论文信息

Francis Quayson / Xiaoli Ding / Mahmoud Abdallah 等 7 人

Water

论文解读

AI 辅助整理
方法
Google Earth Engine处理Sentinel-2影像,随机森林分类
发现
Keta水域与农业、建成区减少而植被增加;Muni-Pomadze水域略增,年际波动更大。
机会点
可引入实地校准水质浓度与因果归因分析,弥补光学指数仅作筛查基线的不足。

论文摘要

沿海潟湖是环境敏感的湿地系统,其水文和土地覆盖对气候变率和流域活动作出响应。本研究利用在Google Earth Engine中处理的Sentinel-2影像,比较了2018—2024年加纳Keta和Muni-Pomadze拉姆萨尔潟湖流域制图得到的表层水体范围、光学水质指标和土地利用/土地覆盖。研究生成了2018年、2021年和2024年的随机森林土地覆盖分类,并采用共同覆盖区面积和历史参考特征留出诊断。Keta制图结果显示农业(−52.70 km²)、建成区(−30.83 km²)和水体(−31.53 km²)减少,同时植被(+60.09 km²)和灌丛(+54.97 km²)增加。Muni-Pomadze制图结果显示水体增加1.00 km²,建成区减少5.29 km²。两个站点三个年份的留出总体精度为82.81%至87.50%;参考标签的时间独立性仍是一项局限。另外,经一致重处理的七年水体掩膜序列显示,Muni-Pomadze的年际水体面积比例变率大于Keta。NDTI、NDCI和悬浮沉积物指数描述光学条件,并不代表经野外校准的浓度。研究结果提供了比较性筛查基线,但并未证明特定土地覆盖转换导致了观测到的光学变化。

全文

Water

Academic Editor: Cesar Andrade
Received: 24 September 2026
Revised: 29 October 2026
Accepted: 13 November 2026
Published: 16 November 2026
Copyright: © 2026 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license.
Article
Remote Sensing Assessment of Hydrological Dynamics and
Environmental Indicators in Two Contrasting Ramsar Coastal
Lagoons of Ghana
Francis Quayson 1 , 2 , *
, Xiaoli Ding 1 , 2 , 3
, Mahmoud Abdallah 1 , 2
, Ishmael Yaw Dadson 4 , Kofi Adu-Boahen 4 ,
Yaw Asamoah 4 and George Owusu Amoah 5
1 Department of Land Surveying and Geospatial Science, The Hong Kong Polytechnic University,
Hong Kong, China; xl.ding@polyu.edu.hk (X.D.); mahmoud.abdallah@connect.polyu.hk (M.A.)
2 Research Institute for Land and Space, The Hong Kong Polytechnic University, Hong Kong, China
3 Shenzhen Research Institute, The Hong Kong Polytechnic University, Shenzhen 518057, China
4 Department of Geography Education, University of Education, Winneba P .O. Box 25, Ghana;
iydadson@uew.edu.gh (I.Y.D.); kadu-boahen@uew.edu.gh (K.A.-B.); yasamoah@uew.edu.gh (Y.A.)
5 Department of Geography and Regional Planning, University of Cape Coast, Cape Coast CC-192-2032, Ghana;
george.amoah003@stu.ucc.edu.gh

