Довгостроковий моніторинг якості поверхневих вод та потенціалу підземних вод із використанням обчислювального інтелекту, GIS-технологій та дистанційного зондування
Water scarcity and declining water quality due to population growth, urbanization, industrialization, and climate change highlight the importance of effective water management. Advances in remote sensing, cloud computing, and computational intelligence underscore the need to utilize modern technolog...
Збережено в:
| Дата: | 2026 |
|---|---|
| Автори: | , |
| Формат: | Стаття |
| Мова: | Англійська |
| Опубліковано: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2026
|
| Теми: | |
| Онлайн доступ: | https://journal.iasa.kpi.ua/article/view/358082 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | System research and information technologies |
| Завантажити файл: |  |
Репозитарії
System research and information technologies| _version_ | 1862949218429173760 |
|---|---|
| author | Klimov, Serhii Starovoit, Tetiana |
| author_facet | Klimov, Serhii Starovoit, Tetiana |
| author_sort | Klimov, Serhii |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2026-04-19T21:53:19Z |
| description | Water scarcity and declining water quality due to population growth, urbanization, industrialization, and climate change highlight the importance of effective water management. Advances in remote sensing, cloud computing, and computational intelligence underscore the need to utilize modern technologies for monitoring surface water quality. This research involves the development of hybrid intelligent models using Landsat and Sentinel-2 images and WISE data with hybrid deep learning networks to evaluate surface water quality and groundwater potential. Correlation analysis revealed strong connections between remote sensing data and water quality parameters (such as chlorophyll-a, dissolved oxygen, nitrogen, and phosphorus). The hybrid models surpassed traditional machine learning methods, demonstrating their effectiveness in real-world water management. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2026.1.09 |
| first_indexed | 2026-04-20T01:00:22Z |
| format | Article |
| fulltext |
S.V. Klimov, T.V. Starovoit, 2026
124 ISSN 1681–6048 System Research & Information Technologies, 2026, № 1
UDC 004.89:556.3/5:528.8
DOI: 10.20535/SRIT.2308-8893.2026.1.09
LONG-TERM MONITORING OF SURFACE WATER QUALITY
AND GROUNDWATER POTENTIAL USING COMPUTATIONAL
INTELLIGENCE, GIS TECHNOLOGIES,
AND REMOTE SENSING
S.V. KLIMOV, T.V. STAROVOIT
Abstract. Water scarcity and declining water quality due to population growth, ur-
banization, industrialization, and climate change highlight the importance of effective
water management. Advances in remote sensing, cloud computing, and computational
intelligence underscore the need to utilize modern technologies for monitoring surface
water quality. This research involves the development of hybrid intelligent models
using Landsat and Sentinel-2 images and WISE data with hybrid deep learning net-
works to evaluate surface water quality and groundwater potential. Correlation anal-
ysis revealed strong connections between remote sensing data and water quality pa-
rameters (such as chlorophyll-a, dissolved oxygen, nitrogen, and phosphorus). The
hybrid models surpassed traditional machine learning methods, demonstrating their
effectiveness in real-world water management.
Keywords: computational intelligence, fuzzy logic, remote sensing, satellite imagery,
surface water quality monitoring, groundwater potential assessment, hybrid neural
networks, NEFCLASS-EM, TS-FNN, Fuzzy C-Means, K-Means.
INTRODUCTION
Water is crucial for human health, food security, economic growth, energy produc-
tion, and ecosystems. However, factors such as population growth, urbanization,
industrial development, increased demand, and water misuse have made water
scarce and expensive, particularly in developing countries. To address this issue,
various strategies have been developed to improve water quality and quantity by
2030 [1]. In Europe, the Water Framework Directive (WFD) [2] aims to achieve
good status for water resources. To assess the status, it’s essential to monitor bio-
logical, hydro morphological, and physicochemical water quality indicators. Ac-
cording to this directive, rivers with a catchment area of more than 10 km2 and lakes
with an area of more than 0.5 km2 should be included in the assessment and moni-
toring of water status [2].
Water quality parameters are traditionally determined by collecting samples
on-site and analyzing them in the laboratory [3]. This method provides high accu-
racy, but is also labor-intensive and time-consuming, requiring significant financial
investment. In addition, the traditional method determines the concentration of the
required indicators only at the point of sampling. Meanwhile, the water quality in
water bodies is rarely constant due to unpredictable events, such as accidental or
deliberate leaks from industrial facilities and other factors. This makes accurate
water quality monitoring a challenging task.
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 125
To overcome these limitations, we used remote sensing technology (RS),
which has the advantage of large spatial coverage and high temporal resolution,
which has been used to identify and monitor water resources more efficiently and
effectively [4–6]. Remote monitoring of water quality indicators is based on estab-
lishing a correlation between the monitoring data and the corresponding surface
reflection. Spectral characteristics of water are functions of hydrological, biologi-
cal, and chemical characteristics of water [7]. Specifically, the amount of wave ra-
diation at different wavelengths reflected from the water surface can be used di-
rectly or indirectly to detect water quality indicators [8].
Pure water can reflect light with a wavelength of more than 600 nm, which
provides a high blue-green reflectance while absorbing radiation in the near-infra-
red (NIR) spectrum and beyond. Increasing the chlorophyll concentration increases
the absorption of red (R) light and strongly absorbs blue (B) light, while the peak
of reflection is located in the green (G) part of the spectrum [9]. The transparency
of water depends on the total concentration of suspended solids. This concentration
is a measure of the weight of inorganic particles suspended in the water column and
is responsible for most of the scattering. By affecting the scattering of light, the
suspended solids concentration (SSC) in water directly controls the transparency
and oxygen content of a water body [10]. An increased concentration of SSC causes
a shift in the peak from the G to the R region and increases the reflectance of water
in the NIR region.
The relationship between surface reflectance and the concentration of water
quality parameters is indirect and non-linear. This makes their estimation problem-
atic, especially when based on traditional empirical algorithms. Over the past dec-
ade, the advancement in computing power and the development of artificial intelli-
gence and machine learning (ML) algorithms have led to an increased use of these
technologies to solve this problem. The most common machine learning models
used in water quality assessment tasks are Random Forest (RF), Support Vector
Machine (SVM), and Artificial Neural Network (ANN).
Studies [11–14] have demonstrated that Artificial Neural Networks (ANNs)
and Support Vector Machines (SVMs) deliver excellent performance in monitoring
both optically active and inactive water quality indicators. Generally, artificial neu-
ral networks, as a linear approximation method, offer greater flexibility for moni-
toring water quality indicators. However, the accuracy of machine learning models
typically depends on the chosen model and the quality of the training data. Devel-
oping ANN models requires large training datasets and significant experience to
construct the optimal architecture for artificial neural networks. Using too many
layers can lead to overfitting, which involves fitting noise in the training data [15].
