A global high resolution coastline database from satellite imagery

Abstract

In this dataset, we present global coastlines, water probability maps, and intertidal zones derived from a large collection of multispectral images acquired by Maxar (formerly DigitalGlobe) satellites between 2009 and 2023. The extracted coastlines correspond to the median tidal height of all image acquisitions at a location, with the modeled tidal height included in the product. These products are provided at a high spatial resolution of 2 m across the globe. Coastline products are compared with the standard coastline products from NOAA. The water probability and coastline data are used to generate coastal intertidal zones, which represent the horizontal extent water covers during the transition between low tide and high tide. Intertidal zones are dynamic regions that could be recognized as sensitive coastal areas to sea level variations. We detect the largest intertidal zone in south-central Alaska, with a total area of 124.7 km² and a width of 3.8 km. The high resolution coastline product can support coastal resources management and planning and coastline changes corresponding to sea level rise.

Background & Summary

Coastal regions are important environments that require detailed study and consistent monitoring, and remote sensing techniques have been widely used as a cost-effective method for many coastal applications. Coastline mapping is one such remote sensing application that has been widely studied. Normalized Difference Water Index (NDWI) can be used to separate water features and non-water features, thus facilitating the identification of coastlines¹. Dai et al.² developed a coastline extraction method based on NDWI applied to a stack of satellite images, demonstrating its feasibility for generating 2-meter resolution coastlines and water probability maps. This study extends the algorithm to approximately 500 TB of multispectral images on a global scale, including the Arctic and Antarctica. In addition, implementation schemes to effectively use these datasets to derive other coastal parameters such as intertidal zones are presented.

The coastline products are compared against the NOAA Continually Updated Shoreline Product (CUSP - https://shoreline.noaa.gov/cusp.html), which is updated regularly. The comparison was also carried out using the Global Self-consistent, Hierarchical, High-resolution (GSHHS - https://www.soest.hawaii.edu/pwessel/gshhg/) shoreline product³ to emphasize our products’ validity against widely available shoreline products. Remotely sensed data capturing the constantly changing coastal zone are pre-processed to improve their quality and interpretability. This includes orthorectification to remove terrain induced distortions⁴. A reference frame in the form of a Digital Elevation Model (DEM) is generally used for orthorectification. In the present study, the orthorectification effect on satellite imagery capturing the coastal zone is evaluated using the Geoid (EGM2008)⁵ and Space Radar and Topography Mission (SRTM-https://www.earthdata.nasa.gov/data/instruments/srtm)⁶ DEMs with the shorelines providing reference. In addition, we introduce a water probability and shoreline-based approach to identify low-lying coastal areas that are vulnerable to coastal flooding and storm surges. These regions encompass the intertidal zone that lies between the highest level enveloped by the waves and the lowest level exposed by the tide⁷.

To the best of our knowledge, there is no complete global dataset to inform shorelines, intertidal zones, and water probability percentages at meter-scale resolution. We aim to fill this gap by providing a comprehensive coastal dataset in one data descriptor and platform. The main focus of this data descriptor is providing a global coastline database rather than the temporal evolution of the coastline. These data, derived from large inventories of multispectral images, will be a critical benchmarking dataset for global coastal mapping. Each of these products and methods are described in detail in Methods below.

Methods

This study adopts the shoreline mapping method by Dai et al.² to produce a global shoreline product from a large volume of repeat, high-resolution multispectral satellite imagery (WorldView-2 and WorldView-3). Dai et al.² evaluated the statistics of NDWI over GSHHS coastline-based water and land areas on a large collection of test images from Iceland and Alaska to estimate the water classification threshold. The images were preprocessed to correct for distortions and to standardize them for analysis. This involved radiometric correction, atmospheric correction, and geometric correction. Based on the NDWI statistics generated for test images, an adaptive thresholding method was used for water classification² with the threshold, T, for each individual image is defined as:

