See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/358594839

INVESTIGATION OF THE THRESHOLDING METHODS AND WATER INDICES

FOR WATER SURFACE MAPPING USING LANDSAT-8 OLI DATA

Conference Paper · January 2022

DOI: 10.15659/isag2021.12508

CITATIONS READS

0 243

2 authors, including:

Gizem Senel
Istanbul Technical University
16 PUBLICATIONS 158 CITATIONS

SEE PROFILE

All content following this page was uploaded by Gizem Senel on 22 February 2022.

The user has requested enhancement of the downloaded file.

 International Symposium on Applied Geoinformatics (ISAG2021)

INVESTIGATION OF THE THRESHOLDING METHODS AND WATER

INDICES FOR WATER SURFACE MAPPING USING LANDSAT-8 OLI DATA

Gizem Senel 1*, Cigdem Goksel 2

1IstanbulTechnical University, Graduate School, Department of Geomatics Engineering, Istanbul, Turkey

([email protected]); ORCID 0000-0001-6697-9213
2 Istanbul Technical University, Faculty of Civil Engineering, Department of Geomatics Engineering, Istanbul, Turkey

([email protected]); ORCID 0000-0001-8480-1435

ABSTRACT

Water surfaces are facing unprecedented pressure due to climate change and anthropogenic activities. Determination of
water surfaces is necessary for monitoring these areas and helpful for many disciplines. Remote sensing technology
provides cost-effective, time-saving, and reliable data with different spatial or spectral resolutions to determine water
surfaces. In this paper, the water surface detection ability of three thresholding algorithms, namely Minimum, Li, and Otsu
thresholding methods and the normalized difference water index (NDWI) and modified NDWI (MNDWI), have been
investigated with Landsat-8 data. The algorithms and indices have been tested in the Gala Lake in Meriç River Basin in
Turkey. After the threshold values of the index images were determined, binary images representing the water and other
classes were created. The results were evaluated by quantitative statistical analysis, including overall accuracies and Kappa
coefficients. The highest overall accuracy has been obtained as 97.14% with the Li thresholding method for MNDWI in
the selected test area. This study revealed the best index and threshold method for determining the water surface in the
selected test area located within the agricultural areas. Future studies should consider different indices and sensors in test
areas with different geographical characteristics and sizes.

Introduction

Water bodies, including rivers, lakes, seas, and wetlands, are vital for ecosystem balance, the hydrological cycle, and
biodiversity. Determination of the spatial distribution of water bodies provides valuable geo-information for different
applications such as lake or coastline erosion monitoring, coastal zone management, watershed definition, or flood
prediction (Dörnhöfer and Oppelt, 2016). However, traditional survey techniques might be costly and time-consuming;
moreover, the accessibility of water bodies is another challenge. In this context, remote sensing imagery systems provide
timely, accurate, and excessive data for detecting and extracting water bodies thanks to their ability to collect images with
different spatial, spectral, and temporal resolutions. Landsat-8 Operational Land Imagery (OLI) provides a tremendous
advantage since they are easily accessible and free of charge.
Numerous image-processing algorithms have been proposed to extract water bodies from satellite images. Among the
methods, spectral water indices are widely used in water body delineation because of their simplicity. Water indices take
advantage of the reflectivity differences of each involved band and water body information based on the investigations of
spectral differences between water and other surfaces. An appropriate threshold should be established to distinguish water
bodies from other land cover features from index outputs. Although thresholding is a simple and efficient tool, determining
the threshold value is the main issue. Ji et al. (2009) stated that the threshold is a dynamic value, which changes depending
on the subpixel land-cover characteristics. Sezgin and Sakur (2004) investigated and classified different thresholding
methods. They pointed out that the Otsu thresholding method is one of the most referenced thresholding methods. Otsu
thresholding method has been successfully used in many studies for different purposes, such as to separate built-up and

1
 International Symposium on Applied Geoinformatics (ISAG2021)
2-3 December 2021, Riga, Latvia

non-built-up land features or water bodies and background features. Sekertekin (2020) investigated the water surface
detection ability of fifteen automatic thresholding methods with normalized difference water index (NDWI) derived from
Sentinel-2 image. The author reached promising mapping accuracy results while the Minimum thresholding method was
the best among algorithms.
The multispectral advantage of Landsat-8 provides an effective water surface detection ability with various indices
with threshold adjustment. For example, NDWI utilized the spectral difference between green and near-infrared (NIR)
bands, while modified normalized difference index (MNDWI) utilized the spectral difference between green and shortwave
infrared (SWIR) bands. These indices are broadly and successfully used water indices in water surface detection studies.
Therefore, the main aim of this study is to investigate the water surface detection potential of three thresholding algorithms,
Minimum, Li, and Otsu thresholding methods, with NDWI and MNDWI derived from Landsat-8 image. The Gala Lake in
Meriç River Basin in Turkey was chosen as the study area to examine the algorithms. The methods were evaluated by
constructing an error matrix, and the results were comparatively evaluated.

