Introduction
The Mekong Delta (MK) has endured two significant droughts and saltwater intrusion events in 2016 and 2020. In 2016, saltwater intrusion affected approximately 160,000 hectares of rice and vegetable fields. The 2020 event, however, was even more severe, causing wide-spread devastation, particularly in the provinces of Ca Mau and Ben Tre. According to the Ben Tre Department of Science and Technology (2023), saltwater intrusion in the province results from the interaction of natural and geographical factors. The fan-shaped terrain gradually slopes toward the sea, causing approximately 94.2% of the total provincial area to be affected by tidal activity, which facilitates the deep penetration of saltwater into inland areas. The region’s hot climate, low annual precipitation, and prevailing monsoon winds exacerbate the extent of saltwater intrusion. Additionally, the dense network of rivers and canals, combined with a significant reduction in freshwater inflow during the dry season, further intensifies seawater intrusion. The semi-diurnal tidal regime of the East Sea, characterized by large tidal variations and high tide levels, contributes to the worsening of this phenomenon. Furthermore, wave action from the Northeast, East, and Southeast, along with the province’s flat topography and weak river flows, facilitates the inland movement of saltwater, exacerbating its impact on agricultural activities and water resources. In Ben Tre, the 2019–2020 saltwater intrusion had a profound impact on all aspects of daily life and agricultural activities. The agricultural sector suffered substantial losses, with 5,400 hectares of Winter-Spring rice destroyed and nearly 28,000 hectares of fruit orchards severely damaged. Furthermore, approximately 87,000 households experienced acute drinking water shortages. The total economic losses in the agricultural sector alone were estimated at 1,660 billion VND (IPCC, 2023; Ben Tre DARD, 2022; MONRE, 2020). Recent studies reveal that saltwater is increasingly infiltrating groundwater systems, exacerbating the vulnerability of this critical resource and heightening the risk of soil salinization in agricultural areas. This escalating crisis poses a serious challenge to the region’s capacity to meet rising freshwater demands, jeopardizing the sustainability and future development of agriculture in the MK as a whole and in Ben Tre province in particular (Smajgl et al., 2015; Tran et al., 2020).
To elucidate the intricate interplay between saltwater intrusion, water resource degradation, and agricultural risks in Ben Tre province during the period 2014–2023, this study employed the DRASTIC method, modified by substituting the “depth to water” parameter with “electrical conductivity,” to evaluate the susceptibility of groundwater to salinization. Furthermore, the NDVI was employed to examine changes in agricultural land use, while the GWR was applied to examine the spatial variability in the relationship between saltwater intrusion and changes in agricultural vegetation.
Various methods have been developed to assess the risks of saltwater intrusion and can be grouped into three main types including process-based methods, statistical methods, and overlay/index methods. Process-based models use complex simulations to predict the movement of saltwater within aquifers but often require substantial data, which is frequently estimated through indirect means, limiting their application in data-scarce areas (Barbash and Resek, 1996). Statistical methods depend on correlations between groundwater quality and environmental factors, proving effective in identifying patterns in specific regions but lacking general applicability due to their regional specificity (Babiker et al., 2005). Overlay index methods, which incorporate factors influencing the infiltration of saltwater into the groundwater by determining the soil’s EC index, are widely favoured for their reliance on readily available data, making them suitable for large-scale regional assessments (Kumar et al., 2014).
Among these, the DRASTIC method stands out as a powerful and widely utilized tool for assessing vulnerability to saltwater intrusion. This method has been applied in both its original and modified forms in various regions worldwide (Neshat et al., 2014; Yin et al., 2013; Bai et al., 2012; Akhavan et al., 2011). Its effectiveness is attributed to the use of readily accessible data and the integration of multiple input data layers, which minimizes the impact of individual parameter errors on the final outcomes (Thapinta and Hudak, 2003; Yin et al., 2013). In this study, the adjusted DRASTIC values are classified into risk levels (very low, low, medium, or high) using ArcGIS’s quantile classification method to create a saltwater intrusion risk map for Ben Tre Province. However, a major limitation of the DRASTIC method is its inherent subjectivity in assigning numerical values to descriptive entities and determining the relative weights of different attributes (Chitsazan and Akhtari, 2009; Gogu et al., 2003; Hamza et al., 2015). To address these limitations, the results of the DRASTIC method will be compared with those derived from the classification of agricultural land-use objects based on NDVI calculations. The NDVI, a widely used remote sensing index (Bhandari et al., 2012), is strongly correlated with the severity of saltwater intrusion and drought (Wardlow et al., 2007; Drisya & Roshni, 2012). As water resources decline, vegetation health and density also decrease, as reflected by lower NDVI values. Through NDVI analysis, areas with low NDVI values indicate reductions in both surface water and vegetation cover, whereas areas with high NDVI values reflect conditions that are maintained or improved.