$$T=({\mathit{mean}}_W+{\mathit{mean}}_L)/2$$ 1

where mean_W and mean_L are the mean NDWI values within the known water and land areas, respectively². Dai et al.² defined the water probability in each pixel as the ratio of the number of measurements that classify a pixel as water over the total number of measurements. The final water classification was determined based on the water probability map and the 50th percentile threshold². Both the use of stacking multiple repeat images and water probability maps mitigated the misclassification of shadowed surfaces and cloudy regions. The stacking image method is highlighted also to prevent errors from temporal anomalies such as clouds, shadows, and tidal heights. Image coregistration was not implemented to save computation time, also considering the fact that coregistration errors (1 ± 4 m) can be mitigated through stacking (Fig. 5 in Dai et al.²). It should be noted that our mapped coastlines are tagged with the image acquisition date (accurate to seconds) and the modeled tidal height (at the acquisition time) using the 1/6° resolution TPXO-9.1 global inverse tide model (an update to Egbert and Erofeeva (2002)⁸).

Fig. 5.

The largest intertidal zones (red) are detected in high-altitude regions. They are scattered in Alaska, Northern Canada and Svalbard. The areas of these intertidal regions in square kilometers are highlighted in the top-left corner of each zoomed-in map.

Datasets used

The multispectral satellite images from WorldView-2 (since 2009) and WorldView-3 (since 2014) used in this study are listed in Table S1. A preview of the imagery archive is available at https://xpress.maxar.com/ (last accessed April 21, 2025). Multispectral imagery can be accessed at no cost to NASA funded researchers through NASA’s Commercial Satellite Data Acquisition (CSDA) Program (Level 1B Multispectral data at https://search.earthdata.nasa.gov/search?q=csda, last accessed April 21, 2025). The multispectral band imagery has swath widths of 13 to 17 km and spatial resolutions of around 1.8 m for WorldView-2 and 1.2 m for WorldView-3 (https://resources.maxar.com/data-sheets/worldview-3, last accessed April 21, 2025). The water mask, derived from adaptive thresholding, for each individual multispectral image is interpolated on a regular 2-m resolution grid. We processed approximately 1.3 million multispectral images (464 TB) acquired between 2009 and 2023.

Coastline comparison

The coastline along the entire continental United States was retrieved and compared against standard coastline products. Our coastline product is evaluated against the CUSP shorelines from NOAA, which are acquired as tiles containing the coastal states along the United States. CUSP provides the most up-to-date shoreline of the United States that is derived from state-of-the-art cartographic technology. In addition, a comparison was carried out against the GSHHS shoreline product³, which is downloaded as a single shapefile covering the entire United States. Each of the extracted shoreline products consists of hundreds of polyline segments, thus an organized method to evaluate their validity is needed.

The extracted shoreline is in geodetic coordinates based on the WGS 84 ellipsoid and the units are in decimal degrees. To formulate the comparison in metric units, the mapped shorelines were projected to the Universal Transverse Mercator (UTM). The comparison is carried out on the outermost shoreline that contains simple straight shoreline segments, avoiding complex geometric shapes. For this task, an AOI polygon was digitized to cover all the coastal states of the continental US, and it was used to capture the outer shorelines. The extracted outer shoreline contained discontinued segments and thus all parts were then merged to yield a single shoreline product.

Point features are generated using a 50 m distance interval along this shoreline that covers the entire continental US. The closest distance from each of these points to the reference shoreline is calculated. The perpendicular distance is the closest. The Analysis-> Proximity -> Near tool in ArcGIS Pro was used for this purpose as it estimates the closest distance between two features. In this case, the distance between a point feature class (points along the mapped shoreline) and a polyline feature class (reference shoreline) is used.

A comparison of our mapped shoreline is carried out against the NOAA CUSP and GSHHS products along each coastal state in the continental US. The statewide shoreline accuracy statistics of mean, median, root-mean-square-error (RMSE), and standard deviation were estimated and reported.

GEOID vs SRTM orthorectification effects on coastal images

This section evaluates the effect of the Digital Elevation Model (DEM) on the orthorectification of WorldView-2 images used for coastline mapping. For this purpose, we used imagery from Goleta, California, which contains steep, coastal slopes reaching ~75 degrees. Steeper coasts facilitate more pronounced orthorectification distinctions generated from different DEMs. Three WorldView-2 images over this area were orthorectified using the SRTM DEM and the geoid (EGM2008). The geoid is the equipotential surface of the Earth that best fits the global mean sea level⁹. The Geoid has been shown to provide better orthorectification along coastlines as DEMs tend to be noisy in coastal areas or lack sufficient spatial resolution².

