Evaluating Landsat Water Indices and Monitoring Long-Term Surface-Water Dynamics in Lake Nasser and the Tushka Lakes in a Hyper-Arid Environment Using Google Earth Engine

Abstract

Long-term monitoring of surface-water dynamics in hyper-arid reservoir systems requires consistent remote-sensing methods that can distinguish open water from bright desert surfaces, shallow water, wet sand, and mixed shoreline pixels. This study evaluates Landsat-derived spectral water indices for delineating surface water in Lake Nasser and the adjacent Tushka Lakes, generates a multi-decadal record of surface-water extent using Google Earth Engine, and places the resulting surface-water patterns in the context of available hydrogeological observations. Landsat TM and OLI surface reflectance imagery was used to compare seven commonly applied water indices (NDWI, EWI, NDX, WRI, AWEI_nsh, TCW, and NWI) based on mapped water area, relative area differences, and classification accuracy metrics derived from 1000 stratified reference samples. Among the tested indices, NDWI provided stable water–land separation (overall accuracy ≈ 93.6%; κ ≈ 0.898) and was selected for long-term mapping. The NDWI-based workflow was implemented in Google Earth Engine to generate quarterly composites of surface-water extent for the period 1987–2026. The resulting time series reveals stable, persistent surface water in the central and southern sectors of Lake Nasser, in contrast to pronounced seasonal and interannual variability in the shallow, intermittently connected Tushka basins. Total mapped water area increased from 2631 km² in 1987 to 8923 km² in early 2026, with Lake Nasser ranging from 2411 to 6060.7 km² and the Tushka Lakes expanding from no mapped water before 1998 to more than 3300 km² during 2025. To assess possible surface–subsurface interaction, daily lake-stage records (1965–2014) and monthly groundwater levels from 44 observation wells were used to estimate potential seepage losses from Lake Nasser to the Nubian Sandstone Aquifer System using Darcy’s law. Annual seepage estimates ranged from 15.58 × 10⁶ to 36.68 × 10⁶ m³/year, suggesting spatial variability in potential lake–aquifer seepage along the western lake margin. The combined remote-sensing and hydrogeologic results provide complementary, non-causal evidence for interpreting where surface-water persistence and estimated seepage may co-occur. Because spatial correlation analysis, calibrated ground-water modeling, full water-budget analysis, and independent field validation were not performed, the inferred seepage–surface-water relation should be regarded as a cautious hypothesis rather than proof of causality.

1. Introduction

Monitoring surface-water dynamics in large desert reservoirs is essential for understanding hydrologic variability, evaluating water-resource security, and supporting sustainable management under changing climatic and upstream regulatory conditions [1,2,3,4]. In Egypt, Lake Nasser and the adjacent Tushka Lakes form a strategically important and highly dynamic surface-water system in which the water extent varies over time in response to reservoir operations, hydrologic forcing, local topography, and surface–subsurface interactions [5,6,7]. Because these water bodies occupy a hyper-arid environment characterized by shallow shorelines, intermittent connectivity, and spectrally complex surrounding terrain, they provide a valuable natural setting for testing geospatial methods for water detection and long-term monitoring [8].

Recent advances in cloud-based geospatial analysis have expanded the capacity to process large Earth-observation archives by integrating remote sensing, automated preprocessing, reproducible scripting, and spatial analytics [9,10,11]. Within this context, the Landsat program offers one of the most valuable resources for long-term inland-water monitoring because it provides multi-decadal, moderate-resolution imagery suitable for consistent change analysis [12,13]. However, reliable extraction of water extent from Landsat imagery depends strongly on the performance of spectral water indices, which may vary across sensors and environmental conditions [14]. In arid reservoir settings such as Lake Nasser and the Tushka Lakes, index performance can be influenced by turbidity, shallow-water effects, mixed shoreline pixels, dark desert surfaces, and seasonal fluctuations in inundation patterns [15].

Previous studies have established a strong foundation for satellite-based monitoring of inland water bodies. Global and regional investigations have shown that multi-temporal satellite imagery can be used to map surface-water occurrence, lake-area change, and reservoir dynamics over long periods [4,9,10]. Other studies have focused on the development and comparison of spectral water indices, including NDWI, MNDWI, AWEI, WRI, TCW, and related formulations, demonstrating that index performance depends on sensor characteristics, threshold choice, water turbidity, background land cover, and shoreline conditions [11,12,13,14,15,16]. Recent cloud-computing studies have further shown that Google Earth Engine and automated workflows can efficiently process large Landsat archives for water mapping, trend detection, and near-operational monitoring [17,18,19,20]. These studies confirm the general value of Landsat and cloud-based analysis for surface-water monitoring, but they also emphasize that no single water index performs optimally in all environments and that local validation remains necessary.

For the Lake Nasser–Tushka region specifically, earlier research has addressed several related aspects of the system. Remote sensing and GIS studies monitored Lake Nasser morphology, surface area fluctuations, and environmental characteristics, while altimetry-based studies evaluated water-level variability and short-term changes in lake area [21,22]. Other studies examined the formation, expansion, contraction, and water-budget behavior of the Tushka Lakes, including the role of evaporation and groundwater recharge in lake-volume loss [5,15]. Hydrogeologic investigations have also examined the interaction between Lake Nasser and the Nubian Sandstone Aquifer System using piezometric data, geophysical surveys, Darcy-based seepage calculations, isotope evidence, and groundwater-flow models [23,24,25,26,27,28,29,30,31,32,33]. Collectively, these studies demonstrate that the Lake Nasser–Tushka system has been studied from remote-sensing, geomorphologic, water-budget, and hydrogeologic perspectives; however, these strands of research have generally been treated separately rather than combined in a single framework that first evaluates multiple Landsat water indices and then applies the most robust index to a long-term, sensor-consistent surface-water record.

Despite these contributions, several gaps remain. First, many previous studies of Lake Nasser and Tushka focused on water-area change, water level, volume, evaporation, or groundwater response, but did not systematically compare the performance of several Landsat-derived water indices under the challenging spectral conditions of a hyper-arid reservoir margin. Second, existing regional studies often emphasized either the main reservoir or the Tushka Lakes, whereas fewer studies treated them as a coupled yet behaviorally contrasting surface-water system with a stable deep-reservoir core and shallow, threshold-controlled peripheral basins. Third, previous hydrogeologic studies have provided important estimates of seepage and groundwater response, but these results have rarely been linked, even cautiously, to multi-decadal maps of surface-water persistence and shoreline variability. Finally, the multi-sensor Landsat record introduces methodological challenges related to band equivalence, sensor transitions, threshold consistency, mixed pixels, and poor-quality observations that require explicit treatment before long-term trends can be interpreted with confidence.

The novelty of the present study is therefore threefold. First, it provides a site-specific comparative evaluation of seven Landsat-derived spectral water indices (NDWI, EWI, NDX, WRI, AWEInsh, TCW, and NWI) for a hyper-arid reservoir environment where bright desert surfaces, shallow water, wet sand, and mixed shoreline pixels complicate water extraction. Second, it develops a reproducible Landsat/GEE workflow that harmonizes Landsat 5 TM and Landsat 8 OLI imagery, applies consistent thresholding and accuracy assessment, and reconstructs quarterly surface-water dynamics for Lake Nasser and the Tushka Lakes from 1987 to 2026. Third, it places the resulting persistence and surface-water variability maps in the context of available lake-stage, piezometric, and Darcy-based seepage information to provide a cautious, non-causal interpretation of possible lake–aquifer interaction. Accordingly, the objectives are to (1) evaluate seven Landsat-derived spectral water indices using mapped-area and accuracy metrics; (2) identify the most robust index for long-term application in a hyper-arid reservoir environment; (3) generate a quarterly surface-water record for 1987–2026 to characterize seasonal and interannual variability in Lake Nasser and the Tushka basins; (4) map surface-water persistence to distinguish stable from episodic inundation patterns; and (5) assess potential lake–aquifer seepage using long-term lake-stage data, piezometric observations, and hydrogeologic parameters. Through this integrated approach, this study fills the gap between previous index-comparison studies, regional Lake Nasser–Tushka monitoring, and hydrogeologic seepage investigations, while providing a transferable workflow for monitoring large desert reservoirs.

2. Study Area and Geology

2.1. Study Area

The study area comprises Lake Nasser and the adjacent Tushka Lakes in southern Egypt (Figure 1). Lake Nasser is a large artificial reservoir formed by impounding the Nile River following completion of the Aswan High Dam in the late 1960s [21]. Extending approximately 500 km from southern Egypt into northern Sudan, the lake typically ranges from 10 to 30 km wide. At high operational levels, Lake Nasser stores as much as 10¹¹ m³ of water and inundates several thousand square kilometers, making it Egypt’s primary strategic surface-water storage facility [22]. The hydrogeologic setting of the Lake Nasser region is dominated by the Nubian Sandstone Aquifer System (NSAS), one of the largest fossil aquifers in the world, extending beneath much of southwestern Egypt, southeastern Libya, northeastern Chad, and northern Sudan. In this study area, the aquifer consists of sandstone interbedded with shale and clay, with a total thickness ranging from 390 to 592 m [23]. Hydrogeological investigations indicate that the Nubian Sandstone sequence can be subdivided into several water-bearing layers. These horizons are separated by semi-confining units, with local hydraulic connections through interfingering and facies changes [24]. It consists of sandstone interbedded with shale and clay. Unconfined (shallow) and confined (deep) groundwater conditions are present, and earlier studies have documented hydraulic connectivity between upper and lower aquifer layers in the Lake Nasser region [25,26,27].