Alternatively, this study employed the GWR model, a robust spatial analytical tool, to explore the intricate relationship between saltwater intrusion and changes in agricultural land use. GWR dynamically assigns spatially weighted correlations between variables, reflecting the geographical distance between independent and dependent variables. This approach enhances the representation of spatial heterogeneity in the data, providing a more nuanced analysis compared to traditional regression methods (Fotheringham et al., 2002). Evaluating the performance of spatial regression models for datasets with inherent spatial variability is crucial. Research indicates that GWR consistently outperforms the Ordinary Least Squares (OLS) regression in modeling complex spatial phenomena, offering superior predictive capabilities (Brown et al., 2012; Pratt and Chang, 2012; Foody, 2003). Moreover, GWR is particularly effective in assessing saltwater intrusion, as it allows for the precise delineation of affected areas and quantification of the severity of intrusion. This capability significantly aids in planning agricultural land use and managing water resources. By not only improving predictive accuracy but also uncovering spatial heterogeneity, GWR provides critical insights that empower researchers and policymakers to make informed, region-specific decisions.
Therefore, the objective of this study is to evaluate agricultural land use changes (ALUC) by classifying the NDVI index for the period 2014–2023, utilizing the DRASTIC model to construct a saltwater intrusion risk map, and ultimately applying the GWR model to analyze the spatially varying relationship between intrusion risk levels and changes in agricultural land use patterns. This approach enables an assessment of the impacts of saltwater intrusion on agricultural land use changes in Ben Tre province over time. The integrated methodology allows the province to identify vulnerable areas for agricultural development and formulate adaptive strategies to address saltwater intrusion challenges.
Research Methods
Fig. 1 presents the methodological framework of the study, which integrates remote sensing, GIS, and advanced modeling techniques to assess salinity intrusion risks on agricultural land use changes in Ben Tre Province (2014–2023). Landsat 8 imagery, NDVI analysis, and ISODATA classification were utilized to detect land use dynamics, while the GRASTIC model was applied to map salinity intrusion risk. The spatial correlation between salinity intrusion and agricultural land use changes was examined using the GWR model, with OLS adjustments to enhance analytical precision.
Study area and data collection
Ben Tre Province, located in Vietnam’s MK, covers an area of 2,379.7 km2, accounting for 5.8% of the total MK. The province consists of three major islets, An Hoa and Bao Minh, shaped by the confluence of four significant rivers: Tien, Ba Lai, Ham Luong, and Co Chien. The topography is predominantly flat, with an average elevation ranging from 1 to 2 meters, gradually declining from the northwest to the southeast. Land use is dominated by agriculture, which occupies 75.64% of the total area, followed by non-agricultural land (23.8%) and unused land (0.56%).
As a coastal province, Ben Tre is highly susceptible to saltwater intrusion, particularly during the dry season from November to June. Salinity intrusion occurs via major river mouths, including Co Chien, Ham Luong, Ba Lai, and Cua Dai. Salinity levels of 4‰ have been recorded as far as 60 to 70 km inland, with 1‰ salinity affecting nearly the entire province. This severe and prolonged intrusion has caused significant agricultural losses, estimated at 1,794 billion VND, severely impacting local farmers’ livelihoods and the regional economy (Ben Tre DOST, 2023; MONRE, 2020; Eslami et al., 2019).
This study utilized Landsat 8 OLI satellite imagery provided by the United States Geological Survey (USGS) via http://earthexplorer.usgs.gov/ (Table 1) for processing and analysis based on the NDVI index to identify ALU types and evaluate their transformations. Imagery from February 2014, representing the pre-salinity intrusion period, and December 2023, reflecting severe salinity intrusion conditions, was analyzed to assess changes in ALU types over time.
Secondary data
Using the 1:50,000 scale digital mapping database provided by the Ben Tre Department of Science and Technology (Ben Tre DOST), it was used to develop a salinity intrusion risk assessment map for the study area, aiming to accurately identify regions highly vulnerable to salinity intrusion risk. The datasets employed in this analysis include maps of electrical conductivity, recharge capacity, aquifer media characteristics, soil media properties, topography, vadose zone impacts, and aquifer hydraulic conductivity.
Primary data
To support image interpretation and processing for identifying salinity intrusion conditions, field surveys were conducted in December 2023 and April 2024. A total of 200 samples were collected, encompassing information on various land-use types, including rice, cash crops, fruit trees, coconut, mangrove forests, aquaculture, and urban land within the study area.
