Ecological Quality Assessment of the Semipalatinsk Nuclear Test Site Using RSEI Based on Sentinel-2 Data: A Comparison of 2020 and 2024 Summer Conditions

2.1 Study area

The Semipalatinsk Nuclear Test Site (SNTS) is located in northeastern Kazakhstan, centered approximately at 50.4 °N, 78.0 °E (Figure 1). The site covers an area of approximately 18,500 km² spanning the territories of the former Semipalatinsk (now East Kazakhstan), Pavlodar, and Karaganda oblasts. The terrain is predominantly flat to gently undulating steppe, with elevations ranging from approximately 150 to 500 m above sea level. The Irtysh River forms the northern boundary of the test site, while the Chagan River flows through the central test area. The city of Kurchatov (formerly Semipalatinsk-21), the former administrative center of the SNTS, is located on the northern site boundary [7].

The regional climate is strongly continental, characterized by hot and dry summers (mean temperatures of 25–35 ℃) and cold winters. Annual precipitation ranges from 200 to 300 mm, with most rainfall occurring in spring [8].

Natural vegetation is dominated by dry steppe communities, including grasses such as Stipa and Festuca, and drought-resistant shrubs of the genus Artemisia. Riparian vegetation, including willow and poplar, occurs along river valleys.

Large parts of the SNTS are characterized by sparsely vegetated or disturbed surfaces, particularly within former test zones. These environmental conditions are influenced by a combination of historical disturbances and natural factors such as aridity, soil properties, and climatic variability [7].

Figure 1. Geographic location of the study area within Kazakhstan, including the Semipalatinsk Nuclear Test Site (SNTS)

2.2 Data sources

Satellite data from Sentinel-2 and Landsat 8 were used to derive the input indicators for the RSEI model. Sentinel-2 Level-2A surface reflectance imagery from the COPERNICUS/S2_SR_HARMONIZED collection was used for the calculation of NDVI, Wetness, and NDBSI. Data were selected for the summer growing seasons (1 June to 31 August) of 2020 and 2024 to ensure temporal consistency.

To ensure data quality, scenes were filtered using a cloud cover threshold of less than 10%. Cloud masking was performed using both the QA60 band and the Scene Classification Layer (SCL), allowing the removal of clouds, cirrus, cloud shadows, and other low-quality pixels. The filtered image collections were composited using pixel-wise median aggregation to generate spatially continuous seasonal composites.

Landsat 8 Collection 2 Tier 1 Level-2 data (LANDSAT/LC08/C02/T1_L2) were used for LST derivation, as Sentinel-2 does not include thermal infrared bands. The same seasonal filtering (June–August) was applied for both 2020 and 2024. Cloud and shadow masking were performed using the QA_PIXEL band, with pixels flagged as dilated cloud (bit 1), cloud shadow (bit 3), snow (bit 4), or cloud (bit 5) excluded from analysis. The median composite of the thermal band (Band 10, ST_B10) was used to estimate LST [8, 9].

All data processing, including filtering, masking, compositing, and index calculation, was conducted within the Google Earth Engine (GEE) cloud computing platform, which enables efficient processing of large-scale geospatial datasets and is widely applied in ecological monitoring studies [2, 8].

The pixel-wise median compositing strategy, applied independently to each year's filtered collection, ensured full spatial coverage across the entire 18,500 km² study area for both 2020 and 2024, with no data gaps remaining after processing.

The main characteristics of the satellite data used in this study, including data source, processing level, spatial resolution, temporal period, cloud-cover threshold, and derived indicators, are summarized in Table 1.

Table 1. Characteristics of satellite data used in the study

2.3 Calculation of ecological indicators

The Normalized Difference Vegetation Index (NDVI) was calculated from the Sentinel-2 median composite using the standard formula:

$N D V I=\frac{B 8-B 4}{B 8+B 4}$                  (1)

where, B8 is the near-infrared band (842 nm, 10 m spatial resolution) and B4 is the red band (665 nm). NDVI values range from –1 to +1, with higher values indicating denser and more photosynthetically active vegetation. In the SNTS, NDVI reflects spatial variability in vegetation cover influenced by environmental conditions such as soil properties, moisture availability, and climatic variability [7].

Wetness was derived using the tasseled cap transformation adapted for Sentinel‑2:

$\begin{aligned} \text { Wetness }=0.1509 B 2+0.1973 B 3 & +0.3279 B 4+0.3406 B 8 -0.7112 B 11-0.4559 B 12\end{aligned}$                        (2)

where, B2, B3, B4, B8, B11, and B12 correspond to the blue, green, red, near-infrared, and shortwave infrared bands. Higher wetness values indicate increased surface and soil moisture conditions [9].

LST was derived from Landsat 8 thermal infrared data (Band 10, ST_B10) using the Collection 2 Level-2 scaling factors:

$L S T=D N \times 0.00341802+149.0-273.15$                               (3)

where, DN is the digital number of the thermal band. The resulting LST values represent surface temperature conditions during the summer composite period. Higher LST values indicate increased thermal stress and are typically associated with bare or sparsely vegetated surfaces [10].

The Normalized Difference Built-up and Soil Index (NDBSI) was calculated from Sentinel-2 data as a composite of the Soil Index (SI) and the Built-up Index (BI):

$S I=\frac{(B 11+B 4)-(B 8+B 2)}{(B 11+B 4)+(B 8+B 2)}$                                (4)

$B I=\frac{(B 11+B 3)-(B 8+B 4)}{(B 11+B 3)+(B 8+B 4)}$                            (5)

$N D B S I=\frac{S I+B I}{2}$                                (6)

where, SI is the Soil Index, BI is the Built-up Index, and NDBSI is the average of these two components. Higher NDBSI values indicate greater surface bareness and soil exposure, which are characteristic of degraded or unvegetated terrain. NDBSI is particularly sensitive to extensive bare ground and disturbed soils found in the core test areas of the SNTS.

