Challenges and Opportunities in Remote Sensing for Soil Salinization Mapping and Monitoring: A Review

: Meeting current needs without compromising future generations’ ability to meet theirs is the only path toward achieving environmental sustainability. As the most valuable natural resource, soil faces global, regional, and local challenges, from quality degradation to mass losses brought on by salinization. These issues affect agricultural productivity and ecological balance, undermining sustainability and food security. Therefore, timely monitoring and accurate mapping of salinization processes are crucial, especially in semi-arid and arid regions where climate variability impacts have already reached alarming levels. Salt-affected soil mapping has enormous potential thanks to recent progress in remote sensing. This paper comprehensively reviews the potential of remote sensing to assess soil salinization. The review demonstrates that large-scale soil salinity estimation based on remote sensing tools remains a signiﬁcant challenge, primarily due to data resolution and acquisition costs. Fundamental trade-offs constrain practical remote sensing applications in salinization mapping between data resolution, spatial and temporal coverage, acquisition costs, and high accuracy expectations. This article provides an overview of research work related to soil salinization mapping and monitoring using remote sensing. By synthesizing recent research and highlighting areas where further investigation is needed, this review helps to steer future efforts, provides insight for decision-making on environmental sustainability and soil resource management, and promotes interdisciplinary collaboration.


Introduction
Soil salinization is a major environmental hazard affecting agricultural productivity and food security worldwide [1].It adversely influences soil structure, nutrient availability, and plant growth, leading to reduced crop yields and, in extreme scenarios, desertification [2,3].The increasing levels are caused by diverse natural and anthropogenic factors, such as inadequate irrigation practices, fertilizer overuse, and land use changes [4].In addition, climate change impacts on soil salinization are a significant concern, with weather patterns playing a fundamental role in increasing salt content around the rhizosphere [5,6].This is particularly noticeable in areas with shallow water tables and degraded groundwater quality [7].Therefore, real-time monitoring of soil salinity levels is essential for effective soil management and sustainable agriculture [8,9].
Subsequently, remote sensing has proven to be an attractive alternative for mapping and monitoring salinization in large-scale and heterogeneous landscapes, especially under different land use and land cover types and areas where socio-culturally different farming cultivation techniques are maintained [10,11].Remote sensing data from satellite imagery and aerial photography offer valuable information on various environmental Remote Sens. 2023, 15, 2540 2 of 20 parameters, including vegetation cover, soil composition, and moisture content, which are interconnected to salt content [12].By analyzing and interpreting such data, researchers and practitioners can generate detailed maps and spatial models of salinity distribution, which inform land management decisions and support the development of effective strategies for risk mitigation [13,14].Over the past few decades, remote sensing has undergone significant advancements, enabling the collection of high-resolution data on various scales [15,16].As research progresses, various tools have emerged [17][18][19], including multispectral imaging sensors which capture information at different wavelengths, leading to more accurate results with higher spatial resolution [20].This enables the extraction of auxiliary data on soil properties such as moisture content, organic matter, and salt content by analyzing the reflected radiation from the surface [21].To map soil salinization, many researchers have used the concept of spectral index, a combination of pixel values from two or more spectral bands [22][23][24].As they rely on the variance in reflectance between the visible (VIS), near-infrared (NIR), and shortwave infrared (SWIR) bands, they can be useful in detecting changes in salt content [25].
In addition to multispectral imaging, synthetic aperture radar (SAR) has recently become one of the most efficient remote sensing tools for soil salinity detection due to its insensitivity to weather conditions, unlike optical remote sensing [26].SAR uses microwave signals to penetrate the soil and retrieve information on soil moisture [27] and structure [28].Therefore, it can generate relevant information on soil's electrical conductivity (EC), which is closely related to its salt content [29,30].The backscatter coefficient (sigma-0), as a measure of the microwave energy reflected back to the sensor, is a commonly used parameter for mapping soil salinity by integrating it into empirical models [31].
The accuracy of salinity mapping has been significantly improved by the fusion of multiple data sources, such as optical and SAR data [32].As optical sensors can capture surface reflectance and vegetation cover, SAR sensors penetrate the vegetation and retrieve information on soil properties [33].Moreover, integrating remote sensing data with other data types, such as those on land cover, land use and topographic features, provides even more accurate estimations [4,34].By combining data from different sensors and platforms, researchers can take advantage of the complementary strengths of each data source and overcome their limitations.
As the field of salinization mapping continues to progress, it has become increasingly evident that integrating remote sensing data with machine learning offers a more robust framework for effectively processing large datasets and generating more accurate products [35].Machine learning algorithms, such as random forest [36] and support vector machines [37], have been remarkably efficient at data processing and analysis, enabling prediction models to learn from the spectral and spatial patterns and produce estimations based on input features [38,39].
Despite the remarkable growth, remote sensing applications for salinization assessment pose significant challenges, including issues with data resolution, spatial and temporal coverage, acquisition costs, data processing and storage.In this regard, a comprehensive review of remote sensing's current state can help identify areas for further research and technological development while providing a valuable resource for researchers, policy-makers, and stakeholders concerned with environmental sustainability and land management.