Application of NDVI for assessing agricultural land-use changes under salinity intrusion
NDVI is a widely recognized indicator calculated from satellite imagery, such as Landsat 8 data from 2014 and 2023, to quantify vegetation health and coverage on Earth’s surface. NDVI values range from −1 to +1, where values closer to +1 indicate robust and healthy vegetation, while values near 0 or negative suggest vegetation degradation or absence (Tucker, 1979; Pettorelli et al., 2005; Gitelson et al., 1996).
This index is extensively utilized to monitor environmental dynamics, particularly the impacts of salinity intrusion on vegetation. As salinity intrusion depletes soil fertility and reduces its water-holding capacity, vegetation health diminishes, reflected in a decrease in NDVI values (Singh et al., 2010; Mafi-Gholami et al., 2019). In coastal regions heavily affected by salinity intrusion, temporal fluctuations in NDVI offer clear insights into the degradation of agricultural ecosystems (Huete et al., 2002; Joiner et al., 2013).
By leveraging spatiotemporal changes in NDVI, this study effectively captures vegetation dynamics over time and assesses the vulnerability of agricultural crops to salinity intrusion. This approach not only highlights the extent of salinity’s impact but also provides critical information for mitigating risks to agricultural sustainability (Zhou et al., 2022; Kogan, 1995).
NDVI calculation and classification of agricultural land-use categories
Landsat 8 satellite imagery provides two critical spectral bands: the red band (Red) and the near-infrared band (NIR). These bands are utilized to compute the NDVI for each pixel, based on the range of NDVI density slices corresponding to different ALU types (Table 2). The calculation is performed using ArcGIS software, following the established formula (Gorelick et al., 2017).
The unsupervised ISODATA classification method was applied to determine mean threshold values derived from field samples, corresponding to various agricultural land-use types distributed spatially within the study area. The thresholding process utilized a minimum distance classification technique, which involved splitting, merging, and removing classes based on specified input threshold parameters (Jensen, 1996; Richards & Jia, 2006). ALU classifications derived from NDVI imagery for 2014 and 2023 were directly compared to assess temporal changes using the following formula:
Where:
ALUC represents agricultural land use changes, ALU2023 and ALU2014 represent agricultural land use in 2023 and 2014
The results derived from Equation (2) indicate changes in agricultural land use during the 2014–2023 period. Positive ALUC values reflect improvements, whereas negative ALUC values indicate declines, often attributed to factors such as salinity intrusion, drought, or climate change (Pettorelli et al., 2005; Singh et al., 2010). This analysis allows for the identification of vulnerable areas, highlighting regions where agricultural land-use degradation has occurred, and offers a scientific basis for developing targeted mitigation strategies.
DRASTIC index calculation
To assess the impact of salinity intrusion on agricultural land use, this study applied the DRASTIC model framework (Aller et al., 1987; Javadi, 2011; Chalá et al., 2024), tailoring its parameters to reflect the specific conditions of coastal areas that are significantly affected by salinity intrusion. A pivotal modification involved replacing the depth to water parameter with EC, enhancing the model’s precision in evaluating and predicting salinity intrusion. In regions like Ben Tre, where salinity intrusion occurs frequently, the depth to water parameter is insufficiently sensitive to variations in soil salinity levels. In contrast, soil conductivity, which correlates directly with salt concentrations, offers a more accurate and responsive measure for identifying affected areas. By substituting depth to water with EC, the focus of the DRASTIC model shifts from assessing aquifer vulnerability to conventional pollutants toward evaluating risks associated with salinity intrusion. This adaptation not only improves the accuracy of salinity risk mapping but also provides a robust scientific basis for agricultural land-use planning. The refined DRASTIC model allows for more precise predictions of salinity intrusion impacts, supporting strategic decision-making to mitigate risks and enhance agricultural sustainability in salinity-affected regions (Pacheco and Fernandes, 2015; Al-Adamat et.al., 2003).
