On the Possibilities of Using Bidirectional Long Short-Term Memory for Predicting Heavy-Metal Transport with Bottom Sediments of the Terek River

Abstract

The rivers of the North Caucasus, particularly the Terek, experience significant anthropogenic loads of heavy metals (Pb, Cd, Zn), which accumulate in bottom sediments and, during flood periods, pose a risk of secondary pollution of floodplain agricultural lands. This paper presents a comparative analysis of the hydrological and sedimentological characteristics of the Terek and Yangtze rivers (Hunan Province, China). It is shown that both rivers exhibit similarities in three key parameters: a pronounced summer flood regime, the dominant role of fine suspended solids in pollutant transport, and the presence of anthropogenic sources of heavy metals (mining activities, industrial discharges). Based on the established similarities, the feasibility of adapting hybrid architectures with bidirectional long short-term memory networks and preliminary signal decomposition, which have been successfully tested on the Yangtze River, to predict Pb, Cd, and Zn concentrations in the Terek River was substantiated. The results of the study provide a scientific and methodological basis for subsequent calibration and verification of the model using our own field data, which it is planned to collect in 2026.