 years of data (2002~2008) and verified with 1 year of data (2009). Water quality forecasts from the particle swarm optimization + k-nearest neighbors algorithm was observed to be better when compared with the results obtained from the classification and regression trees algorithm. Finally, the category of the water quality was evaluated using 3 years of water level data (2010~2012) as an input to the particle swarm optimization + k-nearest neighbors algorithm.