This study, based on data collected from the Ben Tre DOST, quantitatively evaluates the impact of salinity intrusion on various types of agricultural land use. The adjusted DRASTIC model for the study area incorporates seven key parameters: Electrical Conductivity, Recharge, Aquifer Media, Soil Media, Topography, Impact of the Vadose Zone, and Hydraulic Conductivity of the Aquifer (Rupert, 2001; Hasan et al., 2023; Momejian al., 2019). The attribute layers for the seven DRASTIC parameters were evaluated by assigning rating scores ranging from 1 to 10 and weights from 1 to 5 for each factor (Table 3) to assess the potential for salinity intrusion within the study area. The data were structured in a GIS compatible format and processed using ArcGIS software. Fig. 2 illustrates the preparation of data layers, where parameters were systematically classified, evaluated, and assigned appropriate weights to calculate indices within the DRASTIC model. Subsequently, risk maps were developed by integrating seven layers, as designed in Table 3, and salinity intrusion risk levels (RLs) were classified and detailed in Table 4. The DRASTIC model was applied to assess salinity intrusion risk using the following equation (Aller et al. 1987, Babiker et al. 2005):
Explanation of variables:
W: The weight assigned to each factor; R: The rating score for each factor;
Corresponding indices for the factors: Wc, Rc: Electrical conductivity; WR, RR: Recharge; WA, RA: Aquifer media; WSo, RSo: Soil media; WT, RT: Topography; WI, RI: Impact of vadose zone; WH, RH: Hydraulic conductivity
Soil electrical conductivity (EC) measures the soil’s capacity to conduct electricity, serving as a direct indicator of the concentration of dissolved ions in soil water (Estefan et al., 2013). The EC map was interpolated from soil electrical conductivity data provided by Ben Tre DOST in 2023. EC is a critical parameter for assessing soil salinity, as dissolved salts in the soil, composed of charged ions, significantly influence its electrical conductivity (Richards, 1954). In the study area, salinity intrusion occurs when saline water from the sea or other sources infiltrates agricultural land, increasing soil salt concentrations (Machado et al., 2017). As salts dissolve in the soil, the resulting rise in ion concentration leads to higher electrical conductivity. This change in EC caused by salinity intrusion has a profound impact on crop growth and health. Salinity stress reduces agricultural productivity, especially for salt-sensitive crops such as rice, maize, beans, and fruit trees. Affected crops often display symptoms such as leaf scorching, stunted growth, and slower development, ultimately leading to diminished yields (Drechsel et al., 2023).
Recharge is a vital parameter in the DRASTIC model (Aller et al., 1987; De Vries & Simmers, 2002; Scanlon et al., 2002) for evaluating groundwater vulnerability to salinity intrusion. This factor represents the volume of water from precipitation or other sources, such as irrigation, rivers, or streams, that infiltrates the soil to replenish groundwater reserves. The Recharge index serves as an essential metric for assessing the sensitivity of groundwater to salinity impacts. Higher Recharge values indicate a greater capacity to mitigate salinity intrusion, while lower values suggest increased susceptibility to salinization. In areas with a high risk of salinity intrusion, such as Ben Tre Province, recharge plays a critical role in maintaining groundwater quality. Insufficient recharge allows saline water from the sea or adjacent rivers to infiltrate groundwater systems, resulting in salinization. Thus, ensuring adequate recharge is fundamental to protecting groundwater resources in vulnerable regions (Le, T.T.H. and T.T. Nguyen, 2017; Nguyen Thanh., et al., 2023; Barlow, P.M., et al., 2010).
In the DRASTIC model, the aquifer media parameter plays a crucial role in assessing the protective capacity of aquifer materials against surface influences, such as pollution and salinity intrusion (Foster, 1987; Aller, et al., 1987; Melloul and Collin, 1998; Fidelibus and Paniconi, 2000). In the study area, data extracted from hydrogeological and environmental geological maps indicate that aquifers are typically composed of a mixture of sand, silt, and clay. These materials exhibit varying permeability, ranging from high (sand) to low (clay), particularly in riverine and coastal regions (Masaaki et al., 2007). Aquifers dominated by sandy materials allow saline water to infiltrate more rapidly into the groundwater, increasing the risk of salinity intrusion. Conversely, clay-dominated areas impede water infiltration, offering greater protection to the groundwater from salinity intrusion (Werner et al., 2009). Furthermore, aquifers with higher storage capacities, such as those containing substantial amounts of sand and silt, enhance the dilution of saline water, mitigating the impacts of salinity intrusion (Al-Adamat et al., 2003; Babiker et al., 2005). This variability in material composition and permeability underscores the importance of Aquifer Media in evaluating groundwater vulnerability in salinity-prone regions.
Soil media is a critical factor in assessing the vulnerability of aquifers to salinity intrusion (Aller et al., 1987). Soil media influences the permeability of water into aquifers and, consequently, affects the susceptibility of groundwater to salinity intrusion (Piscopo, 2000). Soil maps provided by the Ben Tre DOST classify soils into five categories based on their permeability. Alluvial soils, typically deposited by rivers, exhibit high permeability, while acid sulfate soils are less permeable and highly acidic. This low permeability leads to surface water accumulation, increasing the risk of salinity intrusion as saline water remains undiluted (Eslami et al. 2019. In Ben Tre, acid sulfate and saline soils significantly reduce crop growth potential, particularly during the dry season when salinity intrusion becomes more severe. These soil types exacerbate agricultural challenges, underscoring the importance of soil media in evaluating and managing salinity risks in groundwater systems.