NDWI maps were derived from the two WorldView-2 orthorectified images obtained in 2020. A modified NDWI index, which was introduced for WorldView-2 and 3 images¹⁰ using Band 1 (Coastal blue) and Band 8 (NIR2) (Dai et al. ² (Eq. (2)).

$$\mathit{NDWI}=(\mathit{Coastal}-\mathit{NIR}2)/(\mathit{Coastal}+\mathit{NIR}2)$$ 2

The NDWI map from the geoid-orthorectified image shows agreement with the extracted shoreline along the yellow-orange pixels, which represent the water-land boundary (Fig. 1b). In contrast, the NDWI map of SRTM shows the green-yellow pixels (water-land boundary) shifting toward seaward (Fig. 1d). The comparison of NDWI from two orthorectified images (also Figure S1) reveals that there is a consistent 2–3-pixel shift around the shoreline, which is likely caused by the higher surface elevation values of SRTM DEM near coastlines (Fig. 1c). Thus, in our data processing, 2.5 arc-minute Earth Gravitational Model 2008 (EGM2008)⁵ is selected to orthorectify images for terrain correction.

Fig. 1.

NOAA topo-bathy lidar DEM (spatial resolution = 10 m) (a), NDWI map of the geoid orthorectified image (b), SRTM DEM (spatial resolution = 30 m) (c) and NDWI map of the SRTM orthorectified image (d). The orthorectified image for both cases was from the WorldView-2 satellite acquired on 7 October 2011 (https://discover.maxar.com/e3f0141a-1ecc-11f0-a88a-c3c20e6c4080, last accessed April 21, 2025). Our mapped shoreline is indicated with the black line as a reference. The red dot in the inset marks the location of the study area near Gaviota, California. The lidar DEM (a) provides an accurate representation of topography and bathymetry along our mapped shoreline. In contrast, SRTM DEM model (c) shows 10 m value pixels beyond the mapped shoreline, indicating the errors in elevations around the selected shoreline. NDWI map derived from SRTM orthorectified image reflects a seaward shift of the green-yellow pixels (water-land boundary) (see also Figure S1).

Identification of intertidal zones/coastal low-lying areas

The water probability mapping method developed by Dai et al.² is based on the ratio of the number of measurements that classify a pixel as water, according to the NDWI, over the total number of measurements. This approach also yields information about the extent of the intertidal zone. The water probability maps generated by Dai et al.² provide examples of the exposed land area due to low tide, as well as the same areas, inundated due to high tide. These findings reveal coastal areas of large variability that are perceived to be the most vulnerable to sea level changes².

Dai et al.² used multiple images denoted in the ‘nov’ file for each area to derive the water probability. The higher the number of images, the more accurate the estimated water probability percentage value. For instance, for a 20% water probability, there needs to be at least five images. Thus, as an initial step, the water probability maps were filtered based on the minimum number of images of five using the ‘nov’ file. This step ensures the regions that utilize at least five images are only considered and less accurate water probability maps without a sufficient number of repeats are filtered out from the analysis.

The identification of coastal intertidal zones is conducted using a robust and methodical approach. By adopting suitable thresholds of water probabilities, the horizontal coastal dynamic range is initially filtered. For this purpose, minimum and maximum water probabilities of 20% and 80%, respectively were used. It should be noted that using a low water probability like 5% for the minimum threshold would not be feasible in regions where only a small number of images (e.g. <5 images) were used to derive water probability. Therefore, a 20% threshold was selected, and the experiments carried out in random coastal regions yielded good results, confirmed by visual comparison. The resulting probability masks were then eroded and filtered to remove spurious, isolated clusters. Based on the test data from the extensive coastal low-lying regions in Alaska and Bangladesh, we find that a minimum pixel width of 2500 is appropriate.

