Comparison of Surface Water Volume Estimation Methodologies That Couple Surface Reflectance Data and Digital Terrain Models

Fuentes Fuentes,Padarian Padarian,Van Ogtrop Van Ogtrop,Vervoort Vervoort

doi:10.3390/w11040780

Abstract

Uncertainty about global change requires alternatives to quantify the availability of water resources and their dynamics. A methodology based on different satellite imagery and surface elevation models to estimate surface water volumes would be useful to monitor flood events and reservoir storages. In this study, reservoirs with associated digital terrain models (DTM) and continuously monitored volumes were selected. The inundated extent was based on a supervised classification using surface reflectance in Landsat 5 images. To estimate associated water volumes, the DTMs were sampled at the perimeter of inundated areas and an inverse distance weighting interpolation was used to populate the water elevation inside the flooded polygons. The developed methodology (IDW) was compared against different published methodologies to estimate water volumes from digital elevation models, which assume either a flat water surface using the maximum elevation of inundated areas (Max), and a flat water surface using the median elevation of the perimeter of inundated areas (Median), or a tilted surface, where water elevations are based on an iterative focal maximum statistic with increasing window sizes (FwDET), and finally a tilted water surface obtained by replacing the focal maximum statistic from the FwDET methodology with a focal mean statistic (FwDET_mean). Volume estimates depend strongly on both water detection and the terrain model. The Max and the FwDET methodologies are highly affected by the water detection step, and the FwDET_mean methodology leads to lower volume estimates due to the iterative smoothing of elevations, which also tends to be computationally expensive for big areas. The Median and IDW methodologies outperform the rest of the methods, and IDW can be used for both reservoir and flood volume monitoring. Different sources of error can be observed, being systematic errors associated with the DTM acquisition time and the reported volumes, which for example fail to consider dynamic sedimentation processes taking place in reservoirs. Resolution effects account for a fraction of errors, being mainly caused by terrain curvature.

Highlights

Changes in global climate and population result in increased uncertainty in relation to production and resource exploitation [1,2,3]
As the LiDAR digital terrain models (DTM) used in the volume estimates of the Menindee lakes was developed when no water was stored, quantifying water stored in the reservoirs does not require further processing of the terrain model
The DTM can be extrapolated to quantify water volumes associated with flood events in areas that are dry most of the time

Summary

Introduction

Changes in global climate and population result in increased uncertainty in relation to production and resource exploitation [1,2,3]. This is relevant for water resources, whose availability and projections have recently been disputed [4,5]. The advantages of remote sensing data compared to other hydrological data lie in the opportunity to account for the spatial variability of processes [8,13]. Satellites regularly pass over the same location, which provides a time series of the images, catching the temporal variability of some processes [7,15]

Objectives

Methods

Discussion

Conclusion