Remote Sensing Applications: Society and Environment 23 (2021) 100547

Contents lists available at ScienceDirect

Remote Sensing Applications: Society and Environment

journal homepage: www.elsevier.com/locate/rsase

Monitoring of dynamic wetland changes using NDVI and NDWI based

landsat imagery
Amgoth Ashok a, *, Hari Ponnamma Rani b, K.V. Jayakumar a
a
Department of Civil Engineering, National Institute of Technology, Warangal, India
b
Department of Mathematics, National Institute of Technology, Warangal, India

A R T I C L E I N F O A B S T R A C T

Keywords: Recent initiatives using the GIS tools for mapping and analysis of wetland dynamics bring the hope for an
Wetlands effective wetland management. Degraded or modified wetlands are more sensitive and less resilient to the
Google earth engine climate change. This paper focusses on a study of water bodies in Sirmaur District, Uttarakhand, India with the
Harmonic time series analysis
special focus on Renuka wetland. The Renuka wetland, a Ramsar site is affected by global warming, floods and
Image classification
Global water occurrence
storms and changes due to precipitation. The updated or accurate maps and spatial and temporal changes of
Renuka wetland are not available. Only the details about the flora and fauna have been recorded in this region.
Traditional methods in mapping and monitoring, like field survey and sampling, are usually labour intensive,
time consuming, and expensive and often fail to detect the changes over the regions of wetland zones. The
present study is carried using the online interactive cloud-based planetary processing open source platform,
Google Earth Engine (GEE). The wetland maps considered are Landsat 7 and 8 images with cloud cover less than
30% during the summer (May) and winter (September) seasons. The suitability index is also calculated based on
the area and perimeter of the water body in study region to demonstrate the characteristic change of Renuka
wetland over the selected time period. The missing data of NDVI and NDWI is fitted using the harmonic time
series and the resultant map with the supervised classification Random Forest shows a high overall accuracy of
90% with a reasonable Kappa coefficient of 0.8. Using the Global Water Occurrence Explorer, the water changes
intensity and the seasonality water change over the wetland for the period of 1984–2019 are studied. The results
showed the significant insight into the changes over the study region in terms of pixel range with the presence of
water and the change in locations in terms of seasonality and persistence.

1. Introduction Water body extraction plays a vital role in the management of wet­
lands, the evaluation of water cover status and the detection and
Wetland is a valued resource for the groundwater recharge and flood monitoring of surface water change (Ouma and Tateishi 2006). GIS and
control (Rundquist et al., 2001). The shrinkage and degradation of RS are used in the timely monitoring of water bodies based on the
wetlands are major threat for the conservation of biodiversity and the availability of satellite data and aid in the decision and policy making
ecological environment. Mapping of wetland plays an important role in processes (Verpoorter et al., 2012). Such types of spatiotemporal
planning, conservation and management of wetlands. In general, wet­ changes in lakes can be calculated by using the multi-temporal Landsat
lands are part of the earth structure and considered to be one of the Thematic Mapper (TM) and Landsat Enhanced Thematic Mapper Plus
important ecosystems consisting of water, vegetation, soil and micro­ (ETM+) images and can be quantified in the high-resolution mapping of
organism systems and they play a significant role in the maintenance of global surface water (Pekel et al., 2016) and temperature (Orhan et al.
aquatic ecosystem (Mitsch and Gosselink 2007; Lee and Yeh 2009). The 2014, 2019; Kirtiloglu et al., 2016). The medium spatial resolution im­
water content in the vegetation, the existence of surface water body and agery of Sentinel-2 and Landsat-8 Operational Land Imager (OLI) initi­
the soil water status can be detected and extracted based on the specific ated an era for land resource monitoring (Roy et al., 2014). Landsat-8
reflective solar spectrum bands of Landsat images (Sarp and Ozcelik OLI derived burn area fitted better with the fire product of Climate
2017; Naik and Anuradha 2018). Change Initiative European Space Agency (Fire-CCI of ESA) and

* Corresponding author.
E-mail addresses: [email protected] (A. Ashok), [email protected] (H.P. Rani), [email protected] (K.V. Jayakumar).

https://doi.org/10.1016/j.rsase.2021.100547
Received 25 December 2020; Received in revised form 23 May 2021; Accepted 24 May 2021
Available online 2 June 2021
2352-9385/© 2021 Elsevier B.V. All rights reserved.
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