Conversely, a small number of layers can lead to underfitting, where the model
cannot adequately represent the complexity of the data [15].
This study aims to develop and compare hybrid computational intelligence mod-
els, specifically neuro-fuzzy neural networks, using spatial data from remote sensing
and geographic information systems (GIS). It proposes combining neuro-fuzzy neural
networks with metaheuristic and remote sensing algorithms to assess water quality
from satellite images and evaluate their effectiveness. These hybrid models, integrating
remote sensing and GIS, offer innovative methods for assessing water quality and
groundwater potential. They can accurately identify factors contributing to water qual-
ity deterioration and unexpected surface formations in inland water bodies, while also
examining their long-term impact on ecological status.
S.V. Klimov, T.V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 1 126
MATERIALS AND METHODS
Research area
We chose the territory of Kyiv and the Kyiv region (Ukraine) as the research area
(Fig. 1). The study area is located in south-eastern Europe. Most of the rivers belong
to the Black Sea basin. The largest river is the Dnipro, which we chose for our
research. We used data from the Water Information System of Europe (WISE) to
train artificial neural networks and neuro-fuzzy neural networks.
Fig. 1. Location of the study area in Kyiv and Kyiv region, Ukraine. The green rectangle
marks the region of interest including key sampling sites along the Dnipro River
Data Preparation
Remote monitoring of water quality indicators is based on the correlation between
on-site measurements and the corresponding surface reflectance. For this study,
Landsat and Sentinel-2 satellite images of the surface over Europe from 2010 to
2024 were used. In total, more than two thousand images were analyzed to create
time series and train monitoring models. Also used some of the materials from the
DHI educational resource [16].
Landsat satellites achieve maximum surface coverage once every 16 days, with a
spatial resolution of 30 m for multispectral bands. The Google Earth Engine API, inte-
grated into Google Colab, was used as an access point to the imagery.
The surface reflectance values for each point were obtained from available Landsat
Surface Reflectance Level 2A images. Cloud and shadow masking were performed to
ensure clear water pixels. The resulting table included the identifier of the monitoring
stations, the corresponding surface reflectance value, and the date of the survey. The sur-
face reflectance was filtered by date to match the on-site data, with a maximum time
interval of 3 days between the on-site sampling and the satellite overpass.
We used Pearson correlation analysis to explore the connection between re-
mote sensing and in situ data, using the correlation coefficient (r). Based on the
correlation, we identified a specific set of input data for each water quality indica-
tor. The data was then standardized to a normal distribution with a mean of 0 and a
standard deviation of 1, and then divided into training and test sets (80 % and
20 %, respectively).
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 127
Fuzzy neural network model
Fuzzy theoretical TS systems can apply fuzzy mathematical rules to generate more
complex nonlinear functions. This allows the system to reduce the number of fuzzy
rules needed when dealing with problems involving multiple variables [17]. The
fuzzy theoretical system TS is typically defined using "if-then" logic, and its fuzzy
conclusion is expressed as follows [18]:
𝑅 ∶ 𝐼𝑓 𝑥 𝑖𝑠 𝐴 , 𝑥 𝑖𝑠 𝐴 , … 𝑥 𝑖𝑠 𝐴 , (1)
𝑇ℎ𝑒𝑛 𝑦 = 𝑝 + 𝑝 𝑥 + ⋯ + 𝑝 𝑥 ,
where 𝐴 is a fuzzy
set of a fuzzy system; 𝑝 (j = 1, 2, …, 𝑘) are the parameters of the fuzzy system; 𝑦 is the initial value obtained by the fuzzy rule; input part i.e. 𝐼𝑓 is fuzzy and output
part i.e. 𝑇ℎ𝑒𝑛 is deterministic part. This vague conclusion indicates that the output
is a linear combination of the inputs [18].
Suppose that for inputs 𝑥 = [𝑥 , 𝑥 , … , 𝑥 ], the degree of membership of each
input variable 𝑥 is first calculated by a fuzzy rule [18]: 𝜇 = exp(−( 𝑥 − 𝑐 ) / 𝑏 ), (2) (𝑗 = 1, 2, … , 𝑘; 𝑖 = 1, 2, … , 𝑛),
where 𝑐 and 𝑏 are the center and width of the membership function, k is an input
parameter, and n is the number of fuzzy subsets [18].
Fuzzy operations are performed on each of the above degrees of membership,
and the fuzzy operator is used as a concatenated multiplicative operator [18]: 𝜔 = 𝑢 (𝑥 ) × 𝑢 (𝑥 ) × 𝑢 (𝑥 ), (3) (𝑖 = 1, 2, … , 𝑛).
According to the results of the fuzzy calculation, the initial value of the model 𝑦 is obtained [18]: 𝑦 = ∑ ( ⋯ )∑ . (4)
A Fuzzy Neural Network (FNN) is a hybrid model that integrates the learning
capabilities of artificial neural networks with the interpretability of fuzzy logic sys-
tems. The network typically consists of four layers: an input layer, a fuzzification
layer, a rule base, and an output layer.
The algorithm of this type of neural network is described using equation (5).
𝜇 (𝑥 ) = exp − ( ) , (5)
S.V. Klimov, T.V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 1 128
where 𝑥 = (𝑥 , 𝑥 … , 𝑥 ) input vector; and 𝜇 (𝑥 ) is the membership function.
The activation of rule j is described by equation (6). The output of the network is
described by equation (7). 𝑤 = ∏ 𝜇 (𝑥 ) , (6)
𝑦 = ∑ ( )∑ , 𝑤ℎ𝑒𝑟𝑒 𝑓 (𝑥) = 𝑎 𝑥 + 𝑏 . (7)
A Takagi–Sugeno Fuzzy Neural Network (TS-FNN) extends the FNN by us-
ing linear functions in the consequents of fuzzy rules. Instead of constant outputs,
each rule produces an output of the form: 𝑦 = 𝑎 𝑥 + 𝑎 𝑥 + ⋯ + 𝑎 𝑥 + 𝑏 . (8)
The final output is a weighted sum of these rule outputs, normalized by the
total rule strength. TS-FNNs offer higher precision and faster convergence com-
pared to classical FNNs.
In the fuzzy neural network NEFCLASS (Neuro-Fuzzy CLASSification) clas-
sification is performed with automatic rule learning. Inputs are fuzzified by linguis-
tic variables (low, medium, high) [19]: 𝜇𝐴 (𝑥) = 𝑚𝑎𝑥 0.1 − . (9)
The rules are presented in the form: 𝐼𝑓 𝑥 𝑖𝑠 "High" and
x2 is "Low" 𝑡ℎ𝑒𝑛 𝐶𝑙𝑎𝑠𝑠 = 𝑊𝑎𝑡𝑒𝑟
The process of class estimation through fuzzy inference and max-aggregation
is described in equation (10), [20]. 𝑦 = arg max max min 𝜇 (𝑥 ) . (10)
The NEFCLASS-EM model enhances NEFCLASS with a metaheuristic opti-
mizer, such as the Electromagnetic Algorithm (EM). NEFCLASS-EM is calculated
in a similar way as in equations (9), (10), but with optimization of parameters (cen-
ters, widths) through metaheuristics (equation (11)).
