Satellite Estimation of River Water Level from Shoal Monitoring Data: the Case of the Transboundary Ili River (Central Asia)

Lowland rivers flowing through the loess rocks carry large quantities of suspended materials and experience active processes of channel deformation (meandering). A large number of shoals, as a result of sediment deposition, exist in the riverbeds. Shoals often change, especially during the period of high water. Such rivers do not have stable hydrological profiles, and so satellite estimates of their water content are of great practical interest. During the period of low water, the morphology of the riverbed is relatively stable and the shoals are in a semisubmerged state. Under these conditions, the proportion of flooding of shoals is very sensitive to the river water level, which creates favorable conditions for the development of satellite methods for assessing water levels. The study examined the Ili River in Central Asia with an annual outflow of about 12 km(3), which is main tributary of the large Lake Balkhash. The informative value of Sentinel-2 data (resolution 10 m) in the task of assessing water level in the river in conditions of relative low water was shown. 179 cloudless Sentinel-2 satellite images were selected for the period from April to November 2018-2022. An additional criterion for choosing the image was the water level at the Ili-Dobyn hydropost of Kazhydromet less than 280 cm. This water level was typical for about 80% of the time of the low-water period of 2018-2022. The Ili riverbed on the border of China and Kazakhstan with a length of about 40 km was used to create the shoal (sediments) mask with a total area above 4 km(2). The MNDWI1 (Modified Normalized Difference Water Index) water index and its threshold value of +0.25 were used to identify the "water" class on the shoal mask. The proportion of flooded shoals was compared with ground data from the Ili-Dobyn hydrographic station on the water level during the satellite's flight. For the April-November period, linear dependences were obtained between the proportion of shoals flooding and the river water level. The reliability of the linear approximation for seasonal data was high and varied from R-2 = 0.83 (2022) to 0.96 (2020). The entire period (2018-2022) of the analysis was characterized by R-2 = 0.93. The use of the linear regression equation as a model for predicting the water level in the river based on the proportion of flooding of test shoals provided a sufficiently high value (0.74) of the Nash-Sutcliffe hydrological forecasting efficiency parameter, which indicates the practical significance of the developed methodology.