MCD64A1 of the Moderate Resolution Imaging Spectroradiometer v. Calculate the spatial and temporal distribution of water surface at
(MODIS) burn area product Landsat-8 than Sentinel-2 (Bar et al., 2020). global scale from the year 1984–2019 and provide the wetland
The unique features of wetlands are highlighted through the algorithms surface water changes.
by Xu et al. (2019). On a regional scale they help to map the annual
actual wetland area. Using the multi temporal Landsat 5-TM, 7-ETM+ 2. Study area
and 8-OLI images of lake Urmia, the spatiotemporal water changes are
extracted by Rokni et al. (2014) for the period from 2000 to 2013 with Renuka wetland, located in the Sirmaur district of Himachal Pradesh,
the applicability of different satellite-derived indexes including the India is chosen as the study area. This wetland, with the circumference
Normalized Difference Water Index (NDWI) and the Normalized Dif­ of about 3210 mand water depth varying from 1 to 13 m, is one of the
ference Vegetation Index (NDVI). important attractive natural wetlands in India with a home for at least
The NDVI denotes a non-dimensional index which defines the 443 species of fauna and is declared as a Ramsar site in November 2005
vegetation cover with the difference between the visible and near- (Fig. 1(a)). It is located at about 37 km from Nahan district at an altitude
infrared reflectance. This index is one of the most commonly used of 660 m above MSL. The State Government also declared an area
indices for monitoring the dynamics of vegetation at the regional and comprising of 402 ha in and around this wetland as a wild life sanctuary.
global scales (Tucker (1979)). The NDVI values are used in the estima­ The major problem identified in this wetland is the continuous
tion of density of greenery (Tucker 1979; Vrieling et al., 2013; Zhu et al., shrinking in its size due to the sedimentation and dumping of con­
2013; Weier and Herring 2000). struction materials in it. Thus, the present study is taken up to focus on
In mapping and monitoring of wetland zones the traditional the temporal and spatial changes of Renuka wetland during the period
methods, like field survey and sampling, are usually labour intensive, 1984 to 2019 to understand the significant water change intensity using
time consuming, expensive and often fail to detect the changes over the the histogram of surface water occurrences. In this period, the analysis
large regions (Ozesmi and Bauer 2002; Ghosh et al., 2016). Processing about the global surface water occurrences and water seasonality
the digital images of satellite data provides the tools for analysing the change will help to understand the significance of water level in the
images through the mathematical indices and algorithms (Eid et al., study region.
2020). The satellite based remote sensing studies provide images in
monitoring the land use and land cover changes in approximate 2.1. Data collection
real-time at the different spatial scales and resolutions (Darby and
Turner 2008). Landsat data availability, especially in the flooding and The NDVI and NWDI time series between the years 2013 and 2019 of
transplanting phase, is critical for guaranteeing a high-quality map of the Landsat 7 and 8 Collection 1 Tier 1 calibrated Top of Atmosphere
paddy rice planting area (Dong et al., 2015). (TOA) reflectance are employed. The details are given in Table 1. Bands
In general processing of downloaded remote sensing images are 1 to 7, with an original spatial resolution, are used. The GEE contains the
time-taking, especially for the large-scale long time series data, which Landsat specific processing methods to compute at-sensor radiance, TOA
often involve thousands of gigabytes. Since the Landsat series data is reflectance, surface reflectance (SR), cloud score and cloud-free com­
freely available, all these data sets have been archived by Google and posites. It also provides a simple cloud scoring algorithm, viz. ee.Algo­
they are linked with the Google cloud computing engine so that they can rithms.Landsat.simpleCloudScore( ) for scoring the Landsat pixels based
be used as an open source. With a time period spanning over 30 years, on their relative cloudiness. In this platform the processing of Landsat
more accurate and refined 30 m NDVI composite was developed using data is carried using the cloud computing technology (https://eart
the Google Earth Engine (GEE) and one such similar study was carried hengine.google.org/). This enables the big data processing and paral­
on cloud-based planetary processing platform by Robinson et al. (2017). lel computing feasible in the study area.
The GEE is the online cloud computing platform in the planetary-scale
environmental data analysis. The major focus of GEE is to accomplish 2.2. Methods
the highly-interactive online algorithm, to utilize the knowledge of big
data analytics in remote sensing with high impact enabled data driven 2.2.1. NDVI and NDWI
science and to make the substantive progress based on the global chal­ The value of NDVI lies between − 1 and + 1 and the values less than
lenges involving the large geospatial dataset. The GEE is, thus, designed zero are observed during the dormant season, i.e., no vegetation cover,
to store and process the huge data set images at the petabyte scale. like bare earth, desert, cloud, icepack, glacier, snow, water body, etc.
Thus, the major aim of the present paper is to develop and analyse Values greater than zero identify the vegetation cover and are observed
the wide-range of Landsat time series algorithms for mapping of wet­ in the gowning season. The NDVI is computed based on surface reflec­
lands using the GEE. The suggested algorithm in the present study, is tance bands of Landsat 7 and 8. and is given by
able to capture the changes of wetland characteristics based on NDVI
NIR − RED
and NDWI. The study involves the quantitative and qualitative temporal NDVI = (1)
NIR + RED
and spatial changes of the chosen wetland during the period 2013 to
2019. The outcome of the study can provide a basis in better under­ where NIR and RED denote the reflectance in Band 5 and Band 4 of
standing of changes in the wetland regime and a potential guide for Landsat 7, respectively. The changes in the vegetation over the time is
further research in these areas. The objectives of the present study are to: obtained with the atmospheric correction.
The NDWI values are strongly related to the water content of plant
i. Evaluate the wetland areas based on NDVI and NDWI Landsat sensitiveness to the built-up land and hence it is a good representation
images. for plant water stress measurement. These values are useful in the cal­
ii. Identify the presence of permanent water bodies and calculate the culations of remote sensing vegetation liquid water status from the
Suitability Index (SI) over the years using the estimated area and space. This status is centered on the reflectance properties of green and
perimeter. dry vegetation. The reflectance index ranges from − 1 to +1 and is used
iii. Predict missing data over the lake using the harmonic time series in the extraction of the water bodies from the satellite imagery. The
analysis. NDWI is calculated based on the surface reflectance of NIR and SWIR
iv. Perform the image classification using the supervised classifica­ bands of Landsat 7 and 8. and is given by
tion (Random Forest). Check the overall accuracy and validation
accuracy using the Kappa coefficient. NDWI =
NIR − SWIR
(2)
NIR + SWIR