𝜃( ) = 𝜃( ) + 𝐹 𝜃( ) , (11)
where 𝐹 – is the vector of the force of attraction/repulsion between candidates.
A schematic representation of the architecture of the four hybrid neural network
models is shown in Fig. 2.
Fig. 2 illustrates the structural differences between four neuro-fuzzy models:
FNN, TS-FNN, NEFCLASS, and NEFCLASS-EM. Each model includes key func-
tional components such as input processing, fuzzification, rule evaluation, and out-
put generation. The FNN and TS-FNN architectures are organized and layered,
with TS-FNN producing linear outputs. NEFCLASS and NEFCLASS-EM focus
on classification tasks, where NEFCLASS-EM integrates a metaheuristic optimizer
(EM) to enhance the rule parameter optimization.
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 129
Fig. 2. Architectures of the fuzzy neural network models used in the study
In this study, fuzzy neural networks were applied to predict water quality in-
dicators and groundwater potential. A fuzzy neural network model usually has four
levels: an input level, a fuzzification level, a fuzzy rule calculation level, and an
output level [21]. The input layers of the network model are connected through the
vector 𝑥 , so the number of nodes of the model network is consistent with the di-
mension of the input vector. The fuzzification layer uses the membership function
of equation (3) to fuzziness the input values to obtain the membership value 𝑢 .
The value of 𝜔 at the fuzzy computing level is obtained by using equation (4) of
successive phase multiplication, and then the output value of the output data level
in this fuzzy model system is obtained by equation (8). The fuzzy neural network
learning algorithm is as follows [21]:
Step 1. Calculation error: 𝑒 = (𝑦 − 𝑦 ) , (12)
where the expected output of the network is 𝑦 and the error between the expected
output and the actual output is e [21].
Step 2. Correction of the coefficient:
𝑝 (𝑘) = 𝑝 (𝑘 − 1) − 𝛼 , (13) 𝜕𝑒𝜕𝑝 = (𝑦 − 𝑦 )𝜔∑ 𝜔 ∙ 𝑥 ,
where 𝑝 is the coefficient of the neural network; 𝛼 is the learning rate of the net-
work; 𝑥 is the input parameters of the network; 𝜔 is the continuous product of the
membership of the input parameters [21].
Step 3. Correction of parameters: 𝑐 (𝑘) = 𝑐 (𝑘 − 1) − 𝛽 , (14) 𝑏 (𝑘) = 𝑏 (𝑘 − 1) − 𝛽 ,
where 𝑏 and 𝑐 are the width and center value of the representative membership
function in the fuzzy rule [21].
S.V. Klimov, T.V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 1 130
Indicators of assessment of accuracy of models
We used four metrics to evaluate the accuracy of the models: overall accuracy
(OA), misclassification error (ME), omission error (OE), and ROC-AUC value. OA
is the sum of pixels correctly classified as water divided by the total number of
water pixels represented by the confusion matrix. OE is the number of pixels that
belong to water but are classified as other surface types that can be identified by
the error matrix column. The ROC-AUC value [22] is the area under the curve of
the ratio of sensitivity (equation 15) to specificity (equation 16). This value ranges
from 0.50 to 1. The higher the value, the better the performance. If the value ex-
ceeds 0.70, the classification result is reliable [20]. 𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 = , (15)
𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑖𝑡𝑦 = , (16)
where TPR (true positive) and FNR (false negative) are pixels correctly and
incorrectly classified as water, and TNR (true negative) and FPR (false positive)
are pixels correctly and incorrectly classified as non-water [23].
To assess water quality, a water quality model’s performance can be measured
using various metrics. These include the coefficient of determination (𝑅 , Eq. 17),
the mean absolute error (MAE, Eq. 18), the root means square error (RMSE, Eq.
19), the mean square error (MSE, Eq. 20), residual prediction deviation (RPD, Eq.
21), and confidence interval (CI, Eq. 22), [23].
𝑅 = 1 − ∑ ∑ , (17)
𝑀𝐴𝐸 = 1𝑛 |𝑦 − 𝑦 | , (18)
𝑅𝑀𝑆𝐸 = 1𝑛 (𝑦 − 𝑦 ) , (19)
𝑀𝑆𝐸 = ∑ (𝑦 − 𝑦 ) , (20)
𝑅𝑃𝐷 = ∑ ( )∑ , (21)
𝐶𝐼 = 1 − ∑ | |∑ (| | | |) × 1 − ∑ ( )∑ ( ) . (22) 𝑅 values should not be too high, as excessively high values lead to over-
fitting and lack of model portability, since this metric is sensitive to outliers
[23]. Therefore, in practice, 𝑅 is often used in combination with RMSE, RPD,
CI, and other indicators to balance the fitting accuracy and computational com-
plexity [23, 24].
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 131
RESULTS OF SIMULATION
Aerospace research and geoinformation modeling
Satellite technology has a significant advantage in its ability to capture light beyond
the visible spectrum, which is undetectable to the human eye. Infrared light, a type
of radiation, can be detected by satellites like Sentinel-2. This infrared radiation can
provide valuable information about surface temperature, vegetation conditions, and
atmospheric conditions. For instance, healthy vegetation reflects more infrared
light than unhealthy vegetation or non-vegetated surfaces.
The raster image measures 1009 pixels in height and 1014 pixels in width,
covering approximately 10 kilometers in both dimensions. It contains four spectral
bands: blue, green, red, and near infrared. The satellite captures data across
11 different wavelengths, and we specifically selected the blue, green, red, and in-
frared spectra for analysis.
The near-infrared range is very helpful for analyzing vegetation. In this setup,
healthy vegetation appears bright red, while non-vegetated surfaces appear in other
colors. This method improves the visibility of different ground cover types and al-
lows us to see details that are not visible in normal light.
We first rendered the image in different color spaces and then focused on the
red and infrared ranges. To do this, we created a scatter diagram where the reflec-
tion coefficient of the red pixel is shown on the x-axis and the infrared pixel on the
y-axis (Fig. 3). Next, we converted the images into tabular data. Each row in the
resulting table represents one pixel for a specific date. For instance, the first row
corresponds to the pixel (1, 1) on February 22, 2022. Each column represents the
intensity of a spectral band (blue, green, red, infrared).
a b
Fig. 3. Visualization of the results of clear and fuzzy clustering, where a – clear clustering;
b – fuzzy clustering of pixels
To recognize clusters that correspond to water bodies, we applied the follow-
ing intelligent algorithms: Fuzzy Neural Networks (FNN), Takagi–Sugeno Fuzzy
Neural Networks (TS-FNN), NEFCLASS, and NEFCLASS-EM.
