Electrical conductivity and spectral indices of ground points
The maximum EC recorded was 7.8 dS/m, while the minimum value was 2.8 dS/m. The average EC across all sample points was found to be 4.25 dS/m, with a standard deviation of 1.30 dS/m. Eight maps describing vegetation and salinity indices are presented in Fig 2. The data from Fig 2 shows that the SI_T, SI1 and SI3 indices are significantly lower in vegetated areas compared to other land types, while the SI2 index is lowest in water-covered areas. The vegetation indices (NDVI, SAVI, EVI and CRSI) indicate the extent of vegetation cover across the study area, highlighting that the southwestern region-primarily used for double-cropped rice cultivation-has the highest vegetation density.
The data corresponding to these eight indices collected from 50 sampling points are detailed in Table 3. The standard de
viation (SD) values, which have been calculate based on packages in R studio software, of the indices for rice land use are consistent. This consistency can be attributed to the synchronized implementation of the crop calendar, which has been applied for rice cultivation. In contrast, other LUTs exhibit higher SD values due to the diversity of crops and varying planting times.
The correlation between the EC values and the salinity and vegetation indices is illustrated in Fig 3. The Pearson correlation indicates that the EC index is positively correlated with the salinity indices (SI_T, SI1, SI2, SI3) and negatively correlated with the vegetation indices (NDVI, SAVI, EVI, CRSI). Most correlation values are greater than 0.7, with the exception of the SI2 index, which has a correlation value of 0.35.
The nature of soil salinity is divided into two types: primary and secondary, in which primary is the salinity process generated by the nature of the elements that make up the soil, while secondary is mainly formed from land use, especially for agricultural land
Gojiya et al., (2023). Our study area is located near the sea and lagoons and is characterized by active agricultural practices. As a result, the mechanisms of soil salinization are quite complex, involving both primary and secondary processes. In addition, ecological conditions also affect the EC concentration in soil. Some recent studies have shown that vegetation indices can be used to determine EC content in agricultural land
(Haq et al., 2023), while others have suggested that SI indices can be used to determine EC content in bare, grass land and uncovered areas
(Luo et al., 2025). Recent study used NDVI or EVI as referent classification for soil salinity levels
(Djuraev et al., 2021). On the contrary, others
(Zhang et al., 2011) stated that these indices completely unable to detect soil salinity. In our viewpoint, the lack of significant vegetation cover in their arid study area likely explains this finding. Our results also found that the SI3 index has a strongest correlation to EC value, so it is a good indicator to estimate soil EC. This finding was confirmed by the previous research
(Douaoui et al., 2006a).
The regression models to estimate soil saline for coastal and lagoon areas
Fig 4 presents scatter plots comparing predicted and measured EC values using various salinity and vegetation indices. Each plot includes a regression equation, coefficient of determination (R
2), Akaike Information Criterion (AIC) and root mean square error (RMSE), which collectively assess model accuracy and fit. SI3 yielded the highest prediction accuracy, with an R
2 of 0.82, AIC of 88.22 and RMSE of 0.55, indicating a strong correlation between predicted and observed EC value while SI2 exhibited the poorest model fit (R
2 = 0.12, RMSE = 1.22), indicating it is not a reliable indicator for predicting EC in this study area.
Table 4 presents the MLR models developed using stepwise selection based on the AIC. The MLR models exhibit significantly higher accuracy compared to SLR models. Although the model combining SI3, EVI and NDVI (Model 6) shows an approximately 2% higher RMSE and a 1.1% lower R² compared to the others, it achieves the lowest AIC score of 83.43 and uses the fewest variables. The technique to mitigate multicollinearity among the variables in the model for the model of SI3, NDVI and EVI variables.
Fig 5 shows six scatter plots, each representing a different model (Model 1 to Model 6). Each plot compares predicted EC (X-axis) values to measured EC values (Y-axis). As illustrated in the figure, the differences among the models are not markedly significant. A notable feature across the models is that data points corresponding to the highest values de
viate substantially from the 1:1 reference line. This suggests that for extreme EC values, factors beyond those captured by remote sensing data may be influencing the results, indicating the need to incorporate additional variables or contextual information to improve model accuracy.
Fig 6 shows that only two of four models, those combining SI3 with NDVI and SI3 with EVI, exhibited a multicollinearity index of less than 3, which meets the criteria for acceptable use. The model combining SI3 with NDVI, (EC= 2.22+11.47*SI3-1.78*NDVI) has the lowest AIC and RMSE coefficients (84.6 and 0.52), as well as the highest R2 value with a value of 0.84. We recommend this model for estimating salinity in the coastal sandy areas and lagoons of Hue city. Linear regression models are widely utilized in soil salinity mapping
Al-Ali et al., (2021), among them MLR models tend to be more accurate than SLR models
Al-Khuzaie et al., (2022). This is because MLR models take into account a greater number of factors, reducing the likelihood of overlooking important variables that affect soil salinity compared to simpler models. However, in certain situations, simple models can still yield satisfactory results, serving as a foundation for further analysis or the application of more complex models
Hihi et al., (2019).
Agricultural land management adapt to soil saline situation
Fig 7 illustrate the spatial distribution of soil salinity across agricultural land areas, as determined by the selected MLR model (EC= 2.22+11.47*SI3-1.78*NDVI). The findings indicate that all agricultural land within the study site exhibits salinity.
A Dunn depth analysis was conducted at a 95% confidence level on all points (where each pixel represents a point) across various LUTs as shown in Fig 8. The results indicated a statistically significant difference in the mean EC values among the LUTs.
Irrigation and fertilization significantly impact the soil EC content
(Thiam et al., 2021). Our results show that two-seasons paddy rice areas have the lowest EC due to frequent flooding in November, which washes salts away, consistent with recent study
Shirokova et al., (2024). Single-season rice fields, adjacent to lagoons with higher salinity water, have higher EC than two-seasons paddy rice areas. Coastal forests dominated by Acacia, Melaleuca and Casuarina show the highest salinity, as these species tolerate saline soils
(Khalil et al., 2016). We suggest planting coconut trees in these areas for economic and salinity tolerance benefits
(Sun et al., 2024). Abandoned lands accumulate surface salts with high EC; revegetation with native grasses is recommended to improve the soil gradually. Vegetable and annual crop fields show varied EC due to diverse crops and irrigation, but crop selection often ignores salinity tolerance. Current crops such as sweet potato, chili pepper and corn are not suitable for saline soil, so the shifting to more salt-tolerant crops like garlic and pitaya should be considered
(Cavalcante et al., 2008). Overall, varying salinity across land uses highlights the need for tailored management to minimize salinity impacts and support sustainable agriculture.