Lake Nasser and its surroundings are situated within a hyper-arid desert environment characterized by extremely sparse vegetation, high evaporative demand, and extensive aeolian sand sheets and dune fields, particularly along the northern shoreline [15]. These environmental conditions pose well-known challenges for satellite-based water mapping, including strong land–water spectral contrast, mixed pixels along irregular shorelines, shallow-water and turbidity effects, and spectral confusion with dark soils, wet sand, and shadowed terrain. The surrounding geology consists of Nubian Sandstone plateaus, localized basement rocks, and volcanic outcrops, which influence shoreline morphology and embayment geometry and contribute to spatial variability in shoreline dynamics [21]. The Tushka Lakes are a group of shallow depressional basins found west and southwest of the main Lake Nasser channel. These basins are intermittently connected to Lake Nasser via natural spillways and the topography of the Tushka Depression [5]. During periods of elevated lake levels or exceptional flood events, water spills into the Tushka basins, leading to rapid lake expansion; during drier periods, the basins are partially or completely desiccated. This episodic connectivity, combined with shallow bathymetry and complex shoreline geometry, results in pronounced seasonal and interannual variability in water extent. Consequently, the Tushka Lakes provide a stringent test case for evaluating the performance of spectral water indices under highly dynamic inundation and mixed-shoreline pixels. Together, Lake Nasser and the Tushka Lakes form a coupled surface-water system whose spatial and temporal variability reflects both upstream hydrologic forcing and local geomorphic and hydrogeologic controls. The contrasting behaviors of the stable main reservoir and the highly variable Tushka basins make this region well-suited for comparative evaluation of Landsat-derived water indices and for long-term monitoring of surface-water dynamics in hyper-arid environments.

2.2. Surface and Subsurface Geology

In the Nasser Lake area, the Nubian Sandstone represents a small portion of the well-known Nubian Sandstone Aquifer System in the Eastern Sahara, which covers the entire area of southwest Egypt, southeast Libya, northeast Chad, and northern Sudan and thus constitutes the main source of groundwater in this area (Figure 2 and Figure 3). The aquifer is composed of sandstone intercalated with shale and clay, with a total thickness ranging from 390 to 592 m [23]. The hydrogeology of the study area was studied by several authors [5,28,29,30,31,32,33]. Hydrogeologically, El-Shazly et al. [24] proposed that the sediments comprising the Nubian aquifer system may be subdivided regionally into two or three water-bearing layers, with intervening shale or sandy shale confining units, some of which are hydraulically connected through interfingering and local facies changes. The shallow or upper water-bearing horizons exist under the free water table, while the lower or deeper water-bearing horizons exist under confined conditions. In the lake region, El Ramly [23] noticed a clear hydraulic connection between the upper and lower layers of the Nubian Sandstone aquifer. Subsurface geology of the study area, the west side of Nasser Lake, was inferred from lithological logs of various wells drilled there (Figure 1 and Figure 3 [34]). The Nubian Sandstone aquifer includes water-bearing sands and sandstones, separated by semi-confining clay and sandy-clay intercalations. It directly overlays the basement complex.

3. Data Used

3.1. Remote Sensing Data

The core dataset used in this study is the Landsat surface reflectance archive, which provides the temporal foundation for evaluating spectral water indices and monitoring long-term surface-water dynamics in Lake Nasser and the Tushka Lakes. The remote-sensing component was designed around a cloud-based analytical workflow implemented in Google Earth Engine (GEE) [17,18], in which all image access, filtering, preprocessing, and multi-temporal analysis were performed. Landsat Collection 2, Tier 1, Level 2 Surface Reflectance (SR) products were selected because they provide radiometrically and geometrically corrected observations appropriate for consistent cross-date analysis [35]. Landsat 5 TM provides the historical record from 1987 to 2013, whereas Landsat 8 OLI provides the post-2013 record. The main spectral difference between the two sensors lies in the positions and widths of their visible, near-infrared, and shortwave-infrared bands; therefore, only spectrally equivalent 30 m bands were used for index calculation. In this study, TM bands 1, 2, 3, 4, 5, and 7 were matched, respectively, to OLI bands 2, 3, 4, 5, 6, and 7 for blue, green, red, NIR, SWIR1, and SWIR2 reflectance. OLI coastal aerosol and cirrus bands were not used in the water-index formulas. Landsat 7 ETM+ was not included in the long-term reconstruction because the scan-line corrector failure after 2003 produces systematic data gaps that complicate shoreline extraction and quarterly compositing; excluding ETM+ also avoided introducing a third sensor transition into the time series. Visible, near-infrared (NIR), and shortwave-infrared (SWIR) bands at 30 m spatial resolution were used because they are required to compute the seven Landsat-derived water indices evaluated in this study.

Table 1. Description and characteristics of Landsat time-series data (1987–2026).

Satellite Sensor and Seams	Spatial Resolution (m)	Bands of Interest/Range (Micrometers)	Bandwidth (Micrometers)	Launch Date and Revisit Time
Landsat-5 Landsat 5, level 2, collection 2, tier 1	30	Band 1 Visible Blue	(0.450–0.520 µm)	Launched 1 March 1984 Revisit time 16 days
		Band 2 Visible Green	(0.520–0.600 µm)
		Band 3 Visible Red	(0.630–0.690 µm)
		Band 4 Near-Infrared	(0.760–0.900 µm)
		Band 5 Near-Infrared	(1.550–1.750 µm)
		Band 6 Thermal	(10.40–12.50 µm)
		Band 7 Mid-Infrared	(2.080–2.350 µm)
Landsat-8 OLI Landsat 8, level 2, collection 2, tier 1	30	Band 1 Coastal Aerosol	(0.430–0.450 µm)	Launched 11 February 2013 Revisit time 16 days
		Band 2 Blue	(0.450–0.510 µm)
		Band 3 Green	(0.530–0.590 µm)
		Band 4 Red	(0.640–0.670 µm)
		Band 5 Near-Infrared	(0.850–0.880 µm)
		Band 6 SWIR 1	(1.57–1.65 µm)
		Band 7 SWIR 2	(2.11–2.29 µm)

summarizes the Landsat datasets and their relevant characteristics, and Figure 4 provides representative images from the time series.

Within the Landsat/GEE workflow, Landsat scenes were filtered to the study area boundary and screened for atmospheric contamination using the quality-assurance information supplied with the Level 2 SR products (QA_PIXEL) [36]. Pixels affected by cloud, cloud shadow, cirrus, or dilated cloud were masked before computing the water indices to improve classification reliability. To support long-term monitoring, the processed imagery was aggregated into quarterly composites (January–March, April–June, July–September, and October–December) using a median reducer, thereby reducing residual cloud effects and providing a temporally consistent basis for seasonal and interannual comparison. This Landsat-based quarterly archive constitutes the primary input for evaluating index performance, selecting the most robust water-extraction approach, and reconstructing long-term surface-water dynamics across the Lake Nasser–Tushka system. Additional preprocessing and analytical details are presented in the Methodology section.

3.2. Observation Wells Data

In addition to the Landsat archive, the study uses supporting hydrogeologic datasets to interpret possible surface–subsurface interactions along the western margin of Lake Nasser. These data include groundwater levels from 44 observation wells (piezometers) distributed across five sectors west of the lake—Garf Hussein, Tushka, Abu Simbel, Adindan, and Argeen (Figure 1 and

Table 2. Elevation and total depth of the piezometers around Nasser Lake.

Region	Observation Wells	Ground Elevation (m)	Total Depth (m)
Garf Hussein	PW-7	193.8	102
	PW-8	196.1	107.6
	PW-8A	197.3	101
	PW-8B	190.4	100
Tushka	TU-1	187.1	100
	TU-2	197.9	98.0
	TU-3	199.9	111
	TU-4	210.8	125
	W5 deep	225.1	386
Abu Simble	AS-1	188.3	98.4
	AS-2	185.6	95.0
	AS-3	193.2	101
	AS-4	198.3	108
	W4 DEEP	188.6	431
Adindan	ADW-1	180.5	82.0
	ADW-2	195.5	97.0
	ADW-3	212.9	114.6
	ADW-4	239.1	140.1
	W3 DEEP	245.5	390.3
Argeen	PW-1	187.7	100
	PW-2	204.9	105
	PW-3	219.3	120
	PW-4	237.5	140
	W2 DEEP	244.5	299

). The wells provide a spatial framework for assessing groundwater response relative to long-term changes in surface-water extent. The five monitoring sectors represent different hydrogeologic settings and distances from the High Dam, allowing comparison of hydraulic conditions along the lake margin. The well network includes both shallow and deep piezometers, thereby providing information relevant to vertical and lateral variations in groundwater conditions within the Nubian Sandstone Aquifer System.