S.V. Klimov, T.V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 1 132
DEVELOPMENT OF COMPUTATIONAL INTELLIGENCE MODELS
Fuzzy neural network model
Fuzzy neural networks (FNNs) are hybrid models that combine fuzzy logic with
the computational power of neural networks. They are designed to handle uncer-
tainties and imprecise data in tasks such as classification, clustering, and regression.
In the context of satellite imagery, FNNs are particularly useful because satellite
data often contain noise, incomplete information, and inherent fuzziness. This is
especially true when dealing with natural phenomena such as cloud cover, land-
forms, vegetation, and water bodies.
Fig. 4. Visualization of water recognition results on satellite images using a fuzzy neural
network (FNN) is visualized
Fig. 4 shows the water extraction result, where we can see that the fuzzy neural
network recognized water in satellite images very well. FNNs allow the use of
fuzzy membership values (rather than binary solutions) to model these ambiguous
or mixed pixels. Unlike traditional neural networks, FNNs assign membership de-
grees to different classes, so a single pixel can belong to multiple categories with
different probabilities or degrees (e.g., 70 % forest, and 30 % water).
Takagi–Sugeno fuzzy neural network (TS-FNN) model
Takagi–Sugeno models are a type of fuzzy logic inference system. In these models,
the output of fuzzy rules can be a linear combination of input variables or a con-
stant. In TS-FNN, each fuzzy rule corresponds to a linear model or constant that is
learned, and the output is calculated as the weighted average of the rule’s output.
The implementation of the TS-FNN model is depicted in Fig. 5. The model is
trained to minimize the mean squared error (MSE) between the predicted outputs
and the actual labels. Post-training, the network predicts cluster membership for
each pixel in the satellite image, and the assigned clusters are then
visualized.
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 133
Fig. 5. Visualization of water recognition results on satellite images using Takagi–Sugeno fuzzy neu-
ral network (TS-FNN)
Fuzzy rules in TS-FNN are better interpreted because the output is a linear
function of the input variables, making it easier to understand how the model makes
its decisions.
NEFCLASS fuzzy neural network model
NEFCLASS is a neuro-fuzzy system specifically created for classification tasks. It
integrates fuzzy logic with a feed-forward neural network framework. In this sys-
tem, fuzzy rules are acquired from input data. The model generates fuzzy classifi-
cation rules automatically, which are then refined using neural network training
methods. The neural network adapts the fuzzy rules to minimize classification er-
rors.
Fig. 6. Visualization of the results of water recognition on satellite images using the
NEFCLASS fuzzy neural network
S.V. Klimov, T.V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 1 134
For simplicity, we implemented NEFCLASS-like functionality using a neural
network to represent fuzzy rules, but we manually defined the fuzzification process.
The result is shown in Fig. 6.
To implement this model, we utilized the same approach as in our previous
models. This involved defining fuzzy sets for the red and infrared ranges, which
were classified as low, medium, and high. The input layer represents the fuzzy in-
puts or fuzzy characteristics, and the subsequent layers model the combination of
these fuzzy rules.
The main advantage of this model is its ability to learn fuzzy rules during train-
ing, making it suitable for handling complex decisions, such as those involved in
satellite image classification. NEFCLASS can scale to larger data sets due to its
neural network structure, which allows it to handle multidimensional data more
efficiently.
Hybrid model NEFCLASS-EM
To achieve the highest clustering accuracy, we also tested the NEFCLASS network
with the electromagnetic metaheuristic’s algorithm. The result is displayed in Fig.7.
Fig. 7. Visualization of the results of water recognition on satellite images using the
NEFCLASS-EM hybrid neural network
We combined a neural network with an electromagnetic metaheuristic (EM)
algorithm to improve the network’s performance. We also used the EM algorithm
to optimize the hyperparameters of the NEFCLASS model, which enhanced its
classification efficiency. The EM algorithm works by modeling the attraction and
repulsion forces between different solutions (weight configurations) based on elec-
tromagnetic principles. After optimizing the weights, we further trained the
NEFCLASS network using the Adam optimizer. This hybrid approach led to faster
convergence and improved classification accuracy by leveraging the global search
capabilities of the EM algorithm and the fine-tuning ability of gradient-based opti-
mization.
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 135
Comparison of obtained results and assessment of the accuracy of models of fuzzy
neural networks
The FNN fuzzy network shows a significant improvement in accuracy, quickly ex-
ceeding 91% by the second epoch and stabilizing around 96% by the 20th epoch.
Losses also steadily decrease, indicating efficient learning and reduced errors. TS-
FNN shows a rapid loss reduction, reaching 0.0114 by the 20th epoch. This rapid
convergence indicates high accuracy. NEFCLASS shows a result like FNN but
starts with slightly less accuracy. By the 20th epoch, it reaches over 95% accuracy.
Loss reduction is slower compared to FNN, but still significant. NEFCLASS-EM
starts with the lowest accuracy and highest losses but stabilizes quickly. By the end
of training, the model achieves performance like NEFCLASS, with an accuracy of
over 95% and a significant loss reduction. The results of model training accuracy
are shown in Table 1.
So, from the obtained fuzzy neural network training results, we can see that FNN
exhibits the best overall balance of rapid accuracy improvement and stable loss reduction,
making it the most efficient model in terms of both learning speed and final performance.
TS-FNN has the fastest loss reduction, indicating a very accurate model. NEFCLASS is
comparable to FNN, but slightly slower in terms of increasing accuracy and reducing
loss. NEFCLASS-EM starts with the lowest performance but catches up to achieve ac-
curacy levels close to FNN and NEFCLASS by the last epoch.
The accuracy values reported in Table 1 were obtained from the training da-
taset over 20 epochs. To better assess the training dynamics, Figs. 8 and 9 show the
accuracy and loss values over 20 epochs for each model.
Fig. 8 shows the “Accuracy vs Epoch” curve for four models: FNN, TS-FNN,
NEFCLASS, and NEFCLASS-EM, based on the results presented in Table 1. FNN
demonstrates the fastest accuracy growth and stable performance
(~96–97 %).
TS-FNN starts slightly slower but quickly stabilizes at the same level.
NEFCLASS and NEFCLASS-EM initially show lower accuracy but gradually
catch up.
NEFCLASS-EM exhibits powerful dynamics after the 10th epoch. The plot
shows that TS-FNN and FNN achieve high accuracy early in training, while
NEFCLASS and NEFCLASS-EM improve gradually and stabilize after 10 epochs.
