Exploiting Sentinel-2 dataset to assess flow intermittency in non-perennial rivers

Area of Science:

• Hydrology and Remote Sensing
• Environmental Monitoring

Background:

• Limited data on non-perennial river flow hinders water management.
• Existing hydrological models struggle to accurately predict surface water presence.

Purpose of the Study:

• To develop a method for monitoring water surface presence in non-perennial rivers using satellite imagery.
• To classify river flowing status (flowing, ponding, dry) and predict daily occurrence.
• To assess the duration of non-flowing periods in Mediterranean rivers.

Main Methods:

• Utilized multispectral Sentinel-2 images to analyze water, sediment, and vegetation reflectance.
• Identified optimal Sentinel-2 bands (SWIR, NIR, RED) for distinguishing water surfaces.
• Developed and trained Random Forest (RF) models using satellite data and field observations.
• Incorporated cumulative rainfall and air temperature as key predictors in RF models.

Main Results:

• Sentinel-2 false-color composites effectively differentiated water surfaces.
• RF models achieved high accuracy (0.82-0.97) and true skill statistics (0.64-0.95) in predicting river status.
• Annual non-flowing periods ranged from 5 to 192 days, varying by river reach.
• Identified key predictors for RF models: cumulative rainfall and air temperature.

Conclusions:

• Sentinel-2 imagery provides a robust method for monitoring non-perennial river flow status.
• The developed RF models offer accurate daily predictions of river conditions.
• This technique has significant potential for improving understanding and management of surface water resources.