Remote Sensing for Mapping Soil Salinization
Not only do environmental factors, such as soil type, land use, topography, and climate, play a leading role in salinization expansion, but anthropogenic actions, such as the inadequacy of drainage systems and ineffective irrigation activities over an extended period, also have a direct impact on this dynamic process [4,40,41].In light of this, the availability of spaceborne and airborne platforms has significantly facilitated the monitoring of environmental hazards by providing vast amounts of data that can be applied to diverse fields, from sustainable agriculture, land surveys, and climate change to risk mitigation [42].Enhanced data in terms of spatiotemporal and spectral resolutions offered by these systems have enabled researchers to monitor changes inland and identify salinity patterns at various spatial scales.In addition, when combined with geospatial data, ground-based systems such as the electromagnetic induction instrument (EMI) give valuable insights into the salinization status at both local and canopy scales, allowing policy-makers to gain a more comprehensive understanding of the complex dynamics of salinization at the field level [43].
Integrating remote sensing data with robust analytical techniques has shown great promise in salinization mapping, as suggested by many researchers.A study conducted in Qom Valley in Iraq demonstrated that combining Landsat 8 OLI's spectral indices and topographic features can accurately predict and map soil salinity [44].Further, jointly using Sentinel-2 Multispectral Imager (MSI) data with laboratory measurements to build a machine learning model for soil salinity estimation in the northern margin of the Tarim Basin (China) provided a timeless scientific reference for futuristic scenarios related to salinization expansion in arid areas [45].Field observations, Landsat 5 TM and radar data retrieved from ALOS (Advanced Land Observing Satellite) and PALSAR (Phased Array L-Band Synthetic Aperture Radar) have provided a promising solution for salinity monitoring in central Iraq with lower costs, as suggested by the authors of [46].In the Great Hungarian Plain, the authors of [47] employed spectral indices and principal components derived from Landsat 8 OLI data coupled with multiple linear regression analysis to map salt content distribution in the area.The study proved the potential of multispectral data, with the outperformance of ridge regression, yielding an overall accuracy of 75%.Thus, linear regression modeling using remote sensing-based variables can be significantly effective for locally assessing soil salinity.
As pattern changes in land use and land cover supposedly vary with salinization magnitude, relevant data can be effectively employed to predict soil salinity levels [48,49].In Europe, among several land cover inventories, the CORINE system has solely provided this information for over two decades, which fortunately could be used to map salinization by many researchers [4,50,51].In addition, a study conducted in Dakhla Oasis, located in the western desert of Egypt, showed a discrepancy in soil salinity estimations based on the linear spectral unmixing (LSU) related to land surface temperature over different land cover types and altitudes [52].These findings are consistent with another study conducted in Korat province (Thailand), emphasizing the importance of vegetation cover, soil characteristics, and seasonal fluctuations in mapping soil salinization via remote sensing [53].
Over the past decade, research focus has shifted from traditional, labor-intensive methods of measuring salt content through field surveys and laboratory analysis towards a greater reliance on remotely sensed data often used with limited reference datasets for calibration purposes [54].Based on a qualitative analysis of the Scopus database, we have run an advanced search query to find available peer-reviewed research papers, with the following terms: soil salinization, monitoring, and remote sensing.Significant progress was made in spaceborne and airborne remote sensing systems between 2014 and 2023, which was fundamentally driven by the launch of Landsat 8 in 2013 and the subsequent launches of Sentinel 1 and Sentinel 2 in 2014 and 2015.Figure 1 demonstrates a positive trend in research studies that employed remote sensing for salinization assessment in the same timeframe.However, the sudden drop in 2021 is attributed to a shift in research focus toward other areas or technologies, funding limitations, and reference data unavailability due to the geographic inaccessibility caused by extreme global events such as COVID-19.Given the rise in technology availability and professional knowledge worldwide, this increasing trend is expected to continue in 2023.
Extensive data have been used to assess salinization by measuring the changes in reflectance along the spectrum from visible to infrared (IR) for different salinity levels accompanied by vegetated and sparsely vegetated profiles to distinguish the disparities.Usually, the interaction between soil and spectral energy differs based on the emitted radiation and the surface properties, whereas salt-affected soils often exhibit whitish or grayish crust on the topsoil [55,56].In this regard, the authors of [57] found an increase in reflectance in the visible range, particularly in the blue (450 nm) range with excessively saline soils.Moreover, higher reflectance occurs in the SWIR region (1100-3000 nm), revealing more sensitivity to salt content, as discovered by the authors of [58] and [59].According to the authors of [60], alterations in surface roughness caused by salinity induce shifts in spectral reflectance.Many studies have used SWIR and thermal infrared (TIR) spectroscopy to quantify salt content [61,62].This means various properties influence salt-affected land identification, including soil color, texture and moisture content.Extensive data have been used to assess salinization by measuring the changes in reflectance along the spectrum from visible to infrared (IR) for different salinity levels accompanied by vegetated and sparsely vegetated profiles to distinguish the disparities.Usually, the interaction between soil and spectral energy differs based on the emitted radiation and the surface properties, whereas salt-affected soils often exhibit whitish or grayish crust on the topsoil [55,56].In this regard, the authors of [57] found an increase in reflectance in the visible range, particularly in the blue (450 nm) range with excessively saline soils.Moreover, higher reflectance occurs in the SWIR region (1100-3000 nm), revealing more sensitivity to salt content, as discovered by the authors of [58] and [59].According to the authors of [60], alterations in surface roughness caused by salinity induce shifts in spectral reflectance.Many studies have used SWIR and thermal infrared (TIR) spectroscopy to quantify salt content [61,62].This means various properties influence saltaffected land identification, including soil color, texture and moisture content.
Consequently, hyperspectral remote sensing has facilitated salt characterization by establishing predictive models to estimate salt content more accurately [63] or through creating spectral libraries for various salt types based on their narrow continuous bands [64], which cover the entire spectrum from visible, near-infrared to SWIR and TIR [64].The situation is quite different in vegetated areas since reduced crop growth induced by saline stress causes chlorophyll and other pigment alterations and increases heat emissions.This presents a potential avenue to use thermal remote sensing for qualifying vulnerable areas based on vegetation status, except for halophytes [65].
While remote sensing has provided new opportunities for soil salinization assessment, this task complexity highlights the need for addressing data availability, investigation scale, and mapping approaches, which are discussed in the following section.

