Dynamic River Masks from Multi-Temporal Satellite Imagery: An Automatic Algorithm Using Graph Cuts Optimization

Abstract

Our knowledge of the spatio-temporal variation of river hydrological parameters is surprisingly poor. In situ gauge stations are limited in spatial and temporal coverage, and their number has been decreasing during the past decades. On the other hand, remote sensing techniques have proven their ability to measure different parameters within the Earth system. Satellite imagery, for instance, can provide variations in river area with appropriate temporal sampling. In this study, we develop an automatic algorithm for water body area monitoring based on maximum a posteriori estimation of Markov random fields. The algorithm considers pixel intensity, spatial correlation between neighboring pixels, and temporal behavior of the water body to extract accurate water masks. We solve this optimization problem using the graph cuts technique. We also measure the uncertainty associated with the determined water masks. Our method is applied over three different river reaches of Niger and Congo rivers with different hydrological characteristics. We validate the obtained river area time series by comparing with in situ river discharge and satellite altimetric water level time series. Along the Niger River, we obtain correlation coefficients of 0.85-0.96 for river reaches and 0.65 for the Congo River, which is demonstrably an improvement over other river mask retrieval algorithms.

Spatio-Temporal Change of Lake Water Extent in Wuhan Urban Agglomeration Based on Landsat Images from 1987 to 2015

“…In the random forest model, the number of classification trees (T) and the number of features (m) at each node for spilling, is critical to the results [36]. This study employed an RF classifier with 5,10,15,20,25,30,35,40,45, and 50 classification trees and chose the optimal item based on the OOB accuracy, using water and non-water samples as the input data. Our sensitivity tests indicate that 20 classification trees are best because, beyond that, the classification gains little performance improvement.…”

“…Remote-sensing data have been widely used for mapping the lake-water extent over time and space [10][11][12][13][14][15][16][17], with Landsat image data being one of the most common data types for monitoring and analyzing long-term lake-water extent changes, due to their high spatial resolution (30-60 m) and long data record [11][12][13][14][15][16].…”

Estimating River Depth from SWOT-Type Observables Obtained by Satellite Altimetry and Imagery

“…The work in [38] developed a Markov Random Fields (MRF) model, which considers spatial correlation between neighboring pixels and the long-term temporal behavior of the river. In this method, to extract the river mask in each image, we define the Maximum A Posteriori (MAP) estimate of the MRF that models the interaction between different constraints and auxiliary sources of information.…”

“…In order to derive the effective width (W) at different river reaches, we apply the algorithm introduced by [38] on MODIS images. The work in [38] developed a Markov Random Fields (MRF) model, which considers spatial correlation between neighboring pixels and the long-term temporal behavior of the river.…”

“…In this way, an energy minimization frame is developed regarding the problem and solved by the graph cuts technique, which is a powerful solution for a wide range of image segmentation problems. We followed the approach by [38] and used an undirected graph with two terminals to extract the river mask by finding the max-flow solution on the graph. Figure 5 presents an example of a derived water mask.…”