Assessment and prediction of shoreline change using multi-temporal satellite data and geostatistics: A case study on the eastern coast of India

Shoreline change is a long-term natural process and it is continuously changing due to erosion or accretion (Kannan et al. 2016; Ranasinghe 2016). However, anthropogenic activities in the fragile coastline have also significantly affected the global coastline. In the past few decades, coastlines around the world have witnessed rapid urbanization and the subsequent population growth (Kilibarda et al. 2014). Therefore, proper monitoring and management of coastal zones at micro level will assist in identifying the nature and process of the changes in fragile coasts, and help in management strategies (Micallef et al. 2018). Coastal landforms are very sensitive to coastal process enforced by waves, off shore current, wind, sediment transport by rivers, and anthropogenic activities (Carter 1988; Bird 2011; Kaliraj et al. 2017; Basheer Ahammed & Pandey 2019a).

Changes of coastal regions are a dynamic process, thus, regular monitoring of coastal zones is very important (Nayak 2002). Moreover, preparation of an appropriate coastal zone management plan as well as implementation of regulations in the coastal zone require spatial information on the coastal land use and land cover, the inventory and status of coastal habitats and information on ecologically sensitive areas (Camilleri et al. 2017; Khamis et al. 2017; Ramesh et al. 2017).

Coastal researchers have to overcome many challenges over the coming century, including those related to climate change and human activities like sea level rise, increased wave heights, elevated frequency and intensity of storm events, unexpected swells and depressions, sediment runoff, water quality deterioration, salt water intrusion, etc. (Thampanya et al. 2006; Jha et al. 2012). The development and implementation of effective strategies to manage these challenges is dependent on a better scientific understanding of how they will manifest along coastal zones (Cooper & Zmud 1990; National Research Council et al. 2000; Bodin et al. 2017). The geospatial technology has currently reached the edge of a technological revolution in which the accuracy, precision and affordability of geospatial technologies such as GIS, global positioning system (GPS), and remote sensing (RS) provide opportunities for their application to coastal zones (Shamsi 2005; Blankespoor et al. 2012; Craglia et al. 2012).

Coastal ecosystem management involves the procedures of monitoring and modeling which require a reliable information base and robust analytical technologies (Yang 2005). Conventional field-based mapping methods can still be vital but often logistically constrained, thus remote sensing and geospatial technologies, given their cost-effectiveness and technological soundness, are increasingly being used to develop useful sources or information supporting decision-making for a wide array of coastal applications (Wright 2009; Yang 2009). Although the coastal environment, because of its complex and dynamic landscapes, challenges the applicability and robustness of remote sensing and geospatial technologies (Wang et al. 2010). However, recent innovations in data, technologies, and theories in the wider arena of remote sensing and geospatial technologies have provided scientists with invaluable opportunities to advance their studies on the coastal environment (Chang & Lai 2014; Burgan & Aksoy 2018; Ehteram et al. 2018; Nabipour et al. 2020). Increased coastal population and intense development have threatened and degraded global coastal ecosystems, placing an elevated burden on those organizations responsible for the planning and management of these sensitive areas (Small et al. 2000; Small & Nicholls 2003; Li et al. 2010; Ahmed 2011).

The present study utilized the remote sensing data and GIS to predict future shoreline changes for a short-term (2025) and long-term (2050) period using historical shoreline change data and mathematical models.

Andhra Pradesh is one among the nine coastal states of India, located on the eastern coast of the Indian subcontinent and lies between 12° 35′ 23″ to 19° 10′ 15″ N latitude and 76° 44′ 32″ to 84° 47′ 33″ E longitude. Andhra Pradesh shares its state boundaries with Odisha in the northeast, Telungana in the northwest, Karnataka in the west and Tamilnadu in the south. The Bay of Bengal marks the eastern boundary of the state (Figure 1). Andhra Pradesh has the second longest coastline in the country which is 972 km in length and the coast plays an important role in coastal ecology and the Indian economy as well. Coastal dynamics by erosion and accretion is gradually changing the coastline position with resultant inundation in the low laying areas (Basheer Ahammed & Pandey 2019a) besides accelerated coastal changes caused by deforestation of large mangrove swamps (Radhakrishna Murthy & Bangaru Babu 2006). Most parts of the study area exhibit low elevated topography and deltaic environment, rendering the study area highly sensitive to shoreline positional changes.