To complement the Landsat-based surface-water analysis, daily Lake Nasser stage records for 1965–2014 and monthly groundwater-level measurements from the observation wells were used to evaluate potential lake–aquifer seepage. These records do not replace the remote-sensing data; rather, they provide independent hydrogeologic context for interpreting long-term shoreline behavior and water persistence derived from Landsat. Lithological and hydrogeological information from the drilled wells in each sector was also used to estimate hydraulic conductivity, seepage-face thickness, and hydraulic gradients needed for the Darcy-based seepage calculations. Accordingly, the data used in this study can be grouped into two complementary categories: (1) Landsat imagery processed in Google Earth Engine for water-index evaluation and long-term surface-water monitoring, and (2) lake-stage, piezometric, and hydrogeologic data for supporting interpretation of surface–subsurface interaction in the Lake Nasser–Tushka system.

4. Methodology

The methodology was designed as a Landsat/GEE-based workflow to evaluate water indices and monitor long-term surface-water dynamics in Lake Nasser and the Tushka Lakes. The workflow integrates cloud-based image processing in Google Earth Engine (GEE), multi-index spectral analysis, threshold-based water extraction, accuracy assessment, temporal reconstruction, and hydrogeologic interpretation [19,20]. In sequence, the workflow included (1) acquisition of Landsat Collection 2 Level 2 surface-reflectance imagery for 1987–2026; (2) preprocessing and quality masking using QA_PIXEL; (3) generation of quarterly median composites; (4) computation of seven Landsat-derived water indices (NDWI, EWI, NDX, WRI, AWEInsh, TCW, and NWI); (5) extraction of binary water masks using index-specific thresholds; (6) comparative evaluation of index performance using mapped area and classification accuracy metrics; (7) selection of the most robust index for long-term application; (8) re-construction of quarterly surface-water dynamics and persistence patterns; and (9) qualitative interpretation of potential lake–aquifer seepage using lake-stage, piezometric, and hydrogeologic data.

4.1. Data Preprocessing

Preprocessing was applied to create a temporally consistent Landsat archive suitable for both index evaluation and long-term monitoring. Landsat 5 TM and Landsat 8 OLI Collection 2 Tier 1 Level 2 surface-reflectance imagery were accessed and processed in Google Earth Engine [18]. Scenes were filtered to the Lake Nasser–Tushka study area, and cloud-, cloud-shadow-, cirrus-, and dilated-cloud pixels were removed using the QA_PIXEL quality band [36]. Sensor harmonization was handled at four levels [37,38]. First, both datasets were taken from the same Collection 2 Level 2 surface-reflectance processing framework, so reflectance values were atmospherically corrected and distributed in a comparable scaled-reflectance format. Second, equivalent 30 m spectral bands were renamed to a common band scheme before index computation: blue, green, red, NIR, SWIR1, and SWIR2. Third, all composites were generated on the native 30 m Landsat grid and clipped to the same study-area boundary, ensuring that mapped area was computed from a consistent pixel size. Fourth, the same index equations, thresholding logic, masking procedure, and area-calculation method were applied to both TM and OLI composites. The Landsat 5–Landsat 8 transition was therefore treated as a controlled sensor change rather than a methodological break: the time series ends for TM in 2013, resumes with OLI after its operational availability in 2013, and no artificial adjustment was applied beyond the surface-reflectance harmonization and common-band mapping described above. Landsat 7 ETM+ was excluded from the long-term reconstruction to avoid scan-line corrector gaps after 2003 and to maintain a simpler two-sensor TM–OLI record. Quarterly median composites were then generated for January–March, April–June, July–September, and October–December. This aggregation reduced residual atmospheric contamination, minimized the influence of individual outlier scenes, and provided a consistent seasonal sampling unit for comparing observations before and after the Landsat 5–Landsat 8 sensor transition. Quarters with insufficient valid observations after QA masking, severe residual contamination, or incomplete spatial coverage over the water bodies were treated as poor-quality composites and were not used for quantitative interpretation; where necessary, the nearest valid quarterly or annual observation was used only to preserve temporal context in figures and tables, not to replace the original mapped value. These preprocessed composites formed the base dataset for computing all evaluated water indices and reconstructing long-term surface-water dynamics.

4.2. Spectral Water-Index Computation

A central component of the proposed framework is the comparative evaluation of Landsat-derived spectral water indices for water extraction in a hyper-arid reservoir environment [39,40]. Seven indices were computed from the quarterly Landsat composites: the Normalized Difference Water Index (NDWI) [41]; Equation (1), the normalized difference index (NDX) [42]; Equation (2), the New Water Index (NWI) [43]; Equation (3), the Water Ratio Index (WRI) [44]; Equation (4), the Automated Water Extraction Index (AWEI_nsh) [45]; Equation (5), the Tasseled Cap Wetness (TCW) transformation [46]; Equation (6); and the Enhanced Water Index (EWI) [47], Equation (7). Each index was calculated using the standard Landsat band formulation after preprocessing. Binary water masks were then produced using an index-specific thresholding procedure consisting of two stages: histogram-guided threshold calibration, followed by the application of a fixed threshold [48]. In the calibration stage, index-value histograms were generated from representative quarterly composites over the study area, the dominant water and non-water modes were identified, and an initial candidate threshold was placed at the local minimum between these modes. Candidate thresholds were then checked against stratified visual reference samples and representative shoreline transects. If a candidate value produced systematic commission errors over dark desert surfaces, wet sand, shadowed terrain, or omission errors over shallow water and mixed shoreline pixels, it was adjusted within the same histogram-defined separation interval. The term “global threshold” therefore refers to the final index-specific threshold selected after this calibration and validation process; it does not mean that a new unconstrained threshold was recalculated independently for every image or quarter. Once selected, the same final threshold and decision rule were applied consistently to all Landsat 5 TM and Landsat 8 OLI quarterly composites to preserve temporal comparability across the multi-decadal record. For indices in which larger values indicate water, pixels equal to or greater than the selected threshold were classified as water; for TCW, pixels greater than the selected wetness threshold were classified as water. Sensor comparability was supported by using Landsat Collection 2 Level 2 surface reflectance products, equivalent spectral bands, a common 30 m grid, identical masking logic, and quarterly median compositing. Mixed and shallow shoreline pixels were not treated as a separate class because the mapping objective was binary water/non-water delineation. Instead, these ambiguous pixels were handled conservatively through threshold calibration, visual reference checks, shoreline transects, and quarterly compositing, and they are explicitly recognized as a principal source of shoreline uncertainty when interpreting the long-term water-area estimates. The goal of this step was not only to extract water but also to determine which index is most robust for long-term monitoring of surface-water dynamics in Lake Nasser and the Tushka Lakes.

$$NDWI={\displaystyle \frac{({\mathrm{G}}_{\mathrm{b}3}-{\mathrm{N}\mathrm{I}\mathrm{R}}_{\mathrm{b}5})}{({\mathrm{G}}_{\mathrm{b}3}+{\mathrm{N}\mathrm{I}\mathrm{R}}_{\mathrm{b}5})}}$$

(1)

$$NDX={\displaystyle \frac{({\mathrm{R}}_{\mathrm{b}4}-{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}_{\mathrm{b}6})}{({\mathrm{R}}_{\mathrm{b}4}+{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}_{\mathrm{b}6})}}$$

(2)

$$NWI={\displaystyle \frac{({\mathrm{B}}_{\mathrm{b}2}-{({\mathrm{N}\mathrm{I}\mathrm{R}}_{\mathrm{b}5}+\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}_{\mathrm{b}6}+{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}_{\mathrm{b}7}))}{({\mathrm{B}}_{\mathrm{b}2}+{({\mathrm{N}\mathrm{I}\mathrm{R}}_{\mathrm{b}5}+\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}_{\mathrm{b}6}+{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}_{\mathrm{b}7}))}}$$

(3)

$$WRI={\displaystyle \frac{(G_{\mathrm{b}3}+{\mathrm{R}}_{\mathrm{b}4})}{({NIR}_{\mathrm{b}5}+{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}_{\mathrm{b}6})}}$$

(4)

$${\mathrm{A}\mathrm{W}\mathrm{E}\mathrm{I}}_{\mathrm{n}\mathrm{s}\mathrm{h}}=4\times \left(G_{\mathrm{b}3}-{SWIR1}_{\mathrm{b}6}\right)-0.25 \times {NIR}_{\mathrm{b}5}+2.75\times {SWIR2}_{\mathrm{b}7}$$

(5)

$$\begin{gathered} TCW={0.0315\times B}_{\mathrm{b}2}+0.2021\times G_{\mathrm{b}3}+0.3102\times R_{\mathrm{b}4}+0.1595\times {NIR}_{\mathrm{b}5} \\ -0.6806\times {SWIR1}_{\mathrm{b}6}-0.6109\times {SWIR2}_{\mathrm{b}7} \end{gathered}$$

(6)

$$EWI={\displaystyle \frac{(G_{\mathrm{b}3}-{\mathrm{N}\mathrm{I}\mathrm{R}}_{\mathrm{b}5}-{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}_{\mathrm{b}6})}{(G_{\mathrm{b}3}+{\mathrm{N}\mathrm{I}\mathrm{R}}_{\mathrm{b}5}+{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}_{\mathrm{b}6})}}$$