2
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

Fig. 1. (a) Ramsar sites in India (Source: Wetlands International South Asia Annual report 2019–2020) and (b) Study area - clipped geometry (shaded polygon
region) over the Renuka wetland.

wetland plants are in the budding period i.e., they are in the vegetative
Table 1 growth, reproductive and ripening stages. They are characterized with
Description of Landsat data.
the NDVI values greater than the NDWI values. This implies that any
ID Description pixel covered by the vegetation has a relationship 0 < NDVI-NDWI < 1,
LANDSAT/LC08/C01/T1_RT Landsat 8, Collection 1, Tier 1 + Real Time while the permanent water bodies have − 1 < NDVI-NDWI < 0. It is
LANDSAT/LE07/C01/T1_TOA Landsat 7, Collection 1, Tier 1 aimed to calculate the area related to the vegetation and water body
during the summer season, viz., late April to early May from 2013 to
2019 as shown as algorithm.
In the above expression NIR and SWIR denote the reflectance of Band
Algorithm Proposed for workflow for mapping of wetland areas
5 and Band 6 used in Landsat 8; Band 4 and Band 5 used in Landsat 7,
using GEE.
respectively. √̅̅̅̅̅̅̅̅̅̅
The index-based classification methodology is used to study the The following Workflow for mapping of wetland areas (SI = Area/
temporal variations of NDVI and NDWI, derived from the Landsat im­ Perimeter) are mentioned below.
ages. The multi-decadal trends are documented and the timing of events
such as changes in the wetland area are identified. From the multi- Step 1: Extract Landsat 7 and 8 Images of Renuka wetland
temporal Landsat images an algorithm based on the methodology pro­ Step 2: Calculate NDVI and NDWI
posed by Dong et al. (2014) is developed for mapping in and around Step 3: Calculate the water covered area: − 1 < NDVI-NDWI< 0
Renuka wetland region. Wetland unique features like shallow flooding Step 4: Extract − 1 < NDVI< 0
√̅̅̅̅̅̅̅̅̅̅
water and annual phonological land cover change of the exposed soils Step 5: Calculate SI = Area/Perimeter
and plants are paid importance. In the ante-tillering phase, the NDVI Step 6: If (SI > 0.18) Classify the rivers and ponds
value is greater than zero and is less than the NDWI value due to the Else If (SI < = 0.18) Classify Paddy and Wetlands
flooding of shallow water bodies. In the reproductive and ripening Step 7: Display the result
phases, the NDVI values are higher than the NDWI values and are greater
than zero. In the post-harvest wetland planting phase, the NDVI value is 2.3. Time-series analysis
greater than the NDWI and NDWI is less than zero in this phase.
The clipped geometry (the considered study region) in and around The GEE web portal provides the satellite imagery time series and the
the Renuka wetland shown in Fig. 1(b) consists of the area of 312499 m2 corresponding vector data with the access to algorithms for processing
and perimeter of 3760 m. Landsat 7 cloud-free TM images of late May the data. The time series image collections are complex because each
and early June are captured to mask the wetland, based on the difference pixel contains different types of data. The length of the time series varies
between the NDVI and NDWI values. An automatic cloud masking across the pixels with a chance of missing the data due to the cloud
procedure was applied to the Landsat data. The chosen images of the masking. Due to these complex factors, the time series is analysed by
using joins to define the temporal relationship between the collections of