We find that some inland water bodies and shadowy areas were also filtered in for the low-lying coastal area maps. To address this, the extracted outer shorelines were incorporated to fine-tune the algorithm by distinguishing the actual intertidal zones (low-lying tidal regions) from the inland water bodies. The derived coastal low-lying areas are tested and qualitatively visually validated in known high-density coastal low-lying regions such as the Yukon delta in Alaska and Bangladesh. The refined algorithm was adopted to yield coastal intertidal zones around the globe at a 2 m resolution (Fig. 2).

Fig. 2.

Workflow of methodology to map intertidal zones. The water probability maps (prob), image number files (nov) and shoreline maps (coast_tide) were incorporated in generating these intertidal zone maps.

Data Records

The data products are available at NSF Arctic Data Center Repository¹¹. The extracted shorelines are provided in shapefile format, containing line segments in WGS 84 geodetic coordinates (EPSG:4326) with a resolution of 2 m and with attributes of modeled tidal height and image acquisition date (Fig. 3). Considering that the modeled tidal height might have errors for deep estuaries or narrow fjords, the accurate image acquisition time is provided to allow future reestimation of better tidal information from other sources. The water probability maps and coastal intertidal zone maps are provided in GeoTIFF format. For mid-to-low latitude regions, the data are in WGS84 UTM projection EPSG:326NN for the Northern Hemisphere and EPSG:327NN for the Southern Hemisphere, where NN represents the UTM zone number. Polar stereographic projections are applied for high-latitude regions, specifically EPSG:3413 for the Arctic and EPSG:3031 for Antarctica. The ‘nov’ file indicates the number of overlapping images used in the analysis within each area. The ‘bound’ file specifies the boundaries of each image used in any specific area. The ‘low’ files indicate the coastal intertidal zones (low-lying areas) along the global coastline. The naming convention is consistent with all five files in each tile indicating the corresponding UTM zone followed by the corresponding tile ID (Fig. 4). The tile ID is consistent with the naming convention used in the ArcticDEM¹², EarthDEM, and REMA products¹³. The dataset is provided in a total of 64 folders for each sub-region covering the whole world (Table S1).

Fig. 3.

Mapped global coastline with blue, orange, and green colors representing the Arctic, Earthdem and Antarctic regions, respectively (refer Table S1). These 2 m resolution coastlines are derived using satellite images acquired between 2009 and 2023. The acquisition date and time of the images used for each segment are included in the attributes of the coastline products (Fig. 4).

Fig. 4.

Comparison of mapped coastline against the NOAA coastline along the continental United States. The coastline attributes (tile ID, tidal height, and source date), water probability, and number of repeat maps are also highlighted.

Technical Validation

Coastline comparison

Our mapped shoreline is compared against the NOAA CUSP product and GSHHS product along each coastal state of the United States (Table 1). The statewide shoreline difference statistics include mean, median, root-mean-square-error (RMSE), and standard deviation. Table 1 presents the comparison of the shoreline product against the NOAA CUSP and the GSHHS. The NOAA CUSP product provides a multibeam and bathymetric lidar assimilation, which is regularly updated and thus a more accurate source of data. On the other hand, the GSHHS product is based on two databases namely, World Vector Shorelines (WVS) and CIA World Data Bank II. It is constructed entirely from hierarchically arranged closed polygons and is perceived to be less accurate. Therefore, the comparison against the NOAA CUSP product is more relevant than the GSHHS product. These factors are reflected in the coastline comparison results listed in Table 1. For instance, the RMSE values lie in the ranges of 14–27 m and 53–171 m for NOAA CUSP and GSHHS, respectively. The standard deviations against NOAA CUSP and GSHHS are 9–24 m and 36–138 m, respectively.

Table 1.

Coastline comparison against the NOAA CUSP product and GSHHS product based on each coastal state along the continental United States.

