Introduction

The utilization of remote sensing technology has demonstrated efficacy in enhancing the administration of natural resources, encompassing land, soil, water, forests, and mountains. Additionally, it has been important in the surveillance of desertification, floods, droughts, and alterations in landforms. According to Cloude (2009), the utilization of this technology facilitates the process of investigating, categorizing, and assessing the natural resources present in less developed regions (Cloude, 2009). The utilization of remote sensing, a sophisticated technique for data acquisition through the reflection of radiation employing satellites, airplanes, and ground-based devices, is experiencing significant growth in the field of agriculture (Shanmugapriya et al., 2019). As a potent and non-destructive means of aid in agriculture, this technique should be utilized (Yones et al., 2023). Airborne and satellite imagery serve as valuable mapping instruments, facilitating the classification of crops, assessment of their condition and sustainability, and surveillance of agricultural methodologies.

The use of remote sensing in agriculture includes (Campbell & Wynne, 2011):

  • Classification of crops.

  • Condition assessment and estimation of crop yield.

  • Soil characteristics mapping and management processes.

  • Farming practices.

Farming practices refer to the methods and techniques employed in agricultural activities. These practices encompass a wide range of activities, including cultivation, irrigation, and pests.

The primary threats to the sustainable growth of agriculture are land degradation and desertification (Israr et al., 2017). Salinity exerts a significant influence on food production (Cui et al., 2023), hence posing dangers that detrimentally affect contemporary crop growth (Lhissoui et al., 2014). Soil salinization impacts around 100 countries, as reported by Shahid et al. (2013) and Shahid et al. (2013). Salinization denotes the accumulation of salts in soil, causing detrimental impacts on agricultural productivity. It is crucial to address this phenomenon promptly to impede its progression. To foster sustainable land utilization and enhance efficient reclamation and rehabilitation management strategies, meticulous monitoring and mapping of soil salinity are essential. This is especially critical in tropical and semi-tropical regions, where climate change is anticipated to amplify the challenges posed by intensification and population growth. The examination of the literature demonstrates that numerous approaches have been utilized and proposed for the quantification and visualization of soil salinity. Field measurements and laboratory analysis, despite their widespread usage, include several limitations that make them less than ideal for studying variations in soil salinity (Allbed et al., 2014). These drawbacks include their high costs, time-intensive nature, and lack of suitability for accurately evaluating changes in soil salinity.

In the present era, the most recent information concerning changes in land cover is communicated by employing remote sensing data and Geographic Information System (GIS) methodologies. As Çullu (2003) asserts, these techniques can also play a role in monitoring ongoing land degradation and fluctuations in soil salinity (ÇULLU, 2003). Singh proposed an augmentation of sensing methodologies, in conjunction with retrieval systems, to streamline the direct detection of salinization and waterlogging parameters on a broader scale (Singh, 2021). Research outcomes have indicated that employing band ratios, specifically those involving the visible to near-infrared and infrared bands, has shown improved effectiveness in discerning salts in soils and crops under salt stress conditions. The application of remote sensing has been widespread in the identification of soil salinity (Singh et al., 2017).

The Landsat satellite series provides the most extensive chronological collection of surface observations from orbit, spanning a period of more than four decades. These observatories offered improved spectral and geographic precision in both reflecting and thermal wavelengths (Lauer et al., 1997; Loveland & Dwyer, 2012; Williams et al., 2006). The Operational Land Imager (OLI) and the Thermal Infrared Sensor (TIRS), both incorporated into the Landsat 8 satellite, are comprehensively elucidated in a publication authored by Irons et al. (2012) and Irons et al. (2012).