The present study addressed two major objectives, the empirical shoreline change rate, and the prediction of future shoreline position along the east coast of India, Andhra Pradesh. Remote sensing technology is widely utilized based on the purpose of use as there are different satellite systems with a wide range of resolutions. However, Landsat satellite images are commonly used in coastal change studies with geospatial techniques. In this study, six periods of Landsat satellite images for the Andhra Pradesh coast were acquired from US Geological Survey (USGS) Earth Explorer covering a span of 43 years between 1973 and 2015. The major properties of the acquired satellite images are given in Table 1.

In the present study, satellite images were processed using ArcGIS 10.5 software, and after the geometric correction and image enhancement, shoreline corresponding to each individual period was generated using on-screen point mode digitization technique by using standard false color composite (FCC) with blue, green, and near infra-red bands to separate the land–water boundary distinctly. Many studies have been carried out using Landsat satellite images to analyze the coastal changes based on various interpretation techniques including visual interpretation and object oriented classification techniques (Barnhardt & Schwab 2009; Mahendra et al. 2011; Basheer Ahammed & Pandey 2019a). The principal shoreline was extracted and further utilized for shoreline change analysis and its future prediction. The detailed methodology is discussed in the following sections, and comprehensive methodology is shown in Figure 2.

In this study, shoreline was extracted for the years 1973, 1980, 1990, 2000, 2010, and 2015 and utilized for the shoreline change analysis using Digital Shoreline Change Analysis (DSAS v5.0). Using the tool, coastal change analysis was performed at 95% confidencel by the end point rate (EPR), least-squares regression rate (LRR), net shoreline movement (NSM), and shoreline change envelope (SCE) statistical methods. EPR and LRR were utilized for further analysis, such as empirical assessment and future prediction. EPR was calculated by dividing the distance of shoreline movement by the time elapsed between the earliest and latest measurements, i.e., the oldest and the youngest shorelines, while the LRR (represented by the slope of the line) of change statistic was determined by fitting a least-squares regression line to all shoreline points for transect. The method of linear regression includes these features: (1) all the data to be used, regardless of changes in trend or accuracy, (2) the method is purely computational, (3) the calculation is based on accepted statistical concepts, and (4) the method is easy to employ (Dolan et al. 1991).

Three components which are required to estimate shoreline changes include base line, shorelines, and transect line. DSAS application uses a measurement baseline method (Clow & Leatherman 1984) to calculate rate-of-change statistics for a time series of shorelines. In the analysis, baseline was formed parallel to the shoreline and at a distance of approximately 5,000 m in the landward direction, which is the starting point for all transects cast by the DSAS application. Transects were generated at 100 m intervals on the baseline which intersects each shoreline at the measurement points that are used to calculate shoreline change rates. Further, using the LLR method, which gives the most reliable accuracy among these methods, shorelines extracted using the model on the same date were compared with the predicted shoreline. Afterwards, the changes in shoreline through processes of accretion and erosion were analyzed in a GIS by measuring differences in shoreline locations by the DSAS derived results (Thieler et al. 2009).

To process the prediction model, two shoreline positions were required to estimate the shoreline position in the EPR model, i.e., the earliest (y₁) and the recent positions (y₂) and a minimum of three shoreline positions were required to estimate the shoreline position in the LRR model. Further, the model was used to estimate future shoreline for a short-term change (2025) and a long-term change (2050).

Climate change and increasing temperature are likely to cause global sea level rise with concomitant increasing trend in the frequency and magnitude of storm surges. When these two factors are combined, it will have the effect of focusing wave energy closer to the shore, leading to increased rates of coastal erosion (Linham & Nicholls 2010). This present study estimated the historical and future shoreline change rate and its future prediction along the east coast of India, Andhra Pradesh. The study observed that the coastal tracts in Andhra Pradesh will undergo high shoreline movement in future decades where major parts of the shoreline will experience erosion in comparison to accretion.