Error metric		TX	LA	FL	SC	NC	VA	MA	DE	NJ	NY
Number of samples		10007	1514	10010	6247	9995	2282	2152	1915	1883	9752
CUSP	Mean (m)	15	27	10	21	19	15	13	11	11	17
	Median (m)	10	19	8	18	16	13	12	10	7	14
	RMSE (m)	20	37	14	27	24	19	16	15	17	22
	Std Dev (m)	14	24	9	17	14	12	9	9	13	13
GSHHS	Mean (m)	97	126	63	100	87	67	66	47	98	37
	Median (m)	64	124	44	38	78	66	50	43	96	24
	RMSE (m)	149	147	87	171	117	76	85	59	112	53
	Std Dev (m)	113	75	59	138	78	36	53	35	54	39

Intertidal zones

Over the coming decades, more coasts worldwide will be exposed to increasing risks including coastal erosion due to climate change and sea-level rise. In this context, having a global shoreline and intertidal zone products (low-lying coastal regions) on a single platform is encouraging for future research. Like any global product, this dataset contains outliers and artifacts. Overall, it offers a valuable resource to benchmark global shorelines and intertidal zones at a 2 m resolution with additional data such as the water probability maps, and the image data. The largest intertidal zones were detected in high-altitude regions such as Alaska, Northern Canada, and Svalbard (Fig. 5). We detect the largest intertidal zone in south-central Alaska, with a total area of 124.7 km² and a width of 3.8 km, and median intertidal zones as 1.06 km². This finding echoes Archer 2013¹⁴ classification of south-central Alaska as a region with the world’s highest tides with hypertidal tidal ranges exceeding 10 m. It should also be noted that some of the mid-water probability values reflect water-ice in the high-latitude areas.

The resource-rich coastal zones attract large populations, and these increasing coastal communities at risk are expanding the stress due to land use and hydrological variants in these tidal deltas^15,16. In this context, low-lying coastal urban areas and populated deltas (e.g. Asian megadeltas) can be identified as key hotspots of coastal vulnerability¹⁷. Currently, South, South-east, and East Asia and small and barrier islands are labeled as high-risk regions regionally¹⁷.

The remote sensing mapping of the intertidal zones is challenging owing to the rapidly changing tidal conditions and the image-tide time sensitivity¹⁸. The coastal low-lying areas are a revealing dataset derived from satellite images for coastal risk assessment in the dynamic coastal zone. The risk is specifically substantial for regions with extensive low-lying coastal regions such as Bangladesh and Alaska. For instance, the western coastal zone of Bangladesh is highly vulnerable to surge flooding due to its low-lying coastal land and poor defense against surge waves¹⁹. Our results show that high-density coastal low-lying regions along Bangladesh are located right next to communities, indicating high risk (Fig. 6). These tidal regions implicate the highest variability within the already dynamic nearshore region. These low-lying coastal zones also play a major role in ecosystem balance²⁰, storm surge protection¹⁷, and aiding in coastline stabilization²¹.

Fig. 6.

The demarcation of intertidal zones (red) which is in close proximity to highly populated Bangladeshi coastal suburbs indicates the significance of risk. AOI A provides an example of many suburban houses located less than 100 m distance to the low-lying coastal regions. The intertidal zone covers 12.9 km².

Usage Notes

The coastline, water probability, and intertidal zone data are provided in a total of 64 folders covering the whole globe including the Arctic and Antarctic regions. It is worth noting that most published coastal datasets are of coarse resolution with limited coverage. This dataset introduces a resource to benchmark satellite-derived coastlines at a 2 m spatial resolution. The future reuse of these data is encouraged as the mapped coastlines are tagged with the image acquisition date (accurate to seconds) and the modeled tidal height (at the acquisition time). Thus, the data are intended to encourage research that focuses on the dynamic coastal zone and extend further to develop temporal coastlines or water probability datasets in any region worldwide.

Author contributions

S.D.M.: Methodology, Data Processing, Validation, Visualization, Writing-Original Draft. C.D.: Conceptualization, Methodology, Data Processing, Writing-Original Draft, Writing-Review & Editing, Supervision. I.M.H.: Conceptualization, Methodology, Writing-Review & Editing. E.L.: Methodology, Writing-Review & Editing. E.H.: Methodology, Data Transfer Support, Writing-Review & Editing. All authors reviewed the manuscript.

Code availability

All codes to derive coastlines and water probability maps are made public. These are available at: https://github.com/Chunli-Dai/CoastlineMapping. The codes to derive coastal intertidal zones from water probability and coastline maps are available at: https://github.com/sanduni30/IntertidalZoneMapping.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41597-025-05180-9.