  • Correspondence: francis.quayson@connect.polyu.hk
    Abstract
    Coastal lagoons are environmentally sensitive wetland systems whose hydrology and land
    cover respond to climatic variability and catchment activity. This study used Sentinel-2
    imagery processed in Google Earth Engine to compare mapped surface-water extent, optical
    water-quality indicators and land use/land cover in the Keta and Muni-Pomadze Ramsar
    lagoon catchments, Ghana, over 2018–2024. Random forest land-cover classifications were
    generated for 2018, 2021 and 2024, with common-footprint areas and historical-reference
    feature holdout diagnostics. Keta mapped decreases in agriculture (−52.70 km2 ), built-up
    (−30.83 km2 ) and water (−31.53 km2 ), alongside increases in vegetation (+60.09 km2 )
    and shrubland (+54.97 km2 ). Muni-Pomadze mapped a 1.00 km2 water increase and a
    5.29 km2 built-up decrease. Holdout overall accuracies ranged from 82.81% to 87.50%
    across the two sites and three years; temporal independence of reference labels remains a
    limitation. Separately, the consistently reprocessed seven-year water-mask series showed
    greater proportional interannual water-area variability at Muni-Pomadze than at Keta.
    The NDTI, NDCI and suspended-sediment indices describe optical conditions and do not
    represent field-calibrated concentrations. The results provide a comparative screening
    baseline but do not demonstrate that particular land-cover conversions caused observed
    optical changes.
    Keywords: Ramsar wetlands; environmental indicators; Sentinel-2; Google Earth Engine;
    hydrological variability; sediment dynamics
  1. Introduction
    Wetlands are among the most productive ecosystems on Earth and provide numerous
    ecological, hydrological, and socio-economic services, including flood attenuation, ground-
    water recharge, shoreline stabilization, nutrient retention, carbon sequestration, biodiversity
    conservation, fisheries production, and support for agriculture and local livelihoods [1,2].
    Coastal lagoons constitute a unique class of wetlands because they occur at the interface
    between terrestrial, freshwater, and marine environments, where relatively small changes
    Water 2026, 18, 2856 https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 2 of 29
in hydrological connectivity or catchment conditions can produce substantial ecological
responses [3,4]. Their ecological functioning depends on complex interactions among fresh-
water inflow, tidal exchange, sediment transport, nutrient cycling, climatic variability, and
anthropogenic activities, making them among the most environmentally sensitive coastal
ecosystems worldwide [3,4].
During recent decades, coastal lagoons have experienced increasing environmental
pressure arising from rapid urbanization, agricultural expansion, shoreline modification,
pollution, altered freshwater inflows, and climate change [2,4–6]. These pressures mod-
ify catchment hydrology, increase sediment and nutrient loading, alter water residence
time, and accelerate ecological degradation through habitat loss and eutrophication [4,7].
Assessments of coastal lagoons in Europe and Ghana indicate that cumulative anthro-
pogenic disturbance, when combined with hydrological variability, can substantially reduce
ecosystem resilience and ecological integrity [4,7–9]. Consequently, continuous monitoring
has become a priority for wetland conservation and sustainable management. A recent
scientometric and systematic review identifies coastal-wetland vulnerability, adaptation,
and advanced monitoring and modeling as major research themes, reinforcing the need for
integrated observations under climate change [10].
Traditional monitoring programs rely primarily on field measurements of water level,
turbidity, chlorophyll-a, suspended sediment concentration, dissolved oxygen, nutrients,
salinity, and other physicochemical parameters. Although these observations provide direct
environmental information, they are often expensive, labor-intensive, spatially restricted,
and difficult to maintain over long periods, particularly in developing countries where
routine monitoring networks remain limited [11–13]. Satellite remote sensing therefore
provides an attractive complementary approach because it enables consistent, synoptic, and
cost-effective observations across large spatial and temporal scales [11,12,14–16].
The Sentinel-2 mission has considerably improved the capacity for monitoring inland
and coastal waters owing to its high spatial resolution (10–20 m), five-day revisit interval,
and spectral bands covering the visible, red-edge, near-infrared, and shortwave-infrared
regions [17]. These characteristics allow the derivation of several spectral indices that de-
scribe hydrological and environmental conditions. The Normalized Difference Water Index
(NDWI) has been widely applied to delineate surface-water bodies [18], while the Modified
Normalized Difference Water Index (MNDWI) improves water extraction in heterogeneous
landscapes by reducing confusion between water and built-up surfaces [19]. Additional
indices, including the Normalized Difference Turbidity Index (NDTI), the Normalized Dif-
ference Chlorophyll Index (NDCI), and red-to-near-infrared suspended sediment proxies,
have been used to investigate spatial variability in turbidity-related, chlorophyll-related,
and sediment-related optical responses within aquatic environments [20–24].
Although these indices have become widely adopted for wetland monitoring, they
represent remotely sensed optical proxies rather than direct measurements of water qual-
ity [13,21,25]. Their interpretation depends on atmospheric conditions, water depth, bottom
reflectance, suspended material, colored dissolved organic matter, aquatic vegetation, and
sensor characteristics [11,12,25,26]. Consequently, satellite-derived indicators should be
interpreted cautiously where concurrent field observations are unavailable. Nevertheless,
numerous studies have shown that integrating multiple spectral indicators provides a ro-
bust framework for evaluating hydrological and ecological variability in wetlands where
conventional monitoring data are scarce [12,13,25–27].
Ghana contains six internationally designated Ramsar wetlands, including five coastal
sites along the Gulf of Guinea that provide critical ecological, economic, and cultural ser-
vices [9,28]. Among these, the Keta Lagoon Complex and the Muni-Pomadze Ramsar Site
represent two contrasting coastal lagoon systems with markedly different geomorphological,
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 3 of 29
hydrological, and ecological characteristics. Keta Lagoon is the largest lagoon system in
Ghana and exhibits relatively persistent open-water conditions associated with its extensive
surface area and hydrological connectivity [28,29]. In contrast, Muni-Pomadze is consider-
ably smaller, shallower, and more responsive to seasonal rainfall, evaporation, intermittent
ocean exchange, and wetland hydrology [9,30]. These contrasting characteristics provide
an ideal opportunity to examine how lagoon morphology, hydrological connectivity, and
catchment transformation influence remotely sensed environmental indicators.
Previous investigations in Ghana have examined land-use and land-cover dynamics,
shoreline evolution, wetland degradation, and selected hydrological characteristics of in-
dividual lagoon systems [9,28–31]. Recent comparative work has assessed hydroclimatic
variability and anthropogenic land transformation in Keta and Muni-Pomadze during
2000–2024 [32], while an integrated assessment combined land-cover change, hydroclimatic
variables, and satellite-derived indicators of human activity to characterize relative environ-
mental pressure at both sites [8]. Building on these studies, the present analysis focuses
on Sentinel-2-derived surface-water extent and turbidity-, chlorophyll-, and suspended-
sediment-related optical proxies during 2018–2024, together with catchment land-cover
change. This comparison examines how the optical indicators vary between lagoons with
contrasting size, hydrological connectivity, and catchment conditions.
This study therefore presents a comparative remote sensing assessment of hydro-
logical dynamics and environmental indicators in the Keta and Muni-Pomadze Ramsar
lagoons using Sentinel-2 imagery acquired between 2018 and 2024. Specifically, the study
aims to (i) quantify spatiotemporal variations in surface-water extent using NDWI and
MNDWI, (ii) evaluate turbidity-related, chlorophyll-related, and suspended sediment opti-
cal responses using NDTI, NDCI, and SSP , (iii) assess land-use and land-cover dynamics
within the surrounding catchments using random forest classification, and (iv) examine
how differences in hydrological connectivity and landscape transformation influence the
observed environmental patterns. By integrating hydrological, ecological, and landscape
information within a unified Google Earth Engine framework, the study provides a practical
methodology for long-term monitoring of tropical Ramsar wetlands and contributes to a
better understanding of environmental variability in coastal lagoon ecosystems.
Three null hypotheses specify measurable comparisons. H01 (interannual water-area
variability): the difference between the coefficients of variation (CV) of annual mapped
open-water area at Muni-Pomadze and Keta during 2018–2024 is zero (CVMP − CVK = 0);
the directional alternative is CVMP > CVK. H02 (agriculture and sediment-sensitive
response): the slope relating mapped agricultural area to annual mean suspended-
sediment optical proxy (SSP) at Keta, using temporally matched observations, is zero
(βagriculture = 0); the directional alternative is a positive slope. H03 (water-mask thresh-
old sensitivity): the difference between lagoons in the maximum absolute percentage
deviation in mapped open-water area from the joint zero-threshold baseline across the
same NDWI/MNDWI scenarios is zero (max|ΔAMP| − max|ΔAK| = 0); the directional
alternative is greater sensitivity at Muni-Pomadze. These null values make the comparisons
falsifiable. H01 and H03 are evaluated against observed descriptive statistics; the limited
LULC snapshots do not permit a robust test of the H02 slope. No formal hypothesis test or
statistical rejection is claimed.
2. Materials and Methods
2.1. Study Area
The study was conducted in the Keta Lagoon and Muni-Pomadze Lagoon Ramsar
sites located along the Gulf of Guinea in southern Ghana (Figure 1). Both wetlands are
internationally recognized under the Ramsar Convention because of their ecological impor-
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 4 of 29
tance, biodiversity, fisheries resources, and role in supporting migratory birds and local
livelihoods [8,9,28,30,32].
Figure 1. Geographical location and spatial context of the Keta Lagoon Complex and Muni-Pomadze
Ramsar sites in Ghana. The map presents the lagoon boundaries, surrounding land-cover features,
regional locations, scale bars, north arrows, and national context. Orange denotes the Volta Region,
and pink denotes the Central Region in the inset maps.
The Keta Lagoon Complex is situated within the Volta Region between approximately
5°45′–6°10′ N and 0°40′–1°05′ E. It is the largest lagoon system in Ghana and consists of
an extensive shallow lagoon connected to wetlands, marshes, salt pans, and seasonally
inundated floodplains. Freshwater inputs originate primarily from rainfall, seasonal streams,
and surrounding catchments, while occasional exchange with the Atlantic Ocean influences
salinity and water-level fluctuations. The surrounding landscape consists of agricultural
land, shrubland, settlements, wetlands, and salt-production areas. The region experiences
a tropical coastal savanna climate characterized by bimodal rainfall and relatively high
evaporation rates [28,29].
The Muni-Pomadze Ramsar Site is located near Winneba in the Central Region between
approximately 5°18′–5°24′ N and 0°37′–0°45′ W. Compared with Keta, Muni-Pomadze is
considerably smaller and shallower, comprising a lagoon connected to freshwater marshes
and seasonally flooded wetlands. The lagoon is influenced by seasonal rainfall, runoff,
evaporation, and intermittent exchange with the Atlantic Ocean. Land cover surrounding
the lagoon consists mainly of cultivated land, herbaceous wetlands, natural vegetation,
settlements, and small forest patches. Because of its relatively shallow water depth, envi-
ronmental conditions respond rapidly to seasonal hydrological fluctuations [9,30].
The contrasting geomorphological and hydrological characteristics of the two lagoons
provide an ideal opportunity to investigate how lagoon morphology, hydrological connec-
tivity, and catchment transformation influence remotely sensed environmental indicators.
2.2. Overall Methodological Framework
The methodological framework integrated satellite image processing, land-use and
land-cover classification, hydrological mapping, environmental indicator extraction, statisti-
cal analysis, and uncertainty assessment (Figure 2). Sentinel-2 imagery acquired between
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 5 of 29
2018 and 2024 was processed within Google Earth Engine (GEE) [16], while statistical
analyses and graphical outputs were generated using Python 3.12 within Google Colab
(cloud-based notebook environment; accessed in 2026).
Figure 2. Analytical framework for the comparative assessment of hydrological dynamics, remote
sensing-derived environmental indicators, and catchment land-cover changes in the Keta and Muni-
Pomadze lagoon systems. The null hypotheses specify zero-difference or zero-association bench-
marks, but the available evidence is evaluated descriptively, not through formal null-hypothesis
significance tests. Mapped annual surface-water area is a satellite-derived extent, not a direct measure
of depth or stored water volume; optical indices are uncalibrated proxies rather than measured water-
quality concentrations. Seven annual observations per lagoon cannot establish long-term trends or
causal mechanisms.
The workflow consisted of six principal stages. First, Sentinel-2 Level-2A surface
reflectance imagery was acquired and preprocessed through cloud masking and annual
compositing. Second, annual land use and land cover maps were generated for 2018, 2021,
and 2024 using the random forest classifier. Third, NDWI, MNDWI, NDTI, NDCI, and
SSP were calculated from annual Sentinel-2 composites. Fourth, binary water masks were
generated using NDWI and MNDWI thresholding to estimate annual water extent. Fifth,
annual descriptive statistics were extracted for each lagoon. Finally, temporal comparisons
were performed to evaluate hydrological dynamics and environmental variability.
Google Earth Engine was selected because it provides direct access to extensive satellite
archives and enables efficient cloud-based geospatial analysis without local computational
constraints [16]. The satellite datasets used in this study, including their spatial and temporal
resolutions, study periods, and respective analytical purposes, are summarized in Table 1.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 6 of 29
Table 1. Satellite datasets used in this study.
Dataset Spatial Resolution Temporal Resolution Study Period Purpose
Sentinel-2 MSI Level-2A 10–20 m 5 days 2018–2024
Spectral indices and
LULC classification for
2018, 2021, and 2024
Sentinel-2 Scene
Classification Layer 20 m 5 days 2018–2024 Cloud and shadow masking
Google Earth Engine
Feature Collections — Static 2018–2024 Lagoon boundary extraction
