Comparing Thresholding with Machine Learning Classifiers for Mapping Complex Water

Abstract

S.M.alfieri-1@tudelft.nl (S.M.A.); M.Menenti@tudelft.nl (M.M.)Abstract: Small reservoirs play an important role in mining, industries, and agriculture, but storage levels or stage changes are very dynamic. Accurate and up-to-date maps of surface water storage and distribution are invaluable for informing decisions relating to water security, flood monitoring, and water resources management. Satellite remote sensing is an effective way of monitoring the dynamics of surface waterbodies over large areas. The European Space Agency (ESA) has recently launched constellations of Sentinel-1 (S1) and Sentinel-2 (S2) satellites carrying C-band synthetic aperture radar (SAR) and a multispectral imaging radiometer, respectively. The constellations improve global coverage of remotely sensed imagery and enable the development of near real-time operational products. This unprecedented data availability leads to an urgent need for the application of fully automatic, feasible, and accurate retrieval methods for mapping and monitoring waterbodies. The mapping of waterbodies can take advantage of the synthesis of SAR and multispectral remote sensing data in order to increase classification accuracy. This study compares automatic thresholding to machine learning, when applied to delineate waterbodies with diverse spectral and spatial characteristics. Automatic thresholding was applied to near-concurrent normalized difference water index (NDWI) (generated from S2 optical imagery) and VH backscatter features (generated from S1 SAR data). Machine learning was applied to a comprehensive set of features derived from S1 and S2 data. During our field surveys, we observed that the waterbodies visited had different sizes and varying levels of turbidity, sedimentation, and eutrophication. Five machine learning algorithms (MLAs), namely decision tree (DT), k-nearest neighbour (k-NN), random forest (RF), and two implementations of the support vector machine (SVM) were considered. Several experiments were carried out to better understand the complexities involved in mapping spectrally and spatially complex waterbodies. It was found that the combination of multispectral indices with SAR data is highly beneficial for classifying complex waterbodies and that the proposed thresholding approach classified waterbodies with an overall classification accuracy of 89.3%. However, the varying concentrations of suspended sediments (turbidity), dissolved particles, and aquatic plants negatively affected the classification accuracies of the proposed method, whereas the MLAs (SVM in particular) were less sensitive to such variations. The main disadvantage of using MLAs for operational waterbody mapping is the requirement for suitable training samples, representing both water and non-water land covers. The dynamic nature of reservoirs (many reservoirs are depleted at least once a year) makes the re-use of training data unfeasible. The study found that aggregating (combining) the thresholding results of two SAR and multispectral features, namely the S1 VH polari...

Support Vector Machine Versus Random Forest for Remote Sensing Image Classification: A Meta-Analysis and Systematic Review

“…The large volume of RS data [15], the complexity of the landscape in a study area [22]- [24], as well as limited and usually imbalanced training data [25]- [28], make the classification a challenging task. Efficiency and computational cost of RS image classification [29] is also influenced by different factors, such as classification algorithms [30]- [33], sensor types [34]- [37], training samples [38]- [41], input features [42]- [46], pre-and post-processing techniques [47], [48], ancillary data [49], [50], target classes [22], [51], and the accuracy of the final product [21], [50], [52]- [54]. Accordingly, these factors should be considered with caution for improving the accuracy of the final classification map.…”

Automatic Inundation Mapping Using Sentinel-2 Data Applicable to Both Camargue and Doñana Biosphere Reserves

“…Several approaches use information from Shortwave infrared (SWIR) spectral ranges to identify shallow inundated wetland areas, since it is less sensitive to sediment-filled waters and, hence, more efficient for registering the boundaries between water and dry areas in shallow wetlands [13,[22][23][24]. Automatic thresholding approaches can be applied to different areas and are computationally inexpensive, but they may wrongly classify dark objects (i.e., shadows and buildings) as water when their spectral characteristics are similar [25]. Automatic thresholding approaches are distinguished into: (a) global approaches [15,17,19,20,26], which estimate thresholds based on the histogram analysis of the complete image, and (b) local thresholding approaches [23], which estimate local thresholds for image subsets containing high percentages of pixels belonging to the water and non-water classes, and then may take into consideration subsets' thresholds to estimate an overall threshold.…”

“…Even though machine-learning approaches usually exhibit higher accuracy than thresholding ones [30], they have several limitations. (a) In case reference data is not available, supervised approaches require collection of training samples, which is a time-consuming and tedious task that requires expert knowledge and/or validation in the field [25,40]. (b) Supervised methods may meet problems when mapping water bodies over large scale areas [41].…”

A Comparative Study of Water Indexes and Image Classification Algorithms for Mapping Inland Surface Water Bodies Using Landsat Imagery

“…Most researches related to mapping inland surface water bodies using Landsat imagery consist three steps: 1) using the spectral reflectance captured by Landsat to compute one type of water index at each pixel; 2) using one type of image classification algorithm (unsupervised or supervised) to identify water and non-water pixels; 3) using ground truth or alternative data to assess accuracy of the extracted water body features. However, all these three steps have some unsolved issues, and inconsistent conclusions can be found across the literature [32][33][34][35][36][37]. Therefore, this study is dedicated to address these problems so that our knowledge and techniques in optical remote sensing of inland surface water bodies using Landsat imagery could be advanced.…”