The “Loss vs Epoch” graph (Fig. 9) shows that TS-FNN achieves the fastest
loss reduction, reaching approximately 0.0114 by the 20th epoch. FNN steadily
reduces loss, although at a slightly slower pace. NEFCLASS and NEFCLASS-EM
start with higher loss values but gradually decrease them to an acceptable level,
demonstrating stable learning behavior.
TS-FNN shows the fastest reduction in loss, reaching approximately 0.0114
by epoch 20. FNN steadily reduces loss, while NEFCLASS and NEFCLASS-EM
start from higher values but show gradual and stable convergence.
The results show that FNN and TS-FNN achieve the highest accuracy and the fastest
convergence, while NEFCLASS and NEFCLASS-EM demonstrate stable learning with
gradual improvement. The integration of fuzzy logic and neural networks enables accu-
rate modeling of nonlinear processes in aquatic environments, offering promising oppor-
tunities for intelligent environmental monitoring.
S.V. Klimov, T.V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 1 136
T a b l e 1 . Comparison of the obtained results of training accuracy of fuzzy neural
networks
Fig. 8. Accuracy vs Epoch curves for all models Fig. 9. Loss vs Epoch curves for all models
Assessment of water quality and groundwater potential
Based on our research, we have developed a methodology for evaluating water
quality by utilizing satellite images and computational intelligence techniques, spe-
cifically fuzzy neural networks. This method integrates remote sensing and geoin-
formation modeling with sophisticated machine learning and artificial intelligence
No.
Epoch
Fuzzy Neural
Network
(FNN)
Takagi–Sugeno
Fuzzy Neural Net-
work (TS-FNN)
NEFCLASS Fuzzy
Neural Network
Hybrid
NEFCLASS-EM
1 0.5781 0.5710 0.5630 0.4789
2 0.9162 0.9113 0.8616 0.8452
3 0.9488 0.9363 0.9102 0.9068
4 0.9578 0.9368 0.9177 0.9237
5 0.9594 0.9391 0.9227 0.9289
6 0.9582 0.9382 0.9279 0.9355
7 0.9584 0.9404 0.9330 0.9425
8 0.9647 0.9563 0.9398 0.9463
9 0.9630 0.9590 0.9369 0.9464
10 0.9662 0.9562 0.9466 0.9478
11 0.9642 0.9558 0.9437 0.9498
12 0.9639 0.9539 0.9445 0.9505
13 0.9639 0.9539 0.9513 0.9553
14 0.9640 0.9540 0.9444 0.9531
15 0.9626 0.9526 0.9506 0.9547
16 0.9643 0.9543 0.9488 0.9587
17 0.9643 0.9543 0.9485 0.9569
18 0.9664 0.9564 0.9452 0.9544
19 0.9619 0.9519 0.9543 0.9534
20 0.9679 0.9579 0.9512 0.9560
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 137
(AI) models. This approach enables the automated analysis of water quality param-
eters on a large scale, including turbidity, chlorophyll-a concentration, and total
suspended solids (TSS).
To reflect water quality using this method, we have identified the following
four spatial resolution bands:
1. Blue (450–500 nm): sensitive to chlorophyll-a and water clarity.
2. Green (500–600 nm): reflects organic matter and suspended particles.
3. Red (600–700 nm): Helps detect deposits.
4. Near infrared (700-1100 nm): useful for turbidity and TSS determination.
In the next phase, we extracted spectral and spatial features from satellite im-
ages that are important for assessing water quality. These features serve as input
data for computational intelligence models. We calculated the chlorophyll estimate
using the normalized difference chlorophyll index (NDCI) with formula 16.
𝑁𝐷𝐶𝐼 = . (23)
To determine the turbidity, we utilized the turbidity index (NTU) [22] derived
from the red and near-infrared spectrums [23]. The combination of red and green
wavelengths assisted in estimating total suspended solids (TSS). To enhance our
analysis and capture spatial variations in water bodies, we employed spatial texture
functions, which involve a matrix of adjacent gray levels. Additionally, to gain a
better understanding of the water body characteristics, we calculated statistical
measures such as the mean value, variance, and entropy of pixel values.
Data from ground-based measurements of water quality parameters such as
chlorophyll, turbidity, and suspended solids were utilized to calibrate and validate
computational intelligence models.
The idea of using fuzzy neural networks produced better results than those
based on traditional neural networks and traditional machine learning methods.
Fuzzy logic systems are ideal for handling uncertainty in water quality assessment,
especially when the boundaries between quality classes are unclear. Fuzzy neural
networks enable the determination of fuzzy membership functions for
water quality parameters based on satellite features (for example, a pixel can belong
to both “clean water” and “polluted water” with different degrees of membership).
Once the model is trained and tested, it can be used to predict water quality
parameters based on new satellite images and to create maps of water quality.
By applying the models to satellite images over different periods, temporal changes
in water quality can be monitored. This is particularly useful for detecting trends
such as algal blooms, pollution, or seasonal variations in water quality. Continu-
ously monitoring water quality using computational intelligence models can help
detect early signs of pollution or algal blooms and trigger warnings.
Obtaining the correct characterization from satellite imagery is critical to as-
sessing groundwater potential. These features often serve as input to computational
intelligence models. For example, the Normalized Difference Vegetation Index
(NDVI) helps assess the health and density of vegetation, which is related to the
availability of groundwater
S.V. Klimov, T.V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 1 138
𝑁𝐷𝑉𝐼 = . (24)
Digital relief models (Fig. 10, a) can help us understand the topography of an
area. Low-lying areas and valleys have a higher potential for groundwater recharge.
The slope affects water runoff, while the height (slope direction) affects moisture
retention. Analyzing river patterns and drainage patterns using satellite data can
help identify areas with high infiltration potential.
Surface soil moisture (SSM): obtained from satellites such as Sentinel-1
(Fig. 10, b) shows how much water the soil holds. High soil moisture provides good
conditions for groundwater replenishment. The water content of vegetation can
indirectly indicate the level of soil moisture and the potential presence of ground-
water.
Satellite radar and optical images can help identify geologic lineaments (faults,
cracks) that act as conduits for groundwater. Radar data from satellites such as Sen-
tinel-1 can be useful for this purpose. Determining different types of rock or for-
mations (e.g., porous rock, fractured aquifers) from satellite data is critical because
certain geological formations are more favorable for groundwater storage. Mapping
lakes, rivers, and wetlands from satellite imagery provides insight into groundwater
recharge zones (Fig. 11), as surface water bodies are often associated with aquifers.
To train and test the computational intelligence models, we collected ground-
based data such as groundwater levels from boreholes, soil moisture profiles, geo-
logical and hydrogeological surveys, and basic climate data (precipitation, and
evaporation rates) to determine recharge potential.
a b
Fig. 10. Visualization of the results of obtaining a digital model of the relief – a; and soil
moisture – b
After extracting features from satellite imagery, we applied computational
analysis techniques to model the relationships between these features and ground-
water potential. Characteristics such as NDVI index, DEM, soil moisture, land use,
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 139
slope, and drainage density were used as input variables. As a result, we received a
model capable of predicting areas with high or low groundwater potential based on
input characteristics.