2.3. Satellite Data Acquisition and Pre-Processing
Sentinel-2 MultiSpectral Instrument (MSI) Level-2A imagery was obtained from the
harmonized Sentinel-2 Surface Reflectance archive available within Google Earth Engine.
Level-2A products provide atmospherically corrected bottom-of-atmosphere reflectance
suitable for quantitative environmental applications [33].
All available Sentinel-2 images intersecting each lagoon between 1 January 2018 and
31 December 2024 were retrieved. Images were filtered using the study-area polygons and
annual acquisition dates. Scenes containing excessive cloud contamination were excluded
before cloud masking. Clouds, cirrus, cloud shadows, saturated pixels, and defective
observations were removed using the Scene Classification Layer (SCL) supplied with the
Sentinel-2 Level-2A product [33].
Following cloud masking, annual median composites were generated to minimize
residual atmospheric contamination while preserving representative annual surface condi-
tions. The annual composite was computed as
𝐼𝑡 = median(𝐼1, 𝐼2, … , 𝐼𝑛) (1)
where It represents the annual composite, Ii denotes an individual cloud-free Sentinel-2
observation, and n is the number of valid observations acquired during year t.
Median compositing substantially reduces residual cloud contamination and sen-
sor noise while providing consistent annual imagery for temporal comparison [16,34].
However, annual composites may suppress short-duration hydrological events such as
flooding, lagoon-mouth opening, sediment pulses, and phytoplankton blooms. Conse-
quently, the composites represent typical annual conditions rather than instantaneous
environmental states.
2.4. Land-Use and Land-Cover Classification
Land-use and land-cover classification was performed for 2018, 2021, and 2024 using
annual Sentinel-2 surface-reflectance composites. These years correspond directly with
the environmental indicator analysis period and permit assessment of recent catchment
transformation at the beginning, middle, and end of the study period.
A random forest classifier with 300 trees, a minimum leaf population of 2, and bag frac-
tion of 0.7 was applied separately to each year and lagoon in Google Earth Engine [35,36].
Predictor variables included Sentinel-2 blue, green, red, three red-edge, near-infrared and
shortwave-infrared bands, together with NDVI, MNDWI, NDBI, NDRE and SA VI. Har-
monized Sentinel-2 Surface Reflectance (COPERNICUS/S2_SR_HARMONIZED) annual
median composites were filtered by scene-level cloudiness (≤80%) and masked using SCL
classes 0, 1, 3, 8, 9, 10 and 11. Samples and area reductions were evaluated at 20 m.
Keta was classified into water, built-up, vegetation, agriculture and shrubland, and
Muni-Pomadze was classified into water, built-up, herbaceous, vegetation and culti-
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 7 of 29
vated/planted. Reference features from previously developed historical Landsat classifica-
tions were reused as provisional class labels rather than newly digitized target-year ground
truth. Keta used common reference collections across 2018, 2021 and 2024; Muni-Pomadze
used candidate reference collections from 2015, 2020 and 2024 for the respective 2018, 2021
and 2024 images. This temporal mismatch may bias classification and change estimates.
Reference features were split (70% training; 30% holdout) before sampling Sentinel-2
pixels. Holdout confusion matrices were used to compute overall accuracy and Cohen’s
Kappa; they are internal, feature-disjoint diagnostics, not temporally independent field
validation. Keta overall accuracy was 85.94%, 87.50% and 82.81% in 2018, 2021 and 2024, with
Kappa 0.815, 0.837 and 0.774. Muni-Pomadze accuracy was 87.30%, 87.72% and 86.21%, with
Kappa 0.819, 0.841 and 0.815. After validation diagnostics, classifiers were retrained using
all candidate reference samples to generate final provisional maps. Class-specific confusion
and field/reference verification remain necessary. Spatially grouped holdout validation in
the Volta River Basin flood-susceptibility study [37] provides a methodological comparator;
the present feature-disjoint split does not demonstrate equivalent spatial transferability.
For square pixels of uniform area, the area occupied by class c was calculated as
𝐴𝑐 = 𝑁𝑐𝑟2 (2)
where Ac is the mapped area of class c, Nc is the number of pixels assigned to that class,
and r is the spatial resolution of the classified image.
Percentage coverage was calculated as
𝑃𝑐 = 𝐴𝑐𝐴𝑇 × 100 (3)
where AT is the total mapped catchment area.
Net land-cover change between 2018 and 2024 was calculated as
Δ𝐴𝑐 = 𝐴𝑐,2024 − 𝐴𝑐,2018 (4)
and relative change was calculated as
𝑅𝑐 = 𝐴𝑐,2024 − 𝐴𝑐,2018𝐴𝑐,2018 × 100. (5)
The 2021 classification was treated as an intermediate observation for determining
whether the major class changes appeared progressive across the study period. The LULC
results should be interpreted with the validation limitations described above.
Classification reliability: The new Sentinel-2 maps have historical-label feature holdout
accuracy diagnostics but lack temporally independent reference observations. Their class-
area and change statistics are unadjusted map estimates, not unbiased estimates of land-
cover conversion. Transition interpretations among agriculture, shrubland, herbaceous
cover and vegetation remain provisional. Area statistics and transitions were computed on
pixels valid in all three year-specific maps, preventing unequal data coverage from creating
spurious changes. Dynamic World is a separate classification system and must not be pooled
with these five-class maps.
2.5. Remote Sensing-Derived Environmental Indicators
Hydrological and environmental variability was characterized using five spectral
indices derived from annual Sentinel-2 composites. These indices describe relative optical
conditions and were not interpreted as direct measurements of physicochemical water
quality [20–22,25].
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 8 of 29
The Normalized Difference Water Index (NDWI) was calculated following McFeeters [18]:
NDWI =
𝜌Green−𝜌NIR
𝜌Green+𝜌NIR (6)
The Modified Normalized Difference Water Index (MNDWI) followed Xu [19]:
MNDWI =
𝜌Green−𝜌SWIR
𝜌Green+𝜌SWIR (7)
The Normalized Difference Turbidity Index (NDTI) was calculated as
NDTI =
𝜌Red−𝜌Green
𝜌Red+𝜌Green (8)
where less negative values indicate relatively stronger turbidity-related optical responses [20].
The Normalized Difference Chlorophyll Index (NDCI) was calculated following Mishra
and Mishra [21]:
NDCI =
𝜌RedEdge−𝜌Red
𝜌RedEdge+𝜌Red (9)
Higher NDCI values indicate relatively stronger chlorophyll-related spectral responses
but do not represent direct chlorophyll-a concentrations [21,27].
The suspended sediment proxy (SSP) was calculated as
SSP =
𝜌Red
𝜌NIR (10)
SSP was used here as a relative sediment-related optical indicator; quantitative
suspended-sediment retrieval requires calibration and validation [22,24].
These indices were interpreted collectively because each respond to multiple envi-
ronmental processes, including suspended sediments, phytoplankton, colored dissolved
organic matter, aquatic vegetation, water depth, and bottom reflectance [11,12,25].
2.6. Surface Water Extraction and Water Area Estimation
Annual surface-water extent was extracted from the NDWI and MNDWI images to
quantify the hydrological variability within each lagoon. Binary water masks were generated
by applying a threshold of zero to both indices. Pixels with index values greater than zero
were classified as water, whereas pixels with values less than or equal to zero were classified
as non-water. This threshold has been widely applied for surface-water mapping because
positive NDWI and MNDWI values generally correspond to open-water surfaces, while
negative values represent vegetation, exposed soil, or built-up land [18,19].
The extracted water masks were visually compared with the original Sentinel-2 com-
posites and high-resolution reference imagery to ensure that the threshold adequately
represented lagoon boundaries and minimized misclassification along shoreline margins.
Particular attention was given to shallow water, salt-production areas, emergent vegeta-
tion, and exposed sediments where mixed pixels frequently occur. Comparison between
NDWI and MNDWI also provided an internal consistency check because MNDWI generally
performs better in areas where built-up surfaces occur adjacent to water [19].
Annual water area was calculated by summing all pixels classified as water within
each lagoon boundary according to
𝐴𝑤 = ∑𝑛𝑖=1 𝑎𝑖 (11)
where Aw is the annual water area (km2 ), ai represents the area of each water pixel, and n
denotes the total number of pixels classified as water. The resulting values formed the basis
for evaluating interannual hydrological variability between 2018 and 2024.
Because threshold-based water extraction is sensitive to shoreline complexity, shal-
low water, aquatic vegetation, and tidal conditions, the calculated water extents should
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 9 of 29
be interpreted as remotely sensed estimates rather than absolute hydrological measure-
ments [18,19,25].
Threshold sensitivity was assessed for each lagoon and each year (2018–2024) using ten
alternative water-mask definitions: the original joint NDWI > 0 and MNDWI > 0 baseline,
four joint threshold pairs (−0.10, −0.05, +0.05 and +0.10), four single-index threshold
perturbations (±0.05, holding the other index at zero) and an OR instead of an AND
baseline rule. The same annual Sentinel-2 SR Harmonized median of per-scene indices,
matched cloud-probability mask (<40%), and SCL exclusions were retained across scenarios.
Relative area difference (%) was computed against the baseline, and map agreement was
evaluated using intersection-over-union (Jaccard similarity). These are sensitivity, not
accuracy, tests.
2.7. Extraction of Environmental Indicator Statistics
Annual spectral statistics were calculated separately for the Keta and Muni-Pomadze
lagoons. NDWI and MNDWI statistics were derived for the relevant lagoon analysis extents,
while NDTI, NDCI, and SSP were extracted only from pixels classified as water. Restrict-
ing the aquatic indicators to the water masks reduced contamination from neighboring
vegetation, built-up surfaces, bare ground, and exposed wetland substrates.
For each lagoon, indicator, and year, the mean, standard deviation, minimum, maxi-
mum, coefficient of variation, and number of valid pixels were calculated. The arithmetic
mean was estimated as
𝑥 = 1𝑛 ∑𝑛𝑖=1 𝑥𝑖 (12)
where xi is an individual valid pixel value and n is the number of valid pixels.
The standard deviation was calculated as
SD = √ ∑𝑛𝑖=1(𝑥𝑖−𝑥)2𝑛−1 (13)
and the coefficient of variation was calculated as
CV = SD∣𝑥∣ × 100. (14)
The absolute mean was used in the denominator because some indices contained neg-
ative values. Minimum and maximum values described the observed spectral range, while
the valid-pixel count documented the number of pixels contributing to each annual estimate.
These statistics were used for descriptive comparison between lagoons and among years.
Valid-pixel support was independently checked for each annual median composite.
Water was defined by NDWI > 0 AND MNDWI > 0 using consistent cloud and data masks;
valid pixels were counted separately for NDTI, NDCI and SSP on the 10 m analysis grid. All
three optical indices had equal valid-water counts within each site-year. Keta ranged from
2,793,142 (2018) to 3,573,900 (2023) contributing composite-grid pixels; Muni-Pomadze
ranged from 9390 (2020) to 25,974 (2022). Counts represent composite-grid samples, not
independent scenes or independent 10 m physical measurements of 20 m Sentinel-2 bands.
Full annual sample sizes are documented in Supplementary Table S1.
2.8. Statistical Analysis
Statistical processing and visualization were conducted in Python within Google
Colab using Pandas v3.0.6, NumPy v2.5.3, SciPy 1.18.1, Matplotlib v3.11.2, and Scikit-learn
v1.9.1 [38–42]. Annual values of water extent, NDWI, MNDWI, NDTI, NDCI, and SSP were
organized separately for the two lagoons for the 2018–2024 period.
The null hypotheses were compared with the available descriptive evidence. For H01,
coefficients of variation computed from the seven annual zero-threshold water extents
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 10 of 29
were compared between lagoons. For H02, the proposed test quantity is the regression
slope between the annual mapped Keta agricultural area and matched-year SSP , but the
LULC dataset has only three temporal snapshots (2018, 2021 and 2024); no reliable slope
inference was attempted. For H03, the maximum absolute percentage change from the
joint zero-threshold baseline was compared between lagoons over the common water-mask
sensitivity scenarios. These observed contrasts do not constitute inferential tests or justify
statistical rejection of the null hypotheses.
The analysis focused on descriptive comparison because the annual record contained
only seven observations per lagoon and the environmental indicators had not been validated
against field measurements. For each lagoon and indicator, the mean, standard deviation,
minimum, maximum, coefficient of variation, and annual range were calculated. Absolute
and proportional water-area variability were also calculated to compare the hydrological
behavior of the two systems.
Temporal variables were transformed using min–max normalization to permit direct
graphical comparison among variables measured on different scales:
𝑥′ = 𝑥−𝑥min𝑥max−𝑥min (15)
where x
′
is the normalized value, x is the original annual observation, and xmin and xmax
are the minimum and maximum values of the corresponding series. A normalized value
of zero represents the lowest observation within that variable and does not indicate the
complete absence of water or the environmental indicator.
The seven annual observations per lagoon and uncalibrated optical proxies did not
support reliable inferential interpretation. No formal significance test of H01–H03 is re-
ported; the null hypotheses are therefore not rejected or accepted on statistical grounds.
Observed increases, decreases, peaks, fluctuations, and threshold-scenario contrasts are
interpreted descriptively. In particular, scenario-based sensitivity is not a sampling-based
test of classification accuracy.
2.9. Reliability and Uncertainty Assessment
The reliability assessment considered uncertainties associated with atmospheric cor-
rection, cloud masking, temporal compositing, spectral-index formulation, water extraction,
mixed pixels, lagoon boundaries, and LULC classification. Sentinel-2 Level-2A surface
reflectance reduced atmospheric effects through operational correction, while the Scene
Classification Layer was used to remove clouds, cirrus, cloud shadows, saturated pixels,
and defective observations [33]. Residual haze, thin clouds, illumination differences, and
atmospheric-correction errors may nevertheless remain.
Annual median compositing reduced residual cloud contamination and anomalous
observations but may have smoothed short-lived hydrological and ecological events. Flood
pulses, lagoon-mouth opening, rapid recession, sediment mobilization, and phytoplankton
blooms may therefore be underrepresented in the annual composites [34]. The derived
images represent typical annual surface conditions rather than seasonal maxima or minima.
Mixed pixels were especially important within Muni-Pomadze because its narrow and
shallow margins frequently contained water, emergent vegetation, exposed sediment, and