In the study, shoreline change rate for the period 1973–2010 was calculated for the prediction of future shoreline change rate, based on the historical assessment. The study observed higher erosion and accretion in the Krishna Godavari deltaic environment, whereas less dynamic coast was exhibited in the northern and southern coast of Andhra Pradesh. The study estimated shoreline change rate by using EPR, LRR, and NSM statistical tools and depicted high erosion in the coast of Krishna with −5,807.34 m positional shift. Similarly, highest accretion was observed along the Godavari coast with 4,611.95 m positional shift between 37 years of observation period. Based on the LRR estimation, average shoreline change rate experienced in the east coast between 1973 and 2010 was −0.69 m/y, whereas, EPR estimated shoreline change rate was −1.3 m/y between 1973 and 2010 (Table 2). Similarly, average erosion and average accretion were also estimated for the entire coastline with estimated values of −247 m (LRR −6.14 m/y, EPR −6.72 m/y) and 243 m (LRR 7.17 m/y, EPR 6.59 m/y), respectively. The results depicted that the east coast of Andhra Pradesh had experienced the highest rate of erosion as well as accretion during the observed period and the highest change was exhibited in the Krishna Godavari deltaic region (Figure 3). Based on the assessment, future shoreline positional shift was calculated for the year 2015 using the estimated values.

In the present study, shoreline change analysis was carried out along the 972 km-long coastline of Andhra Pradesh and calculated change rate for the period 1973–2015, and observed significant changes in the shoreline position. During the period 1973–2015, the average rate of shoreline change was 1.26 m per year (LRR = 0.46 m/y) and NSM 256 m. The highest rate of erosion was observed in the southern Godavari river mouth near Brahmasamedayam village with −174 m/y (LRR = 133.65 m/y), whereas the highest accretion was observed in the Krishna river mouth near Orangoli RF with 138.78 m/y (LRR = 109.62) (Table 2). Both high erosion and accretion were experienced in the river mouth, therefore, both oceanic and fluvial processes were involved in the huge shoreline movement. Orangoli RF, one of the largest mangrove forests in the Andhra Pradesh, witnessed accretion of 5.6 km length between 1973 and 2015, whereas, Brahmasamedayam coast observed erosion with a 5 km shift towards the land area between 1973 and 2015.

The present study also analyzed the average erosion and accretion between 1973 and 2015. Average erosion was observed as −6.83 m/y (LRR = −6.73) and average accretion as 6.24 m/y (LRR 5.75 m/y). This indicates that the changes along the coastline are irregular with a dominance of erosion. Different from the Krishna–Godavari deltaic plain, southern and northern parts of the study area exhibited low erosion or accretion (Figure 4). Further, the result was utilized for the future shoreline prediction of 2025 and 2050 shoreline position.

In the present study, EPR and LRR statistical tools were utilized for predicting the future shoreline position. Based on the historical shoreline change rate, shoreline was predicted for the year 2015 using EPR and LRR models and, further, the RMSE values were obtained by comparing the actual shoreline with the predicted shoreline of the same date. Both actual and predicted shoreline changes projected similar results in the southern and northern part of the study area, whereas higher rates of error were observed in the deltaic plain and river creeks. Considering the EPR and LRR models, RMSE and r² values obtained in the study area were examined which revealed that the LRR method has the highest accuracy, and therefore the results of the analysis were interpreted by considering the LRR method (Figure 5). However, as the model can produce a high number of errors in the predicted values, the model predicted positions were validated with the observed shoreline positions. The model-derived shoreline positional shift was validated with the observed positional scenario of the 2015 shoreline. Both observed and predicted shoreline positional shifts between 2010 and 2015 were used for the validation of the EPR and LRR models. In the study, RMS error for the LRR model varies from 0 m to 159 m. The overall error for the entire predicted shoreline was observed to be 31.17 m (RMSE), whereas RMSE received for the EPR model was observed to be 79.96 m for the entire predicted shoreline. It has been observed that model prediction error is higher in the Krishna and Godavari deltaic plain where both rivers meet the Bay of Bengal. Based on the correlation between actual and predicted shoreline through r², a positive correlation with 94.8% on the LRR model-based assessment was achieved while the EPR-based prediction model provided accuracy with 82.2% (Table 3). EPR utilizes the oldest and newest shorelines for estimating the slope of the shoreline shift, while LRR utilizes the entire shoreline given for the analysis (Basheer Ahammed & Pandey 2019a), since the shoreline is dynamic and the positional shift may be uncertain from year to year, therefore, predicting the shift in shoreline position is very unpredictable. However, LRR estimates the slope of the positional shift with calculating the average positional shift by utilizing the available shoreline data in the model. It can be seen that the LRR model gives accurate information in comparison to the EPR model, and the uncertainty of the predicted and observed positional shift is very much less in the LRR model (Figure 5).