Machine learning assumes a substantial role in the burgeoning field of data science, involving the training of algorithms for tasks such as classification, prediction, and extraction of pivotal insights in data mining endeavors (Salehi & Burgueño, 2018). In contemporary times, the field of remote sensing has become increasingly reliant on machine learning techniques. This reliance stems from the ability of machine learning to facilitate the rapid and efficient evaluation of vast quantities of data. By leveraging machine learning algorithms, researchers can extract valuable information from remotely sensed data in a manner that is both expedient and effective. The utilization of remote sensing and machine learning methodologies has been extensively employed across diverse academic disciplines (Fazelpoor et al., 2022; Reddy & Kumar, 2022). In recent times, scholarly investigations have focused on the precise assessment of soil salinity through the utilization of machine learning models in areas characterized by substantial spatial heterogeneity and diverse topographies. This issue has been the focus of numerous inquiries. Nguyen et al. (2021) developed an innovative framework in the Mekong Delta of Vietnam to monitor the intrusion of salinity through the utilization of remote sensing techniques and machine learning algorithms (Nguyen et al., 2021). In their study, Naimi et al. (2021) employed remotely sensed data, proximal sensing, and a digital elevation model to ascertain environmental parameters to predict regional soil salinity (Naimi et al., 2021). This prediction was achieved through the utilization of machine learning algorithms. Cheng et al. (2022) utilized Landsat satellite imagery to construct an improved remote sensing-driven framework for the detection and characterization of soil salinity within the Yellow River Delta region in China (Cheng et al., 2022). Zhang et al. (2022) developed a methodology utilizing sparse cone penetration test data to estimate the horizontal scale of fluctuation in soil spatial variability. The approach employed a convolutional neural network-based model for this purpose (Zhang et al., 2022). Aksoy et al. (2022) conducted a comparative analysis of three distinct machine-learning methodologies to assess their efficacy in mapping soil salinity through the integration of satellite imagery and ground-based electrical conductivity data (Aksoy et al., 2022). Furthermore, it has been observed that the categorization of fractional vegetation cover enhances the robustness and forecasting capability of the models. These methodologies offer cost-efficient and dependable techniques for evaluating soil salinity and facilitating decision-making in agricultural management and land use planning.

Nonetheless, there remains a need for a comprehensive assessment of the individual contributions of different factors related to soil salinization through the utilization of machine learning techniques. To evaluate the spatial and temporal variability of soil salinity, it is imperative to conduct thorough assessments of environmental factors that are closely linked to this phenomenon. These evaluations should be carried out in conjunction with appropriate remote sensing imagery and adaptable machine learning models. Within the present context, the primary goal of this investigation is to formulate a model employing advanced machine learning algorithms for the digital mapping of soil salinity. This model seeks to establish a correlation between the bands derived from Landsat 8 satellite image data and the Total Dissolved Solids (TDS) salinity within the designated study area.

Region of Investigation and Data Compilation

Study Area

Sharqiyah governorate in northeast Egypt is known as one of Egypt's key agricultural regions, often referred to as the "breadbasket" of the country. The governorate boasts fertile soil, which supports the cultivation of a variety of crops essential for Egypt's food supply. Studying the effects of salinity on the soil in the Sharqiyah Governorate is essential for developing targeted strategies that promote sustainable agriculture, preserve soil health, and contribute to the overarching goal of food sustainability in Egypt. This knowledge empowers stakeholders to make informed decisions that balance agricultural productivity with environmental conservation. In recent investigations, scholarly inquiries have focused on salinity and its impact on soil within diverse Egyptian regions. Notably, Shaddad and colleagues introduced a probabilistic methodology to assess and cartographically depict a risk index. This approach involves the utilization of comprehensive auxiliary data to estimate the risk index within a specific locale in the Sharkia Governorate (Shaddad et al., 2020). The research site is situated within the Sharqiyah, delineated geographically by longitudes 31° 05′ E to 32° 10′ E and latitudes 30° 10′ N to 31° 10′ N, as depicted in Fig. 1. The soil texture within this area encompasses a spectrum from sand to clay.

Fig. 1
figure 1

Map of the study area

Full size image

Ground Truth Measurements

A thorough gathering of 69 soil samples from the surface was executed at a depth of 0–50 cm using field augers. These augers were strategically dispersed over the study area to accurately assess and document the soil salt levels at each specific location, as visually depicted in Fig. 2. In January 2018, field measurements were conducted to acquire data about soil salinity. The augers' location was rectified to UTM (Universal Transverse Mercator) coordinates, as depicted in the accompanying Figure. The field investigation and sample analysis were undertaken by the Drainage Investigation Institute (DRI) in Egypt.

Fig. 2
figure 2

Locations of soil sampling sites

Full size image

Salinity Computations:

Salinity serves as a measure of the quantity of soluble salts present in water or soil. Any compound comprising a cation, such as sodium + , potassium + , or calcium2 + , along with an anion, like chloride- and sulfate2-, is denoted as salt. The estimation of salinity can be accomplished through the following methods:

  • Utilizing the electrical conductivity (EC) of a solution or a mixture of soil and water, either in the field or laboratory, to approximate or assess salinity.