Topography, expressed through terrain slope, is a key factor in assessing water infiltration and surface movement dynamics. In the case of salinity intrusion in Ben Tre Province, the region’s flat terrain with very low average slopes (generally below 5%) and its location between three major rivers (Tien, Ham Luong, and Co Chien) create conditions that intensify the effects of tidal forces. These characteristics allow saline water to penetrate deeply into soil layers, leading to severe salinity intrusion (Smajgl et al., 2015). The low slope gradient prevents saline water from being readily flushed back to the sea or other areas, instead causing it to stagnate and migrate slowly. This significantly increases the risk of salinity spreading across the region (Saidi et al., 2011; Shirazi et al., 2012). The use of a digital elevation model (DEM) facilitates the precise determination of surface slopes within the study area. Based on topographic and slope data, areas with minimal slopes are identified as high-risk zones, as they are more susceptible to retaining and accumulating saline water (Li and Heap, 2011; Bhuiyan and Dutta, 2012).
The impact of the vadose zone refers to the layer between the ground surface and the aquifer, which plays a critical role in regulating the movement of saline water before it reaches the groundwater (Nachshon, 2018; Javadi et al., 2011). In Ben Tre Province, situated in the low lying MKD, the vadose zone is notably thinner compared to regions with higher elevation. This characteristic allows surface water to rapidly percolate into the aquifer, especially during the dry season when groundwater levels recede (Minderhoud et al., 2019). The vadose zone in Ben Tre predominantly comprises alluvial soils and clay, which have a relatively high capacity to retain water. However, during the dry season, intense evaporation can cause the soil to harden and crack, significantly increasing its permeability. This facilitates the intrusion of saline water from nearby rivers and the sea into the groundwater system, exacerbating salinity risks.
Conductivity of the aquifer is a pivotal parameter in the DRASTIC model, essential for evaluating the vulnerability of groundwater to salinity intrusion (Secunda et al., 1998; Fidelibus and Paniconi, 2000; Javadi et al., 2011; Al Ahmadi, 2013). Aquifers with high hydraulic conductivity are at greater risk of salinity intrusion as saline water can infiltrate more rapidly and penetrate deeper into the aquifer system (Nichols et al., 2015, Piscopo, 2001; Rome and ITFAO, 1997). In Ben Tre Province, the aquifers are composed of porous formations within alluvial deposits and sandy gravel sediments from the Tien and Ham Luong Rivers. These aquifers are predominantly shallow to intermediate in depth, ranging between 20 and 50 meters. Deeper aquifers lie beneath layers of clay and sandy clay. The aquifers are primarily composed of fine sand, alluvial sediments, and sandy clay with medium to high permeability (Van Pham et al., 2019). Due to the high permeability of sandy and alluvial soils, the aquifers in Ben Tre exhibit elevated hydraulic conductivity, significantly increasing their susceptibility to salinity intrusion as saline water from rivers and the sea infiltrates groundwater systems (Vu et al., 2018). Ben Tre’s geographical characteristics, its coastal proximity and location within vulnerability is particularly pronounced during the dry season when reduced freshwater inflow amplifies the likelihood of salinity intrusion (Eslami et al. 2021; Le et al. 2017).
Spatial regression analysis between ALUC and RLs for assessing salinity intrusion risk
Spatial regression analysis between ALUC and RLs serves as a crucial method for evaluating the impact of salinity intrusion on agricultural land use. The application of GWR enables a deeper understanding of the relationship between ALUC and salinity-related factors in the DRASTIC model. Furthermore, GWR model captures spatial variability, providing insights that traditional regression methods cannot achieve (Leung et al., 2000; Fotheringham et al., 2002; Páez et al., 2002).
In this study, ALUC and RLs data were mapped using the WGS84 Zone 48N coordinate system with a 1:50,000 scale, ensuring precise and consistent spatial analysis (Fotheringham et al., 2022; Lloyd, 2010). GWR offers a more detailed analysis of spatial variation in regression coefficients, elucidating the relationship between RLs and ALUC across specific geographic locations (Leung et al., 2000; Wu and Li, 2013). A key advantage of GWR lies in its ability to calculate regression coefficients specific to each location, reflecting spatial heterogeneity between ALUC and RLs. This means that the relationship between ALUC and RLs can vary significantly across different areas, helping to identify stark differences in the impact of salinity intrusion across regions (Páez et al., 2002; Fotheringham et al., 2002). The DRASTIC model categorizes RLs into four levels (Table 4), with each level exerting varying degrees of impact on eight ALU types (Table 2) depending on location. By employing spatial regression, GWR provides detailed insights into how land-use types are affected by salinity intrusion across different regions. This analysis highlights areas most vulnerable to salinity intrusion, where crop health is likely to decline more severely, offering valuable data for developing effective land management and utilization strategies. The results of GWR model reveal clear patterns of spatial variability, enabling the study to not only quantify the risks of salinity intrusion but also gain a comprehensive understanding of how agricultural land use evolves in response to salinity pressures at a localized level. This spatially explicit analysis is critical for informed decision-making and targeted interventions.