Fuzzy logic is particularly useful in groundwater assessment because it can
handle uncertainty and variability in data (e.g., variable soil moisture). The spatial
and temporal resolution of satellite data can affect the accuracy of groundwater
estimation. Higher resolution data increases accuracy but may not be available in
all regions.
Fig. 11. The result of mapping lakes, rivers, and wetlands from satellite images
Groundwater potential assessment using satellite imagery and computational
analysis techniques combines the strengths of remote sensing, machine learning,
and data science. By combining key characteristics such as soil moisture, vegeta-
tion indices, topography, and geological structures, these methods provide a pow-
erful, cost-effective approach to groundwater investigation and monitoring.
CONCLUSION
Water quality assessment from satellite images using computational intelligence
methods is a powerful approach for large-scale automated monitoring of water bod-
ies. Combining remote sensing data with techniques such as artificial neural net-
works, fuzzy logic, decision trees, and support vector machines enables accurate
prediction of key water quality parameters such as chlorophyll-a, turbidity, and to-
tal suspended solids. By continuously monitoring water quality using satellite im-
ages, computational intelligence techniques can help detect environmental changes,
manage water resources, and prevent water pollution.
As a result of the conducted research, the following key points and advantages
of the methods proposed in the article can be noted:
Remote sensing allows continuous coverage of a wide area of water bodies
and terrestrial landscapes, enabling large-scale monitoring that would be impracti-
S.V. Klimov, T.V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 1 140
cal using traditional terrestrial methods. Satellite data provide important infor-
mation about water quality and groundwater potential over vast regions, including
remote and hard-to-reach areas.
Computational intelligence techniques such as fuzzy neural networks can
automate the process of analyzing complex satellite data, greatly reducing the time
and manpower required for water resources assessment. These models can quickly
identify patterns, anomalies, and trends, providing real-time or near-real-time in-
formation critical to timely decision-making.
Fuzzy neural networks excel at dealing with the uncertainty and impreci-
sion inherent in the natural environment. Water quality parameters and groundwa-
ter potential often exhibit complex, non-linear relationships influenced by multiple
factors (e.g., land use, vegetation, climate). FNNs effectively model these complex-
ities and provide soft classifications that allow for more flexible predictions.
The use of satellite data and AI-based models minimizes the need for ex-
tensive field studies, reducing the costs associated with traditional water and
groundwater quality assessments. Thanks to the use of freely available satellite plat-
forms (e.g. Sentinel, Landsat) and powerful computing tools, environmental mon-
itoring is becoming more accessible and scalable.
Geo-information modeling allows the integration of spatial and tem-
poral data, allowing water quality trends and groundwater potential to be
tracked over time. It helps identify seasonal patterns, long-term environmental
changes, and the effects of human activities such as agriculture, urbanization,
or pollution.
Therefore, the combination of fuzzy neural networks, remote sensing of the
Earth and geoinformation modeling offers a reliable, dynamic, and effective frame-
work for the management and protection of water resources, ensuring their sustain-
ability in the face of growing environmental problems.
ACKNOWLEDGMENTS
The ideas for this study are based on the knowledge of water quality assessment
deepened and expanded during the course “Operation and Maintenance of Ur-
ban Water Supply System (Water Quality and Purification) (B)” in
Japan, with the financial support of the Japan International Cooperation Agency
(JICA).
REFERENCES
1. UN General Assembly. Transforming Our World: The 2030 Agenda for Sustainable
Development. 21 October 2015. Available: https://www.refworld.org/legal/
resolution/ unga/2015/en/111816
2. Directive 2000/60/EC of the European Parliament and of the Council of 23 October
2000 establishing a framework for Community action in the field of water policy; Eu-
ropean Parliament: Bruxelles, Belgium, 2003. Available: https://eur-lex.europa.eu/le-
gal-content/EN/TXT/?uri=CELEX:32000L0060
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 141
3. F. Faur, M. Lazăr, I.-M. Apostu, O. Pinchuk, S. Klimov, “Monitoring the water quality
of Jiu River in Dolj County,” E3S Web Conf., vol. 280, article no. 10002, 2021. doi:
https://doi.org/10.1051/e3sconf/202128010002
4. J. He, Y. Chen, J. Wu, D.A. Stow, G. Christakos, “Space-Time Chlorophyll-a Re-
trieval in Optically Complex Waters that Accounts for Remote Sensing and Modeling Un-
certainties and Improves Remote Estimation Accuracy,” Water Research, vol. 171, 115403,
2019. doi: https://doi.org/10.1016/j.watres.2019.115403
5. B. Nas, S. Ekercin, H. Karabörk, A. Berktay, D.J. Mulla, “An Application of Landsat-
5TM Image Data for Water Quality Mapping in Lake Beysehir, Turkey,” Water, Air,
& Soil Pollution, vol. 212, pp. 183–197, 2010. doi: https://doi.org/10.1007/s11270-
010-0331-2
6. M. Govedarica, G. Jakovljevic, “Monitoring spatial and temporal variation of water
quality parameters using time series of open multispectral data,” in Seventh Interna-
tional Conference on Remote Sensing and Geoinformation of the Environment, Pa-
phos, Cyprus, 18–21 March 2019, vol. 11174. doi:
https://doi.org/10.1117/12.2533708
7. C. Wu et al., “Empirical estimation of total phosphorus concentration in the mainstream of
the Qiantang River in China using Landsat TM data,” Int. J. Remote Sens., vol. 31, issue 9,
pp. 2309–2324, 2010. doi: https://doi.org/10.1080/01431160902973873
8. Gordana Jakovljevic, Flor Álvarez-Taboada, Miro Govedarica, “Long-Term Monitoring of
Inland Water Quality Parameters Using Landsat Time-Series and Back-Propagated ANN:
Assessment and Usability in a Real-Case Scenario,” Remote Sensing, vol. 16, issue 1, 68,