  • Determining ion concentrations through a laboratory chemical analysis of Total Dissolved Solids (TDS) from water or soil.

  • Employing an electromagnetic induction (EM) instrument to measure the apparent electrical conductivity of soil.

Calculating salinity and converting units:

  1. 1.

    Electrical conductivity (EC) is commonly employed for the assessment of salinity in both soil and water.

  2. 2.

    Simple relationships are essential for the conversion of EC to Total Dissolved Solids (TDS) or vice versa. Specifically:

  3. 3.

    TDS (mg/L or ppm) equals EC (dS/m) multiplied by 640 (for EC values ranging from 0.1 to 5 dS/m).

  4. 4.

    TDS (mg/L or ppm) equals EC (dS/m) multiplied by 800 (for EC values exceeding 5 dS/m).

In this study the soil samples underwent laboratory investigation to determine their EC using a volumetric soil: water ratio of 1:2.5. Table 1 illustrates the descriptive statistics for salinity of the study samples.

Table 1 Descriptive statistics for the salinity of the samples
Full size table

Satellite Images Data

Landsat-8, a satellite jointly operated by the United States Geological Survey (USGS) and NASA, boasts a sun-synchronous orbit, affording a temporal resolution of 16 days. Its payload, the Operational Land Imager (OLI), comprises 11 spectral bands with varying wavelengths, which are described in detail in (Irons et al., 2012) and are summarized in Table 1. With a radiometric resolution of 12 bits per pixel, Landsat-8 provides crucial data for applications such as agricultural monitoring, land cover classification, and environmental assessment, making it an invaluable tool for Earth observation (Table 2).

Table 2 Landsat-8 bands detail (Roy et al., 2014)
Full size table

The study utilized satellite pictures from the Landsat-8 platform with a spatial resolution of 30 m. The photos underwent a conversion process to obtain UTM coordinates, specifically assigned to the north UTM zone 36 route 176 Row 39 inside the WGS84 (World Geodetic System) 1984 coordinate system. The photograph included in this study was captured in January 2018 during the period of active field research. The photos depicted in Fig. 3 were obtained from the official website of the USGS (https://earthexplorer.usgs.gov/). Specifically, images were downloaded from the aforementioned website, ensuring that only those with cloud coverage below 10% were selected.

Fig. 3
figure 3

Landsat-8 reduced image from the USGS website for study area

Full size image

Research Methodology

The methodology employed in the study encompasses four distinct phases: preprocessing of both the auger map and Landsat image, data processing, machine learning modelling, and the creation of a salinity map utilizing the most optimal model, as illustrated in Fig. 4.

Fig. 4
figure 4

Flowchart depicting the utilized methodology

Full size image

Preprocessing of Data

Initially, an Arc Map is employed to generate the shape file representing the soil map. Subsequently, the shape file is exported to the ERDAS IMAGINE software to do image analysis. Following this stage, to reduce the error rate associated with the downloaded Landsat pictures, it becomes necessary to implement radiometric and atmospheric corrections.

Processing of Image

The DN (digital number) for each soil auger site is determined through the utilization of ERDAS Envision software. The study conducted by Panah and Goossens (2001) and Tran et al. (2018) involves the comparison of field truth data and spectral salinity to establish a correlation between soil digital numbers (DNs) and electrical conductivity (EC) (Panah & Goossens, 2001; Tran et al., 2018).

Machine Learning Algorithms

Since no one modeling technique works well in every region, soil salinity mapping accuracy depends on the modeling technique utilized (Jiang et al., 2019); therefore, in this study, regression models were trained using the Regression Learner App, and predictions were made using supervised machine learning. Regression models were trained and validated using a variety of models.

Regression Learner:

The Regression Learner software is designed to train regression models for data prediction. Utilizing this tool involves navigating through your data, selecting features, specifying validation procedures, training models, and evaluating results. Automated training facilitates the identification of the most suitable regression model type. Supervised machine learning is performed by providing a known set of observations of input data (predictors) and known outcomes, training a model using these observations to predict responses to new incoming data. The model can be exported to the workspace or MATLAB® code can be generated to replicate the trained model for use with new data or for programmatic understanding of regression. To guard against overfitting, the software employs cross-validation, although holdout validation is also an available alternative. Validation errors of each model are examined side by side before selecting the most appropriate model. The app presents the validated model's results, including diagnostic metrics like model accuracy and graphical representations such as response plots or residual plots. Multiple regression models can be automatically trained, and the optimal model for the regression problem is selected based on a comparison of validation results. A flowchart illustrating the typical approach of the Regression Learner app for training regression models is depicted in Fig. 5.