The general formula for GWR is expressed as follows (Leung et al., 2000; Fotheringham et al., 2002; Wu and Li, 2013)
Where:
y(u, v): Represents the spatial variation in agricultural land-use changes at a specific location (u, v).
xk(u, v): Denotes the explanatory variables capturing salinity intrusion risk levels derived from the DRASTIC model at the spatial location (u, v).
βk(u, v): Refers to the regression coefficient for the k-th explanatory variable, which varies across spatial locations, reflecting spatial heterogeneity in the relationship between land-use changes and salinity risk factors
ε(u, v): Represents errors or noise in the model at the spatial location.
Calibrating the weighting function of GWR
In this study, the GWR model was utilized, with the calibration of the weighting function being a critical step to ensure the model accurately reflects the spatial distribution of the data (Gyawali et al., 2013; Fotheringham et al., 2002). To calibrate and select the optimal model, the Akaike Information Criterion corrected for small sample sizes (AICc) was applied to compare the performance of the GWR model against the OLS model (Azua et al., 2020; Ehlkes et al., 2014).
Through this process, the AIC value of the GWR model was analyzed alongside the AIC value of the OLS model to account for model complexity. A model with a lower AIC value is deemed to provide a better fit to the observed data. Comparing the AIC values of GWR and OLS serves as a reliability test for transitioning from the global OLS model to the spatially localized GWR model.
The formula for calculating AIC is as follows:
Where n is the sample size, σ is the estimated standard deviation of the residuals, and tr(S) is the trace of the hat matrix, which is a function of the bandwidth.
To achieve this, the GWR tool in ArcGIS and Python was utilized to compute the regression coefficients for each salinity intrusion risk level in the DRASTIC model and to elucidate the spatial variations in agricultural land-use categories affected by salinity intrusion (Páez et al., 2008; Fotheringham et al., 2022). Areas with strongly negative regression coefficients indicate significant adverse impacts of salinity intrusion, leading to changes in land-use area, while positive coefficients suggest stability or recovery in agricultural land-use categories (Wu and Li, 2013; Fotheringham et al., 2002).
These findings not only enhance the assessment of salinity intrusion impacts but also assist land-use managers in identifying high-risk areas, enabling the implementation of more effective intervention strategies in the face of increasing climate change and salinity intrusion challenges (Leung et al., 2000; Fotheringham et al., 2022; Páez et al., 2002).
Results and Discussion
Comprehensive assessment of ALUC
Using Landsat data from 2014 and 2023, this study evaluated changes in agricultural land use under the influence of salinity intrusion. Digital number (DN) values were converted into NDVI (Equation 1) to classify land-use categories based on thresholds derived from 200 field samples collected in 2023 (Table 2). The ISODATA classification method was employed to segment ALU types (Fig. 3), leveraging field-validated NDVI ranges for accuracy. Categories such as water surfaces, bare land, aquaculture, rice, cash crops, fruit trees, coconut trees, and mangroves were identified, reflecting diverse responses to salinity conditions.
Equation 2 was used to evaluate ALUC. The results in Table 5, Figs. 3 and 4 indicate significant changes in ALU between 2014 and 2023, driven by salinity intrusion and local adaptation strategies. Salinity-sensitive categories such as fruit trees (−37,784.79 ha), rice (−4,450.32 ha), annual crops (−2,730.6 ha), and freshwater aquaculture (−8,225.82 ha) experienced substantial declines due to their inability to withstand increased salinity levels. Conversely, salinity-tolerant categories such as mangroves (+51,757.29 ha) and coconut trees (+7,634.79 ha) expanded significantly, indicating their resilience in adapting to saline environments. An increase in water surfaces (+3,941.73 ha) further highlighted the extent of salinity intrusion into inland areas, while a marked reduction in barren land (−10,154.79 ha) reflected efforts to adopt salinity-resilient farming practices.
These findings underscore the critical impact of salinity intrusion on agricultural land use, driving a shift towards more adaptive and resilient systems. The study highlights the importance of understanding land-use dynamics in saline environments to support strategic planning and sustainable agricultural practices in regions increasingly affected by climate change and salinity pressures.
Assessment of salinity intrusion risk using the DRASTIC model
To assess and develop a salinity intrusion risk map for Ben Tre Province, the Drastic model was established using Equation 3 in conjunction with seven key indicators, integrating spatial data through ArcGIS 10.3 software (Table 3). The primary data supporting this analysis were collected from the Ben Tre DOST (Fig. 2). Salinity intrusion risk levels were calculated based on the values of the seven Drastic indicators and classified using an interval scale ranging from low to very high (Table 4). The resulting risk map (Fig. 5) highlights the diverse spatial distribution of salinity intrusion risks, accurately reflecting the geographic characteristics of each region.