(7)

4.3. Accuracy Assessment

To evaluate the suitability of each Landsat water index, classification outputs were assessed against reference data derived from visual interpretation of Landsat false-color composites and high-resolution imagery available in Google Earth. Accuracy assessment was performed on the seven index-based water masks using an error matrix approach [49]. A stratified random sample of 1000 reference points was generated to represent the two target classes, water and non-water, while also ensuring coverage of the main temporal, sensor-specific, seasonal, and shoreline conditions encountered in the Lake Nasser–Tushka system. The validation samples were distributed across representative Landsat 5 TM and Landsat 8 OLI composites to include both the historical TM period and the more recent OLI period. Temporal coverage included early, middle, and recent parts of the 1987–2026 record, while seasonal coverage included both high-water and low-water quarters where valid cloud-free or cloud-minimized composites were available. Spatially, the reference points were allocated across persistent open-water areas of Lake Nasser, intermittently inundated Tushka basins, stable dry desert surfaces, wet or recently exposed sediments, and mixed shoreline zones. Attention was given to shoreline buffers, shallow margins, narrow embayments, and wet-sand transition zones because these areas are the main sources of classification uncertainty in hyper-arid reservoir environments. For each point, the mapped class from the index-based classification was compared with the manually assigned reference class, and the resulting confusion matrix was used to compute producer’s accuracy, user’s accuracy, overall accuracy, and Cohen’s kappa coefficient [50]. These metrics provide the quantitative basis for comparing index performance and identifying the most reliable method for subsequent long-term application within the Landsat/GEE workflow.

$$\mathrm{PA} (\%)={\displaystyle \frac{Xt}{Xb}}\times 100$$

(8)

$$\mathrm{UA} (\%)={\displaystyle \frac{Xt}{Xa}}\times 100$$

(9)

$$\mathrm{Overall} \mathrm{accuracy} (\mathrm{OA})={\displaystyle \frac{1}{N}}{\displaystyle {\sum }_{p=1}^c}ni$$

(10)

$$\mathrm{Kappa} \mathrm{coefficient} (\mathrm{Ka})={\displaystyle \frac{N{\displaystyle {\sum }_{p=1}^c}Xt-{\displaystyle {\sum }_{p=1}^c}(Xa . Xb)}{N^2-{\displaystyle {\sum }_{p=1}^c}(Xa . Xb)}}$$

(11)

where N stands for the total number of pixels, c for the number of classes, x_t for the total number of pixels in row “p” and column “p,” x_a for the total number of samples in row “p,” and x_b for the total number of samples in column “p” in the error matrix.

4.4. Comparative Evaluation

The evaluated methods were compared using two complementary criteria: (1) mapped surface-water area and its absolute and percentage difference among indices, and (2) classification accuracy statistics derived from the error matrix. This comparative step was designed to identify the index that best balances spatial realism, quantitative agreement, and transferability across the multi-decadal Landsat archive. Indices showing minimal area deviation, together with high overall accuracy and strong kappa agreement, were considered more suitable for long-term monitoring. On this basis, the index with the most stable and interpretable performance in the Lake Nasser–Tushka environment was selected for full time-series reconstruction and persistence mapping.

4.5. Trend Analysis (Mann–Kendall and Sen’s Slope)

Monotonic trends in the NDWI-derived surface-water record were evaluated using the non-parametric Mann–Kendall (MK) test and Sen’s slope estimator [51,52]. The datasets analyzed were the mapped water-area time series for three spatial units: Lake Nasser, the Tushka Lakes, and the combined Lake Nasser–Tushka surface-water system. The input values were derived from the binary NDWI water masks produced from the Landsat composites. For 1987–1997, annual observations were used due to the lower temporal density of usable scenes in the early archive, whereas for 1998–2026, quarterly observations were used to retain seasonal information. For trend testing, quarterly values were also aggregated to an annual mean surface-water area, where a single long-term monotonic trend was required, thereby reducing the influence of intra-annual seasonality. The MK test was applied to determine the direction and statistical significance of monotonic trends without assuming normality, and Sen’s slope was used to estimate the median rate of change in the mapped area through time. Statistical significance was evaluated at α = 0.05. Results were interpreted separately for Lake Nasser, the Tushka Lakes, and the total water area because these units differ strongly in hydrologic behavior and temporal variability.

4.6. Seepage Estimation from Lake Nasser to the Nubian Sandstone Aquifer (Darcy’s Law)

The workflow also incorporates a supporting hydrogeologic component to interpret possible lake–aquifer interaction. Seepage losses from Lake Nasser to the Nubian Sandstone Aquifer were estimated for the western shoreline using a hydraulic-gradient approach based on Darcy’s law. The analysis used (i) daily lake-stage observations for 1965–2014, which were aggregated to annual mean lake levels; (ii) monthly groundwater heads from 44 observation wells distributed across five sectors (Garf Hussein, Tushka, Abu Simbel, Adindan, and Argeen), which were aggregated to representative annual heads for each sector; and (iii) sector-specific hydrostratigraphic information, including lithologic logs, aquifer thickness, and hydraulic-conductivity estimates. For each sector and year, seepage discharge was calculated as Q = KIA, where Q is seepage discharge, K is hydraulic conductivity, I is the hydraulic gradient between the lake and groundwater observation point, and A is the effective seepage cross-sectional area. The hydraulic gradient was estimated as I = Δh/L, where Δh is the difference between the annual mean lake level and the representative groundwater head, and L is the approximate horizontal flow-path distance from the lake margin to the observation well or well group. The effective seepage area was approximated as A = B × W, where B is the saturated aquifer thickness or hydraulically active seepage thickness and W is the representative shoreline length assigned to each sector. The resulting values provide sector-scale approximations rather than direct flux measurements, calibrated groundwater model outputs, or closed-water-budget terms.

The Darcy-based estimates require several simplifying assumptions. First, flow was treated as predominantly lateral from the lake toward the adjacent Nubian Sandstone Aquifer along representative shoreline-to-well flow paths. Second, hydraulic conductivity within each sector was assumed to be representative of the permeable units intersected by the observation wells, although local heterogeneity due to shale and clay interbeds is expected. Third, the effective seepage area was estimated from the sectoral shoreline length and the saturated aquifer thickness, rather than measured directly. Fourth, hydraulic gradients were derived from contemporaneous lake-stage and representative groundwater-head differences normalized by assumed flow-path length. Finally, the estimates were interpreted as potential seepage under the measured hydraulic conditions, not as calibrated groundwater model outputs. This approach treats seepage primarily as lateral flow through permeable shoreline sediments driven by stage–head differences and provides a complementary hydrogeologic context for interpreting long-term surface-water persistence and shoreline behavior derived from the Landsat analysis.

The seepage component is therefore included as complementary hydrogeologic context for interpreting possible surface–subsurface interaction, rather than as direct quantitative confirmation of lake–aquifer exchange. The Darcy-based estimates are sector-scale, assumption-dependent approximations and should not be interpreted as direct measurements, calibrated groundwater-model outputs, or closed water-budget terms. Because the seepage dataset ends in 2014 and only partially overlaps the Landsat-derived surface-water record, comparisons between estimated seepage and mapped water persistence are qualitative and hypothesis-generating. Establishing quantitative lake–aquifer exchange would require additional field measurements, spatial correlation analysis, water-budget closure, tracer evidence, and calibrated groundwater-flow modeling.

5. Results

5.1. Remote Sensing Results (Landsat/GEE)

5.1.1. Statistical Evaluation of Water-Extraction Indices

The first set of results addresses the evaluation of Landsat-derived water indices within the Landsat/GEE workflow. Index performance was assessed using mapped surface-water area (km²), absolute and percentage area difference relative to the NDWI classification, and classification metrics derived from the error matrix (overall accuracy and Cohen’s kappa). As summarized in

Table 3. Statistical comparison of the evaluated water-extraction indices.

Index Number	Water Extraction Techniques	Water Area (km2)	Area Difference Based on the NDWI Method (km2)	% Difference Based on the NDWI Method	Overall Accuracy %	Kappa Index
1	NDWI	8935.7	0	0	93.6	0.898
2	EWI	8942.2	−49.9	−0.55	92.7	0.851
3	WRI	8915.9	6.5	0.073	90.9	0.831
4	NDX	8983.4	−41.2	−0.46	90.1	0.829
5	AWELnsh	8992.1	26.3	0.295	89.3	0.799
6	TCW	8808.3	133.9	1.52	88.4	0.794
7	NWI	8793.7	148.5	1.69	83.9	0.773

and Figure 5, all tested indices could separate water from non-water across most of the study scene, but their performance differed along shallow shorelines and spectrally complex land–water boundaries. These differences are especially important in the Lake Nasser–Tushka system, where mixed pixels, dark desert surfaces, and variable inundation create a demanding test environment for long-term water mapping.