3
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

items. The traditional time series analysis methods are used as the training data by spcifying the features of the landcover categories as the
mapping functions over the joined collections. In the analysis of time class property. The output results show that all the eight datasets
series data, the following steps are followed: showed moderately to highly accurate land cover maps, with overall
accuracy over 89.9%.
i. Search and import the Landsat 7 and 8 Surface Reflectance Tier 1 The classification accuracy of each dataset scenario was assessed
and Tier 2 collection. Make single point geometry in the study using the Confusion matrix (Stehman 1997) and the derived accuracy
region. indices. If the training data set is of polygons representing the homog­
ii. Pre-process and filter the Landsat imagery, mask clouds and add enous regions, every pixel in each polygon is classified as a training
variables. point. These polygons are then used to train through the machine
iii. Plot time series. learning algorithms.
iv. Add linear trend line for reference. To get statistics of pixel values in the study region, ee.Image, image.
reduceRegion() algorithm was used. This reduces all the pixels in the
The following linear model is considered to estimate the linear trend region(s) to a compact representation of the pixel data in the region. The
pt over time, t: region is represented as a geometry, which might be a polygon, con­
taining many pixels, or it might be a single point, in which case there
pt = β0 + β1 t + et (3)
will only be one pixel in the region. Pixels are determined to be in the
where et is a random error and the values for β0 and β1 are obtained from region and weighted according to the following rules, applied in the
each pixel of the two-band coefficients image. This model is convenient specified scale and projection: (i) Unweighted reducers represents pixels
in detrending the data and in reducing the stationarity in time series are included if their centroid in the region and image mask are non-zero.
(Shumway and Stoffer (2019)). (ii) Weighted reducers represent pixels, that are included if at least 0.5%
of the pixel lies in the region and the image mask is non-zero. Their
weight is the minimum of the image mask and the fraction of the pixel
2.4. Harmonic time series covered by the region.
Confusion matrix is a technique to summarise the performance of the
The following linear model (Shumway and Stoffer, 2019) is classification algorithm and gives better idea about the classification
considered model along with the errors involved in the model. Confusion matrix,
pt = β0 + β1 t + A cos(2πωt − φ) + et also known as the error matrix, used as the quantitative method of
characterizing the image classification accuracy. It shows the corre­
= β0 + β1 t + β2 cos(2πωt) + β3 sin(2πωt) + et (4) spondence between the classification result and the reference image.
This matrix requires the ground truth data, like cartographic informa­
Note that β2 = A cos φ and β3 = A sin φ. In Equation (4) A = tion and with the manual digitizing an image result. By using the
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ( )
β22 + β23 , ωand φ = atan ββ32 denote the amplitude, frequency and Landsat 7 and 8 images, the image classification of Renuka wetland is
carried using the GEE. The classification is performed by giving the
phase, respectively. To fit this model to the time series, ω = 1 (one cycle training samples of polygon feature class or shape file. The format of
per unit time) was chosen and the least square regression approach is feature class attribute table requires to be matched with the table of
used. The procedure for fitting the model is initially add the harmonic training samples attributes. For ensuring this, the reference dataset is
variables (third and fourth terms of Equation (4)) to the image collec­ created using the ‘Training Samples Manager’ for reading and writing
tion. The model is fitted with the linear trend by using the linear the dataset. Kappa statistic of agreement gives the overall assessment of
regression reducer function. The coefficients are then plugged in Equa­ accuracy of the classification. The statistical values are presented in
tion (4) in order to get the time series of fitted values. These fitted values Table 2, with accuracy in the range of 0–1, with 1 being the 100 percent
will be shown graphically and analysed in the results and discussion accuracy.
section.
3. Results and discussion
2.5. Image classification
Use of lengthy time series images might increase the ability to
Image classification is used to extract the information classes from characterize the temporal variability of wetlands and help to map and
the multiband raster images. Classifier packages, like, CART, Random monitor the changes along the time period. The shape file for Sirmaur
Forest, Naive Bayes and SVM, are used to handle the supervised classi­ district is extracted from Landsat 7 and 8 images. Landsat 8 images are
fication with the help of the traditional machine learning algorithms available starting from 2013 onwards and hence to study the decadal
running in GEE. Here Random Forest classifier is used over the period of changes from 2000 to 2010, the Landsat 7 data was considered. The
2013–2019. The workflow for the classification is given by: results shown in Fig. 2 are based on the NDVI values with cloud cover
less than 30% during the period starting from 2000 to 2019 for the
• Collecting training data. summer (May) and winter (September) seasons. The blue and green
• Assembling features of known class label and properties. colour regions show the presence of water and vegetation, respectively.
• Instantiating classifier and set its parameters. The vegetation and water covered areas are calculated as explained
• Training classifier using training data. in section 2 in the GEE platform with the Java Script programming. The
• Classifying image or feature collection. calculated year-wise vegetation area and the SI values based on area and
• Estimating classification error with independent validation data.

Table 2
The Landsat 8 surface reflectance data with eight different combi­
Comparison accuracy between Landsat 7 and Landsat 8 image collections over
nation strategies was used to produce and evaluate land cover maps over
the Renuka wetland (2013–2019).
the Renuka lake. The MODIS land cover IGBP is used for training.
Landsat 7 Landsat 8
Random Forest (RF) classifier is used to train the sample data. Different
random seeds were given to get the validation data. Then the result was Overall accuracy 0.90 0.97
filtered to get rid of any null pixels. Based on the smile RF, the data is Validation accuracy 0.80 0.86
Kappa Coefficient 0.85 0.905
validated. The RF classifer ee.Classifier.randomForest() is used on the

4
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

Fig. 2. NDVI values of Sirmaur District from 2000 to 2019 using Landsat 7 imagery.