Fig. 5
figure 5

Workflow for training regression models using the Regression Learner application

Full size image

Regression Model Types:

Regression models are a statistical technique employed to assess the magnitude and nature of the relationship between a dependent variable and a set of independent variables in many fields such as engineering, finance, investment, and other industries. The regression learner app provides an interactive approach for constructing a regression model. The Regression Learner is a tool utilized for training predictive models, encompassing tree-based ensembles, support vector machines (SVM), regression trees, and Gaussian linear regression. The examination of our dataset, the selection of features, the construction of an assessment approach, and the discovery of information contribute to the augmentation of our training sets. Additionally, it is possible to store a trained model in the workspace and afterward employ it for analysis with fresh data. Table 3 displays the attributes of the different types of regression models employed in the study.

Table 3 Characteristics of regression model types used
Full size table

Visualize and evaluate model performance in regression learners:

The purpose of this task is to visually represent and evaluate the performance of a regression learner model. Once an individual has received training in Regression Learner, it becomes possible to assess the effectiveness of a regression model by examining its model statistics. The outcomes of this evaluation can be visually represented through a response plot or by graphing the actual response against the anticipated response. Additionally, the residual plot can be utilized as a means of evaluating the model's performance. The model statistics were generated using the data from the k-fold cross-validation folds, and the average values were afterward reported. The model produces predictions that are visually represented in the plots, utilizing the data from the validation folds. Furthermore, the residuals of the data from the validation folds are also computed.

Verify The evaluation of model performance:

After the completion of training in Regression Learner, the model that exhibits the highest overall score is presented in the History list. The box exhibiting the maximum score is visually emphasized. The score represents the root mean square error (RMSE) of the validation set. The score assesses the performance of the trained model when evaluated on new and unseen data. The outcome of cross-validation yields a score, which is computed by utilizing the RMSE of all observations, exclusively considering those that were present in the held-out folds. The RMSE calculated on the hold-out observations serves as the metric for evaluating the performance of hold-out validation. The statistical measures employed to assess the performance of the algorithms are presented in Table 4.

Table 4 Model statistics
Full size table
$$RMSE= \sqrt{\frac{1}{n}{\sum }_{i=1}^{n}{\left({TDS}_{{actual }_{i}}-{TDS}_{{predicted }_{i}}\right)}^{2}},$$
(1)
$$R-squared=1- \frac{\sum_{i}{\left({TDS}_{{actual }_{i}}-{TDS}_{{predicted }_{i}}\right)}^{2}}{\sum_{i}{\left({TDS}_{{actual }_{i}}-\overline{TDS }\right)}^{2}},$$
(2)
$$\overline{TDS }= \frac{1}{n}{\sum }_{i=1}^{n}{TDS}_{{actual }_{i}},$$
(3)
$$MAE= \frac{\sum_{i=1}^{n}\left|{TDS}_{{actual }_{i}}-{TDS}_{{predicted }_{i}}\right|}{n},$$
(4)

Results and Discussion

Preprocessing of Data (Stage 1)

In the first step of the research process, the software Arc Map was utilized to create the Soil Augers Map Shape. Subsequently, this shape file was exported to the ERDAS IMAGINE software for the purpose of doing image analysis.

The second step involves the preprocessing of the Landsat image. The study region was covered using an image from the Landsat-8 satellite, which was obtained in January 2018. The image was obtained from the Geologic Survey archive of the United States Geological Survey (USGS) (http://earthexplorer.usgs.gov/), with a specific path (176) and row (39). The Landsat image that was obtained through downloading underwent corrections to account for radiometric and atmospheric effects. Figure 6 depicts the precise reflectance of land cover in the processed image after undergoing essential atmospheric and topography modifications.

Fig. 6
figure 6

Study area Landsat-8 image after radiometric and atmospheric corrections

Full size image

Processing of Image (Stage 2)

Proceeding of the satellite image using ERDAS Envision software was conducted to get the digital numbers corresponding to the salinity auger’s locations and these DNs ranges as shown in Table 5.