Figure 5. Comparison of water-extraction models based on the percent area difference relative to the reference classification (NDWI) (red line). For each spectral index, the plotted value represents. Negative values indicate that an index mapped less water than NDWI, while positive values indicate over-mapping relative to NDWI. The figure summarizes the relative agreement among indices and complements the accuracy statistics reported in Table 4.

Figure 5. Comparison of water-extraction models based on the percent area difference relative to the reference classification (NDWI) (red line). For each spectral index, the plotted value represents. Negative values indicate that an index mapped less water than NDWI, while positive values indicate over-mapping relative to NDWI. The figure summarizes the relative agreement among indices and complements the accuracy statistics reported in Table 4.

Table 4. Surface area changes with time for Tushka lakes, Lake Nasser, and the total water surface.

Date (m/y)	Tushka Lakes (km2)	Nasser Lake (km2)	Total Water (km2)	Date (m/y)	Tushka Lakes (km2)	Nasser Lake (km2)	Total Water (km2)	Date (m/y)	Tushka Lakes (km2)	Nasser Lake (km2)	Total Water (km2)
12/1987	0	2631	2631	10-12/2005	1016.9	4286.1	5303	04-06/2016	155.6	4337.8	4493.4
12/1988	0	2411	2411	01-03/2006	975.2	3877.8	4853	07-09/2016	140.8	4398.6	4539.4
12/1989	0	3544	3544	04-06/2006	932.2	3630.8	4563	10-12/2016	169.1	4814.7	4983.8
12/1990	0	3579	3579	07-09/2006	861	4895	5756	01-03/2017	158.5	4714.1	4872.6
12/1991	0	3485	3485	10-12/2006	886.6	4775.4	5662	04-06/2017	129.4	4360.3	4489.7
12/1992	0	3656	3656	01-03/2007	861.5	4568.5	5430	07-09/2017	120.6	4323.6	4444.2
12/1993	0	3819	3819	04-06/2007	800	4466	5266	10-12/2017	157.7	5014.7	5172.4
12/1994	0	4405	4405	07-09/2007	769.9	4580.1	5350	01-03/2018	153.5	5025.6	5179.1
12/1995	0	4779	4779	10-12/2007	750	4750	5500	04-06/2018	129.7	4730.4	4860.1
12/1996	0	4772	4772	01-03/2008	754.2	4907.8	5662	07-09/2018	132.4	4834	4966.4
12/1997	0	5076.2	5076.2	04-06/2008	730.6	4999.4	5730	10-12/2018	153.4	5390.9	5544.3
01-03/1998	0	4920.8	4920.8	07-09/2008	700	5173	5873	01-03/2019	150.5	5283.7	5434.2
04-06/1998	120	4995.7	5115.7	10-12/2008	697.5	4802.5	5500	04-06/2019	126.4	4992.5	5118.9
07-09/1998	220	5280.2	5500.2	01-03/2009	630	4527	5157	07-09/2019	123.9	5010.9	5134.8
10-12/1998	442.2	5140.3	5582.5	04-06/2009	596.5	4480.5	5077	10-12/2019	370	5778.1	6148.1
01-03/1999	680.3	5330.5	6010.8	07-09/2009	571.3	4528.7	5100	01-03/2020	991.4	5383	6374.4
04-06/1999	846.4	5430.6	6277	10-12/2009	574	4550	5124	04-06/2020	955.8	5003.8	5959.6
07-09/1999	815.2	5770.3	6585.5	01-03/2010	553.3	4349.7	4903	07-09/2020	1015	5169.4	6184.4
10-12/1999	1111.5	6060.7	7172.2	04-06/2010	484.6	3887.4	4372	10-12/2020	1616.5	5631.9	7248.4
01-03/2000	1276.2	5766.8	7043	07-09/2010	467.6	4232.4	4700	01-03/2021	1770.3	5486.8	7257.1
04-06/2000	1369.3	5052.7	6422	10-12/2010	491.8	4704.2	5196	04-06/2021	1716.1	5292.5	7008.6
07-09/2000	1324.3	4827.6	6151.9	01-03/2011	484.7	4634.3	5119	07-09/2021	1800	5047.3	6847.3
10-12/2000	1421.9	5585.1	7007	04-06/2011	430	4410	4840	10-12/2021	2468.7	5532.9	8001.6
01-03/2001	1720.1	5405.5	7125.6	07-09/2011	400.2	4189.8	4590	01-03/2022	2486.8	5517.8	8004.6
04-06/2001	1700.3	4714.7	6415	10-12/2011	380	4308	4688	04-06/2022	2378.1	5289.7	7667.8
07-09/2001	1730.5	4559.5	6290	01-03/2012	374.7	4375.3	4750	07-09/2022	2435.6	5121.9	7557.5
10-12/2001	1741.7	5516.3	7258	04-06/2012	322	4528	4850	10-12/2022	2920.1	5556.4	8476.5
01-03/2002	1738.6	5375.4	7114	07-09/2012	314.8	4585.2	4900	01-03/2023	3282.4	5458.1	8740.5
04-06/2002	1612.3	4740.7	6353	10-12/2012	300	4200	4500	04-06/2023	3220.9	5340.5	8561.4
07-09/2002	1566.3	4538.7	6105	01-03/2013	269.1	4218	4487.1	07-09/2023	3153.3	5034.1	8187.4
10-12/2002	1600	4520	6120	04-06/2013	276.8	4216.1	4492.9	10-12/2023	3100.4	5275.3	8375.7
01-03/2003	1500	4500	6000	07-09/2013	255.1	4274.3	4529.4	01-03/2024	3000.5	5514.7	8515.2
04-06/2003	1400	4450	5850	10-12/2013	260.5	4747.7	5008.2	04-06/2024	2941.7	5092.2	8033.9
07-09/2003	1359.1	4151.9	5511	01-03/2014	246.1	4717.7	4963.8	07-09/2024	2915.2	4975.9	7891.1
10-12/2003	1373.1	4126.9	5500	04-06/2014	203.6	4391.9	4595.5	10-12/2024	2862.6	5611.3	8473.9
01-03/2004	1328.3	4271.7	5600	07-09/2014	198.4	4415.3	4613.7	01-03/2025	3288.4	5390.2	8678.6
04-06/2004	1263.9	4394.1	5658	10-12/2014	235.4	5346.6	5582	04-06/2025	3386.4	5060.5	8446.9
07-09/2004	1198.4	4211.6	5410	01-03/2015	231.9	5319	5550.9	07-09/2025	3303.8	4991.1	8294.9
10-12/2004	1169.6	4658.4	5828	04-06/2015	207.9	4988.2	5196.1	10-12/2025	3273.1	5471.4	8744.5
01-03/2005	1145.3	4553.7	5699	07-09/2015	192.9	4799.3	4992.2	01-03/2026	3288.4	5634.6	8923
04-06/2005	1038.1	4002.9	5041	10-12/2015	185.1	4771.4	4956.5
07-09/2005	1003.7	3736.3	4740	01-03/2016	174.3	4381	4555.3

Table 4. Surface area changes with time for Tushka lakes, Lake Nasser, and the total water surface.