2024. doi: https://doi.org/10.3390/rs16010068
9. N. Ha, K. Koike, M. Nhuan, “Improved Accuracy of Chlorophyll-a Concentration Es-
timates from MODIS Imagery Using a Two-Band Ratio Algorithm and Geostatistics:
As Applied to the Monitoring of Eutrophication Processes over Tien Yen Bay (Norther
Vietnam),” Remote Sensing, vol. 6, issue 1, pp. 421–442, 2013. doi: https://doi.org/
10.3390/rs6010421
10. B. Nechad, K. Ruddick, Y. Park, “Calibration and validation of a generic multisensor
algorithm for mapping of total suspended matter in turbid waters,” Remote Sensing of
Environment, vol. 114, issue 4, pp. 854–866, 2010. doi: https://doi.org/
10.1016/j.rse.2009.11.022
11. K.T. Peterson, V. Sagan, J.J. Sloan, “Deep learning-based water quality estimation
and anomaly detection using Land-sat-8/Sentinel-2 virtual constellation and cloud
computing,” GIScience & Remote Sensing, vol. 57, issue 4, pp. 510–525, 2020. doi:
https://doi.org/10.1080/15481603.2020.1738061
12. S. Hafeez et al., “Comparison of Machine Learning Algorithms for Retrieval of Water
Quality Indicators in Case-II Waters: A Case Study of Hong Kong,” Remote Sensing,
vol. 11, issue 6, 617, 2019. doi: https://doi.org/10.3390/rs11060617
13. L.F. Arias-Rodriguez, U.F. Tüzün, Z. Duan, J. Huang, Y. Tuo, M. Disse, “Global Wa-
ter Quality of Inland Waters with Harmonized Landsat-8 and Sentinel-2 Using Cloud-
Computed Machine Learning,” Remote Sensing, vol. 15, issue 5, 1390, 2023. doi:
https://doi.org/10.3390/rs15051390
14. D. Gómez, P. Salvador, J. Sanz, J.L. Casanova, “A new approach to monitor water
quality in the Menor sea (Spain) using satellite data and machine learning meth-
ods,” Environmental Pollution, vol. 286, 117489, 2021. doi: https://doi.org/10.1016/
j.envpol.2021.117489
S.V. Klimov, T.V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 1 142
15. G. Jakovljevic, M. Govedarica, F. Alvarez-Taboada, “Water body mapping: A com-
parison of remotely sensed and GIS open data sources,” International Journal of Re-
mote Sensing, vol. 40, issue 8, pp. 2936–2964, 2018. doi: https://doi.org/10.1080/
01431161.2018.1538584
16. DHI Education Resource. Available: https://raw.githubusercontent.com/
DHI/Intro_ML_course/main/module_6/Brahmaputra_images.zip
17. Z. Zhou, The Application of Fuzzy Neural Network Based on T-S Model in Water
Quality Evaluation. Master’s Thesis. East China University of Political Science and
Law, Nanjing, China, 2007.
18. X. Wang, MATLAB Neural Networks 43 Case Studies; 1st ed. Beijing, China: Beijing
University of Aeronautics and Astronautics Press, 2013, pp. 288–289.
19. O.I. Chumachenko, “Deep learning classifier based on nefclass neural network,” Elec-
tronics and Control Systems, vol. 3, no. 49, 2016. doi: https://doi.org/10.18372/1990-
5548.49.11242
20. Jamileh Yousefi, “A modified NEFCLASS classifier with enhanced accuracy-interpretability
trade-off for datasets with skewed feature values,” Fuzzy Sets and Systems, vol. 413, pp. 99–
113, 15 June 2021. doi: https://doi.org/10.1016/j.fss.2020.07.011
21. H.Y. Hong, A. Jaafari, E.K. Zenner, “Predicting spatial patterns of wildfire suscepti-
bility in the Huichang County, China: An integrated model to analysis of landscape
indicators,” Ecological Indicators, vol. 101, pp. 878–891, 2019. doi:
https://doi.org/10.1016/ j.ecolind.2019.01.056
22. Jinyue Chen et al., “Remote Sensing Big Data for Water Environment Monitoring:
Current Status, Challenges, and Future Prospects,” Earth’s Future, vol. 10, issue 2,
2022. doi: https://doi.org/10.1029/2021EF002289
23. L.L. Tang, S. Zhang, J.H. Zhang, Y. Liu, Y. Bai, “Estimating evapotranspiration based
on the satellite-retrieved near-infrared reflectance of vegetation (NIRv) over
croplands,” GIScience and Remote Sensing, vol. 58, issue 6, pp. 889–913, 2021. doi:
https://doi.org/10.1080/15481603.2021.1947622
24. Normalized difference turbidity index. Available: https://developers.arcgis.com/ py-
thon/latest/samples/river-turbidity-estimation-using-sentinel2-data-/
25. Turbidity Index (NTU). Available: https://www.summerland.ca/docs/default-source/
works-and-utilities/water/turbidity-index.pdf?sfvrsn=2
26. Nazarij Buławka, Hector A. Orengo, Iban Berganzo-Besga, “Deep learning-based de-
tection of qanat underground water distribution systems using HEXAGON spy satel-
lite imagery,” Journal of Archaeological Science, vol. 171, 106053, 2024. doi: https://
doi.org/10.1016/j.jas.2024.106053
27. Bingxue Zhao, Lei Wang, “Surface water monitoring from 1984 to 2021 based on
Landsat time-series images and Google Earth Engine,” Heliyon, vol. 10, issue 17,
e36660, 2024. doi: https://doi.org/10.1016/j.heliyon.2024.e36660
Received 28.11.2024
INFORMATION ON THE ARTICLE
Serhii V. Klimov, ORCID: 0000-0002-5993-847X, National University of Water and En-
vironmental Engineering, Ukraine, e-mail: s.v.klimov@nuwm.edu.ua
Tetiana V. Starovoit, ORCID: 0009-0008-6335-7679, National Technical University of
Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail:
starovoyt.tania@lll.kpi.ua
Long-term monitoring of surface water quality and groundwater potential using computational…
Системні дослідження та інформаційні технології, 2026, № 1 143
ДОВГОСТРОКОВИЙ МОНІТОРИНГ ЯКОСТІ ПОВЕРХНЕВИХ ВОД ТА
ПОТЕНЦІАЛУ ПІДЗЕМНИХ ВОД ІЗ ВИКОРИСТАННЯМ ОБЧИСЛЮ-
ВАЛЬНОГО ІНТЕЛЕКТУ, GIS-ТЕХНОЛОГІЙ ТА ДИСТАНЦІЙНОГО
ЗОНДУВАННЯ / С.В. Клімов, Т.В. Старовойт
Анотація. Дефіцит води і зниження її якості через зростання населення, урба-
нізацію, індустріалізацію й зміну клімату підкреслюють важливість ефектив-
ного керування водними ресурсами. Досягнення в дистанційному зондуванні,
хмарних обчисленнях та обчислювальному інтелекті підкреслюють необхід-
ність використання сучасних технологій для моніторингу якості поверхневих
вод. Це дослідження містить розроблення гібридних інтелектуальних моделей
із використанням зображень Landsat та Sentinel-2 і даних WISE із гібридними
мережами глибокого навчання для оцінювання якості поверхневих вод та поте-
нціалу підземних вод. Кореляційний аналіз виявив сильні зв’язки між даними
дистанційного зондування та параметрами якості води (такими як хлорофіл-а,
розчинений кисень, азот та фосфор). Гібридні моделі перевершили традиційні
методи машинного навчання, продемонструвавши свою ефективність у реаль-
ному керуванні водними ресурсами.