The study analyzed the shoreline positional shift between 2010 and 2015 for the model validation, and observed an average shoreline positional shift with −2.39 m for the entire shoreline. Considering the model prediction, the LRR model produced a better result in comparison to the EPR model (Table 3). Based on the prediction, LRR estimated an average shoreline positional shift with −2.11 m, whereas EPR estimated −5.74 m for the entire shoreline between 2010 and 2015. During 2010 and 2015, EPR and LRR models estimated higher accretion with 678.6 and 897.36 m, respectively, whereas observed shoreline shift was estimated at 889.38 m for the same period. Similarly, LRR and EPR models estimated higher erosion with −463.23 m and −439.8 m, respectively, while observed positional shift indicated −792.364 m during five years of duration (Figure 6).

The river mouth is a natural and strategic checkpoint that exhibits a large scope of physical, sedimentological processes, whereas the estuary entrance is one of the most critical variables controlling the hydrodynamics and broader environmental processes of the deltas (Tay et al. 2013; Rajasekaran et al. 2014; Sreenivasulu et al. 2016). Owing to the Krishna and Godavari rivers' influence, the central part of the Andhra Pradesh coastline is dynamic in nature and, therefore, RMS error was higher in these particular areas. The positional error adjustment has been functioned with the predicted shoreline change rate. The positional shift of the shoreline in x and y directions (latitudinal and longitudinal) was calculated by comparing the actual and predicted shoreline of 2015. The calculated RMS error was adjusted to the predicted shoreline of 2025 and 2050.

Based on the empirical shoreline change, the study investigated the future shoreline positional change for the years 2025 and 2050 in order to understand the short-term and long-term shoreline movements. Coastlines around the world are often affected by extreme climate conditions with events like tsunami, cyclone, and associated storm surges. The intensity and frequency of these extreme events are increasing gradually (Basheer Ahammed & Pandey 2019b). Although the impacts of these extreme events are not considered for the prediction of future shoreline, they will provide baseline information for adaptation strategies and policy framework.

The shoreline positional change predicted using the LRR model depicted the average rate of shoreline positional change during the 2015–2025 period as −4.64 m, whereas between 2015 and 2050 it will increase to −16.25. In the study area, the highest rate of accretion will take place on the western coast of Kakinada Bay with a 1,272 m positional shift towards the ocean, while during 2015–2050, the coast will undergo erosion with a 4,455 m positional shift (Table 4). However, considering the long-term change (2015–2050), higher erosion will occur in the Godavari river mouth with a −6,898 m positional shift towards the land, whereas during 2015–2025 the coast will experience a 1,971 m shoreline shift (Figure 7).

The rate of change in shoreline position is an important factor for the prediction of the future trend of shoreline movement (Maiti & Bhattacharya 2009). An increasing trend of sea level rise and potential impacts of extreme events, including flood and coastal erosion, will likely affect the wide coastal stretches (Bagheri et al. 2019). However, considering the historical shoreline movement without considering the uncertainty due to climate change and extreme events will also provide a significant insight about the shoreline movement (Barman et al. 2015; Esmail et al. 2019). The present study addressed the historical shoreline movement and its future perdition along the Andhra Pradesh coast utilizing the earth observation satellite images between 1973 and 2015. The results obtained from the present study depict that error between predicted and actual shoreline is higher in the river mouth and deltaic plains, while the other parts of the coast exhibited low RMS errors. The results obtained by the LRR model do not match exactly with the actual rate at observed coastal locations; however, a ±10 m difference can be permissible due to different methods used in the estimation of rate of change in shoreline position (Deepika et al. 2014). Since the model is not considering any physical or geological and extreme climatic variable, variability in the predicted shoreline position can be expected. However, the model predicts the future shoreline by using the empirical shoreline change and that will give a generous outline about the future trend. Changes in the deltaic environment are dynamic, therefore, the higher rate of changes observed in the Krishna and Godavari deltaic environments. As fluvial process are also contributing to the shoreline change, prediction in the river mouth will not provide an accurate prediction. The average RMS error observed in the Krishna–Godavari deltaic plain is 39.53 m, whereas in the coastline excluding the Krishna–Godavari deltaic plain the observed value is 14.56 m. Since the RMS error is very low in the northern and southern coast of Andhra Pradesh, it indicates that the LRR model is most accurate and reliable in the non-deltaic coasts. However, the model provides an insight about the future shoreline position. The vulnerable coast observed in the present study can be protected by artificial protection embankments and natural shields like mangroves and wetlands. The eastern coast of India is one of the most vulnerable regions as every year major cyclones affect the region, especially along the Andhra Pradesh coast (Basheer Ahammed & Pandey 2019c). Therefore, the change rates may vary in the areas where the eye of the cyclone passes through.