Table 5 Bands statistics
Full size table

Machine Learning Algorithms

A statistical technique called regression may be used to assess the strength and kind of a connection between a dependent variable and a number of independent variables. To predict the TDS values in the research region, the regression learner app is utilized in this work to train predictive models such as linear regression analysis, regression trees, Gaussian linear regression, SVM, and tree-based ensemble. Follow the steps below to conduct regression analysis using the regression educator application:

  • Select Regression Learner from the Machine Learning group's applications menu.

  • Import data: latitude, longitude, the bands, and salinity TDS values.

  • Specify independent variables (salinity TDS), dependent variables (bands used), and validation strategy, see Fig. 7.

  • Identify model type. Automated regression model training was employed to automatically train all conceivable regression models—comprising linear regression models, regression trees, Gaussian process regression models, support vector machines, and ensembles of regression trees—on the provided dataset.

  • Evaluate the performance of the models using model statistics (RMSE, R-Squared, and MAE) in order to select the best models. Each prediction is derived using a model that was developed without utilizing the matching observation, and the response plot shows the projected response vs the actual number after training a regression model, see Fig. 8.

  • Exporting the best model in order to predict any new values of salinity TDS for any point in the study area.

Fig. 7
figure 7

Pick data from the workspace to begin a session

Full size image
Fig. 8
figure 8

Explore Data and Results in Response Plot

Full size image

The Landsat-8 satellite, equipped with the operational land imager (OLI), produces a total of nine spectral bands denoted as Band 1 through Band 9. Additionally, it features two thermal bands with a spatial resolution of 100 m for the thermal infrared sensor (TIRS). Leveraging the data derived from the band values of Landsat 8 images, various scenarios were systematically generated to formulate an optimal model characterizing the TDS values associated with salinity in the research region. The selection of bands in our study was carefully guided by their correlation with Total Dissolved Solids (TDS) values, as well as by the errors observed in the models based on these bands. The intention was to prioritize spectral features that demonstrated a strong association with TDS levels, ensuring their relevance to the study's objectives. Additionally, the performance metrics and errors associated were taken into consideration with models utilizing these bands, aiming for a robust and accurate representation of the underlying data. The diverse cases comprised the following approaches:

  1. 1.

    Individual computation of salinity values from each of the 11 bands.

  2. 2.

    Calculation of salinity values by combining bands in groups of 2 or 3, 4 or 5. Utilizing more than five bands to construct the model did not lead to improvement in the results.

  3. 3.

    computation of salinity values from the entirety of the 12 bands.

Each of these cases underwent thorough processing, and in each iteration, several machine learning algorithms were employed to forecast the salinity values of Total Dissolved Solids (TDS) in the designated geographical region. These algorithms encompassed support vector machines (SVM), regression trees, Gaussian linear regression, and tree-based ensemble in addition to linear regression analysis. The model with the lowest Root Mean Square Error (RMSE) was selected as the best-performing model. This selection process was repeated for each case, and the three most optimal models from each instance are detailed in Table 6 for comparison and analysis.

Table 6 Model statistics of the best 3 models for the different cases
Full size table

Table 6 displays the top three salinity models for the study region when utilizing the B & R & S2 with SVM-Quadratic SVM model, the B & T1 & S2 and the B & S2 with Gaussian process regression—Squared Exponential GPR model respectively. The blue band consistently produces the greatest results, which shows that there is a significant relationship between this band's value and salinity. In general, Gaussian process regression (GPR) models give the best models when utilized to predict TDS salinity with the different cases. GPR models are nonparametric probabilistic models based on kernels. The expected responses of the GPR fit throughout the observations when the observations are noise-free. The projected response's standard deviation is almost nil. As a result, the prediction intervals are quite small. When noise is present in the observations, the expected responses do not pass through the data, and the prediction intervals widen. The graphic comparing predicted values to actual values was utilized to assess the performance of the model and evaluate the accuracy of the regression model in predicting various response values. The residual plot illustrates the discrepancy between the observed salinity TDS values and the corresponding projected values for each training data point, as depicted in Fig. 9.

Fig. 9
figure 9

Residuals Plot and predicted vs. actual plot from the model of B & R & S2 bands

Full size image

Models Comparison

According to several studies, there is a strong relationship between the reflectance of the soil and its characteristics, including its salt and moisture concentration, mineral composition, and moisture content (Bannari et al., 2008). Various models have already been suggested in order to determine the salinity values from bands. Six typical salinity spectral indices are employed to calculate soil salinity in the research region, to compare them with the best salinity model developed (the B & R & S2 bands with SVM-Quadratic SVM model). Table 7 illustrates the six common salinity spectral indices and their calculation equations. For the 69 points in the research region, the six indices were computed. The descriptive statistics of the soil salinity indices are shown in Table 8.