terrestrial cover within a single 10 m pixel. Such mixing may reduce NDWI and MNDWI
values and alter NDTI, NDCI, and SSP responses. Restricting the aquatic indicators to
classified water pixels and comparing NDWI with MNDWI reduced this uncertainty but
did not eliminate it [18,19,25,26].
The completed threshold-sensitivity experiment found that alternative masks changed
water-area estimates relative to the original joint zero-threshold baseline by −10.86% to
+9.34% across the Keta site-years and −69.92% to +60.30% across Muni-Pomadze site-years.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 11 of 29
The minimum within-year Jaccard similarities across alternatives ranged from 0.891 to 0.937
at Keta and from 0.301 to 0.777 at Muni-Pomadze. Thus, the smaller lagoon is markedly
more sensitive in relative terms, and individual water-area variations must be interpreted
with this uncertainty. The comparison includes the alternative OR rule as well as numerical
threshold perturbations.
Source-data reconciliation: The annual water extents were recalculated from the annual
baseline in the verified threshold-sensitivity export, with matching valid-water pixel counts
and annual index means from the verified composite-grid export. This common processing
version differs from certain earlier water-extent exports, especially at Keta in 2018; earlier
estimates are superseded rather than averaged together. The latest approach requires an
inner join of Sentinel-2 SR with matching cloud-probability data, a 40% cloud-probability
cutoff, SCL-based exclusions and annual medians of per-image index values.
The absence of concurrent field observations was the principal limitation of the aquatic
analysis. NDTI, NDCI, and SSP could not be calibrated against turbidity, chlorophyll-a,
or suspended sediment concentration. They were therefore used only as relative spectral
indicators. They should not be interpreted as direct water-quality measurements, evidence
of eutrophication, or formal classifications of ecological status.
Uncertainty in the LULC analysis arose from spectral similarity among vegetation,
shrubland, herbaceous wetlands, and cultivated land and from the absence of a fully
independent validation dataset. Large and persistent class changes were considered more
reliable than small or isolated transitions.
2.10. Methodological Limitations
The study relied primarily on remotely sensed observations and did not include simul-
taneous field measurements of turbidity, chlorophyll-a, suspended sediment concentration,
dissolved oxygen, nutrients, salinity, discharge, or water depth. Quantitative calibration
and validation of the optical indicators were therefore beyond the scope of the analysis.
The annual composites represent median conditions and do not capture the complete
seasonal cycle or individual hydrological events. The LULC assessment was also limited to
three temporal snapshots, namely, 2018, 2021, and 2024, and therefore does not identify the
exact timing of every land-cover transition.
The random forest classifications were checked visually but were not supported by
a fully independent validation dataset for every year. The maps should consequently be
interpreted as comparative representations of broad catchment change. The methodology
is transferable as a screening framework, but applications in other wetlands should include
independent LULC validation, local threshold testing, seasonal image analysis, and field
calibration of the aquatic indicators.
2.11. AI Tool Disclosure
During the preparation of this manuscript/study, the authors used Grammarly (version
v.1.2.281.1928) for the purposes of language editing and grammatical correction. The
authors have reviewed and edited the output and take full responsibility for the content of
this publication.
3. Results
3.1. Land-Use and Land-Cover Dynamics
The 2018, 2021 and 2024 Sentinel-2 classifications indicate substantial differences be-
tween lagoon catchments (Table 2). Contrary to the earlier provisional results, the newly
generated Keta series maps net losses in agriculture, built-up and water and net gains
in vegetation and shrubland. Muni-Pomadze shows a smaller mapped water increase, a
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 12 of 29
decrease in built-up and increases in herbaceous cover, vegetation and cultivated/planted
land. Because historic reference labels were transferred to the target years, these changes
are map-based estimates requiring independent validation. The spatial distributions of the
mapped land-cover classes for both lagoon systems in 2018, 2021, and 2024 are presented in
Figure 3.
Table 2. Land-use and land-cover changes in the Keta and Muni-Pomadze lagoon catchments between
2018 and 2024.
Lagoon Land-Cover Class 2018 Area (km2 ) 2024 Area (km2 ) Net Change (km2 ) Relative Change (%)
Keta Water 326.20 294.67 −31.53 −9.67
Keta Built-up 224.21 193.38 −30.83 −13.75
Keta Vegetation 201.82 261.91 +60.09 +29.78
Keta Agriculture 311.90 259.20 −52.70 −16.90
Keta Shrubland 294.58 349.55 +54.97 +18.66
Muni-Pomadze Water 3.02 4.01 +1.00 +32.98
Muni-Pomadze Built-up 36.75 31.46 −5.29 −14.39
Muni-Pomadze Herbaceous 29.72 31.44 +1.72 +5.78
Muni-Pomadze Vegetation 7.04 8.23 +1.19 +16.97
Muni-Pomadze Cultivated/Planted 15.55 16.93 +1.38 +8.88
Figure 3. Sentinel-2 random forest land-cover classifications of the Keta Lagoon Complex
(a–c) and Muni-Pomadze Ramsar Site (d–f) for 2018, 2021 and 2024. Class definitions differ be-
tween the sites; historical reference labels were used provisionally, and mapped changes require
independent validation.
At Keta, the common-footprint mapped water class declined from 326.20 km2 in 2018 to
294.67 km2 in 2024 (−31.53 km2 ; −9.67%). Built-up area declined from 224.21 to 193.38 km2
(−30.83 km2 ; −13.75%) and agriculture from 311.90 to 259.20 km2 (−52.70 km2 ; −16.90%).
In contrast, vegetation increased from 201.82 to 261.91 km2 (+60.09 km2 ; +29.78%) and
shrubland from 294.58 to 349.55 km2 (+54.97 km2 ; +18.66%). The common valid classified
area was 1358.71 km2 in all three years.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 13 of 29
The mapped transitions include large exchanges among shrubland, vegetation, agri-
culture and built-up categories; for example, 77.42 km2 was labeled agriculture in 2018
and vegetation in 2024, while 79.75 km2 was labeled built-up and subsequently shrubland.
Such exchanges may partly reflect temporal inconsistencies in training labels and spectral
confusion, so they should not be interpreted as independently verified land conversions.
The intermediate 2021 classifications were not monotonic: Keta built-up declined
to 123.45 km2 in 2021 before increasing to 193.38 km2 in 2024, while shrubland rose to
405.60 km2 and then declined to 349.55 km2 . This non-monotonic behavior requires caution
when attributing changes to directional development pressure. Figure 4 presents the revised
class-area profiles derived from the Sentinel-2 common-footprint analysis.
Figure 4. Radar chart of mapped land-cover class areas (km2 ) in the Keta Lagoon Complex for 2018,
2021 and 2024, derived from the Sentinel-2 classifications over the common valid-data footprint.
At Muni-Pomadze, common-footprint mapped water increased from 3.02 km2 in 2018
to 4.01 km2 in 2024 (+1.00 km2 ; +32.98%). Built-up land decreased from 36.75 to 31.46 km2
(−5.29 km2 ; −14.39%). Herbaceous cover increased from 29.72 to 31.44 km2 (+1.72 km2 ;
+5.78%), vegetation from 7.04 to 8.23 km2 (+1.19 km2 ; +16.97%), and cultivated/planted
land from 15.55 to 16.93 km2 (+1.38 km2 ; +8.88%). All annual comparisons used the same
92.08 km2 valid-classification footprint.
Muni-Pomadze exhibited pronounced intermediate-year shifts: the cultivated/planted
class occupied 29.95 km2 in 2021 but 16.93 km2 in 2024, while built-up area shifted from
36.75 to 23.17 and then 31.46 km2 . These changes should be interpreted as provisional class
allocations rather than confirmed annual land conversion.
Overall, the 2018–2024 mapped changes differed by site. Keta displayed declines in the
water, built-up and agriculture classes and increases in vegetation and shrubland. Muni-
Pomadze displayed a small, mapped water increase, changes among vegetated classes and
a built-up decline. These are mapped class allocations from provisional historically derived
reference labels; the large exchanges between classes and non-monotonic trajectories make
unqualified interpretations of land conversion inappropriate.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 14 of 29
3.2. Spatial Patterns of Surface-Water Presence
The NDWI maps showed a marked difference in the spatial continuity and persistence
of surface water between the two lagoon systems (Figure 5). Keta consistently displayed an
extensive and connected open-water body, with annual mean NDWI values ranging from
0.233 to 0.356 during 2018–2024. Higher values were concentrated within the central lagoon
basin, whereas lower values occurred along shallow margins, vegetated edges, and areas
containing exposed sediment.
Figure 5. Spatial distribution of the Normalized Difference Water Index in the Keta and Muni-Pomadze
lagoons during 2018, 2021, and 2024. Higher positive values represent stronger open-water signals,
whereas lower values indicate shallow water, vegetation interaction, exposed sediment, or non-water
surfaces. Consistent color limits are used to enable interannual comparison.
Muni-Pomadze recorded lower NDWI values, ranging from 0.113 to 0.152, and exhib-
ited a more fragmented water distribution. Positive water signals were concentrated mainly
in the central and southern portions of the lagoon, while shoreline and wetland-transition
zones displayed weaker values. The reduced NDWI magnitude reflected the smaller open-
water area and the greater spectral mixing of water with emergent vegetation, herbaceous
cover, and exposed wetland substrates.
MNDWI produced patterns broadly consistent with those obtained from NDWI but
provided clearer discrimination between water, built-up surfaces, and exposed sediment.
This improvement was most evident around developed sections of the Keta catchment and
along shallow lagoon margins. Agreement between the two indices increased confidence
in the identification of permanent open water in Keta and the more discontinuous water
distribution in Muni-Pomadze.
3.3. Comparative Ranges of Spectral Indicators
Updated annual-composite NDTI, NDCI and suspended-sediment-proxy (SSP) values
differed between the two lagoons (Figure 6; Table 3). The revised comparison uses 2018–2024
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 15 of 29
water-masked annual means from the consistently processed Sentinel-2 export. NDWI
ranges in Table 3 originate from the earlier processing summary and are included only as
contextual spectral observations; they are not directly comparable with the final annual
water-area and water-only optical-index series.
Figure 6. Ranges and temporal means of NDTI, NDCI and suspended-sediment proxy (SSP) from an-
nual water-masked Sentinel-2 composites, 2018–2024. Horizontal segments represent the range across
seven annual means, and circles represent their mean. The three indices are presented on separate
scales and are not direct measures of turbidity, chlorophyll-a or suspended-sediment concentration.
Table 3. Observed ranges of surface-water extent and remotely sensed environmental indicators
during 2018–2024.
Indicator Keta Lagoon Muni-Pomadze Lagoon Comparative Result
Water area (km2 ) 276.01–353.15 0.93–2.57 Greater proportional variability in Muni-Pomadze
Relative water-area range (%) 27.95 176.57 Muni-Pomadze was substantially more variable
NDWI 0.233–0.356 0.113–0.152 Higher and more spatially continuous in Keta
NDTI −0.098 to -0.039 −0.054 to −0.032 Annual water-masked optical-index means
NDCI 0.007 to 0.025 0.032 to 0.058 Annual water-masked optical-index means
SSP 1.437–1.789 1.098–1.252 Keta had the higher SSP proxy range
Keta annual mean NDTI ranged from −0.0982 to −0.0392, compared with −0.0543 to
−0.0316 at Muni-Pomadze. Annual NDCI ranged from 0.0074 to 0.0250 at Keta and 0.0320
to 0.0583 at Muni-Pomadze. The SSP ranged from 1.4369 to 1.7887 at Keta and from 1.0981
to 1.2523 at Muni-Pomadze. These are relative spectral proxies and not field-calibrated
water-quality concentrations. The values summarized in Table 3 are plotted as annual-
mean ranges in Figure 6; both use the final water-masked optical-index series, whereas the
historical NDWI values in Table 3 follow an earlier workflow.
3.4. Turbidity-Related Optical Responses
The spatial distribution of NDTI revealed heterogeneous turbidity-related optical
conditions within both lagoons (Figure 7). Keta recorded values between approximately
−0.098 and −0.039. More negative values occurred across much of the central open-water
area, while less negative values were concentrated along shallow margins, inflow zones,
and areas adjacent to agricultural and built-up land.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 16 of 29
Figure 7. Spatial distribution of the Normalized Difference Turbidity Index in the Keta and Muni-
Pomadze lagoons during 2018, 2021, and 2024. Less negative values indicate a stronger turbidity-
related optical response. Elevated values are concentrated mainly along shallow margins and
catchment-inflow zones.
Muni-Pomadze had annual mean NDTI values between −0.0543 and −0.0316. Less
negative values occurred around lagoon margins, wetland inflows, and shallow-water zones,
with a comparatively stronger signal in 2024. Because less negative NDTI values indicate
stronger red reflectance relative to green reflectance, Muni-Pomadze generally displayed a
stronger turbidity-related optical signal than Keta.
The spatial pattern was not uniform within either lagoon. In Keta, the extensive
central water body displayed relatively weak turbidity-related reflectance, while marginal
zones contained localized areas of higher response. In Muni-Pomadze, localized sediment
inputs and shallow-water effects influenced a larger proportion of the lagoon because of its
smaller size. These results represent relative optical differences and were not converted to
turbidity units.
3.5. Temporal Variability in Lagoon Water Extent
Annual open-water area was estimated from the NDWI > 0 AND MNDWI > 0
baseline mask under the final cloud-probability-matched processing workflow (Figure 8;
Table 4). Keta water extent was 276.01 km2 in 2018, 337.99 km2 in 2019, 318.96 km2 in 2020,
316.76 km2 in 2021, 343.51 km2 in 2022, 353.15 km2 in 2023 and 334.27 km2 in 2024. The
range was 77.14 km2 (27.95% of the minimum), and the sample coefficient of variation
was 7.82%. These values replace the superseded 2018 estimate produced by an earlier
processing version.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 17 of 29
Figure 8. Annual mapped surface-water area at Keta Lagoon and Muni-Pomadze, 2018–2024, derived
from the final cloud-probability-matched Sentinel-2 analysis using NDWI > 0 and MNDWI > 0. Panels
use separate vertical scales because the lagoons differ greatly in size. Water extent is a satellite-derived
estimate rather than an independently measured hydrological quantity.
Table 4. Annual surface-water extent of the Keta and Muni-Pomadze lagoons during 2018–2024.
Year Keta Water Area (km2 ) Muni-Pomadze Water Area (km2 )
2018 276.01 2.17
2019 337.99 1.47
2020 318.96 0.93
2021 316.76 1.71
2022 343.51 2.57
2023 353.15 1.88
2024 334.27 1.01
The Muni-Pomadze water extent was 2.17 km2 in 2018, 1.47 km2 in 2019, 0.93 km2
in 2020, 1.71 km2 in 2021, 2.57 km2 in 2022, 1.88 km2 in 2023 and 1.01 km2 in 2024. Its
range was 1.64 km2 (176.57% of the minimum), with a sample coefficient of variation of