Date (m/y)	Tushka Lakes (km2)	Nasser Lake (km2)	Total Water (km2)	Date (m/y)	Tushka Lakes (km2)	Nasser Lake (km2)	Total Water (km2)	Date (m/y)	Tushka Lakes (km2)	Nasser Lake (km2)	Total Water (km2)
12/1987	0	2631	2631	10-12/2005	1016.9	4286.1	5303	04-06/2016	155.6	4337.8	4493.4
12/1988	0	2411	2411	01-03/2006	975.2	3877.8	4853	07-09/2016	140.8	4398.6	4539.4
12/1989	0	3544	3544	04-06/2006	932.2	3630.8	4563	10-12/2016	169.1	4814.7	4983.8
12/1990	0	3579	3579	07-09/2006	861	4895	5756	01-03/2017	158.5	4714.1	4872.6
12/1991	0	3485	3485	10-12/2006	886.6	4775.4	5662	04-06/2017	129.4	4360.3	4489.7
12/1992	0	3656	3656	01-03/2007	861.5	4568.5	5430	07-09/2017	120.6	4323.6	4444.2
12/1993	0	3819	3819	04-06/2007	800	4466	5266	10-12/2017	157.7	5014.7	5172.4
12/1994	0	4405	4405	07-09/2007	769.9	4580.1	5350	01-03/2018	153.5	5025.6	5179.1
12/1995	0	4779	4779	10-12/2007	750	4750	5500	04-06/2018	129.7	4730.4	4860.1
12/1996	0	4772	4772	01-03/2008	754.2	4907.8	5662	07-09/2018	132.4	4834	4966.4
12/1997	0	5076.2	5076.2	04-06/2008	730.6	4999.4	5730	10-12/2018	153.4	5390.9	5544.3
01-03/1998	0	4920.8	4920.8	07-09/2008	700	5173	5873	01-03/2019	150.5	5283.7	5434.2
04-06/1998	120	4995.7	5115.7	10-12/2008	697.5	4802.5	5500	04-06/2019	126.4	4992.5	5118.9
07-09/1998	220	5280.2	5500.2	01-03/2009	630	4527	5157	07-09/2019	123.9	5010.9	5134.8
10-12/1998	442.2	5140.3	5582.5	04-06/2009	596.5	4480.5	5077	10-12/2019	370	5778.1	6148.1
01-03/1999	680.3	5330.5	6010.8	07-09/2009	571.3	4528.7	5100	01-03/2020	991.4	5383	6374.4
04-06/1999	846.4	5430.6	6277	10-12/2009	574	4550	5124	04-06/2020	955.8	5003.8	5959.6
07-09/1999	815.2	5770.3	6585.5	01-03/2010	553.3	4349.7	4903	07-09/2020	1015	5169.4	6184.4
10-12/1999	1111.5	6060.7	7172.2	04-06/2010	484.6	3887.4	4372	10-12/2020	1616.5	5631.9	7248.4
01-03/2000	1276.2	5766.8	7043	07-09/2010	467.6	4232.4	4700	01-03/2021	1770.3	5486.8	7257.1
04-06/2000	1369.3	5052.7	6422	10-12/2010	491.8	4704.2	5196	04-06/2021	1716.1	5292.5	7008.6
07-09/2000	1324.3	4827.6	6151.9	01-03/2011	484.7	4634.3	5119	07-09/2021	1800	5047.3	6847.3
10-12/2000	1421.9	5585.1	7007	04-06/2011	430	4410	4840	10-12/2021	2468.7	5532.9	8001.6
01-03/2001	1720.1	5405.5	7125.6	07-09/2011	400.2	4189.8	4590	01-03/2022	2486.8	5517.8	8004.6
04-06/2001	1700.3	4714.7	6415	10-12/2011	380	4308	4688	04-06/2022	2378.1	5289.7	7667.8
07-09/2001	1730.5	4559.5	6290	01-03/2012	374.7	4375.3	4750	07-09/2022	2435.6	5121.9	7557.5
10-12/2001	1741.7	5516.3	7258	04-06/2012	322	4528	4850	10-12/2022	2920.1	5556.4	8476.5
01-03/2002	1738.6	5375.4	7114	07-09/2012	314.8	4585.2	4900	01-03/2023	3282.4	5458.1	8740.5
04-06/2002	1612.3	4740.7	6353	10-12/2012	300	4200	4500	04-06/2023	3220.9	5340.5	8561.4
07-09/2002	1566.3	4538.7	6105	01-03/2013	269.1	4218	4487.1	07-09/2023	3153.3	5034.1	8187.4
10-12/2002	1600	4520	6120	04-06/2013	276.8	4216.1	4492.9	10-12/2023	3100.4	5275.3	8375.7
01-03/2003	1500	4500	6000	07-09/2013	255.1	4274.3	4529.4	01-03/2024	3000.5	5514.7	8515.2
04-06/2003	1400	4450	5850	10-12/2013	260.5	4747.7	5008.2	04-06/2024	2941.7	5092.2	8033.9
07-09/2003	1359.1	4151.9	5511	01-03/2014	246.1	4717.7	4963.8	07-09/2024	2915.2	4975.9	7891.1
10-12/2003	1373.1	4126.9	5500	04-06/2014	203.6	4391.9	4595.5	10-12/2024	2862.6	5611.3	8473.9
01-03/2004	1328.3	4271.7	5600	07-09/2014	198.4	4415.3	4613.7	01-03/2025	3288.4	5390.2	8678.6
04-06/2004	1263.9	4394.1	5658	10-12/2014	235.4	5346.6	5582	04-06/2025	3386.4	5060.5	8446.9
07-09/2004	1198.4	4211.6	5410	01-03/2015	231.9	5319	5550.9	07-09/2025	3303.8	4991.1	8294.9
10-12/2004	1169.6	4658.4	5828	04-06/2015	207.9	4988.2	5196.1	10-12/2025	3273.1	5471.4	8744.5
01-03/2005	1145.3	4553.7	5699	07-09/2015	192.9	4799.3	4992.2	01-03/2026	3288.4	5634.6	8923
04-06/2005	1038.1	4002.9	5041	10-12/2015	185.1	4771.4	4956.5
07-09/2005	1003.7	3736.3	4740	01-03/2016	174.3	4381	4555.3

Among the evaluated indices, NDWI, EWI, NDX, and WRI yielded the strongest overall performance, showing small area deviations and high classification agreement. NDWI produced the highest overall accuracy (93.6%) and kappa (0.898), with a zero relative area difference because it served as the reference, whereas EWI achieved 92.7% overall accuracy and 0.851 kappa. NDX and WRI also performed well, indicating that several Landsat formulations can delineate open water reliably in this environment. By contrast, TCW and NWI showed larger mapped-area deviations and weaker agreement statistics, indicating greater sensitivity to shoreline mixing and spectrally confusing non-water targets. Overall, the comparative results demonstrate that index choice has a measurable influence on the mapped extent of water and therefore directly affects the reliability of long-term surface-water reconstruction.

5.1.2. Selection of an Index for Long-Term Time-Series Mapping

A key outcome of the Landsat-based evaluation is the selection of NDWI as the primary index for long-term monitoring. This choice was based on a multi-criteria interpretation rather than a single metric. NDWI provided a particularly strong balance of high classification agreement, stable shoreline delineation, simple formulation, and practical transferability across the multi-sensor Landsat archive. In the context of a long-term monitoring workflow, these characteristics are essential because the selected index must remain interpretable, reproducible, and robust across varying hydrologic conditions and across both TM and OLI imagery. The workflow shown in Figure 6 illustrates how each spectral index was converted into a binary water mask, while Figure 7 provides the spectral basis for water–non-water discrimination in the Lake Nasser–Tushka environment.

NDWI is the most suitable index for this study because it offers the basis for monitoring long-term surface-water dynamics in a hyper-arid reservoir setting. Its performance reflects the strong contrast between water in the green and near-infrared bands, which remains effective in a landscape dominated by dry sand, bare rock, and sparse vegetation. Quarterly compositing in Google Earth Engine further improves the temporal stability of NDWI-based mapping by reducing residual cloud and atmospheric effects. The results, therefore, show that the value of the workflow lies not simply in computing multiple indices but in identifying the one most appropriate for consistent multi-decadal monitoring. On this basis, NDWI was carried forward to reconstruct quarterly surface-water dynamics in Lake Nasser and the Tushka Lakes from 1987 to 2026.

5.1.3. Time-Series Analysis of NDWI (1987–2026)

Application of NDWI to the quarterly Landsat composites produced a multi-decadal record of surface-water dynamics for Lake Nasser, the Tushka Lakes, and their combined water surface from 1987 to 2026 (Figure 8 and Table 4). Rather than simply reflecting a single regional water-level signal, the record reveals two distinct hydrologic behaviors. Lake Nasser serves as the stable core of the system: its mapped area changes over time, but large parts of the central and southern reservoir remain persistently inundated due to greater water depth, longitudinal continuity, and sustained connection to the main Nile reservoir. In contrast, the Tushka basins function as shallow peripheral storage zones, with their surface-water extent responding disproportionately to relatively small changes in lake stage. Once water spills into these low-relief depressions, large horizontal expansions can occur; during drawdown, the same shallow geometry promotes rapid contraction and partial desiccation. The NDWI time series, therefore, indicates that the spatial differences between Lake Nasser and Tushka are controlled not only by total water availability but also by basin morphology, connectivity thresholds, and shoreline slope (Figure 9).

The persistence and trend results refine this interpretation by separating stable inundation from episodic shoreline migration. High-persistence zones in Lake Nasser represent areas where water occurrence is controlled primarily by reservoir storage and channel depth, whereas low-persistence zones along northern margins and within the Tushka basins mark areas where small vertical changes in stage translate into large areal changes. These low-persistence zones are also the most sensitive to mixed pixels and threshold uncertainty, because the mapped shoreline often coincides with shallow water, wet sand, and intermittently exposed sediment. Consequently, the long-term record should be interpreted as a robust indicator of seasonal-to-interannual water-area dynamics rather than as an exact instantaneous shoreline inventory. The Mann–Kendall and Sen’s slope analyses add value by distinguishing broad monotonic tendencies from short-term fluctuations, but the most important hydrologic insight is the persistent contrast between a relatively buffered main reservoir and highly responsive peripheral basins.

5.2. Hydrogeologic (Piezometer) and Seepage Results

Piezometer Data Analysis

The hydrogeologic results provide a complementary, but necessarily more uncertain, interpretation of possible lake–aquifer exchange along the western shoreline. Darcy-based seepage estimates for Garf Hussein, Tushka, Abu Simbel, Adindan, and Argeen indicate spatial heterogeneity rather than a uniform leakage pattern. Garf Hussein shows the largest estimated seepage magnitudes, whereas Abu Simbel shows the lowest, with intermediate values in the remaining sectors. This pattern is consistent with spatial differences in hydraulic gradient, shoreline geometry, aquifer thickness, and the distribution of permeable sandstone units relative to shale and clay interbeds. However, these values should be read as sector-scale potential flux estimates, not direct measurements of seepage. Their magnitude depends strongly on assumed hydraulic conductivity, representative flow-path length, effective seepage area, and the degree to which observation wells capture local head conditions near the lake shoreline.