Uncovering Data Availability: Investigating the Scope and Scale of Accessible Data
While remote sensing has become a valuable tool for salinization assessment, its suc- Consequently, hyperspectral remote sensing has facilitated salt characterization by establishing predictive models to estimate salt content more accurately [63] or through creating spectral libraries for various salt types based on their narrow continuous bands [64], which cover the entire spectrum from visible, near-infrared to SWIR and TIR [64].The situation is quite different in vegetated areas since reduced crop growth induced by saline stress causes chlorophyll and other pigment alterations and increases heat emissions.This presents a potential avenue to use thermal remote sensing for qualifying vulnerable areas based on vegetation status, except for halophytes [65].
While remote sensing has provided new opportunities for soil salinization assessment, this task complexity highlights the need for addressing data availability, investigation scale, and mapping approaches, which are discussed in the following section.

Uncovering Data Availability: Investigating the Scope and Scale of Accessible Data
While remote sensing has become a valuable tool for salinization assessment, its successful implementation depends on data availability and the investigation scale.Many studies have applied thermal, multispectral, hyperspectral, and microwave sensors to quantify salt content [66][67][68][69][70].As these sensors have varying spectral, spatial, radiometric, and temporal resolution properties, this has a massive impact on the scope and relevance of the study [71].Table 1 illustrates the most common sensors used in soil salinization detection and their properties.
Table 1.An overview of commonly used sensors for salinization assessment and their characteristics, including spatial, spectral, and temporal resolutions, acquisition cost, and applicable mapping scale.The listed sensors range from regional/global-scale sensors such as MODIS to high-resolution pixelplot sensors such as UAV-based ones and are used for mapping at different scales depending on their spatial coverage.Up to this point, these sensors have demonstrated their ability to detect patterns at spatiotemporal scales, thanks to their spatial resolution ranging from a few centimeters to several hundred meters with a revisit time from one day to two weeks [72].Therefore, selecting a remote sensing system for salinization assessment depends on the sensor's technical characteristics and data availability.In this regard, Sentinel missions, i.e., Sentinel-1 and Sentinel-2, have recently dominated the European imaging systems as part of the Copernicus program funded and developed by the European Space Agency (ESA).The Copernicus program is an Earth observation program that examines Earth's surface and its environment for the benefit of the European community and provides free information services [73].Due to their high spatial resolution (ranging from 10 to 60 m) and fast revisit time (ranging from five days for Sentinel-2 and six to 12 days for Sentinel-1), Sentinel products have become a valuable data resource for many Earth observation research projects, including salinization risk management [74].
On the other hand, fewer studies have used hyperspectral data to mainly focus on specific areas, usually distinguished by an extremely saline environment, e.g., [75][76][77].As such, the authors of [78] studied salinity variation across the Yellow River Delta region in China using a combination of laboratory and hyperspectral data retrieved from EO-1 Hyperion.A soil salinity spectral index (SSI) was developed from continuum-removed reflectance (CR-reflectance) in 2052 and 2203 nm to examine the spectral absorption properties of salt-affected soils, yielding a correlation coefficient (R 2 ) of 0.91.The Hyperion reflectance image was processed using the final model, resulting in a quantitative salinity map with an RMSE of 1.921 g/kg and an R 2 of 0.63.The study elucidated the reliability of hyperspectral data in estimating salt content due to their high spectral resolution.This aligns with [79], the authors of which used HJ-1A hyperspectral data to detect topsoil salt components across another study area in China.The research successfully established a robust relationship between salt content and reflectance spectra.
We performed a quick query on the Scopus database to identify the most widely used remote sensing platforms for soil salinization assessment in peer-reviewed research papers.We used the following search keywords: Landsat, Sentinel-2, MODIS, UAV, ASTER, IKONOS, hyperspectral, radar, remote sensing, and soil salinization.Figure 2 illustrates the overall distribution of remote sensing data used in soil salinization-related studies.
papers.We used the following search keywords: Landsat, Sentinel-2, MODIS, UAV, AS-TER, IKONOS, hyperspectral, radar, remote sensing, and soil salinization.Figure 2 illustrates the overall distribution of remote sensing data used in soil salinization-related studies.Optical remote sensing has been widely applied due to its practicality in data processing and storage, as well as its moderately extended spatial coverage and high resolution, which is in alignment with [71].In contrast, fewer studies have used high-resolution commercial sensors such as ASTER (~15 m), IKONOS (~4 m), and UAV-based ones (<1 m), which is consistent with [72].The associated high costs and limited coverage are the primary causes why its application is restricted to narrow geographic regions.Nevertheless, accessing valuable information at different scales is essential in adopting appropriate mitigation strategies.