Low-risk areas (85–100) are typically located in elevated terrains or regions well-protected by robust dike systems, where saline water intrusion is minimal. Moderate risk areas (100–115) are vulnerable during the dry season when salinity intrusions may occur but with low frequency, thanks to relatively effective salinity prevention measures. High-risk zones (115–130) are more prone to significant impacts from strong salinity intrusion events, particularly in the absence of sufficient salinity control infrastructure. Very high-risk areas (130–150) are predominantly situated in coastal districts like Ba Tri, Binh Dai, and Thanh Phu, where frequent and severe salinity intrusion is a critical issue. These regions require immediate protective interventions and comprehensive long-term adaptation strategies.
In inland districts such as Mo Cay Nam and Mo Cay Bac, risk levels range from low to moderate. However, proximity to major rivers increases the potential for saline water intrusion during the dry season. Giong Trom and Chau Thanh face moderate to high risks, necessitating targeted measures during periods of freshwater scarcity in the dry season. In Ben Tre City, the irrigation system provides some degree of protection, keeping salinity intrusion risk at low to moderate levels. Nonetheless, low river water levels during the dry season may still enable salinity intrusion to affect the area.
This salinity intrusion risk map provides a clear understanding of priority areas requiring proactive prevention and adaptation efforts. It also serves as a vital tool for guiding appropriate solutions to mitigate the adverse impacts of salinity intrusion on local livelihoods and agricultural production across the province.
Spatial regression analysis between ALUC and RLs Model adjustment
Based on Equation 4 and Equation 5, the evaluation of the effectiveness and optimization of selecting between the GWR and OLS models for spatial analysis was conducted using the AICc (Corrected Akaike Information Criterion) parameter. This metric was employed to assess model performance and guide the selection of the most appropriate model. The results, summarized in Table 6, indicate that the GWR model consistently outperforms the OLS model across key performance indicators. Specifically, the AICc value for GWR is 699.779, which is substantially lower than the average threshold observed for OLS. This result suggests that the GWR model achieves a superior balance between model fit and complexity, thereby making it the more optimal choice for spatial analysis.
The GWR model demonstrates a superior capacity to explain the variation in the dependent variable (ALU) based on the explanatory variables (RLs), achieving an impressive R2 value of 0.806. In comparison, the OLS model yields an R2 of only 0.626, indicating that GWR captures a significantly greater proportion of the variability in the dataset. Additionally, the Adjusted R2 of GWR remains high at 0.763, highlighting its robust explanatory power even after accounting for the complexity of the independent variables.
In terms of prediction accuracy, the GWR model also proves more reliable, with a Sigma value of 1.110, closer to the ideal threshold of 1, compared to the OLS model’s Sigma of 1.31. This suggests that GWR’s prediction errors are less dispersed, enhancing its robustness. Additionally, the Residual Squares value for GWR is substantially lower at 199.828, compared to 373.49 for OLS, affirming GWR’s superior precision in handling complex spatial data. Lastly, the higher Adjusted R2 value in GWR compared to OLS further highlights its advantage, demonstrating that GWR’s localized approach is more effective in capturing data variability without being overly influenced by extraneous independent variables. These findings collectively underscore the GWR model’s clear superiority and suitability for spatial data analysis, with a marked distinction from the OLS model.
Spatial relationship between ALUC and salinity intrusion risk
The analysis of the relationship between the dependent variable (ALU) and explanatory variables (RLs, RLs area, and ALUC area) reveals the significant impact of salinity intrusion on the spatial distribution of various types of ALU. Table 7 shows that the Intercept coefficients range from 4.334 to 5.519, indicating variations in baseline land-use values before the influence of salinity intrusion. Among these, coconut and fruit tree plantations have the highest coefficients, reflecting greater stability before being affected by salinity intrusion.
The β_RLs coefficients, representing the level of salinity intrusion impact, are negative and range from −0.02 to −0.389, indicating increasingly negative effects as salinity intrusion intensifies. Mangrove forests are the most severely affected, while barren land is the least impacted. The β_ALUC area coefficients, ranging from 0.8 to 2.843, indicate that rice, fruit tree, and coconut plantations are the most strongly affected by salinity intrusion, as evidenced by changes in area, while aquaculture is relatively less impacted. This underscores the sensitivity of salinity intrusion in altering the scale of agricultural land use in the study area during the 2014–2023 period.
The β_RLs area coefficients further demonstrate that salinity intrusion significantly influences the use of coconut, fruit tree, and rice land (β = 0.435–0.562), while water bodies are minimally impacted (β = 0.095). This reflects the varying sensitivity of agricultural land types to salinity intrusion levels. Predicted values for each land type range from 1.481 to 7.271, with mangrove forests having the highest values, indicating their resilience as ecosystems in the face of salinity intrusion.