35.46%. The markedly greater proportional variability supports the descriptive comparison
in H01, although both series are affected by water-mask threshold uncertainty, particularly
at Muni-Pomadze.
3.6. Chlorophyll-Related Optical Responses
NDCI maps showed consistently stronger chlorophyll-related spectral response in
Muni-Pomadze than in Keta (Figure 9). Annual NDCI values in Muni-Pomadze ranged
from 0.031 to 0.067, with higher values concentrated in shallow and vegetation-adjacent
sections of the lagoon. The strongest relative response occurred around 2022, followed by
lower values during the subsequent years.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 18 of 29
Figure 9. Spatial distribution of the Normalized Difference Chlorophyll Index in the Keta and Muni-
Pomadze lagoons during 2018, 2021, and 2024. Higher values represent stronger chlorophyll-related
red-edge reflectance but do not constitute direct measurements of chlorophyll-a concentration or
confirmed eutrophication.
Keta recorded substantially lower NDCI values, ranging from 0.007 to 0.024. Higher
values were locally concentrated along shallow western and northern margins, while the
central lagoon displayed a weaker and more diffuse response. The contrast indicates
that chlorophyll-bearing material produced a stronger red-edge signal in Muni-Pomadze,
although the values do not represent measured chlorophyll-a concentration.
The annual NDCI pattern also differed from the SSP sequence. In Muni-Pomadze,
NDCI reached its strongest relative level around 2022, whereas SSP reached its highest
relative level around 2023. In Keta, NDCI displayed comparatively limited variation while
SSP increased toward 2024. The temporal offset indicates that the strongest chlorophyll-
related and sediment-related optical responses did not occur simultaneously.
3.7. Surface-Water Masks
The NDWI- and MNDWI-derived water masks confirmed the contrasting spatial
structures of the two lagoons (Figure 10). Keta retained a large and spatially continuous
water body during 2018, 2021, and 2024, with most changes restricted to shallow peripheral
zones. The central lagoon remained consistently classified as water.
Muni-Pomadze displayed a much smaller and more fragmented water distribution.
Its mapped water surface contracted and expanded among the selected years, particularly
along shallow northern and marginal areas. Water remained concentrated in the central and
southern portions, while the surrounding wetland contained a shifting mixture of water,
vegetation, and exposed substrate.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 19 of 29
Figure 10. NDWI- and MNDWI-derived water masks for the Keta and Muni-Pomadze lagoons during
2018, 2021, and 2024. Blue pixels represent classified water; white denotes non-water or masked
background. Keta retained a large continuous water surface, whereas Muni-Pomadze showed greater
fragmentation and interannual variation.
The MNDWI masks reduced confusion between water and adjacent built-up or ex-
posed surfaces, especially within the Keta catchment. Differences between NDWI and
MNDWI classifications were greatest along vegetated and shallow margins, confirming that
water-area estimates were most sensitive in transition zones rather than within permanent
open water.
The min–max normalized annual series compare water extent, NDTI, NDCI and SSP
using the final 2018–2024 data. Their temporal correspondence is descriptive and does not
demonstrate causal coupling. The former NDWI series was excluded because it had not
been reproduced using the final processing workflow.
3.8. Suspended Sediment Dynamics
The suspended sediment proxy displayed contrasting temporal patterns between the
two lagoons. In Keta, SSP increased toward the latter part of the study period and reached
its highest relative level in 2024. Elevated values were concentrated near inflow zones,
shallow margins, agricultural land, and developed shoreline sections.
Muni-Pomadze displayed stronger short-term variability, with SSP reaching its highest
relative level around 2023 (Figure 11). The SSP pattern was spatially similar to the NDTI
distribution, particularly around shallow margins and inflow zones. The correspondence
between the two indicators supports the identification of areas with stronger sediment-
related optical responses.
The SSP results were interpreted comparatively because red and near-infrared re-
flectance may also be influenced by water depth, bottom material, colored dissolved organic
matter, emergent vegetation, and mixed shoreline pixels. No conversion to suspended
sediment concentration was undertaken.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 20 of 29
Figure 11. Min–max normalized annual variation in water extent, NDTI, NDCI and SSP for the Keta
Lagoon Complex and Muni-Pomadze Ramsar Site, 2018–2024. Each series was normalized indepen-
dently within the corresponding site and variable; zero and one indicate the observed minimum and
maximum, not the absence or saturation of water or optical signals. Values were taken from the final
internally checked annual statistical exports.
3.9. Coupling Between Water Extent and Spectral Indicators
Figure 11 compares min–max normalized water extent and the three verified annual
water-only optical indices (NDTI, NDCI and SSP). Muni-Pomadze shows pronounced
changes in mapped water area and optical responses, but their timing does not establish a
common underlying cause.
At Keta, the mapped water area varies between years while SSP reaches its highest
annual-composite mean in 2024. The other spectral indicators display different trajecto-
ries. The plotted co-variation is exploratory because only seven annual observations were
available for each lagoon.
Hypothesis interpretation: annual mapped water-area variability was 35.46% (CV) at
Muni-Pomadze and 7.82% at Keta, giving an observed CV difference of 27.64 percentage
points in the direction of the H01 alternative. Under the common threshold scenarios, the
maximum absolute area deviation was 69.92% at Muni-Pomadze versus 10.86% at Keta,
a descriptive difference of 59.06 percentage points in the direction of the H03 alternative.
These are observed comparisons, not formal null-hypothesis tests. The Keta LULC record
does not show the proposed expansion of mapped agriculture, and the agriculture–SSP
regression slope for H02 was not estimated reliably from the three snapshots. No mechanism
linking land-cover change to SSP is established.
Anomalous years and causal uncertainty: Keta SSP reached 1.766 in 2023 and 1.789 in
2024, while the annual CHIRPS rainfall totals were 1119.42 and 960.12 mm, respectively;
the annual maxima were 57.79 and 50.80 mm/day. At Muni-Pomadze, the 2023 SSP peak
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 21 of 29
was 1.252, alongside 1268.41 mm annual rainfall and a 69.66 mm daily maximum. These
coincidences warrant event-based inquiry; they do not verify runoff pathways, dredging,
lagoon-mouth opening, or other anthropogenic disturbances. No causal attribution to these
processes is claimed.
Rainfall–SSP exploratory assessment: The daily CHIRPS record was aggregated to
calendar-year totals and annual maximum daily precipitation for each catchment and paired
with the reprocessed annual SSP mean. For Keta (n = 7), annual rainfall correlated weakly
with SSP (r = 0.053, two-sided p = 0.910), whereas annual maximum daily rainfall showed
a stronger exploratory association (r = 0.868, p = 0.011). For Muni-Pomadze, annual
rainfall correlated moderately with SSP (r = 0.580, p = 0.172), and annual maximum daily
rainfall yielded r = 0.480 (p = 0.275). p-values are unadjusted and not proof of causal
sediment loading. All four tests are exploratory, their p-values have not been adjusted
for multiple comparisons, and neither catchment runoff nor river discharge or in situ
suspended-sediment concentration was available for validation. The Keta association must
not be interpreted as a demonstrated rainfall-to-runoff-to-sediment causal pathway. The
four coefficients and unadjusted tests are tabulated in Supplementary Table S3.
The normalized values in Figure 11 describe the relative position of each annual
observation between the minimum and maximum recorded values. A normalized value
of zero therefore represents the lowest annual observation and does not indicate complete
absence of water or the corresponding spectral indicator.
The annual comparisons indicate differences in the timing of inundation and optical
responses across the sites, without identifying an underlying hydrological or sediment-
transport mechanism.
The temporal comparisons were descriptive because each lagoon contributed only seven
annual observations and the spectral indicators were not calibrated using field data. The
reported ranges, proportional changes, and normalized patterns therefore describe observed
differences during 2018–2024 and should not be interpreted as statistically confirmed long-
term trends. Statistical significance was not inferred from the apparent direction of the
plotted series.
4. Discussion
4.1. Catchment Transformation and Environmental Pressure
The new Sentinel-2 common-footprint classifications suggest contrasting but uncertain
land-cover patterns. Keta maps net declines in agriculture, built-up and water alongside
increases in vegetation and shrubland during 2018–2024; Muni-Pomadze maps increased
water and vegetated classes, alongside a built-up decline. These mapped changes differ from
the earlier manuscript results and cannot be treated as confirmed catchment transformation
without independent temporal reference checks.
The new results do not support the earlier manuscript claim that agricultural and
built-up expansion at Keta explains the increasing suspended-sediment proxy. Although
agriculture and built-up activity can affect runoff and sediment mobilization [1,3,4], the
present mapped classes decline over 2018–2024. The SSP pattern must therefore be consid-
ered independently until rainfall/runoff event analysis, spatial overlay and field evidence
establish a possible mechanism. Large bidirectional class transitions also indicate possible
classification uncertainty.
At Muni-Pomadze, the 2018–2024 maps suggest increases in herbaceous, vegetation
and cultivated/planted classes, accompanied by a decline in built-up area. The sizable
intermediate-year class exchanges could reflect phenology, water conditions or reference-
label limitations. They do not establish the ecological restoration or degradation of wetland
habitat without additional validation. Future studies should prioritize contemporaneous,
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 22 of 29
independently collected ground-truth samples for each Sentinel-2 classification year, with
class-balanced and spatially separated validation, to reduce reference-label variance and
test whether apparent bidirectional exchanges represent real transitions.
For both Ramsar sites, management implications should emphasize the continued
need to track catchment and lagoon conditions, rather than attributing site-specific optical
indicators to a particular land-cover conversion. Strong causal statements are premature
because the classification and water-quality proxies lack full independent field validation.
4.2. Lagoon Morphology and Hydrological Variability
The contrast in water-area variability reflects fundamental differences in lagoon mor-
phology and hydrological connectivity. Keta retained a large and comparatively continuous
open-water body, whereas Muni-Pomadze experienced pronounced annual expansion and
contraction. The greater storage capacity and surface area of Keta allow fluctuations in rain-
fall, evaporation, and runoff to be distributed across a much larger basin. In Muni-Pomadze,
small changes in water volume produce much larger proportional changes in mapped
inundation. The contrast is consistent with the previous multi-seasonal assessment [32],
which reported more stable water extent at Keta and greater thermal variability at Muni-
Pomadze over 2000–2024, although neither comparison establishes the causal contribution
of individual hydrological controls.
Coastal lagoon behavior is controlled by the interaction of freshwater inflow, evapora-
tion, tidal or marine exchange, water depth, sedimentation, and outlet configuration [3,9,30].
These controls are especially important in shallow and intermittently connected lagoons,
where seasonal rainfall and mouth conditions may rapidly alter water extent. The strong
variability observed in Muni-Pomadze is therefore consistent with the expected response of
a small, shallow lagoon influenced by wetland inflows and intermittent ocean connectivity.
Keta’s relatively stable water extent should not be interpreted as direct evidence of
ecological resilience. Hydrological persistence only indicates that the open-water area
remained comparatively stable. Sediment loading, nutrient enrichment, habitat degradation,
salinity changes, and biological disturbance can occur without a corresponding decline in
total water area [4,7]. The increase in the suspended sediment proxy demonstrates that
optical conditions may change even where the mapped water surface remains extensive.
At Muni-Pomadze, changes in mapped inundation can alter the proportions of open
water, exposed sediment and vegetated margins. Shifts in shoreline position, water depth
and mixed-pixel composition can consequently affect the spectral indices even without a
corresponding change in measured water-quality concentrations. Without water-level and
flow observations, the relative contributions of these processes cannot be quantified
A multi-source assessment of the Mai Po–Inner Deep Bay wetland system similarly
distinguishes mapped aquatic extent from relative potential hydrological connectivity [43].
This distinction supports interpreting lagoon water area as one component of hydrolog-
ical condition rather than a direct measure of water exchange. Evapotranspiration also
requires a separate assessment: a comparison of six products across Hong Kong, including
their applicability to Mai Po, identified substantial inter-product differences [44]. Future
water-balance monitoring at Keta and Muni-Pomadze should therefore evaluate evapotran-
spiration uncertainty alongside rainfall and water-level observations.
4.3. Interpretation of Surface-Water Indices
NDWI and MNDWI produced broadly consistent representations of surface-water
distribution. Keta recorded higher NDWI values and a more spatially continuous open-
water signal, while Muni-Pomadze displayed lower values and greater fragmentation. This
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 23 of 29
difference reflects not only the amount of water present but also the effects of shallow depth,
emergent vegetation, exposed substrate, and shoreline mixing.
NDWI enhances open water through the contrast between green and near-infrared re-
flectance [18]. MNDWI replaces the near-infrared band with shortwave-infrared reflectance
and generally improves water discrimination where built-up or bare surfaces occur close to
the shoreline [19]. The agreement between both indices supports the identification of the
central water bodies, while their differences along lagoon margins reveal the main areas of
classification uncertainty.
The completed NDWI/MNDWI threshold sensitivity analysis found substantially
larger proportional changes in mapped water extent at Muni-Pomadze than at Keta. Across
2018–2024, alternative masks differed from the joint zero-threshold baseline by −69.92% to
+60.30% at Muni-Pomadze compared with −10.86% to +9.34% at Keta. These variations