Although the potential of active remote sensing in assessing soil salinization is well recognized [80], the search yielded fewer research papers, reflecting the challenges of limited access to such data.A study conducted in Keriya River Basin, Northwest China, highlighted the efficacy of active remote sensing to map soil salinity by applying an integrated threshold of backscattering values from PALSAR and Radarsat-2 polarimetric images.The research yielded enhanced results in separating moderately and extremely saline soils [81].Further, another study by the authors of [33] used ERS-1/2 and Sentinel-1 SAR timeseries data acquired between 2000 and 2018 to identify salinity levels in Hortobágyi National Park in Hungary, one of the most naturally alkaline environments in Europe.The research demonstrated the usefulness of SAR data in salt detection with an outperformance of Sentinel-1 SAR products over ERS-1/2 due to its higher resolution and its operation with a dually polarized antenna which could identify better subtle changes in salinity levels.
Based on Figure 2, limited research focused on radar and hyperspectral data compared to multispectral data application (i.e., MODIS, Landsat, and Sentinel 2 MSI).This can be associated with many factors, including data availability and resolution, spatial Optical remote sensing has been widely applied due to its practicality in data processing and storage, as well as its moderately extended spatial coverage and high resolution, which is in alignment with [71].In contrast, fewer studies have used high-resolution commercial sensors such as ASTER (~15 m), IKONOS (~4 m), and UAV-based ones (<1 m), which is consistent with [72].The associated high costs and limited coverage are the primary causes why its application is restricted to narrow geographic regions.Nevertheless, accessing valuable information at different scales is essential in adopting appropriate mitigation strategies.
Although the potential of active remote sensing in assessing soil salinization is well recognized [80], the search yielded fewer research papers, reflecting the challenges of limited access to such data.A study conducted in Keriya River Basin, Northwest China, highlighted the efficacy of active remote sensing to map soil salinity by applying an integrated threshold of backscattering values from PALSAR and Radarsat-2 polarimetric images.The research yielded enhanced results in separating moderately and extremely saline soils [81].Further, another study by the authors of [33] used ERS-1/2 and Sentinel-1 SAR time-series data acquired between 2000 and 2018 to identify salinity levels in Hortobágyi National Park in Hungary, one of the most naturally alkaline environments in Europe.The research demonstrated the usefulness of SAR data in salt detection with an outperformance of Sentinel-1 SAR products over ERS-1/2 due to its higher resolution and its operation with a dually polarized antenna which could identify better subtle changes in salinity levels.
Based on Figure 2, limited research focused on radar and hyperspectral data compared to multispectral data application (i.e., MODIS, Landsat, and Sentinel 2 MSI).This can be associated with many factors, including data availability and resolution, spatial coverage, acquisition costs, and processing and storage infrastructures since those products require high processing and storage capacities.

Local Scale
Remote sensing has proven valuable for accurately and timely detecting excessive soil salinity, enabling automated and reproducible monitoring of salinization processes in agricultural systems.This is fundamental for soil conservation and agricultural productivity at the local scale.Notably, data retrieved from thermal and hyperspectral sensors have shown particular sensitivity to saline stress, making them advantageous in characterizing vegetation conditions and estimating salt content [82].By capturing synoptic information on various physiological processes under saline conditions, such as thermal energy release and canopy health and vigor, they portray the situation on the surface [83].This information helps detect minimal changes in the landscape early on, allowing prompt interventions to avoid soil fertility and biodiversity losses [84].Intensive research has focused on studying salinity distribution on the field scale [85][86][87].Although each study has been conducted in a specific spatiotemporal context on different cropping systems from rice to olive trees and in diverse geographical locations (i.e., Portugal, China, and Tunisia) under different environmental conditions, from semi-arid to irrigated systems, indicating the versatility and applicability of remote sensing under various agricultural settings.The common conclusion is that remote sensing data, specifically satellite imagery with relatively high spatial resolution, i.e., Landsat and Sentinel-2, can be used to assess salinity in agricultural lands with high accuracy when combined with electromagnetic induction techniques to measure reference values.
The operational capabilities of remote sensing at the farm scale have significantly improved due to the increasing availability of high-resolution platforms such as UAVs [88].In recent years, UAVs have shown their utility in improving accuracy and providing more insights into intensive soil monitoring [89].Likewise, imaging spectroscopy techniques, such as matched filtering (MF) and mixture tuned matched filtering (MTMF) with multispectral advanced spaceborne thermal emission and reflection radiometer (ASTER) images, have been successfully used to map saline soils and determine crop yield reduction and land degradation caused by salinization at the farm level [90].
Integrating remote sensing has become a fundamental component of precision agriculture, enabling the detection and monitoring of changes in salinity levels at the plot scale.Accordingly, accurate monitoring can support soil conservation efforts and mitigate the adverse impacts of climate change on agriculture.By optimizing the use of resources and inputs offered by remote sensing, farmers can effectively manage salt-affected lands and optimize crop yield while contributing to sustainable agriculture practices.

