Can Water-Detection Indices Be Reliable Proxies for Water Discharges in Mid-Sized Braided Rivers Using Coarse-Resolution Landsat Archives?

Abstract

The use of water detection (WD) indices to infer daily discharge (Qd) has a great potential to enrich needed hydrological data for understanding fluvial processes driving the morphological changes of braided rivers. However, no consensus has been reached on which one stands out for use in mid-sized braided rivers. In this study, we compared the physical characteristics of three most commonly used WD indices, the Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), and Normalized Difference Moisture Index (NDMI), for two mid-sized braided reach segments in the Qinghai-Tibet Plateau, China, that have different morphological structures. Relying on the Google Earth Engine web interface, we calculated the total mean water width (WWt) based on the detected surface-water areas (As) and braiding index (BI), as well as the mean values (m) of these indices over about four decades at the braided corridor scale (cs) (mNDWIcs, mMNDWIcs, and mNDMIcs). We then examined different responses of these indices to water and non-water features and their best threshold values for characterizing channel structures. Our analyses demonstrated that (1) NDWI and MNDWI perform well for detecting braided channel structures with the threshold of zero; (2) WWt is generally better correlated to Qd in a linear style than WD indices do, particularly when calculated from MNDWI; and (3) among WD indices calculated at the braided corridor scale, mMNDWIcs shows a better relationship with Qd than mNDMIcs does. Finally, we provided mechanisms that may explain these differences in terms of photometric discrepancies in calculating WWt and WD indices and the impact of image resolution on their calculations.