Ключові слова: обчислювальний інтелект, нечітка логіка, дистанційне зонду-
вання, супутникові знімки, моніторинг якості поверхневих вод, оцінювання по-
тенціалу підземних вод, гібридні нейронні мережі, NEFCLASS-EM, TS-FNN,
Fuzzy C-Means, K-Means.
|
| id | journaliasakpiua-article-358082 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2026-04-20T01:00:22Z |
| publishDate | 2026 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/71/966a54858ba3097c32797e05df4bb271.pdf |
| spelling | journaliasakpiua-article-3580822026-04-19T21:53:19Z Long-term monitoring of surface water quality and groundwater po-tential using computational intelligence, GIS technologies, and remote sensing Довгостроковий моніторинг якості поверхневих вод та потенціалу підземних вод із використанням обчислювального інтелекту, GIS-технологій та дистанційного зондування Klimov, Serhii Starovoit, Tetiana computational intelligence fuzzy logic remote sensing satellite imagery surface water quality monitoring groundwater potential assessment hybrid neural networks NEFCLASS-EM TS-FNN Fuzzy C-Means K-Means обчислювальний інтелект нечітка логіка дистанційне зондування супутникові знімки моніторинг якості поверхневих вод оцінювання потенціалу підземних вод гібридні нейронні мережі NEFCLASS-EM TS-FNN Fuzzy C-Means K-Means Water scarcity and declining water quality due to population growth, urbanization, industrialization, and climate change highlight the importance of effective water management. Advances in remote sensing, cloud computing, and computational intelligence underscore the need to utilize modern technologies for monitoring surface water quality. This research involves the development of hybrid intelligent models using Landsat and Sentinel-2 images and WISE data with hybrid deep learning networks to evaluate surface water quality and groundwater potential. Correlation analysis revealed strong connections between remote sensing data and water quality parameters (such as chlorophyll-a, dissolved oxygen, nitrogen, and phosphorus). The hybrid models surpassed traditional machine learning methods, demonstrating their effectiveness in real-world water management. Дефіцит води і зниження її якості через зростання населення, урбанізацію, індустріалізацію й зміну клімату підкреслюють важливість ефективного керування водними ресурсами. Досягнення в дистанційному зондуванні, хмарних обчисленнях та обчислювальному інтелекті підкреслюють необхідність використання сучасних технологій для моніторингу якості поверхневих вод. Це дослідження містить розроблення гібридних інтелектуальних моделей із використанням зображень Landsat та Sentinel-2 і даних WISE із гібридними мережами глибокого навчання для оцінювання якості поверхневих вод та потенціалу підземних вод. Кореляційний аналіз виявив сильні зв’язки між даними дистанційного зондування та параметрами якості води (такими як хлорофіл-а, розчинений кисень, азот та фосфор). Гібридні моделі перевершили традиційні методи машинного навчання, продемонструвавши свою ефективність у реальному керуванні водними ресурсами. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2026-03-31 Article Article Peer-reviewed Article application/pdf https://journal.iasa.kpi.ua/article/view/358082 10.20535/SRIT.2308-8893.2026.1.09 System research and information technologies; No. 1 (2026); 124-143 Системные исследования и информационные технологии; № 1 (2026); 124-143 Системні дослідження та інформаційні технології; № 1 (2026); 124-143 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/358082/344007 |
| spellingShingle | обчислювальний інтелект нечітка логіка дистанційне зондування супутникові знімки моніторинг якості поверхневих вод оцінювання потенціалу підземних вод гібридні нейронні мережі NEFCLASS-EM TS-FNN Fuzzy C-Means K-Means Klimov, Serhii Starovoit, Tetiana Довгостроковий моніторинг якості поверхневих вод та потенціалу підземних вод із використанням обчислювального інтелекту, GIS-технологій та дистанційного зондування |
| title | Довгостроковий моніторинг якості поверхневих вод та потенціалу підземних вод із використанням обчислювального інтелекту, GIS-технологій та дистанційного зондування |
| title_alt | Long-term monitoring of surface water quality and groundwater po-tential using computational intelligence, GIS technologies, and remote sensing |
| title_full | Довгостроковий моніторинг якості поверхневих вод та потенціалу підземних вод із використанням обчислювального інтелекту, GIS-технологій та дистанційного зондування |
| title_fullStr | Довгостроковий моніторинг якості поверхневих вод та потенціалу підземних вод із використанням обчислювального інтелекту, GIS-технологій та дистанційного зондування |
| title_full_unstemmed | Довгостроковий моніторинг якості поверхневих вод та потенціалу підземних вод із використанням обчислювального інтелекту, GIS-технологій та дистанційного зондування |
| title_short | Довгостроковий моніторинг якості поверхневих вод та потенціалу підземних вод із використанням обчислювального інтелекту, GIS-технологій та дистанційного зондування |
| title_sort | довгостроковий моніторинг якості поверхневих вод та потенціалу підземних вод із використанням обчислювального інтелекту, gis-технологій та дистанційного зондування |
| topic | обчислювальний інтелект нечітка логіка дистанційне зондування супутникові знімки моніторинг якості поверхневих вод оцінювання потенціалу підземних вод гібридні нейронні мережі NEFCLASS-EM TS-FNN Fuzzy C-Means K-Means |
| topic_facet | computational intelligence fuzzy logic remote sensing satellite imagery surface water quality monitoring groundwater potential assessment hybrid neural networks NEFCLASS-EM TS-FNN Fuzzy C-Means K-Means обчислювальний інтелект нечітка логіка дистанційне зондування супутникові знімки моніторинг якості поверхневих вод оцінювання потенціалу підземних вод гібридні нейронні мережі NEFCLASS-EM TS-FNN Fuzzy C-Means K-Means |
| url | https://journal.iasa.kpi.ua/article/view/358082 |
| work_keys_str_mv | AT klimovserhii longtermmonitoringofsurfacewaterqualityandgroundwaterpotentialusingcomputationalintelligencegistechnologiesandremotesensing AT starovoittetiana longtermmonitoringofsurfacewaterqualityandgroundwaterpotentialusingcomputationalintelligencegistechnologiesandremotesensing AT klimovserhii dovgostrokovijmonítoringâkostípoverhnevihvodtapotencíalupídzemnihvodízvikoristannâmobčislûvalʹnogoíntelektugistehnologíjtadistancíjnogozonduvannâ AT starovoittetiana dovgostrokovijmonítoringâkostípoverhnevihvodtapotencíalupídzemnihvodízvikoristannâmobčislûvalʹnogoíntelektugistehnologíjtadistancíjnogozonduvannâ |