Materials and Methods

The Gala Lake is one of the lakes in the Meriç River Basin, located in the southwest part of the Sub Thrace Region of
Turkey, in Edirne province. The Meriç River Basin is a unique ecosystem covering coastal lagoons, freshwater lakes, reed
beds, and extensive agricultural fields, primarily paddy fields. Additionally, this basin has been an Important Bird and
Biodiversity Area, Natural Conservation Area, and Protected Site since 1991. Gala Lake is formed by alluvial deposits
stored by the Meriç River. In an average year, the Gala Lake covers an area of around 5.6 km². However, the size of the
Gala Lake changes substantially due to its shallowness, as the maximum depth of the lake is about 2 m. Since 2005, Gala
Lake has been declared a National Park. This lake contributes to biodiversity significantly, especially as a habitat for the
waterfowls.

Figure 1 The location of the study area. The right image shows the ground control points and the Gala Lake illustrated
with a false-color combination (near-infrared, red, green bands) of the July 2018-dated Landsat-8 image.

2018.07.05 dated Landsat-8 OLI surface reflectance data has been used in the analysis. The Landsat-8, which contains
eleven bands, is equipped with an OLI and Thermal Infrared Sensor (TIRS). OLI sensor collects data for the visible, NIR,
and SWIR with a 30 m resolution and panchromatic wavelength regions with a 15 m resolution, while TIRS collects data
for the two thermal wavelength regions with a 100 m resolution. In the analysis, green, NIR, and SWIR-1 bands were used
to calculate spectral indices.
In the first stage of the research, a satellite image was acquired, and spectral indices, NDWI and MNDWI, were derived.
NDWI (Eq. 1) and MNDWI (Eq. 2) are widely preferred in water body delineation studies. McFeeters (1996) introduced
NDWI to delineate open water features using the green (band 2) and NIR (band 4) of Landsat Thematic Mapper (TM).
NDWI aims to maximize the reflectance of the water by using the green band while minimizing the reflectance of the NIR
band by water features. Vegetation and soil generally have zero or negative values, and water features have positive values.
Xu (2006) claimed that water features are mixed with built-up land features since NDWI cannot adequately suppress the
signal from built-up land, so it is necessary to improve NDWI. Hence, according to this assumption, NDWI was modified
by replacing the NIR band with the MIR band. It is stated that modifying the NDWI using a MIR band instead of a NIR
band can considerably improve the enhancement of open water features. MNDWI can quickly and accurately discriminate
water from non-water features (Xu, 2006).

2
 International Symposium on Applied Geoinformatics (ISAG2021)
2-3 December 2021, Riga, Latvia

𝐺𝑟𝑒𝑒𝑛−𝑁𝐼𝑅
𝑁𝐷𝑊𝐼 = (1)
𝐺𝑟𝑒𝑒𝑛+𝑁𝐼𝑅
𝐺𝑟𝑒𝑒𝑛−𝑆𝑊𝐼𝑅1
𝑀𝑁𝐷𝑊𝐼 = (2)
𝐺𝑟𝑒𝑒𝑛+𝑆𝑊𝐼𝑅1

Secondly, Minimum, Otsu and Li thresholding methods available with python programming language through the
scikit-image library were applied to extract water surfaces from index images. Prewitt and Mendelsohn (1966) proposed
the Minimum thresholding method. In this method, the histogram is smoothed until there are only two maxima, and the
minimum in between these maximum values is set as the threshold value (Prewitt and Mendelsohn, 1966). The Otsu method
was suggested to determine the threshold by minimizing the interclass variance of foreground and background pixels (Otsu,
1979). Li and Lee (1993) proposed the Li thresholding method that employs gradient descent to choose the best threshold
by minimizing the cross-entropy between the image and its segmented form.
Thirdly, binary images were created according to the thresholding values. Lastly, the accuracy of the produced thematic
maps was assessed by constructing the error matrix. There are various quality measures; overall accuracy and Cohen’s
Kappa statistic, commonly used quality measures, were calculated for the research. In many studies, both quality measures
are given together to reveal the accuracy of thematic maps. Thus, the overall accuracy and the Kappa coefficients were
calculated. For the accuracy assessment, random points were created by considering stratified random sampling and labeled
from high-resolution Google Earth images.