The impacts of global sea level rise will affect low lying coastal areas due to the intensified coastal flooding and inundation, erosion of shorelines, salt water intrusion into estuaries and coastal aquifers, and drainage problems and is expected to affect millions of additional people each year by late this century. As per the latest report of the Intergovernmental Panel on Climate Change (IPCC), Visakhapatnam is one among six cities that could be exposed to coastal flooding if the sea levels rise by 50 cm due to global warming, and the city has witnessed a higher rate of sea level rise with above 5 mm/yr (Figure 8). As well, sea level rise is also a higher threat to Kirishna Godavari deltaic plain, Machilipatnam and Kakinada, as the difference between the land mass and the sea is very small. In addition, central Andhra Pradesh has witnessed a sea level rise between 3.5 and 5 mm/y (Figure 8). The trend of sea level rise is estimated using satellite altimetry between 1992 and 2021 and the average trend of sea level rise in the Bay of Bengal is estimated as 3.7 ± 0.4 mm/yr (Figure 9), while the average global sea level rise is 3.2 mm/yr.

In the present study, the highest rate of shoreline change in the central Andhra Coast was observed as the area also witnessed the highest rate of sea level rise, and this indicated the influence of sea level rise in changing shoreline position. The trend of sea level rise is expected to increase in the coming decades, and that would result in permanent inundation in the low lying coastal areas.

This study revealed that the use of remote sensing technology and geospatial techniques are the vital tools for predicting and analyzing the dynamic coast. Digital shoreline analysis system is the best tool for monitoring shoreline movements and although there are many statistical approaches available within the model, EPR and LRR are the simplest and best tools for analyzing historical shoreline change rate. Landsat data with 30 m spatial resolution is widely used for analyzing empirical shoreline change rate as the resolution of the image is not a big challenge in regional level studies. Therefore, the present study also utilized the Landsat series data for shoreline change analysis. The northern part of the study area, especially along the coastal tracts of Vishakhapatnam, is predicted and observed with a lower rate of erosion and accretion. Physiographic characteristics of northern Andhra Pradesh are different from the southern coast, as high elevated topography and the hard rock terrain of the Eastern Ghats protect the northern coast of Andhra Pradesh whereas on the southern coast, the geomorphologic feature is mainly coastal plain, but the absence of river creeks reduced the rate of shoreline change. Both marine and fluvial influences are controlling the shoreline movements in the central parts of the study area, therefore, the highest rate of erosion and accretion is observed and predicted along that coastline. Coastal areas have sand bars, spits and deltaic landscape and topographically delicate stone, which are in fact erodible. Further, this study can be carried out using high resolution satellite images and GPS surveys with advanced technology such as DGPS and geotagging applications to demarcate the shoreline positions more accurately. Littoral drift, near shore bathymetry, tidal action, sea waves, construction of seawalls, groins or breakwaters, etc., are influencing factors that are both natural and manmade, and modify the shoreline configuration. The result of the study can be used as baseline information for coastal engineers, planners, and coastal zone management authorities for coastal conservation and management.

The authors would like to thank the US Geological Survey for the Landsat data. The authors also would like to acknowledge the Central University of Jharkhand for proving the facility and opportunity to do this research work.

Data cannot be made publicly available; readers should contact the corresponding author for details.