reflect sensitivity to threshold choice, including an OR alternative, rather than independently
observed water delineation error. Marginal and mixed pixels remain a limitation. The year-
by-year scenario ranges are reported in Supplementary Table S2.
Annual median compositing improved consistency by reducing cloud contamination
and anomalous observations. However, it may also suppress short-lived floods, dry-season
contraction, lagoon-mouth opening, and other event-scale changes [34]. The resulting
water-area estimates should therefore be interpreted as representative annual conditions
rather than as maximum or minimum seasonal extents.
4.4. Turbidity- and Sediment-Related Optical Responses
Muni-Pomadze recorded less negative NDTI values than Keta, indicating a compara-
tively stronger turbidity-related optical response. In a small and shallow lagoon, sediment
resuspension, runoff inputs, exposed bottom material, and shoreline disturbance can in-
fluence a large proportion of the water surface. In Keta, sediment introduced from the
catchment may be diluted or redistributed over a broader area, producing weaker basin-
wide average responses despite localized sediment accumulation.
StrongerNDTI and SSP showed spatial differences near lagoon margins and inflow
zones, where shallow-water effects can be strong. Suspended mineral particles, colored
dissolved organic matter, aquatic vegetation, water depth and bottom reflectance can all
affect red and near-infrared reflectance [11,22,24,25]. NDTI and SSP are therefore relative
spectral indicators, not validated estimates of turbidity or suspended-sediment concentra-
tion. The observed signals are consistent with several explanations, including mixing and
potential runoff inputs, but runoff was not measured.
At Keta, the annual SSP increased toward 2024 even though the new mapped agri-
culture and built-up classes declined from 2018 to 2024. Maximum daily CHIRPS rainfall
was positively associated with annual SSP in the seven-year exploratory series (r = 0.868;
unadjusted p = 0.011), whereas annual rainfall totals were not (r = 0.053; p = 0.910). These
correlations do not establish sediment transport or runoff causation, and possible effects
of wind-driven resuspension, saltpan activity and optical bottom conditions remain unre-
solved. As four rainfall–SSP correlations were inspected, the nominal p-value should not be
read as confirmatory significance; the record lacks contemporaneous runoff or discharge
and sediment measurements to test the physical mechanism.
No spatially and temporally matched sediment concentration, river discharge, dredg-
ing, lagoon-mouth operation or verified field observations are available to identify the
source of the SSP changes. The agricultural expansion explanation (H02) is not supported
by the revised land-cover series. The observed optical differences should therefore be
interpreted as relative satellite-derived proxies.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 24 of 29
4.5. Chlorophyll-Related Spectral Variability
The sites differ descriptively: Keta had more persistent mapped open water, whereas
Muni-Pomadze showed larger proportional variation. These patterns cannot establish
ecological resilience or a causal contrast between catchment pressure and hydrological
sensitivity, particularly given mixed pixels and fixed water thresholds [18,19,25].
Taken together, the observations indicate that mapped water extent and relative aquatic
reflectance do not vary identically. Water extent alone does not describe the entire set of
remotely sensed lagoon conditions; no formal ecological status assessment was conducted.
The temporal difference between the NDCI and SSP peaks indicates that the strongest
chlorophyll-related and sediment-related signals did not occur simultaneously. This sug-
gests that the optical variability was controlled by more than one process. Hydrologi-
cal expansion, nutrient delivery, water residence time, vegetation interaction, sediment
resuspension, and atmospheric conditions may have influenced the indices differently
across years.
Long-term studies of other coastal lagoons have shown that remote sensing can reveal
ecological deterioration and changes in phytoplankton-related conditions, but reliable
diagnosis requires sustained field calibration and ecological observations [7]. The present
findings therefore provide a comparative warning signal rather than a direct ecological
status classification.
4.6. Relationships Between Catchment Change and Lagoon Conditions
The revised classifications do not support the original inference of widespread agricul-
tural and built-up expansion in Keta during 2018–2024. Instead, the mapped classes suggest
declines in agriculture and built-up cover and gains in vegetation and shrubland. Because
these are maps generated using temporally transferred reference labels, apparent changes
and bidirectional transitions are provisional and cannot identify the causes of SSP variation.
In Muni-Pomadze, the strongest environmental variability was associated with annual
changes in water extent rather than extensive built-up expansion. Seasonal inundation
affects the proportion of open water, vegetation, exposed sediment, and mixed shoreline
pixels. These shifts can modify NDWI, NDTI, NDCI, and SSP even without large catchment-
scale land-cover conversion.
The comparison suggests greater proportional sensitivity of mapped open-water ex-
tent to threshold selection at Muni-Pomadze. At Keta, the evidence supports continued
monitoring of land-cover classification uncertainty and sediment-related optical variabil-
ity, but not a causal claim that expansion of built-up or agricultural land drove increased
sediment transport.
The combined land-cover, water-mask and spectral analyses provide a broader screen-
ing description of lagoon variability, but the available evidence cannot distinguish ecological
change from hydrological fluctuations and optical mixing without field validation.
At a broader West African basin scale, Upper Niger research combines hydroclimatic
stress components with spatial propagation analysis [45]. This approach provides context
for examining multiple climatic and landscape controls together. Its basin-scale indices do
not, however, independently validate the lagoon optical proxies or identify the causes of
the observed SSP changes.
4.7. Implications for Ramsar Wetland Management
Management recommendations for Keta include regular field monitoring of shoreline
and sediment conditions, scrutiny of salt-production and disturbed shoreline surfaces, and
independent verification of agricultural, built-up and vegetation boundaries. The revised
maps do not establish a consistent increase in agriculture or built-up area during 2018–2024.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 25 of 29
Both Ramsar sites would benefit from independent reference samples and locally calibrated
optical monitoring.
Muni-Pomadze requires protection of the remaining natural vegetation and main-
tenance of hydrological connections among the lagoon, surrounding wetlands, and the
Atlantic Ocean. Because the lagoon responds strongly to annual hydrological variability,
monitoring based only on annual composites may overlook short-lived but ecologically
important events. Seasonal monitoring should therefore accompany long-term mapping.
The Google Earth Engine and Sentinel-2 framework provide a practical basis for re-
peated monitoring where field programs are limited [16,17]. It can identify expanding
settlements, vegetation loss, changing water extent, and areas with unusual optical re-
sponses. These outputs can guide field sampling and management inspections, but they
should not replace measurements of salinity, nutrients, dissolved oxygen, turbidity, sus-
pended sediment, chlorophyll-a, and ecological condition [12,13,25].
Management planning should reflect the different monitoring needs of the two sites.
At Keta, priority should be given to verifying provisional land-cover changes and sediment-
related optical responses; at Muni-Pomadze, seasonal hydrological and ecological observa-
tions are particularly useful. Neither priority assumes that a specific land-cover conversion
or sediment-transport cause has been established.
The revised Sentinel-2 common-footprint maps indicate a Keta decrease in mapped
agriculture (−52.70 km2 ) and built-up land (−30.83 km2 ), together with gains in vege-
tation (+60.09 km2 ) and shrubland (+54.97 km2 ); Muni-Pomadze indicates a mapped
water increase of about 1.00 km2 within the LULC series. These are map-class alloca-
tions, distinct from the independently reprocessed NDWI/MNDWI open-water series in
Table 4. Threshold-sensitivity tests show greater proportional instability of water masks at
Muni-Pomadze, consistent with the directional alternative to H03 descriptively.
The directional agriculture–SSP expectation associated with H02 is not supported
by the revised Keta LULC series, which shows a reduction rather than an expansion of
mapped agriculture. Exploratory CHIRPS–SSP comparisons for 2018–2024 show a negligible
association with annual rainfall (r = 0.053) and a positive association with annual maximum
daily rainfall (r = 0.868; unadjusted p = 0.011); neither evaluates the agriculture–SSP slope
specified by H02 or demonstrates causality. Threshold-scenario contrasts are directionally
consistent with the alternative to H03, but no formal rejection of H03 is claimed. All optical
indices remain uncalibrated proxies.
Annual median composites reduced cloud contamination but smoothed event-scale
variability. Short-duration flooding, lagoon-mouth opening, sediment pulses, and phy-
toplankton blooms may therefore be underrepresented [34]. Seasonal or monthly com-
posites would provide additional temporal detail, although cloud availability may limit
consistent coverage.
A complementary research direction is to compare lagoon observations with coastal
hazard chronologies, including marine heatwaves and tropical cyclones. Testing this re-
lationship in Ghana would require local event records and should distinguish temporal
coincidence from evidence of interacting impacts.
Monitoring priorities emerging from these observations are to verify catchment land-
cover changes and sediment-related signals at Keta, and to investigate shoreline mixing
and seasonal inundation at Muni-Pomadze. Such priorities are suggestions for targeted
observation, not evidence that specific land-management interventions have already been
proven necessary.
The LULC analysis used three temporal snapshots: 2018, 2021, and 2024. These maps
show net and intermediate changes but do not identify the precise year in which each
transition occurred. In addition, the absence of a fully independent validation dataset limits
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 26 of 29
the certainty of small class changes. Large and persistent changes were therefore given
greater interpretative weight than isolated or marginal transitions [46,47].
The methodology is transferable as a comparative screening framework, but index
thresholds and spectral relationships should not be assumed to perform identically in other
coastal lagoons. Application elsewhere should include local threshold sensitivity testing,
independent LULC validation, seasonal hydrological assessment, and field calibration of
the aquatic indicators.
5. Conclusions
Sentinel-2 random forest maps for 2018, 2021 and 2024 showed contrasting class alloca-
tions in the two Ramsar catchments. Keta mapped reductions in the agriculture, built-up
and water classes with gains in vegetation and shrubland, while Muni-Pomadze mapped
an increase in water and vegetation-related classes and a decline in built-up area. Because
the reference labels are partly historical and not independently verified for each observation
year, these patterns require cautious interpretation.
Taken together, the water-mask and optical-index results demonstrate the value of
examining mapped inundation and reflectance jointly; neither alone is a field-validated
measure of ecological condition.
The supplementary sensitivity, valid-pixel and rainfall analyses (Tables S1–S3) docu-
ment diagnostic uncertainty but do not replace independent field observations. In particular,
seven annual rainfall–SSP pairs per lagoon, unadjusted correlation p-values, and missing
runoff/discharge and sediment measurements preclude physical causal inference.
Management priorities are to verify the mapped land-cover transitions and sediment-
related optical signals at Keta, and to monitor seasonal inundation, shoreline mixing and
wetland connectivity at Muni-Pomadze. The provisional classifications do not establish
agricultural or built-up expansion at Keta, so interventions should be based on independent
field assessment rather than presumed land conversion.
The integrated Sentinel-2 and Google Earth Engine framework provides a reproducible
approach for detecting broad wetland changes and directing field investigation in data-
scarce coastal environments.
Supplementary Materials: The following supporting information can be downloaded at: https:
//www.mdpi.com/article/10.3390/w18222856/s1, Table S1: Valid water-pixel support (10 m analysis
grid). Table S2. Threshold-sensitivity ranges across alternative rules. Table S3. Exploratory annual
rainfall–SSP associations, 2018–2024. The tables draw on Sentinel-2 annual-composite exports [33]
and CHIRPS daily rainfall.
Author Contributions: Conceptualization, F.Q., Y.A., G.O.A., K.A.-B., X.D. and I.Y.D.; methodology,
F.Q., M.A. and X.D.; software, F.Q. and X.D.; validation, F.Q., X.D. and I.Y.D.; formal analysis,
M.A., F.Q., X.D., G.O.A., Y.A., and I.Y.D.; investigation, F.Q., X.D., Y.A., and I.Y.D.; resources, F.Q.
and X.D.; data curation, F.Q. and I.Y.D.; writing—original draft preparation, F.Q., X.D. and I.Y.D.;
writing—review and editing, F.Q., X.D. and I.Y.D.; visualization, F.Q. and X.D.; supervision, I.Y.D.,
Y.A., K.A.-B., and X.D. All authors have read and agreed to the published version of the manuscript.
Funding: The research was jointly supported by the Research Grants Council (RGC) of the Hong
Kong Special Administrative Region (PolyU 152318/22E, 152344/23E), National Science Foundation
of China (42330717), Innovative Technology Commission (ITC) (K-BBY1—Smart Railway Technology,
and Guangdong-Hong Kong Joint Laboratory for Marine Infrastructure (Hong Kong, China).
Data Availability Statement: The Sentinel-2 Level-2A surface-reflectance imagery and Scene Classifi-
cation Layer used in this study are publicly available through the Google Earth Engine data catalog.
Lagoon boundaries, classified land-cover products, processed indicator tables, and analytical scripts
generated or used during the study are available from the corresponding author on reasonable request.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 27 of 29
Acknowledgments: During the preparation of this manuscript/study, the authors used Grammarly
(version v.1.2.281.1928) for the purposes of language editing and grammatical correction. The authors
have reviewed and edited the output and take full responsibility for the content of this publication.
Conflicts of Interest: The authors declare no conflicts of interest.
Abbreviations
The following abbreviations are used in this manuscript:
NDWI Normalized Difference Water Index
MNDWI Modified Normalized Difference Water Index
NDTI Normalized Difference Turbidity Index
NDCI Normalized Difference Chlorophyll Index
GEE Google Earth Engine
MSI MultiSpectral Instrument
SCL Scene Classification Layer
SSP Suspended sediment proxy
References