Results

NDWI and MNDWI were derived from the 2018.07.05 dated Landsat-8 data. Then threshold algorithms were applied
through the scikit-image library, and binary images were produced. The histograms of NDWI and MNDWI are presented
in Figure 1. For the NDWI, threshold values have been obtained as -0.293, -0,12, and -0.404 for the Otsu, Minimum, and
Li thresholding methods, respectively. Threshold values belonging to the MNDWI have been obtained as -0.069, -0.064,
and -0.191 for the Otsu, Minimum, and Li thresholding methods, respectively.

Figure 1 Histograms of a) NDWI and b) MDWI.

Figure 2 Original NDWI and MNDWI, and the outputs of the thresholded images

3
 International Symposium on Applied Geoinformatics (ISAG2021)
2-3 December 2021, Riga, Latvia

The accuracy of the thresholded maps was evaluated by constructing an error matrix. 70 ground control points were
collected from Google Earth images and the overall accuracy and Kappa statistics were calculated. The water body mapping
results of indices and thresholding methods are shown in Figure 2, whereas overall accuracy and Kappa coefficients are
given in Table 1.

Table 1 Accuracy assessment results of the thresholding methods

Indices Thresholding Methods Overall Accuracy Kappa

Otsu 91.43% 0.83
NDWI Li 94.29% 0.89
Minimum 85.71% 0.71
Otsu 94.29% 0.89
MNDWI Li 97.14% 0.94
Minimum 94.29% 0.89

Kappa values ranging from 0.71 to 0.94 and overall accuracy values ranging from 85.71% to 97.14% were obtained.
MNDWI with the Li thresholding method achieved the best statistically and visually results. The highest overall accuracy
was obtained with the Li thresholding method for MNDWI with 97.14%, while the lowest overall accuracy was achieved
with the Minimum thresholding method for NDWI with 85.71%. The Li method achieved better results than other
thresholding algorithms, whereas the MNDWI achieved better results than the NDWI.

Conclusion

Index algorithms provide mapping water surfaces over time in a fast and easy, which is particularly useful in areas such
as the Meriç River Basin. Regarding the effectiveness of water body extraction based on the quantitative and visual
interpretation of the thresholded images, the extraction of water surfaces with the MNDWI and Li thresholding method
provides the most accurate and visually high-quality representation. However, some issues should be considered, such as
atmospheric conditions, the acquisition dates of images, or the shape and depth of water surfaces. Although the results are
promising in the selected area, the above-mentioned factors can influence the accuracy of the used methods. The mentioned
issues should be examined in different test areas with increasing thresholding methods and different sensor data for future
works.

Keywords: Water Surface Extraction, NDWI, MNDWI, Thresholding Methods, Landsat-8

References

Dörnhöfer, K. and Oppelt, N. (2016). Remote sensing for lake research and monitoring - Recent advances. Ecological
Indicators, 64, 105–122.

Ji, L., Zhang, L.and Wylie, B. (2009). Problems of Dynamic NDWI Threshold and Objectives of the Study The NDWI
data derived from Landsat MSS, TM, and ETM. Photogrammetric Engineering and Remote Sensing, 75(11), 1307–1317.

Li C.H. and Lee C.K. (1993). Minimum Cross Entropy Thresholding. Pattern Recognition, 26(4): 617-625.

Mcfeeters, S. K. (1996). The use of the Normalized Difference Water Index ( NDWI ) in the delineation of open water
features. International Journal of Remote Sensing, 17(7), 1425–1432.

Otsu, N. A. (1979). Threshold Selection Method from Gray-Level Histograms IEEE Transactions on Systems, Man, and
Cybernetics. SMC-9(1), 62-66.

Prewitt, J.M.S.and Mendelsohn, M.L. (1966). The analysis of cell images. Annals of the New York Academy of Sciences,
128, 1035-1053.

Sekertekin, A. (2021). Potential of global thresholding methods for the identification of surface water resources using
Sentinel-2 satellite imagery and normalized difference water index. Journal of Applied Remote Sensing, 13(4), 044507.

Sezgin, M. and Sankur, B. (2004). Survey over image thresholding techniques and quantitative performance evaluation.
Journal of Electronic Imaging, 13(1), 146–168.

Xu, H. (2006). Modification of normalised difference water index (NDWI) to enhance open water features in remotely
sensed imagery. International Journal of Remote Sensing, 27(14), 3025–3033.

View publication stats