Table 7 Spectral salinity indices and their calculation equations
Full size table
Table 8 Accuracy statistics of the soil salinity estimation models
Full size table

According to the results presented in Table 8, in order to determine salinity values from the bands, different models have already been suggested, but the proposed model is more advanced, accurate, and reliable.

Salinity Mapping Using the Best Model

Soil salinity has to be precisely and quickly mapped out as a major environmental issue (Avdan et al., 2022). At this stage, a map was drawn that expresses the salinity values in the study area. The curve fitting app in MATLAB R2019a was used to fit the data from the best model (utilizing the B & R & S2 bands with SVM-Quadratic SVM algorithm) and draw a digital salinity map for the study area, see Fig. 10.

Fig. 10
figure 10

The expected digital map of soil salinity TDS from the best model

Full size image

The Food and Agriculture Organization (FAO) and the United States Department of Agriculture (USDA) use a classification system for assessing soil salinity based on ECe values. The classification is summarized in Table 9, (Shirokova et al., 2000). These classifications provide a guideline for understanding the level of soil salinity and help in making decisions related to crop suitability and management practices. The electrical conductivity values are indicative of the concentration of soluble salts in the soil, with higher values corresponding to higher salinity levels.

Table 9 FAO (USDA) classification used for soil salinity assessment
Full size table

According to the findings presented in Fig. 10 and Table 9, the study region can be delineated into two distinct zones. The initial zone, characterized by dark blue and blue hues, signifies non-saline soil, suitable for the cultivation of highly sensitive crops. The second zone, identified by green, orange, and yellow tones, represents soil with low salinity levels, conducive to the growth of crops sensitive to such conditions. The model, developed utilizing the B, R, and S2 bands through the SVM-Quadratic SVM algorithm, has demonstrated its efficacy in mapping and accurately predicting soil salinity levels across various gradients within the study area. The predictive capacity of the model is reported to be 86%, as indicated by an R-squared value of 0.86.

Conclusions

This investigation employed Landsat 8 satellite imaging data and advanced machine learning techniques to examine soil salinity mapping in the Sharkia governorate of Egypt. The study area is situated within the geographic confines of the semi-arid Nile Delta region. Five machine learning methods, encompassing regression trees, Support Vector Machines (SVM), Gaussian linear regression, linear regression analysis, and tree-based ensemble, were evaluated. Among these, the SVM-Quadratic SVM model, utilizing the B & R & S2 bands, yielded the most favorable results, achieving an R2 value of 0.86 and an RMSE value of 175.98. Notably, the blue band consistently demonstrated optimal outcomes, indicating a robust correlation between salinity and the corresponding value of this band. The findings of this study underscore the efficacy of accurately mapping soil salinity through the integration of Landsat 8 satellite imaging data and machine learning methodologies. The provided data can be used to identify regions characterized by elevated levels of soil salinity and to formulate effective strategies to address this problem. Utilizing remote sensing and machine learning techniques for the purpose of mapping soil salinity provides numerous advantages. The non-destructive nature of this method eliminates the need for soil sampling. In addition, this strategy proves to be a cost-effective method for proficiently and quickly mapping vast geographical regions. In addition, this method is repeatable, allowing for the consistent monitoring of soil salinity across multiple instances. The findings of this investigation have significant implications for the effective management of soil salinity in the Sharkia Governorate region. Utilizing the soil salinity map initially enables the identification of regions with heightened susceptibility to soil salinity problems. In addition, the map could aid in the identification and implementation of targeted mitigation strategies, such as the optimization of drainage and irrigation systems, in regions with the greatest need for such measures. In addition, the use of the map permits the continuous evaluation of the effectiveness of mitigation strategies throughout a given period. This study has effectively demonstrated the capacity of remote sensing and machine learning techniques for mapping soil salinity. Future research efforts should prioritize the improvement of soil salinity mapping precision through the use of sophisticated machine learning algorithms and the incorporation of additional data sources, such as ground measurements and climatic data. It is advisable to employ the models derived from Landsat 8 satellite bands utilizing sophisticated machine learning algorithms for the purpose of mapping soil salinity in Egypt, particularly in the Delta region.