Centenary covariations of water salinity and storage of the largest lake of Northwest China reconstructed by machine learning

Xingan Jiang,Chenyu Fan,Kai Liu,Tan Chen,Zhigang Cao,Chunqiao Song

DOI: https://doi.org/10.1016/j.jhydrol.2022.128095

IF: 6.4

2022-09-01

Journal of Hydrology

Abstract:In the context of accelerating climate changes and economic boom in the past decades, lakes have undergone drasticchanges worldwide. Particularly in arid and semi-arid regions, those lakes were severely impacted and threatened due to their hydrologic sensitivity and ecological vulnerability. As the largest lake in Northwest China with arid climate, Bosten Lake provides precious water resources and ecosystem services to local communities. Although a large quantity of earlier efforts have been paid on Bosten Lake, there is still absence of tracking its changing trajectory at a long timescale (e.g., one century), which restricts the holistic understanding of the decadal periodic lake desiccations and their driving forces. This study employs a machine learning method to reconstruct the centenary covariations of water storage and salinity of Bosten Lake by integrating multi-source data. The results showed that, compared with the high stage of 6.76 × 109 m3 in 1961, the lake water storage substantially dropped twice to 3.96 × 109 m3 in 1987 and 4.67 × 109 m3 in 2013. In recent years, the lake level rose rapidly and recovered back to the comparable stage of the 1960 s by 2020. Four metrics of accuracy evaluation employed in this study indicate the reliability of the XGBoost model, with the mean absolute error of 0.31 m, mean squared error of 0.37 m, r-square of 0.85, and adjusted r-square of 0.84. The centenary reconstruction results reveal that the lake salinity underwent six-phase fluctuations with the water level and storage changes during 1920–2020, with the highest value of 1.87 g/L in 1987 and the lowest value of 1.19 g/L in 2002. During the past century, the water salinity and storage of Bosten Lake were influenced chiefly by vapor pressure and precipitation, followed by wet day frequency, daily mean temperature, and potential evapotranspiration. Moreover, the uncertainty of the machine learning model was also explored and discussed. It could be mainly associated with the data accuracy of input climate variables and the ignorance of environmental impacts from the intense agricultural activities after the 1960 s. This study is expected to advance the scientific understanding of long-term change characteristics of Bosten Lake and to provide a technical reference of reconstructing centenary hydrologic and environmental trajectory for dryland lakes.

geosciences, multidisciplinary,water resources,engineering, civil