perimeter of the water body are tabulated in Table 3. It can be observed are less than zero. Low NDWI values correspond to low vegetation water
that the vegetation area and SI are fluctuating each year. The SI is less content and vegetation fraction cover. According to the literature
than 0.18 over the different years and this value classifies the Renuka (Taufik et al., 2016) during the post-harvest wetland planting phase the
lake under the category of wetland. Comparing the vegetation area be­ NDVI value is higher than the NDWI value. It can be observed that this
tween the years 2013 and 2015, it can be seen that the area had abruptly normalized NDWI data is not continuous.
increased and again decreased in year 2016. The water spread areas
during 2013 and 2019 are shown in Fig. 3(a) and (b). It can be observed 3.2. Trend to fit noise using harmonic time series
that there is a reduction in the area of Renuka wetland over the years.
In the calculations of NDVI and NDWI for the period 2013 to 2019,
there are missing data as shown in Figs. 4 and 5. It can be observed that
3.1. Temporal characteristics of NDVI and NDWI there is an increase in the fluctuations and the corresponding image data
has the noise. In order to fill the missing data and to check whether the
The temporal characteristics of NDVI for the period 2013 to 2019 data is in sequential order, harmonic time series analysis approach is
using Landsat 7 images over the Renuka wetland is shown in Fig. 4. The used. By using the harmonic time series analysis, the fluctuations are
NDVI values lie between 0.2 and 0.75 over the study period and these reduced and the smooth curve is obtained. The observed NDVI values
values depict the presence of dense vegetation and permanent water along with the fitted data is shown in Fig. 6. This figure shows the
bodies. Pettorelli (2013) mentioned that the NDVI values, ranging be­ decreasing and increasing curve over the selected months like May and
tween 0.2 and 0.5, represent the shrubs and grasslands or senescing June (summer), and September and November (winter). Over a decade,
crops. The growth of the vegetation and the increase in leaf widths have in the month of September, the NDVI attains the maximum value and
made the index values high over the Renuka lake. It can be observed that rises until it reaches the value of 0.6 and then slightly decreases. The
this normalized data is not continuous. difference between the observed and fitted NDVI values gives the
The surface water body is detected in the study region from the high- discontinuous sequence of observations.
resolution satellite image data of NDWI for the period starting from Using the Landsat 8 image collection, the NDVI values are calculated
2013 to 2019 and shown in Fig. 5. On an average, the values vary be­ and are shown in Fig. 7. It can be observed this data is more continuous
tween − 0.1 and + 0.4. The high (positive) and low (negative) NDWI with smoother curve but it has more noise in comparison with Landsat 7
values correspond to the high plant water content and low vegetation image collection shown in previous Fig. 6. The NDVI values are in the
content, respectively. In the period of water stress, NDWI values will range of 0–0.25, which implies that there is less vegetation reflectance.
decrease. It can be observed that during the year 2017, the NDWI values The harmonic time series analysis is used to fit the observed data with
the predicted values. These predicted Landsat 8 values show slightly
Table 3 increasing trend starting from the year 2013–2019 while these values
Year-wise vegetation area and water spread area. are decreasing in trend with respect to Landsat 7 image collection. From
Year Vegetation Area Water body − 1 < NDVI < 0.3 these results, the quantitative estimation of vegetation growth can be
(m2) √̅̅̅̅̅̅̅̅̅̅ obtained. In this harmonic time series analysis, slightly increasing pe­
Area Perimeter SI (m) = Area/
0 < NDVI-
(m2) (m)
riodic trend can be observed for 2013 in comparison with 2019 along
NDWI < 1 Perimeter
with the decreasing vegetation growth. The magnitude of observed data
2013 432 119 256 016 14788.37 0.03421 is more deterministic and has the sequential trend. However, the curve
2014 412 817 249 762 16445.85 0.03038 obtained for NDVI using Landsat 8 image (Fig. 7) gives more significant
2015 427 487 253 822 14773.11 0.03410
results when compared with Landsat 7 image collection (Fig. 6).
2016 415 905 205 003 48227.32 0.00938
2017 425 943 252 941 36987.49 0.01359
2018 425 170 249 385 15293.77 0.03265
2019 429 031 243 860 16670.06 0.02962

5
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

Fig. 3. Water covered area in (a) 2013 and (b) 2019. The circled region in (a) and (b) shows the area reduction or shrinking of the wetland.

Fig. 4. NDVI data from 2013 to 2019 using Landsat 7 images.

6
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

Fig. 5. NDWI data from 2013 to 2019 using Landsat 7 images.

Fig. 6. Harmonic time series and observed data of NDVI using Landsat 7 image collection.

3.3. Global Surface Water Explorer (GSWE) system, present in the GSWE. The frequency with which water was
present on the surface from March 1984 to October 2019 was captured
The GSWE is the water dataset developed in the Copernicus Pro­ in a single product called surface water occurrence.
gramme (Pekel et al., 2016). The dataset produced from the Landsat
imagery (USGS and NASA) will support the different applications 3.3.1. Water occurrence
including the water resource management. The dataset maps the loca­ The temporal and spatial variations of surface water are captured in
tion along with the temporal distribution of water surfaces at global the study region. The locations of surface water occurrence between
scale and provides the statistics on the extent and change of water sur­ 1984 and 2019 are shown in Fig. 8. This records the frequency and
faces, variations in persistence between water occurrence, change in­ occurrence of water on the land surface in monthly time-steps and
tensity, seasonal water occurrence, annual seasonal recurrence and provides the information related to the overall water dynamics. The
transition among permanent water bodies. The Global Surface Water process captures both inter and intra-annual variability changes. Some
Occurrence (GSWO) map will provide the information on overall water locations are underwater (i) throughout the observation period i.e.,
dynamics by capturing both the inter and intra-annual variability and 100% of all observations are classified as water; (ii) for a few months of
changes. For the period between 1984 and 2019, using the Landsat 8 every year and (iii) on an episodic basis. Some areas have never been
image collection, the Renuka wetland water dataset is collected. Each underwater i.e., 0% classified as water. The occurrence of permanent
pixel was individually classified as water or non-water, using an expert (100%) water body (marked in blue) is predicted within the Renuka

7
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

Fig. 7. Harmonic time series and observed data of NDVI using Landsat 8 image collection.