Regional Scale
As a practical and cost-effective tool, remote sensing generates valuable information about salinization risks and vulnerable areas requiring intervention, efficiently covering large areas and detecting hotspots that traditional field-based surveys fail to cover [91].Recent research has well established that incorporating it with field surveys can provide a more comprehensive understanding of salinization dynamics, at least on a smaller scale, due to the coarse spatial resolution of most satellite imagery products.As such, many studies have employed it to regionally assess salinization under different scenarios, from complex agricultural systems to drylands [92,93].A study by the authors of [94] used imaging spectroscopy data from an SR-3500 spectrometer combined with machine learning to map soil's electrical conductivity in the northern Yinchuan Plain (China).Among the six models, the extremely randomized trees-based model performed the best, with R 2 values of 0.96 and 0.98.Another study by the authors of [95] assessed soil salinity in San Joaquin Valley (CA, USA) using multi-year Landsat 7 ETM+ data in combination with canopy reflectance and the canopy response salinity index (CRSI).The research yielded an estimation model with R 2 values of 0.61 and 0.73, where CRSI, crop type, rainfall, and temperature were the most influential variables.This further proves that recent research efforts have provided significant theoretical support for salinization mapping and monitoring at the regional scale.
Nonetheless, a more comprehensive understanding of salinization dynamics will enable targeted intervention strategies to mitigate its risks and keep it under control [96].Despite the advantages brought on by remote sensing-based approaches in this regard, fundamental challenges must be addressed to ensure its valid representativeness.Not only must the spatial and spectral resolutions of sensors be carefully selected to ensure that the data obtained are suitable for the study scale, but atmospheric interference and calibration issues should be addressed to optimize the product's accuracy [97].To fully harness the potential of remote sensing, it is crucial to develop robust techniques that can address the associated challenges and ensure the accuracy and validity of acquired data.By doing so, we can access reliable and up-to-date information to develop effective strategies to mitigate soil salinization impacts inland.

Global Scale
Salinization is expanded over more than 1 billion hectares worldwide and constantly increasing, according to FAO and ITPS report [98].Other estimations of total salt-affected lands differ somewhat substantially.According to [99], the disparity in current estimates of saline soils is frequently caused by discrepancies in methodology for data collection and analysis; hence, only a rough estimate can be provided.
Currently, the Harmonized World Soil Database represents the most popular database that delivers a global coverage of soil salinity data [100].While this database represents a valuable resource for global studies, it has a few limitations to consider, such as the discontinuity of pixel values, the coarse spatial resolution (>1 km), and outdated information on soil salinity, with its most recent version (v1.2) released in 2012 [101].A lack of spatial resolution and accurate data highlights an urgent need for an updated worldwide soil salinity map.
To address this issue, the Global Soil Partnership (GSP) and the Food and Agriculture Organization (FAO) have initiated efforts to create a more comprehensive global map of salt-affected soils, the Global Map of Salt-affected Soils (GSASmap V1.0.0), which was released in 2020 [102].The product includes contributions from over 118 countries, with more than 350 national experts involved in the harmonization process [103].Each country subsequently generated its maps following approved technical standards set by FAO.The map illustrates SAS spatial distribution at the topsoil (0-30 cm) and subsoil (30-100 cm).According to GSASmap, which covers 85% of the worldwide land surface, salinization affects around 424 million ha of topsoil and 833 million ha of subsoil [104], with nearly two-thirds of global SAS falling under arid and semi-arid climates [105].
Despite the progress made, monitoring salinization is an ongoing process, and there is still much to be done to improve the accuracy of available maps [106].This emphasizes the necessity for further research and international collaborations to address this growing issue in real-time and promote sustainable land management practices.

