On this page

Is the shoreline near me advancing or retreating?

What you can answer

Multi-date shoreline position for any coastal AOI from a water-index waterline
Erosion vs accretion rate in m/yr along shore-normal transects (DSAS-style)
Where the coast is changing fastest (hotspot map of retreat/advance)
Decadal trend by extending back to Landsat Collection 2 (1982+)
Seasonal vs long-term signal by comparing pre- and post-monsoon scenes

What you can NOT answer with these alone

True erosion vs tide/run-up shift without tidal correction or consistent tide-stage scene selection — the optical waterline is instantaneous.
Sub-pixel change below ~30 m (HLS/Landsat) or ~10 m (Sentinel-2); narrow-beach change can be lost to mixed pixels.
Sediment volume / beach profile — these are 2-D position changes, not 3-D volume; needs lidar/DEM or surveyed profiles.
Cause attribution (groynes, ports, sea-level rise, sand mining) — imagery shows change, not its driver.

Code template

import earthaccess
import xarray as xr
import numpy as np

earthaccess.login(strategy="netrc")

# AOI: Andhra–Tamil Nadu east coast
aoi = (80.0, 12.5, 81.0, 16.0)
epochs = [("2015-01-01", "2015-04-30"),   # dry-season, lower turbidity
          ("2020-01-01", "2020-04-30"),
          ("2025-01-01", "2025-04-30")]

shorelines = {}
for start, end in epochs:
    granules = earthaccess.search_data(
        short_name="HLSS30",          # HLS Sentinel-2, 30 m
        bounding_box=aoi,
        temporal=(start, end),
        cloud_cover=(0, 10),
    )
    # Open green (B03) and SWIR1 (B11) bands; stack to xarray
    # ds = xr.open_mfdataset(...) with the granule assets

    # MNDWI = (green - swir1) / (green + swir1)
    # mndwi = (ds.B03 - ds.B11) / (ds.B03 + ds.B11)

    # Composite over the epoch (median) to suppress clouds/waves
    # mndwi_c = mndwi.median(dim="time")

    # Water mask: MNDWI > 0 is water (verify threshold on a clear scene;
    # tune per coast — turbid/built coasts may need a small positive offset)
    # water = mndwi_c > 0
    # Trace land/water boundary -> waterline vector (e.g. skimage / rasterio)
    shorelines[start] = "<waterline geometry>"

# Cast shore-normal transects from a baseline, intersect each waterline,
# and fit per-transect rates:
#   end-point rate  = (pos_late - pos_early) / years
#   linear rate     = slope of position vs date (>=3 epochs)
# Sign convention: seaward (+) = accretion, landward (-) = erosion

Expected output

Multi-date waterline overlay on a basemap (one colour per year)
Transect map coloured by rate: red = retreat/erosion, blue = advance/accretion
Histogram of m/yr rates across all transects (median + IQR)
Hotspot callouts: fastest-eroding and fastest-accreting transects with values

Caveats

Tide & wave run-up bias: the optical waterline is the instantaneous land/water edge. Without tidal correction, compare scenes at similar tide stage and treat near-pixel rates as uncertain.
Index threshold is not universal: MNDWI > 0 is a common starting point but turbid plumes, foam, wet sand and dark built surfaces shift it — verify the threshold on a clear scene per coast rather than hard-coding it.
Resolution floor: detectable change is bounded by pixel size (~30 m HLS/Landsat, ~10 m Sentinel-2). Use 10 m Sentinel-2 L2A for narrow beaches.
Cloud and monsoon gaps: the east coast is cloudy in NE-monsoon months; prefer dry-season composites and median-stacking to fill gaps.
Validate before publishing rates: cross-check against surveyed shorelines or DSAS reference data; satellite-derived shoreline rates carry several-metre uncertainty.

Cross-DAAC composition

HLS (HLSS30/HLSL30) is served by LP DAAC — single auth. Extending to Landsat Collection 2 L2 pulls from USGS (separate credentials), and 10 m Sentinel-2 L2A from ESA/Copernicus.

Sources

HLS (HLSS30/HLSL30): https://lpdaac.usgs.gov/products/hlss30v002/
Landsat Collection 2: https://www.usgs.gov/landsat-missions/landsat-collection-2
DSAS (Digital Shoreline Analysis System): https://www.usgs.gov/centers/whcmsc/science/digital-shoreline-analysis-system-dsas
Water indices (NDWI/MNDWI/AWEI) — McFeeters 1996; Xu 2006; Feyisa et al. 2014

⚲ How a scientist answers this

Parameters

Instantaneous shoreline position, defined as the land/water boundary extracted per epoch from multi-date optical imagery. The waterline is found by thresholding a water index — NDWI (green vs NIR), MNDWI (green vs SWIR, better over turbid/built coasts), or AWEI — on Landsat Collection 2 (1982+, 30 m), HLS (30 m, 2013+), or Sentinel-2 (10 m). The answer is a change rate in metres per year (accretion positive, erosion negative) measured along shore-normal transects.

Method

For each epoch, compute the water index, threshold it to a binary land/water mask, and trace the boundary to a vector waterline. Cast shore-normal transects at fixed alongshore spacing from a baseline, and at each transect measure the shoreline intercept per date. Fit change rates DSAS-style — end-point rate between two dates, or linear-regression rate across many dates — to get m/yr of advance or retreat. Tidally-aware processing (selecting near-mean-tide scenes or applying a tidal correction with a beach-slope estimate) reduces water-level bias.

Validation

The extracted waterline is the *instantaneous* land/water edge, so tide stage and wave run-up shift it seaward or landward independent of real erosion — state the tide condition and prefer consistent tide stages across epochs. Pixel size sets the floor on detectable change (sub-pixel change is noise), and mixed wet-sand / foam / turbid plumes blur the index. Cross-check against surveyed shorelines, DSAS reference datasets, or higher-resolution imagery before reporting rates.

In plain EnglishTrace where land meets water in satellite pictures from different years, then measure how far that line has moved along lines drawn straight out from the coast. If it moves seaward the beach is growing; if landward, it is eroding. Remember the tide moves the waterline too, so compare like with like.

Make it yours → Pick your stretch of coast and the year range, choose the water index (MNDWI for turbid or built coasts, NDWI for clean water, 10 m Sentinel-2 for finer detail), set transect spacing, and the notebook returns per-transect erosion/accretion rates plus a shoreline-change map. Swap in Landsat Collection 2 to extend the record back to the 1980s.

⌨ Run the core method · no login

The thresholding a measurement into classes at the heart of this question — runnable on synthetic data, right here. The full earthaccess code template further down does it on real NASA data (needs an Earthdata login).

editable · runs in your browser

import numpy as np
# threshold a measurement into a class (here: Sentinel-1 VV backscatter, dB -> water)
vv = np.array([-8, -22, -25, -9, -20])      # one value per pixel
THRESH = -18.0                               # water is LOW backscatter
water = vv < THRESH
print("class per pixel:", np.where(water, "water", "land").tolist())
print("water pixels:", int(water.sum()), "of", water.size)
print(">>> move THRESH and rerun; real life adds look-alikes (roads, sand) to worry about")

Datasets used

HLSS30

Sharp land snapshots (Sentinel-2)

Harmonized Sentinel-2 30m

LP DAAC · 30 m (resampled from 10/20/60 m Sentinel-2 native)

HLSL30

Sharp land snapshots (Landsat)

Harmonized Landsat 30m

LP DAAC · 30 m