Finally, local R2 values range from 0.793 to 0.821, highlighting the model’s strong suitability and reliability for different agricultural land-use types. Overall, the GWR model parameters demonstrate a strong spatial relationship, capturing the variability of ALU types under the influence of salinity intrusion. The analysis also emphasizes the correlation among RLs, RLs area, and ALUC area in shaping agricultural land use dynamics.
The StdResid (Standardized residual) results in spatial regression analysis indicate the extent of the discrepancy between the observed actual values and the predicted values generated by the GWR model
Fig. 6. illustrates the StdResid parameter, reflecting the accuracy of the GWR model in predicting the impact of RLs on ALUC and identifying deviations in agricultural land use changes under salinity intrusion. StdResid values from −3.449037 to −1.622846 indicate under-predicted areas, primarily in the coastal districts of Binh Dai and Thanh Phu, with scattered areas in central Ba Tri. These regions show significant transitions from barren land and aquaculture to other uses (−10,154.79 ha and −8,225.82 ha, respectively), highlighting severe vegetation degradation due to stronger-than-predicted salinity impacts. Values between −1.622845 and −0.597450 reflect shifts from rice and crops to coconut plantations and freshwater reservoirs, concentrated in Giong Trom, Binh Dai, and Mo Cay Nam. Reductions in rice and crop areas (−2,730.6 ha and −4,450.32 ha) indicate lower salinity tolerance than anticipated. The range of −0.597449 to 0.237962 highlights changes in fruit tree areas, primarily in Cho Lach, Chau Thanh, and Ben Tre City. Despite significant area declines (−37,784.79 ha), the moderate adaptability of fruit trees aligns with model predictions. Slightly positive StdResid values (0.3339 to 1.039061) are observed in Mo Cay Bac, Mo Cay Nam, and southern Thanh Phu, where coconut plantations expanded (+7,634.79 ha). These results confirm the model’s accurate prediction of coconut resilience. Highly positive values (1.039062 to 2.025366) are concentrated in coastal Binh Dai and Ba Tri, where mangrove forests expanded significantly (+51,757.29 ha). This indicates mangroves’ exceptional adaptability, surpassing the model’s projections.
Conclusion
Understanding the spatial impacts of salinity intrusion on agricultural land use and their relationship to land adaptation is essential for effective agricultural land management and building resilience to the increasing complexities of salinity intrusion. This study assessed salinity intrusion risks using the DRASTIC model and evaluated agricultural land-use changes from 2014 to 2023 through NDVI indices for eight agricultural land-use categories. By employing the GWR model, the study uncovered spatial heterogeneity in the relationship between RLs and ALUC.
The findings reveal that coconut plantations and mangrove forests in the study area demonstrate superior adaptability to salinity intrusion, even as saline water penetrates further inland. Conversely, GWR analysis shows that RLs negatively impact areas dedicated to fruit trees, rice, and vegetable cultivation, leading to significant reductions, particularly in central districts and coastal regions of Ben Tre Province. A strong correlation was observed between
ALU, RLs area, and ALUC area, especially in regions with high intercept and predicted values. These results suggest that increased soil salinity has severely diminished the areas of salinity-sensitive crops, such as rice, fruit trees, and vegetables. However, salinity tolerant species, like coconuts and mangroves, are expanding in response to these conditions.
In contrast, the correlation between RLs area and ALUC area was weak in barren lands, water bodies, and aquaculture zones near coastal regions, where vegetation is largely absent. The analysis of StdResid parameters in the GWR model further revealed the reliability of predicting the spatial variability of ALUC under salinity intrusion (Fotheringham et al., 2022; Nazeer and Bilal, 2018). These insights offer a novel approach for accurately managing and mitigating the effects of salinity intrusion on agricultural crops in Ben Tre Province.
The study identified three key patterns in StdResid values:
StdResid < 0: Reflects severe crop degradation due to salinity intrusion impacts that exceed the model’s predictions, indicating very low adaptability in these areas.
StdResid ≈ 0: Indicates that the model provides accurate predictions of agricultural crop adaptability to saline environments.
StdResid > 0: Highlights greater-than-predicted adaptability, particularly in salinity-tolerant crops like mangroves and coconuts.
These findings highlight the critical importance of promoting salinity tolerant crops to ensure sustainable agricultural land use and resilience in regions affected by salinity. Through the integration of spatial risk analysis and adaptive modeling, this study offers a comprehensive framework to strengthen land management strategies and effectively address the challenges associated with salinity intrusion.
PDF Links
PubReader
ePub Link
Full text via DOI
Download Citation