Mapping Approaches
Although diverse approaches are available for generating accurate salt content estimates, including in situ and laboratory analysis, they are time-consuming and require intensive sampling and enormous labor costs, making them impractical for larger-scale studies [107].Alternatively, indirect methods such as remote sensing can provide relatively useful information at lower costs based solely on electromagnetic energy, enabling continuous monitoring of saline environments at a broader range [108].Recent progress in machine learning and artificial intelligence has tremendously supported the establishment of more robust digital soil mapping approaches, providing reliable predictive tools for salinization assessment [109].
Substantially, changes in land cover due to excessive salinity, such as deterioration of soil structure, organic matter loss, changes in water balance, and loss of biodiversity, result in detectable differences in reflectance characteristics, which can be detected by sensors in high to extreme scenarios [110].The most commonly used methodologies can be broadly categorized into three main types: statistical models, machine learning algorithms, and physical models.Statistical models such as those based on linear regression [111,112] use an empirical approach to analyze the correlation between variations in remote sensing derivatives such as spectral indices and principal components with field measurements.These models are simple and easy to develop but fail to capture complex interactions between variables.As a result, this affects the methodology's replicability since there may be other factors involved that interact in synergy or solely, which are not captured by simple estimation models.Therefore, the model fails to yield accurate results without considering these influential indicators when replicating the research workflow.In this context, hybrid modeling techniques are increasingly integrated for salinity prediction [113,114], particularly in poorly sampled locations.These methods combine the strengths of various models and sensors to generate more accurate and reliable estimates.Geostatistical models such as co-kriging and regression kriging rely on a spatial correlation, assuming that locations closer to each other have similar properties, while a regression analysis between two variables or more is exploited to predict the dependent variable's distribution in lowly sampled areas [4,115].Although hybrid models' main goal is to overcome individual models' limitations by combining them, their effectiveness highly depends on the nature of the data, the significance of the correlation between covariates, and the research question.Thus, a hybrid technique application becomes irrelevant if no significant correlation exists between the model covariates.Consequently, it is crucial to carefully select and combine the models based on the research scope and available data to obtain the most accurate and reliable estimations.For instance, cubist models are based on a hybrid approach that combines regression analysis and decision tree modeling techniques to enable accurate predictions based on input data while employing boosting with multiple training data points to improve accuracy and balance the variables' weights [116].Through generating a set of rules to combine input variables, each rule is associated with a linear model [117].This method involves constructing a series of decision trees, each with adjusted weights, to produce a model that accurately reflects the patterns in the data [118].It can be advantageous in cases where the relationship between input variables and the predicted outcome is complex and difficult to discern using traditional statistical models such as soil salinity modeling [119,120].Nevertheless, this is quite different for geostatistical models [121][122][123], although they rely on statistical assumptions about underlying spatial variability to estimate salinity at unsampled locations.Some geostatistical methods, such as the stochastic simulation technique [124], can be considered process-based modeling, as it involves using physical parameters and equations to estimate the parameter of interest.On the other hand, machine learning-based models [125], such as neural networks [126], support vector machines [127], random forests [128], and extreme gradient boosting [129], use advanced mathematical models to analyze enormous and complex datasets for prediction.These models can be computationally intensive and may not be as easy to interpret as most statistical models.Physical models, such as the soil, vegetation and atmosphere transfer (SVAT) model [130], are process-based and numerical models that simulate the physical processes controlling salt accumulation and vegetation growth.While they can be computationally demanding, they provide a more in-depth understanding of the underlying physicochemical mechanisms at play.Their simulation capabilities allow physical models to test various hypotheses and scenarios that may not be directly observable in the field.A comparative study conducted by the authors of [131] between a physical model and three ML models, including distributed random forest (DRF), gradient boosting machine (GBM), and deep learning (Deeplearning) for salinity estimation at the canopy scale, found that machine learning-based models have predictive power similar to physical-based models; however, their performance primarily depends on the prediction scenarios and input variables.In the coastal rural areas of Bangladesh, research was carried out by the authors of [132] to explore the potential of salinity using Landsat 8 OLI data.The study used various vegetation and salinity indices in a linear regression analysis-based approach to determine the statistical association between these indices and ground-measured electrical conductivity to yield a low correlation between the ground EC and the pixel values of generated maps, suggesting that the indices are not sufficient to assess salinity.This eventually contradicts other research work carried out in the same context, which suggests differently.In the oasis lands of Egypt, a study conducted by the authors of [133] to map salinity using different statistical models based on Landsat 8 OLI and imaging spectroscopy data revealed that the used spectral indices had low to moderate correlations with EC values, with an R 2 ranging between 0.27 and 0.64.Although this research recommended Landsat-based spectral indices to produce spatial distribution maps, a further investigation is suggested to explore the produced models' dependencies and their validity under other climatic conditions.In the Ebinur Lake region in China, a bootstrap hybrid machine-learning framework was established by combining Sentinel-2 MSI data and environmental covariates [134].The research initially compared four machine learning methods (i.e., bagging, classification and regression tree, random forest, and gradient boosting regression tree (GBRT)) to conclude that the models driven by spectral information and environmental covariates explained up to 88% of data variability, with the superiority of GBRT.The proposed approach offers a soil salinity mapping strategy with a 10 m resolution and high accuracy in poorly sampled locations, which can eventually help in future land restoration projects.
Although several remote sensing methods have been proposed for determining soil salinity, no widely accepted standard can consistently generate accurate data across diverse environmental conditions [135].The accuracy of these methods can vary significantly across different regions, indicating the challenge of creating a globally harmonized data system.Consequently, producing a comprehensive and reliable global soil salinity map remains complicated.
Ideally, combining the abovementioned approaches and fusing multi-source data allow extracting the maximum amount of information from remotely sensed data while considering the complex interactions between environmental factors.The choice usually depends on the research scope, the available data, and the computational resources.Table 2 provides an overview of the most commonly used methods for salinization assessment.[148,149] Table 2 summarizes the main approaches used in recent studies, encompassing the estimated accuracy range, strengths, and weaknesses of each method.Deep learning and artificial neural networks have the highest estimated accuracy ranges of 90-95% and 80-95%, respectively.While multiple linear regression and spectral indices have significantly lower accuracy ranges of 60-80%, linear regression has the lowest expected accuracy.This can be explained by the fact that deep learning approaches focus more on the complex nonlinear connections between covariates, which is more realistic than the assumption of linearity.Researchers must consider the available data, investigation scale, and desired accuracy to choose an appropriate mapping approach for salinization assessment.The specific characteristics of the study area, such as vegetation cover, land use, and soil type, should also be studied.
It is crucial to address the challenges associated with this topic, including data availability, spatial resolution, and investigation scale.By considering the strengths and weaknesses of different methodologies, researchers can build more practical tools for assessing soil salinization based on remote sensing data.