The relation between lake stage and estimated seepage is therefore best interpreted as a hydraulic tendency rather than a calibrated causal model. Higher lake levels increase the potential head difference between the reservoir and adjacent aquifer, which can increase estimated seepage where permeable pathways are present (

Table 5. Water-level and total-seepage calculations for the western side of Lake Nasser.

Year	Total Seepage × 106 m3/Year	Lake Level (Average in m)	Year	Total Seepage × 106 m3/Year	Lake Level (Average in m)
1965	27.07	126.28	1990	21.07	167.3
1966	32.29	131.29	1991	22.05	166.46
1967	28.16	142.90	1992	19.36	167.72
1968	31.3	151.10	1993	24.08	170.53
1969	23.44	156.18	1994	26.1	173.63
1970	22.83	159.83	1995	26.73	175.21
1971	28.67	163.63	1996	26.08	175.45
1972	28.05	165.17	1997	29.13	177.37
1973	30.94	162.85	1998	28.06	178.13
1974	27.79	165.75	1999	31.56	178.91
1975	32.97	170.09	2000	30.25	178.88
1976	36.68	174.83	2001	26.79	178.76
1977	35.27	174.86	2002	35.56	177.66
1978	32.7	175.35	2003	29.46	175.62
1979	30.91	175.26	2004	29	174.86
1980	25.35	174.22	2005	28.13	172.79
1981	26.8	174.13	2006	23.19	173.10
1982	25.34	172.66	2007	23.4	176.20
1983	21.93	169.01	2008	24.28	175.62
1984	24.54	169.34	2009	21.64	177.66
1985	22.65	160.90	2010	19.37	173.59
1986	22.38	161.08	2011	19.06	169.93
1987	18.74	161.66	2012	19.29	174.14
1988	15.58	168.82	2013	20.33	173.23
1989	24.77	169.79	2014	19.56	172.43

, Figure 10). At the same time, local lithologic heterogeneity, delayed groundwater response, vertical leakage between aquifer layers, uncertainty in lake-margin geometry, and uncertainty in hydraulic conductivity can weaken or obscure a simple stage–seepage relationship. The calculated annual seepage values range from 15.58 × 10⁶ to 36.68 × 10⁶ m³/year. In addition, the seepage record ends in 2014, whereas the Landsat-based water-area record extends to 2026. Therefore, only the 1987–2014 overlap period supports qualitative comparison between surface-water extent and seepage estimates. When considered alongside the NDWI persistence maps, the seepage estimates help identify sectors where lake–aquifer exchange is more likely, but they do not independently establish the magnitude of recharge, post-2014 seepage behavior, or detailed flow pathways.

6. Discussion

6.1. Implications of the Surface-Water Record

The discussion confirms that the main contribution of this study lies in the development of a Landsat/GEE-based workflow for evaluating Landsat water indices and applying the most suitable index to monitor long-term surface-water dynamics in Lake Nasser and the Tushka Lakes. The comparative results demonstrate that water mapping in this hyper-arid reservoir environment is strongly influenced by shoreline complexity, mixed pixels, and spectrally confusing dry-land surfaces. In this setting, NDWI proved to be the most suitable index for long-term application because it maintained high agreement with the reference data while also providing stable and interpretable shoreline delineation across the multi-sensor Landsat archive. This outcome is important because the reliability of any long-term surface-water record depends first on selecting a method that remains robust across changing conditions and across decades of imagery.

From a methodological perspective, the Google Earth Engine workflow offers a practical way to transform large Landsat archives into a consistent monitoring product. By integrating cloud-based preprocessing, QA-based masking, quarterly compositing, multi-index calculation, threshold-based water extraction, and accuracy evaluation into a single analytical environment, the workflow reduces technical barriers to long-term reservoir analysis and improves reproducibility. The quarterly compositing strategy is especially useful in the Lake Nasser–Tushka system because it suppresses atmospheric noise and residual cloud contamination while preserving the seasonal-to-interannual variability needed for long-term interpretation. At the same time, quarterly aggregation can smooth very short-lived inundation events, so the mapped series should be interpreted as representative seasonal conditions rather than exact instantaneous shoreline positions.

The long-term NDWI record is most informative when interpreted in terms of system behavior rather than as a sequence of mapped areas. Lake Nasser and the Tushka Lakes respond differently because they occupy different geomorphic and hydraulic settings. The main reservoir has a relatively continuous water body and deeper basin geometry, so changes in stage are expressed mainly as shoreline adjustment around a persistent water core. The Tushka basins, by contrast, are shallow depressions connected by spillways and topographic lows; once connectivity thresholds are crossed, water can spread rapidly across broad, low-gradient surfaces, and when levels fall, those areas recede just as rapidly. This explains why the Tushka record contains sharper expansions and contractions than Lake Nasser, even though both are part of the same regional surface-water system. The persistence maps, therefore, provide hydrologic interpretation by identifying where water occurrence is structurally stable and where it is threshold-controlled and episodic.

More broadly, the findings suggest that the proposed Landsat/GEE workflow can serve as a transferable approach for monitoring other large desert reservoirs and inland basins, provided that local index evaluation is performed before long-term implementation. Because index behavior is landscape-dependent, this study reinforces the conclusion that no single spectral water index should be assumed optimal without site-specific testing. In that sense, the broader value of this work lies not only in the resulting Lake Nasser–Tushka record itself, but also in the demonstration that a combination of Landsat water-index evaluation, cloud-based processing, and multi-decadal surface-water reconstruction can provide a reliable basis for environmental monitoring in hyper-arid settings.

6.2. Implications for Potential Groundwater Recharge

Although the primary emphasis of the study is Landsat water-index evaluation and long-term surface-water dynamics, the hydrogeologic observations add important supporting context for interpreting the surface-water record. Groundwater levels in the Nubian Sandstone Aquifer adjacent to Lake Nasser and the Tushka Lakes can be discussed in relation to the NDWI-derived maps of surface-water extent and persistence, but these links should be interpreted cautiously. NDWI is an optical indicator of surface-water presence and does not directly measure recharge, seepage flux, or hydraulic connectivity. Accordingly, the discussion of groundwater response in this study is best understood as an interpretation supported by independent piezometric and seepage information, rather than as a direct output of the remote-sensing analysis itself.

When interpreted alongside groundwater observations, the persistence maps indicate areas where sustained surface-water presence may create conditions favorable to lake–aquifer interaction, but this interpretation must be treated with caution. Longer inundation duration can maintain hydraulic gradients for longer periods, yet persistence alone does not demonstrate seepage, recharge, or a causal hydraulic connection. Darcy-based calculations are highly sensitive to hydraulic conductivity, which can vary by orders of magnitude in interbedded sandstone, shale, and clay sequences. They are also affected by assumptions about saturated thickness, effective seepage area, shoreline length, flow direction, and the distance over which hydraulic gradients are calculated. Because these parameters are represented at the sector scale, the estimates cannot resolve local preferential pathways, vertical leakage, delayed aquifer response, or spatially variable bank-storage effects. Thus, any spatial coincidence between persistent surface water and estimated seepage should be treated only as evidence of possible association, not as proof of a specific recharge rate, seepage pathway, or causal mechanism.

The Tushka Lakes are especially important for interpreting uncertainty because they combine shallow bathymetry, intermittent hydraulic connection, and spectrally complex wet shorelines. Their rapid expansion during high-water periods does not necessarily imply equivalent groundwater recharge, because part of the mapped increase may represent shallow surface spreading over low-permeability or fine-grained deposits. Conversely, rapid contraction may reflect evaporation, disconnection from the main reservoir, infiltration, or some combination of these processes. The NDWI record identifies the timing and spatial footprint of these changes, but it cannot separate the relative contributions of evaporation, infiltration, seepage, bank storage, and reservoir operation. This distinction strengthens the interpretation of Figure 11: the apparent agreement among water-level, surface-water-area, and seepage estimates should be viewed only as qualitative covariation. Without formal spatial correlation, calibrated groundwater-flow modeling, water-budget closure, tracer evidence, or independent field validation, the magnitude, direction, and mechanism of lake–aquifer exchange remain uncertain.

Overall, the integrated results show that this study’s strongest contribution is not simply documenting that water extent changed over time but explaining how different parts of the Lake Nasser–Tushka system respond differently to the same broad hydrologic forcing. NDWI-based mapping provides a consistent long-term record of surface-water persistence and variability, while the seepage analysis offers a first-order hydrogeologic context for interpreting possible surface–subsurface exchange. The seepage estimates should therefore be used as interpretive indicators rather than definitive water-budget terms. Future work should reduce uncertainty by combining the Landsat-derived persistence record with field-based hydraulic-head monitoring, aquifer tests, geophysical characterization of shoreline sediments, tracer studies, and calibrated groundwater-flow models. Such integration would allow the spatial patterns identified here to be translated into more quantitative estimates of recharge, seepage pathways, and storage impacts.