Fig. 8. Water occurrence over the period 1984–2019.

wetland. The water availability at the edges of the Renuka wetland is pair of months and the average of all of these differences establishes the
less in comparison with other areas. This decrease in frequency with surface water occurrence change intensity. The scores are averaged to
varying colour from pink to purple i.e., 0 < occurrence <100% shows account for the variations in the data distribution over time i.e., seasonal
there is a change in the water body area with respect to time and space. variations and frequency of valid observations and to provide a consis­
The results show the water occurrences in the Renuka wetland is grad­ tent measurement of occurrence of change.
ually decreasing both spatially and temporally. The degree of water occurrence change intensity map in terms of
percentage is shown in Fig. 9. The detected water areas with equal
3.3.2. Water occurrence change intensity occurrence by considering a tolerance of ±15% in both the periods are
The water occurrence change intensity calculations will show depicted in black colour on the map (water presence with no change in
whether the surface water occurrence increases or decreases or area). Locations at which the surface water occurrence gets decreased
remained invariant over a period of time. Direction of change (increase, and increased across the study period are mapped in red and green
decrease or no change) along with its intensity are calculated over the colour, respectively. In both the cases, the brighter tones indicate the
period from 1984 to 2019. Water occurrence change accommodates for large changes in the intensity. Grey coloured areas represent the loca­
the variations in data acquisition over time. This gives the clear picture tions where there are no pair of homologous months between the two
about the temporal deepness and frequency density of the satellite ob­ periods. The water body change with respect to time in the middle of
servations. Change in the water occurrence intensity between the two wetland is in red colour and it may lead to decrease in water occur­
periods is derived from the homologous pair of months i.e., same months rences. The bright red colour shows there is a great loss of water than the
containing valid observations in both periods. The difference due to the light red areas. Some of the areas in the wetland on the western side
occurrence of surface water is calculated for each of the homologous show green in colour representing there is an increase in the water

8
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

Fig. 9. Water occurrence change intensity over a period 1984–2019.

occurrence. Some of the areas are in black colour, since these locations 3.5. Annual water reoccurrence
do not have the sufficient data to compute the temporal change in the
availability of water. However, the results show that there is significant Water occurrence means that water is present every year for the
increase in water level at the eastern side of the wetland. study period and is a measurement of the degree of variability in the
presence of water. As for as the water occurrence, the recurrence is
3.4. Seasonal water occurrence expressed as a percentage. It describes the frequency with which water

The water seasonality study provides the information concerning the

intra-annual behaviour of water surface levels. It separates the map of
‘permanent’ water bodies, which are present throughout the period of
observation nominally for a year, from the ‘seasonal’, which are present
only for a part of the year. This also represents the degree of seasonality
i.e., the ratio of number of months the water is present with respect to
the total number of months. Fig. 10 shows the seasonal water occurrence
in Renuka wetland. It can be noted that there is seasonal increase in the
occurrence of water (dark blue colour) whereas there is seasonal
decrease in the areas of permanent water bodies (light blue colour). The
areas (light blue colour) at the eastern end of the wetland represents the
increase in the seasonal water body or the loss with more rainfall in­
tensity. Some of the areas especially the edges represent the loss of
seasonal water over the Renuka wetland. Fig. 11. Annual water recurrence for the period 1984 to 2019 over the
Renuka wetland.

Fig. 10. Seasonal water occurrences over a period 1984–2019.

9
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

Fig. 12. (a) Transition class over a Renuka wetland and (b) Pie chart of transition class.

returned to a particular location. The water recurrence map documents images the slightly increasing curve is noted. The area and perimeter are
the inter-annual variability of water availability. This map shows the calculated over the study region to locate the presence of permanent
frequency of water availability during each year. In Fig. 11, the annual water bodies in and around the Renuka wetland. Based on these values
water recurrence during 1984–2019 at all the sides of the wetland is the suitability index (SI) is calculated to predict the long-term changes in
shown. The low percentage characterizes the places at which the inun­ the water body and its characteristics over the Renuka wetland. The
dation is far from the systematic, such that the lighter the colour the calculated SI values confirmed the presence of paddy fields and wetlands
greater is the inundated area. The figure shows the discriminate between over the years in this study region. Supervised image classification using
the areas that are flooded on an episodic basis (in orange colour) and Random Forest showed the better accuracy with reasonable Kappa co­
those that are regularly flooded (in blue colour) and therefore predict­ efficient. Using the Global Surface Water Explorer map, over period of
ably inundated (light blue colour). There is a chance of flooding during 1984–2019, water occurrences, change intensity, annual recurrences
the rainfall and the light blue colour area represents the regularity of and water transitions are studied. The water changes over the study
water level by means of inundated area. region in terms of pixel range along with seasonality and persistence is
calculated. The results show the long-term changes in the surface water
occurrence and identify the shrinking of Renuka wetland over the years.
3.6. Water transition maps
Ethical Statement
The changes in the surface water classes are shown in Fig. 12 in terms
of the water transition map. The water classes are classified as, no water, Hereby, I A. ASHOK consciously assure that for the manuscript title
seasonal and permanent water bodies (Pekel et al., 2016). The transition “Use of Landsat derived NDVI, NDWI and Global Surface Water Explorer
changes in the water spread areas in the study period is computed and to Monitor Dynamic Changes in Wetland based on Google Earth Engine”
shown in Fig. 12(a). Temporal profiles recording the full history of each the following is fulfilled:
pixel defines the long-term basis of presence or absence of water
throughout the period of study. However, the water transition will be 1) This material is the authors’ own original work, which has not been
higher in the rainfall seasons and less variations will be there in the previously published elsewhere.
summer seasons. 2) The paper is not currently being considered for publication
In the study region the following transitions were mapped and are elsewhere.
shown as pie chart (Fig. 12 (b)): unchanging permanent water surfaces is 3) The paper reflects the authors’ own research and analysis in a
about 12.7%; new permanent water surfaces is about 1.4% (conversion truthful and complete manner.
of land into permanent water); unchanging seasonal water surfaces is 4) The paper properly credits the meaningful contributions of co-
about 38.8%; new seasonal water surfaces is about 12.5% (conversion of authors and co-researchers.
land into seasonal water); lost seasonal water surfaces is about 30.2% 5) The results are appropriately placed in the context of prior and
(conversion of a seasonal water into land); conversion of permanent existing research.
water into seasonal water is about 1.7%; and the conversion of seasonal 6) All sources used are properly disclosed (correct citation). Literally
water into permanent water is about 2.7%. However, the seasonal copying of text must be indicated as such by using quotation marks
variation of wetland is shown to be frequently occurring throughout the and giving proper reference.
period. This quantified study of water transition over the Renuka 7) All authors have been personally and actively involved in substantial
wetland suggests there is a need for the improvement to maintain the work leading to the paper, and will take public responsibility for its
wetland in good condition. content.