Challenges in Salinization Mapping
Soil salinization assessment via remote sensing poses several challenges that must be addressed to deliver reliable information on a spatiotemporal framework.Substantially, the remote sensing community faces a trade-off between data quality, spatial coverage, acquisition costs, and high accuracy [150].One of the primary challenges is the limited spatial coverage.As most sensors have low spatial coverage, this has restricted soil salinization studies mainly to a local scale [151].
When tracking changes in soil salinity over an extended period, the limited temporal coverage of remote sensing data presents another challenge.Since salinity can frequently fluctuate in response to various factors, such as climate variability, irrigation practices, and land management activities [152], it must be continuously evaluated to identify the causes and impacts in the long term.Collecting and comparing remote sensing data at different time intervals allows the detection of patterns and trends in soil salinization, enabling a better understanding of its dynamics for more effective intervention strategies to be developed.In the Yellow River Delta, the authors of [153] established a novel remote sensing monitoring index of salinization based on a three-dimensional feature space model using Landsat data.The research showed an increasing trend in salinization intensity between 1984 and 2022 due to the inadequate agricultural systems adopted.The authors of [154] extracted Sentinel-2 MSI-based indices to monitor soil salinity in a typical saline zone in the Weigan River-Kuqa River Delta Oasis.The study proved that using remote sensing-based monitoring models in this context is fundamental to comprehensively grasp the salinization magnitude and enhance land management practices on the watershed scale.Therefore, the concept of temporal variation is fundamental for the success of remote sensing-based monitoring efforts, as proven by these studies.
Landsat missions have continuously acquired data since the 1970s, providing extensive temporal coverage for analyzing land use and cover changes [155].This can be leveraged to monitor salinization and even predict future trends based on past data.In comparison, other space missions, such as MODIS and Sentinel 2, have had less archived data since 2002 and 2015, respectively.Although they provide relevant information, more extended coverage remains essential for temporal analysis-related studies, given the climate and land cover changes that have gradually accelerated salinization expansion in some regions in recent decades [156].
While hyperspectral data have shown higher efficiency [157], they are limited in use compared to multispectral data, which have lower efficiency but broader spatial coverage [158].The application of hyperspectral data is constrained by their high cost, low temporal resolution, and restricted spatial coverage.These limitations make it difficult to develop comprehensive and accurate salinization maps that could inform and assist in developing soil management strategies.Nevertheless, the emergence of new generations of hyperspectral satellites such as the Italian PRecursore IperSpettrale della Missione Applicativa (PRISMA), American Hyperspectral InfraRed Imager (HyspIRI), Japanese Hyperspectral Imager Suite (HISUI), and German Environmental Mapping and Analysis Program (En-MAP) is revolutionizing soil mapping by providing higher spatial and spectral resolution data [159][160][161].With their advanced capabilities, these satellites can offer a wealth of information on the composition and properties of the land surface, allowing more detailed and accurate soil salinity mapping.
Additional challenges may arise in some regions, such as those with dense vegetation and high cloud cover [162].Under dense vegetation cover, remote sensing data acquisition and accuracy can be limited, with the obstructed view of the soil surface making it impossible to directly detect salt content variations from bare soil, reducing mapping accuracy.Similarly, cloud cover can hinder the acquisition of optical data, resulting in limited temporal coverage and mediocre accuracy, which can be addressed by implementing radar data instead to avoid the atmospheric effects associated with optical data [163].
Validation based on ground truth data can improve salinization products derived from remote sensing, but obtaining reliable field data can be complex due to factors such as the heterogeneity of soil properties and salinity distribution, the lack of uniformity in the methods used to collect and analyze soil samples, and the inaccessibility of some locations [164].Standardized methods for collecting and analyzing field data are essential to account for soil heterogeneity and salinity distribution across various regions.This task requires the involvement of soil experts and environmental scientists with local knowledge to ensure the representativeness of field surveys.
Moreover, data processing and analysis pose another challenge to accurately assessing soil salinity spatiotemporal distribution.With the enhancement of data quality retrieved from sensors, the amount and dimension of generated information can reach the terabyte scale, making it necessary to adopt efficient methods for processing, analysis, and storage without affecting data quality [165].Integrating multi-sensor data should also be further explored to improve the accuracy of salinization maps, though it requires supplementary efforts in terms of data processing and management, as well as high-performance computing resources.Addressing these challenges involves establishing processing algorithms that meet the operational requirements of soil experts and remote sensing scientists [166].
Despite these challenges, remote sensing offers opportunities to better monitor and map soil salinization.Establishing new data processing and analysis methods, adopting standard variables for monitoring, and integrating multi-sensor data may provide an opportunity to improve our ability to assess salt-affected lands.Further, recent research has demonstrated the potential of using composite images in soil mapping by combining information from multi-temporal data [167].By integrating temporal information with spectral data, such as surface reflectance values obtained from remote sensing, researchers have developed accurate models for mapping soil attributes [168].Composite images have several advantages over single-date images, including noise reduction, the ability to capture the dynamic nature of soil processes [169], and identifying long-term trends and patterns.Thus, it is worth investigating in future studies for soil salinity mapping.
Furthermore, cloud-based systems, such as the Google Earth engine and Microsoft's Planetary Computer, have granted an opportunity to manage and analyze large amounts of data and track temporal changes in soil salinity [170,171].These systems are timeefficient for time series data processing and interpretation, which are difficult to fulfill using conventional computing methods.As such, cloud-based systems enable easier collaboration and data sharing among researchers and policy-makers.For soil salinity monitoring, they merit further investigation to explore their full potential.
Developing innovative remote sensing tools, such as imaging spectroscopy and UAVbased systems, has tremendously improved our capacity to conduct a more thorough and detailed local assessment.Specifically, UAV technology has emerged as an attractive option for acquiring high-resolution data at the field scale and upscaling satellite images covering larger geographic areas [172,173].Future work should examine the efficiency and potential of recent technologies in validating the findings retrieved from space-borne remote sensing systems.