  1. Mitsch, W.J.; Gosselink, J.G. Wetlands, 5th ed.; John Wiley & Sons: Hoboken, NJ, USA, 2015.
  2. Convention on Wetlands. Global Wetland Outlook: Special Edition 2021; Secretariat of the Convention on Wetlands: Gland,
    Switzerland, 2021. [CrossRef]
  3. Kjerfve, B. Coastal lagoons. In Coastal Lagoon Processes; Kjerfve, B., Ed.; Elsevier: Amsterdam, The Netherlands, 1994; Volume 60,
    pp. 1–8. [ CrossRef]
  4. Newton, A.; Icely, J.; Cristina, S.; Brito, A.; Cardoso, A.C.; Colijn, F.; Dalla Riva, S.; Gertz, F.; Hansen, J.W.; Holmer, M.; et al. An
    overview of ecological status, vulnerability and future perspectives of European large shallow, semi-enclosed coastal systems,
    lagoons and transitional waters. Estuar. Coast. Shelf Sci. 2014, 140, 95–122. [CrossRef]
  5. Davidson, N.C. How much wetland has the world lost? Long-term and recent trends in global wetland area. Mar. Freshw. Res.
    2014, 65, 934–941. [CrossRef]
  6. Kirwan, M.L.; Megonigal, J.P . Tidal wetland stability in the face of human impacts and sea-level rise. Nature 2013, 504, 53–60.
    [CrossRef] [PubMed]
  7. Molner, J.V .; Soria, J.M.; Campillo-Tamarit, N.; Pérez-González, R.; Sòria-Perpinyà, X.; Muñoz-Colmenares, M. Ecological Decline
    over a Decade in the Albufera of Valencia Coastal Lagoon (Spain): A Growing Environmental Hazard in a Hypertrophic System.
    Phycology 2026, 6, 27. [CrossRef]
  8. Quayson, F.; Ding, X.; Abdallah, M.; Dadson, I.Y.; Adu-Boahen, K.; Asamoah, Y. Integrated assessment of environmental pressure
    in Ghana coastal wetlands using climate remote sensing and socioeconomic data. Discov. Sustain. 2026. advance online publication.
    [CrossRef]
  9. Davies-Vollum, K.S.; Zhang, Z.; Agyekumhene, A. Impacts of lagoon opening and implications for coastal management: Case
    study from Muni-Pomadze lagoon, Ghana. J. Coast. Conserv. 2019, 23, 293–301. [CrossRef]
  10. Quayson, F.; Ding, X. A scientometric analysis and systematic review on climate change and coastal wetlands. Discov. Sustain.
  11. advance online publication. [ CrossRef]
  12. Mouw, C.B.; Greb, S.; Aurin, D.; DiGiacomo, P .M.; Lee, Z.; Twardowski, M.; Binding, C.; Hu, C.; Ma, R.; Moore, T.; et al. Aquatic
    color radiometry remote sensing of coastal and inland waters: Challenges and recommendations for future satellite missions.
    Remote Sens. Environ. 2015, 160, 15–30. [CrossRef]
  13. Tyler, A.N.; Hunter, P .D.; Spyrakos, E.; Groom, S.; Constantinescu, A.M.; Kitchen, J. Developments in Earth observation for
    the assessment and monitoring of inland, transitional, coastal and shelf-sea waters. Sci. Total Environ. 2016, 572, 1307–1321.
    [CrossRef] [PubMed]
  14. Gholizadeh, M.H.; Melesse, A.M.; Reddi, L. A Comprehensive Review on Water Quality Parameters Estimation Using Remote
    Sensing Techniques. Sensors 2016, 16, 1298. [CrossRef] [PubMed]
  15. Pekel, J.-F.; Cottam, A.; Gorelick, N.; Belward, A.S. High-resolution mapping of global surface water and its long-term changes.
    Nature 2016, 540, 418–422. [CrossRef] [PubMed]
  16. Wulder, M.A.; Loveland, T.R.; Roy, D.P .; Crawford, C.J.; Masek, J.G.; Woodcock, C.E.; Allen, R.G.; Anderson, M.C.; Belward,
    A.S.; Cohen, W.B.; et al. Current status of Landsat program, science, and applications. Remote Sens. Environ. 2019, 225, 127–147.
    [CrossRef]
    https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 28 of 29
16. Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary-scale geospatial
analysis for everyone. Remote Sens. Environ. 2017, 202, 18–27. [CrossRef]
17. Drusch, M.; Del Bello, U.; Carlier, S.; Colin, O.; Fernández, V .; Gascon, F.; Hoersch, B.; Isola, C.; Laberinti, P .; Martimort, P .;
et al. Sentinel-2: ESA ’s Optical High-Resolution Mission for GMES Operational Services. Remote Sens. Environ. 2012, 120, 25–36.
[CrossRef]
18. McFeeters, S.K. The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. Int. J.
Remote Sens. 1996, 17, 1425–1432. [CrossRef]
19. Xu, H. Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery.
Int. J. Remote Sens. 2006, 27, 3025–3033. [CrossRef]
20. Lacaux, J.P .; Tourre, Y.M.; Vignolles, C.; Ndione, J.A.; Lafaye, M. Classification of ponds from high-spatial resolution remote
sensing: Application to Rift Valley Fever epidemics in Senegal. Remote Sens. Environ. 2007, 106, 66–74. [CrossRef]
21. Mishra, S.; Mishra, D.R. Normalized difference chlorophyll index: A novel model for remote estimation of chlorophyll-a
concentration in turbid productive waters. Remote Sens. Environ. 2012, 117, 394–406. [CrossRef]
22. Nechad, B.; Ruddick, K.G.; Park, Y. Calibration and validation of a generic multisensor algorithm for mapping of total suspended
matter in turbid waters. Remote Sens. Environ. 2010, 114, 854–866. [CrossRef]
23. Dogliotti, A.I.; Ruddick, K.G.; Nechad, B.; Doxaran, D.; Knaeps, E. A single algorithm to retrieve turbidity from remotely-sensed
data in all coastal and estuarine waters. Remote Sens. Environ. 2015, 156, 157–168. [CrossRef]
24. Novoa, S.; Doxaran, D.; Ody, A.; Vanhellemont, Q.; Lafon, V .; Lubac, B.; Gernez, P . Atmospheric Corrections and Multi-Conditional
Algorithm for Multi-Sensor Remote Sensing of Suspended Particulate Matter in Low-to-High Turbidity Levels Coastal Waters.
Remote Sens. 2017, 9, 61. [CrossRef]
25. Odermatt, D.; Gitelson, A.; Brando, V .E.; Schaepman, M. Review of constituent retrieval in optically deep and complex waters
from satellite imagery. Remote Sens. Environ. 2012, 118, 116–126. [CrossRef]
26. Toming, K.; Kutser, T.; Laas, A.; Sepp, M.; Paavel, B.; Nõges, T. First Experiences in Mapping Lake Water Quality Parameters with
Sentinel-2 MSI Imagery. Remote Sens. 2016, 8, 640. [CrossRef]
27. Pahlevan, N.; Smith, B.; Schalles, J.; Binding, C.; Cao, Z.; Ma, R.; Alikas, K.; Kangro, K.; Gurlin, D.; Hà, N.; et al. Seamless retrievals
of chlorophyll-a from Sentinel-2 (MSI) and Sentinel-3 (OLCI) in inland and coastal waters: A machine-learning approach. Remote
Sens. Environ. 2020, 240, 111604. [CrossRef]
28. Duku, E.; Mattah, P .A.D.; Angnuureng, D.B. Assessment of Land Use/Land Cover Change and Morphometric Parameters in the
Keta Lagoon Complex Ramsar Site, Ghana. Water 2021, 13, 2537. [CrossRef]
29. Peters, M.K.; Kusimi, J.M. Changes in wetland and other landscape elements of the Keta Municipal area of Ghana. J. Coast.
Conserv. 2023, 27, 1. [CrossRef]
30. Gordon, C.; Ntiamoa-Baidu, Y.; Ryan, J.M. The Muni-Pomadze Ramsar site. Biodivers. Conserv. 2000, 9, 447–464. [CrossRef]
31. Adade, R.; Nyarko, B.K.; Aheto, D.W.; Osei, K.N. Fragmentation of wetlands in the south eastern coastal savanna of Ghana. Reg.
Stud. Mar. Sci. 2017, 12, 40–48. [CrossRef]
32. Quayson, F.; Ding, X. Multi-Seasonal Analysis of Hydroclimate Variability and Anthropogenic Pressures on the Keta and
Muni-Pomadze Wetlands. Hydrology 2026, 13, 249. [CrossRef]
33. Google. Harmonized Sentinel-2 MSI: MultiSpectral Instrument, Level-2A (SR). Available online: https://develop-
ers.google.com/earth-engine/datasets/catalog/COPERNICUS_S2_SR_HARMONIZED (accessed on 24 September 2026).
34. Google. Reducing an ImageCollection. Available online: https://developers.google.com/earth-engine/guides/ic_reducing
(accessed on 24 September 2026).
35. Breiman, L. Random Forests. Mach. Learn. 2001, 45, 5–32. [CrossRef]
36. Belgiu, M.; Drăguţ, L. Random forest in remote sensing: A review of applications and future directions. ISPRS J. Photogramm.
Remote Sens. 2016, 114, 24–31. [CrossRef]
37. Quayson, F.; Ding, X.; Amoah, G.O.; Ajugu, C.A.; Asamoah, Y. Spatially Validated Machine Learning Flood Susceptibility
Assessment in the Volta River Basin Using Extreme Rainfall and Multi-Source Geospatial Data. Hydrology 2026, 13, 268. [CrossRef]
38. McKinney, W. Data structures for statistical computing in Python. In Proceedings of the 9th Python in Science Conference, Austin,
TX, USA, 28 June–3 July 2010; pp. 56–61. [CrossRef]
39. Harris, C.R.; Millman, K.J.; van der Walt, S.J.; Gommers, R.; Virtanen, P .; Cournapeau, D.; Wieser, E.; Taylor, J.; Berg, S.; Smith,
N.J.; et al. Array programming with NumPy. Nature 2020, 585, 357–362. [CrossRef] [PubMed]
40. Virtanen, P .; Gommers, R.; Oliphant, T.E.; Haberland, M.; Reddy, T.; Cournapeau, D.; Burovski, E.; Peterson, P .; Weckesser, W.;
Bright, J.; et al. SciPy 1.0: Fundamental algorithms for scientific computing in Python. Nat. Methods 2020, 17, 261–272. [CrossRef]
[PubMed]
41. Hunter, J.D. Matplotlib: A 2D Graphics Environment. Comput. Sci. Eng. 2007, 9, 90–95. [CrossRef]
42. Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V .; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P .; Weiss, R.; Dubourg, V .;
et al. Scikit-learn: Machine learning in Python. J. Mach. Learn. Res. 2011, 12, 2825–2830.
https://doi.org/10.3390/w18222856