4. Conclusions The violation of the Ethical Statement rules may result in severe
consequences.
The Renuka wetland is one of the important Ramsar sites of India. I agree with the above statements and declare that this submission
The NDVI and NDWI indexes for the Renuka wetland are calculated from follows the policies as outlined in the Guide for Authors and in the
the Landsat 7 and 8 images using the Google Earth Engine over the Ethical Statement.
period 2013 to 2019. Missing data were fitted using the harmonic time
series analysis to reduce the noise from the Landsat images and to obtain
the smooth curve. The computed results of NDVI fitted data of Landsat
images showed gradually decreasing curve while with the Landsat 8

10
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

Author statement Declaration of competing interest

Ashok Amgoth: Formal analysis, Investigation, Methodology. The authors declare that they have no known competing financial
Hari Ponnamma Rani: Project administration, supervision. interests or personal relationships that could have appeared to influence
K.V. Jayakumar: Supervision, Writing-review and editing. the work reported in this paper.

Annexure.

References Mitsch, W.J., Gosselink, J.G., 2007. Wetlands Ecosystems, fourth ed. John Wiley & Sons,
Inc., New York, USA.
Naik, B.C., Anuradha, B., 2018. Extraction of water-body area from high-resolution
Bar, S., Parida, B.R., Pandey, A.C., 2020. Landsat-8 and Sentinel-2 based Forest fire burn
landsat imagery. Int. J. Electr. Comput. Eng. 2088–8708. https://doi.org/10.11591/
area mapping using machine learning algorithms on GEE cloud platform over
ijece.v8i6.pp4111-4119.
Uttarakhand, Western Himalaya. Rem. Sens. Appl.: Soc. Environ. 100324. https://
Orhan, O., Ekercin, S., Dadaser-Celik, F., 2014. Use of landsat land surface temperature
doi.org/10.1016/j.rsase.2020.100324.
and vegetation indices for monitoring drought in the Salt Lake Basin Area, Turkey.
Darby, F.A., Turner, R.E., 2008. Below- and aboveground Spartina alterniflora
Sci. World J. 1–11. https://doi.org/10.1155/2014/142939.
production in a Louisiana salt marsh. Estuar. Coast 31, 223–231. https://doi.org/
Orhan, O., Dadaser-Celik, F., Ekercin, S., 2019. Investigating land surface temperature
10.1007/s12237-007-9014-7.
changes using Landsat-5 data and real-time infrared thermometer measurements at
Dong, Z., Wang, Z., Liu, D., Song, K., Li, L., Jia, M., Ding, Z., 2014. Mapping wetland
Konya closed basin in Turkey. Int. J. Electron. Govern. 4 (1), 16–27.
areas using Landsat-derived NDVI and LSWI: a case study of West Songnen plain,
Ouma, Y.O., Tateishi, R., 2006. A water index for rapid mapping of shoreline changes of
Northeast China. J. Indian Soc. Rem. Sens. 42 (3), 569–576. https://doi.org/
five East African Rift Valley lakes: an emprical analysis using Landsat TM and ETM+
10.1007/s12524-013-0357-1.
data. Int. J. Rem. Sens. 27, 3153–3181. https://doi.org/10.1080/
Dong, J., Xiao, X., Kou, W., Qin, Y., Zhang, G., Li, L., Jin, C., Zhou, Y., Wang, J.,
01431160500309934.
Biradar, C., 2015. Tracking the dynamics of paddy rice planting area in 1986–2010
Ozesmi, S.L., Bauer, M.E., 2002. Satellite remote sensing of wetlands. Wetl. Ecol. Manag.
through time series Landsat images and phenology-based algorithms. Rem. Sens.
10 (5), 381–402. https://doi.org/10.1023/a:1020908432489.
Environ. 160, 99–113. https://doi.org/10.1016/j.rse.2015.01.004.
Pekel, J.F., Cottam, A., Gorelick, N., Belward, A.S., 2016. High-resolution mapping of
Eid, A.N.M., Olatubara, C.O., Ewemoje, T.A., Farouk, H., El-Hennawy, M.T., 2020.
global surface water and its long-term changes. Nature 540 (7633), 418–422.
Coastal wetland vegetation feature and digital change detection mapping based on
https://doi.org/10.1038/nature20584.
remotely sensed imagery: El-Burullus Lake Egypt. Int. Soil Water Conserv. Res. 8 (1),
Pettorelli, N., 2013. Climate and the NDVI: a complex story. In: The Normalized
66–79. https://doi.org/10.1016/j.iswcr.2020.01.004.
Difference Vegetation Index. Oxford University Press.
Ghosh, S., Mishra, D.R., Gitelson, A.A., 2016. Long-term monitoring of biophysical
Robinson, N.P., Allred, B.W., Jones, M.O., Moreno, A., Kimball, J.S., Naugle, D.E.,
characteristics of tidal wetlands in the northern Gulf of Mexico - a methodological
Erickson, T.A., Richardson, A.D., 2017. A dynamic Landsat derived normalized
approach using MODIS. Rem. Sens. Environ. 173, 39–58. https://doi.org/10.1016/j.
difference vegetation index (NDVI) product for the conterminous United States.
rse.2015.11.015.
Rem. Sens. 9 (8), 863–864. https://doi.org/10.3390/rs9080863.
Kirtiloglu, O.S., Orhan, O., Ekercin, S., 2016. A map mash-up application: investigation
Rokni, K., Ahmad, A., Selamat, A., Hazini, S., 2014. Water feature extraction and change
the temporal effects of climate change on salt lake basin. Int. Arch. Photogram. Rem.
detection using multitemporal Landsat imagery. Rem. Sens. 6 (5), 4173–4189.
Sens. Spatial Inf. Sci. XLI-B4, 221–226.
https://doi.org/10.3390/rs6054173.
Lee, T., Yeh, H., 2009. Applying remote sensing techniques to monitor shifting wetland
Roy, D.P., Wulder, M.A., Loveland, T.R., W, C.E., Allen, R.G., Anderson, M.C., Helder, D.,
vegetation: a case study of Danshui river estuary mangrove communities, Taiwan.
Irons, J.R., Johnson, D.M., Kennedy, R., Scambos, T.A., Schaaf, C.B., Schott, J.R.,
Ecol. Eng. 35, 487–496. https://doi.org/10.1016/j.ecoleng.2008.01.007.
Sheng, Y., Vermote, E.F., Belward, A.S., Bindschadler, R., Cohen, W.B., Gao, F.,