6.3. Environmental Problems Associated with High Water Levels in the River Nile

High water levels in the River Nile and its associated systems (e.g., irrigation canals, drains, and large reservoirs such as Lake Nasser) could result in various environmental issues. The main routes and typical examples can be summarized as follows: (1) Floodwater inundation of low-lying floodplains can result in crop damage, delay planting/harvesting, and deposit fine sediments clogging soil pores or covering seedlings. Example: inundation of agricultural fields and adjacent settlements in the Nile Valley and Delta during high-stage events, with subsequent sediment deposition and cleanup needs [53]. (2) Sediment derived from floods can settle in canals, reducing their conveyance capacity and increasing maintenance needs. Example: silt and debris buildup in secondary canals following flood events can impair irrigation delivery and drainage efficiency [53]. (3) Inundation of wastewater facilities, waste dumps, or contaminated soils can mobilize nutrients, pathogens, and metals into surface waters. Example: floodwater transport of heavy-metal-contaminated materials from wastewater disposal areas toward canals and the Nile, thereby increasing downstream environmental and health risks [54]. (4) Enhanced flow velocity and shear stress may favor erosion of alluvial banks, resulting in bank retreat, loss of vegetation, and localized damage of embankments and riverside infrastructure. Example: erosion hotspots at bends where cohesive bank materials are undercut during sustained high stage [55]. (5) Rapid water-level rise inundates exposed shores; repeated wetting–drying and wave action can destabilize sediments and increase turbidity through resuspension. Example: erosion and sediment remobilization along the Lake Nasser shorelines during rapid stage changes, thereby raising nearshore turbidity [56]. (6) A sustained high river stage can elevate adjacent groundwater tables. Waterlogging reduces aeration in the root zone and can depress yields, while evaporation in arid/semi-arid climates concentrates salts at the surface, worsening soil salinity. (7) Floods transport more suspended sediments, nutrients, and organic matter. Decomposition of flooded biomass and high organic loading can increase biochemical oxygen demand and cause localized low-oxygen conditions (i.e., severe waterlogging and salinization reported in parts of the northeastern Nile Delta, associated with shallow saline groundwater and constraints on irrigation water quality [57,58]), particularly in weakly mixed areas. Example: Surveys of Lake Nasser show seasonal contrasts (pre- and post-flood) in nutrient and chlorophyll dynamics indicative of flood-driven material inputs and changes in productivity [56,59,60]. (8) Standing water in depressions, canals and drains can expand mosquito breeding habitat and increase exposure to microbial contamination where sanitation is compromised. Example: stagnant pools that persist after inundation events may increase vector abundance and increase waterborne disease risk [53]. Overall, these examples illustrate that high Nile water levels are not only a flooding hazard; they can also drive geomorphic change, soil and groundwater degradation, and impacts on water quality and health across the connected river–canal–reservoir system. In the Lake Nasser–Tushka setting, these processes are particularly relevant along shallow shorelines and intermittently inundated zones, where stage changes can rapidly modify sediment and water quality conditions.

7. Future Trends and Study Limitations

Future work should build directly on this study’s three central components: the Landsat/GEE workflow, the evaluation of Landsat water indices, and the monitoring of long-term surface-water dynamics in Lake Nasser and the Tushka Lakes. First, the workflow can be expanded through multi-sensor integration by combining Landsat with higher-temporal-frequency optical observations, such as Sentinel-series imagery, to improve detection of short-lived inundation events and rapid shoreline shifts, especially in the Tushka basins. Second, future studies may extend the index-evaluation component by testing adaptive thresholds, machine learning classifiers, or sensor-specific calibration strategies to determine whether water extraction can be made more robust under highly variable illumination, turbidity, and mixed-pixel conditions. Third, the long-term monitoring component can be strengthened through near-real-time implementation in cloud platforms such as Google Earth Engine, allowing routine updates of quarterly or monthly water extent products for water resource assessment and operational reservoir management. In addition, coupling the mapped water record with elevation–area–volume relationships would allow future work to move beyond surface-water area toward more complete estimates of storage variability.

Several limitations should be considered when interpreting the results. First, the mapped water masks are derived from 30 m Landsat imagery, so shoreline position and area estimates remain sensitive to mixed pixels, narrow embayments, and irregular coastlines. Second, the use of quarterly composites improves temporal consistency but can smooth short-duration inundation peaks or abrupt contractions, particularly in episodically connected parts of the Tushka system. Third, the comparative evaluation of indices relies on visually interpreted reference data rather than fully contemporaneous field-based shoreline observations, introducing unavoidable uncertainty into the accuracy assessment. Fourth, the monotonic trend analysis captures broad tendencies in the time series but may be influenced by temporal aggregation, serial correlation, seasonality, and abrupt operational changes. Finally, the seepage analysis contains important methodological and temporal limitations. Darcy-based estimates are highly sensitive to hydraulic conductivity, hydraulic gradient, seepage thickness, shoreline length, and the assumed flow-path distance; these parameters were represented at the sector scale rather than resolved locally. In addition, seepage calculations end in 2014 because the lake-stage and groundwater datasets used for this component do not extend to 2026, whereas Landsat surface-water monitoring continues through 2026. As a result, the 1987–2014 overlap period supports only qualitative comparison between seepage and mapped water extent, while post-2014 Landsat observations should be interpreted only as surface-water dynamics and not as evidence of quantified seepage. This study’s primary contribution, therefore, remains the remote-sensing analysis; the seepage component is complementary, uncertainty-bearing, and hypothesis-generating. Despite these limitations, this study provides a strong and reproducible basis for future refinement of index selection and long-term surface-water monitoring in large desert reservoirs.

8. Conclusions

This study developed a Landsat/GEE-based workflow to evaluate water indices and monitor long-term surface-water dynamics in Lake Nasser and the Tushka Lakes, Egypt, using Landsat 5 TM and Landsat 8 OLI surface reflectance imagery from 1987 to 2026. By integrating Google Earth Engine processing, multi-index comparison, threshold-based water extraction, and temporal analysis, the workflow provided a consistent basis for assessing which Landsat-derived index is most suitable for water mapping in a hyper-arid reservoir environment characterized by shallow shorelines, mixed pixels, and spectrally complex dry-land backgrounds. The analysis evaluated seven spectral water indices and used 1000 stratified reference samples to quantify classification performance.

The comparative evaluation showed that several indices performed well, but NDWI provided the best balance of classification accuracy, mapped-area consistency, interpretability, and transferability across the multi-sensor Landsat archive. NDWI achieved the highest classification statistics in the evaluation scene, with an overall accuracy of 93.6% and a kappa coefficient of 0.898. NDWI mapped 8935.7 km² of water in the comparison scene and served as the reference for mapped-area consistency; NDX and WRI differed from NDWI by only −0.46% and 0.073%, respectively, whereas TCW and NWI showed larger deviations of 1.52% and 1.69%. NDWI was therefore selected as the most robust option for long-term use because it provided reliable shoreline delineation and consistent performance throughout the study period. These findings highlight that the main value of the workflow lies not only in computing multiple water indices, but in identifying the one most suitable for reproducible multi-decadal monitoring.

Application of NDWI to quarterly Landsat composites from 1987 to 2026 generated a multi-decadal record of surface-water behavior that clearly distinguishes the relatively stable core reservoir of Lake Nasser from the more episodic and highly dynamic Tushka Lakes. The mapped total water area increased from 2631 km² in 1987 to 8923 km² in early 2026. Lake Nasser ranged from a minimum mapped area of 2411 km² in 1988 to a maximum of 6060.7 km² in late 1999, whereas the Tushka Lakes expanded from no mapped water before 1998 to more than 3300 km² during 2025. The results show broad zones of persistent water in the central and southern sectors of Lake Nasser, in contrast to strong seasonal and interannual variability in the shallow Tushka basins. Persistence mapping and time-series analysis further demonstrate that the most dynamic water–land transitions occur along intermittently inundated shorelines, emphasizing the importance of using a carefully evaluated index for long-term monitoring in such environments.

The hydrogeologic analysis adds complementary interpretive value by placing mapped persistence and shoreline behavior in the context of possible lake–aquifer seepage along the western margin of Lake Nasser. However, this component should be interpreted as an uncertainty-bearing, sector-scale hydrogeologic interpretation rather than as direct quantitative confirmation of lake–aquifer exchange. The Darcy-based estimates are assumption-dependent and do not represent direct seepage measurements, calibrated groundwater model outputs, or closed water budget terms. The present study therefore identifies only qualitative co-variation between persistent surface water, lake stage, groundwater observations, and estimated seepage during the period of data overlap, but it does not demonstrate causality, quantify post-2014 seepage behavior, or resolve detailed flow pathways. Establishing a quantitative lake–aquifer exchange relationship would require additional spatial correlation analysis, calibrated groundwater modeling, water-budget closure, tracer or field-based validation, and higher-resolution hydraulic monitoring. Taken together, the results demonstrate that Google Earth Engine processing, Landsat water-index evaluation, and multi-decadal surface-water reconstruction provide a strong and reproducible foundation for monitoring large desert reservoirs, while the seepage component should be viewed as complementary, hypothesis-generating hydrogeologic context to guide future validation. This study therefore contributes both a site-specific understanding of the Lake Nasser–Tushka system and a transferable workflow that can support future remote-sensing and hydrogeologic investigations in other arid and semi-arid surface-water systems.