Higher NDBSI values indicate greater surface bareness and soil exposure, characteristic of degraded or unvegetated terrain. NDBSI is particularly sensitive to the extensive bare ground and disturbed soils found in the core test areas of the SNTS [3, 7].

Before RSEI computation, all indicators were normalized to the (0, 1) range using min–max normalization:

$X_{\text {norm}}=\frac{x-x_{\min }}{x_{\max }-x_{\min }}$                  (7)

where, $X_{\text {norm}}$ is the normalized value of a given indicator, $x$ is the original pixel value of the indicator, and $x_{\text {min}}$ and $x_{\text {max}}$ represent the minimum and maximum values of the corresponding indicator, respectively.

The normalized indicators were then integrated using PCA. To ensure consistency and reduce spatial noise, all datasets were resampled to a common spatial resolution of 1000 m prior to PCA. This aggregation improves the stability of covariance estimation and enhances the robustness of the resulting components [2].

The PCA loadings of the ecological indicators used for RSEI construction are presented in Table 2. Since PC1 loads positively on degradation indicators (LST and NDBSI) and negatively on ecological quality indicators (NDVI and Wetness), the index was calculated as:

$R S E I=1-P C 1_{\text {norm }}$                    (8)

Table 2. Principal component analysis (PCA) loadings of ecological indicators used for RSEI construction

This transformation ensures that higher RSEI values correspond to better ecological quality.

Finally, the RSEI was classified into five categories using the equal interval method:

• Poor (0–0.2)
• Fair (0.2–0.4)
• Moderate (0.4–0.6)
• Good (0.6–0.8)
• Excellent (0.8–1.0)

For mapping and statistical analysis, these classes were reclassified into integer values (0–5).

The RSEI [13] assessment of the SNTS for the summer seasons of 2020 and 2024 reveals clear spatial contrasts in ecological conditions across the study area. The results indicate that areas with lower ecological quality are primarily concentrated in the central parts of the site, while relatively better conditions are observed in peripheral and riverine zones [11]. These patterns reflect the combined influence of environmental factors, including land surface properties, moisture availability, and climatic conditions [14].

Although the spatial distribution of lower RSEI values corresponds to historically disturbed areas, it is important to note that RSEI does not directly measure radioactive contamination. Instead, it captures integrated ecological responses observable at the surface, such as reduced vegetation cover, increased surface temperature, and higher soil exposure. Therefore, the observed patterns should be interpreted as indicators of environmental condition rather than direct measures of contamination [15].

The radar chart (Figure 12) illustrates the characteristic profiles of ecological indicators for different RSEI classes. The plotted values represent mean normalized values of NDVI, Wetness, LST, and NDBSI for each class. Areas classified as Poor exhibit high LST and NDBSI values combined with low NDVI and Wetness, reflecting unfavorable surface conditions. In contrast, the Good class is characterized by higher NDVI and Wetness values, indicating more stable ecological conditions. The Moderate class shows intermediate characteristics between these two extremes.

The observed spatial patterns are also influenced by regional climatic conditions. The SNTS is located in a semi-arid zone, where high summer temperatures and limited precipitation naturally result in relatively low vegetation density and moisture availability. As a result, even areas not subject to significant disturbance may exhibit moderate ecological quality. This highlights the importance of considering background environmental conditions when interpreting RSEI results [16].

The relatively higher ecological quality observed along the Irtysh River is consistent with the role of riparian zones as ecological refugia in semi-arid landscapes. Increased soil moisture and vegetation density in these areas contribute to lower surface temperatures and improved ecological indicators [17].

The spatial patterns identified in this study are broadly consistent with findings from previous remote sensing assessments of degraded and post-disturbance landscapes in Central Asia and arid regions. Similar spatial gradients of vegetation degradation have been observed in East Kazakhstan using NDVI and Landsat data, with the most pronounced deficits occurring in historically disturbed areas and relatively stable conditions along river corridors. This is consistent with the low NDVI and elevated NDBSI values observed in the central SNTS in the present study. RSEI effectively captures ecological gradients in desertification-prone arid environments, where the index is sensitive to the combined signal of sparse vegetation, dry soils, and elevated surface temperatures – conditions characteristic of the SNTS interior. The predominance of Good and Excellent classes in peripheral and riverine zones observed here is consistent with previous findings that riparian and low-disturbance areas consistently score higher in RSEI assessments across diverse landscape types [16]. Taken together, these comparisons support the validity of the RSEI approach for the SNTS and suggest that the observed ecological patterns reflect both the legacy of nuclear testing and the background constraints of the semi-arid steppe environment [18].

From a methodological perspective, the integration of Sentinel-2 and Landsat 8 data within the GEE platform proved effective for large-scale ecological assessment. The use of Sentinel-2 for vegetation, moisture, and surface condition indicators, combined with Landsat-derived LST, provides a balanced approach that leverages the strengths of both datasets [16, 19].

Several limitations should be acknowledged. First, the analysis represents seasonal conditions (summer periods of 2020 and 2024) and does not capture inter-annual variability. Second, the RSEI reflects only surface-observable ecological characteristics and does not directly quantify subsurface or radiological processes. Third, the absence of field validation introduces uncertainty in the interpretation of ecological classes.

Future research should focus on multi-temporal analysis and comparison with climatically similar reference areas to better distinguish between natural and disturbance-related ecological patterns.

The authors express their sincere gratitude to Al-Farabi Kazakh National University (KazNU) for providing scientific and technical support during the research, as well as for creating favorable conditions for conducting scientific work within the framework of the Department of Cartography and Geoinformatics.