Conclusions
The significance and relevance of salinization assessment using remote sensing in the scientific community were assessed in this review paper.The number of research papers published on this topic increased over the last decade, highlighting the need for further development.This review underlines the advantages of remote sensing tools in salinization mapping but also emphasizes the need for continued research and technological development to improve the accuracy and effectiveness of salinization products to achieve sustainable land management practices.
While hyperspectral data outperformed multispectral data regarding salt-affected land characterization and detection due to its higher spectral resolution, less research has been carried out using commercial sensors, including most hyperspectral instruments, due to their high costs, limited spatial coverage, and restricted public access.
Given the limitations of current remote sensing systems, it is essential to investigate alternative options.With the growing accessibility of open-access data, such as Sentinel-1 synthetic aperture radar (SAR) data, radar remote sensing has emerged as a promising approach for salinization mapping.Microwave sensors have the potential to provide valuable information due to their capacity to penetrate vegetation and measure soil moisture content, which has been demonstrated to have a strong correlation with salinization processes.
Data retrieved from high-resolution sensors have shown their full potential in identifying plant canopy conditions and detecting soil moisture and salinity-induced stress at the plot and local scale.Therefore, further research should prioritize regional monitoring at the landscape scale to maintain environmental sustainability and explore the efficiency of emerged hyperspectral remote sensing systems such as EnMap, PRISMA, and UAVs to validate the findings from open-access satellite data.
Finally, it should be noted that no universally recognized approaches have been established for estimating soil salinity using remote sensing that could be applied to multiple scenarios and still produce accurate data under various climatic conditions.

21 Figure 1 .
Figure 1.Progress of remote sensing application in salinization mapping based on available studies in Scopus between 2014 and 2023.The yearly number of studies using various sensors has significantly increased from 14 in 2014 to 81 in 2022.The availability of satellite data and the development of sophisticated instruments with higher spectral and spatial resolutions have contributed to this growth.

Figure 1 .
Figure 1.Progress of remote sensing application in salinization mapping based on available studies in Scopus between 2014 and 2023.The yearly number of studies using various sensors has significantly increased from 14 in 2014 to 81 in 2022.The availability of satellite data and the development of sophisticated instruments with higher spectral and spatial resolutions have contributed to this growth.

Figure 2 .
Figure 2. Distribution of data types used in accessible research papers related to soil salinization assessment.Remote sensing data selection considers several factors, such as cost, accessibility, spatial and spectral resolutions, and the research scale.Available studies have used various instruments, such as multispectral, hyperspectral, and microwave sensors.Open-access satellite data has gained popularity among researchers recently due to their cost-effectiveness and widespread availability, unlike those of commercial platforms.

Figure 2 .
Figure 2. Distribution of data types used in accessible research papers related to soil salinization assessment.Remote sensing data selection considers several factors, such as cost, accessibility, spatial and spectral resolutions, and the research scale.Available studies have used various instruments, such as multispectral, hyperspectral, and microwave sensors.Open-access satellite data has gained popularity among researchers recently due to their cost-effectiveness and widespread availability, unlike those of commercial platforms.

Table 2 .
Comparison of modeling techniques for soil salinization mapping based on remote sensing data.This table provides an overview of commonly used modeling techniques for remote sensing data analysis, their estimated accuracy range, and their strengths and weaknesses.The estimated accuracy range of each method is based on the existing literature and may vary depending on the specific application and data used.