11
A. Ashok et al. Remote Sensing Applications: Society and Environment 23 (2021) 100547

Hipple, J.D., Hostert, P., Huntington, J., Justice, C.O., Kilic, A., Kovalskyy, V., Lee, Z. Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring
P., Lymburner, L., Masek, J.G., McCorkel, J., Shuai, Y., Trezza, R., Vogelmann, J., vegetation. Rem. Sens. Environ. 8 (2), 127–150. https://doi.org/10.1016/0034-
Wynne, R.H., Zhu, Z., 2014. Landsat-8: science and product vision for terrestrial 4257(79)90013-0.
global change research. Rem. Sens. Environ. 145, 154–172. https://doi.org/ Verpoorter, T., Kutser, T., Tranvik, L., 2012. Automated mapping of water bodies using
10.1016/j.rse.2014.02.001. Landsat multispectral data. Limnol Oceanogr. Methods 10, 1037–1050. https://doi.
Rundquist, D.C., Narumalani, S., Narayanan, R.M., 2001. A review of wetlands remote org/10.4319/lom.2012.10.1037.
sensing and defining new considerations. Rem. Sens. 20, 207–226. https://doi.org/ Vrieling, A., De Leeuw, J., Said, M.Y., 2013. Length of growing period over Africa:
10.1080/02757250109532435. variability and trends from 30 years of NDVI time series. Rem. Sens. 5 (2),
Sarp, G., Ozcelik, M., 2017. Water body extraction and change detection using time 982–1000. https://doi.org/10.3390/rs5020982.
series: a case study of Lake Burdur, Turkey. J. Taibah Univ. Sci. 11 (3), 381–391. Weier, J., Herring, D., 2000. Measuring vegetation (NDVI & EVI). Earth observatory. Natl
https://doi.org/10.1016/j.jtusci.2016.04.005. Aeronaut. Space Adm.
Shumway, R., Stoffer, D., 2019. Time Series: a Data Analysis Approach Using R. CRC Xu, R.G., Qiao, G., Wu, Y.J., Cao, Y.J., 2019. Extraction of rivers and lakes on Tibetan
Press. https://doi.org/10.1201/9780429273285. plateau based on Google earth engine. Int. Arch. Photogram. Rem. Sens. Spatial Inf.
Stehman, S.V., 1997. Selecting and interpreting measures of thematic classification Sci. 1797–1801. https://doi.org/10.5194/isprs-archives-xlii-2-w13-1797-2019.
accuracy. Rem. Sens. Environ. 65 (11), 77–89. https://doi.org/10.1016/S0034-4257 XLII-2/W13.
(97)00083-7. Zhu, Z., Bi, J., Pan, Y., Ganguly, S., Anav, A., Xu, L., Samanta, A., Piao, S., Nemani, R.R.,
Taufik, A., Ahmad, S.S.S., 2016. Land cover classification of Landsat 8 satellite data Myneni, R.B., 2013. Global data sets of vegetation leaf area index (LAI) 3g and
based on Fuzzy Logic approach. In: IOP Conference Series: Earth and Environmental fraction of photosynthetically active radiation (FPAR) 3g derived from global
Science, p. 12062. https://doi.org/10.1088/1755-1315/37/1/012062. inventory modeling and mapping studies (GIMMS) normalized difference vegetation
index (NDVI3g) for the period 1981 to 2011. Rem. Sens. 5 (2), 927–948. https://doi.
org/10.3390/rs5020927.

12