Water 2026, 18, 2856 29 of 29
43. Quayson, F.; Ding, X.; Abdallah, M.; Dadson, I.Y.; Adu-Boahen, K.; Asamoah, Y.; Amoah, G.O. Environmental Stress and Relative
Hydrological Connectivity Across the Mai Po–Inner Deep Bay Wetland System Using Multi-Source Remote Sensing. Geomatics
2026, 6, 121. [CrossRef]
44. Quayson, F.; Ding, X.; Dadson, I.Y. Spatiotemporal Evaluation and Uncertainty Assessment of Multi-Source Evapotranspiration
Products Across Hong Kong: Implications for Mai Po Wetland Water-Balance Monitoring. Water 2026, 18, 1897. [CrossRef]
45. Amoah, G.O.; Olawoyin, R.; Quayson, F. Hydro-climatic ecosystem risk mapping and spatial propagation analysis in the Upper
Niger Basin, West Africa. Environ. Chall. 2026, 24, 101567. [CrossRef]
46. Congalton, R.G.; Green, K. Assessing the Accuracy of Remotely Sensed Data: Principles and Practices, 3rd ed.; CRC Press: Boca Raton,
FL, USA, 2019. [CrossRef]
47. Olofsson, P .; Foody, G.M.; Herold, M.; Stehman, S.V .; Woodcock, C.E.; Wulder, M.A. Good practices for estimating area and
assessing accuracy of land change. Remote Sens. Environ. 2014, 148, 42–57. [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury
to people or property resulting from any ideas, methods, instructions or products referred to in the content.
https://doi.org/10.3390/w18222856

论文原文与出版信息

Water
查看论文原文

继续探索