Vegetation damage severity mapping (VDSM) has been performed using pre- and post-event satellite images. NDVI and FVC are widely used techniques for measuring the vegetation density, health status, and regional vegetation condition and detecting vegetation disturbances (Xu et al., 2021; Mishra et al., 2021b; Wang et al., 2010; Yang et al., 2018; Wang and Xu, 2018; Carlson and Ripley, 1997). Subsequently, numerous studies (Xu et al., 2021; Mishra et al., 2021a; Charrua et al., 2021; Shamsuzzoha et al., 2021; Kumar et al., 2021; Nandi et al., 2020; Wang and Xu, 2018; Konda et al., 2018; Zhang et al., 2013; Rodgers et al., 2009) have shown that NDVI and FVC are reliable indicators of post-typhoon damage detection. Therefore, in this study, the vegetation damage due to Typhoon Soulik has been determined using the NDVI and FVC approach. The NDVI value has been calculated as follows (Rouse et al., 1974; Filgueiras et al., 2019): 1NDVI=ρNIR-ρRedρNIR+ρRed, where ρNIR and ρRed are the spectral reflectances corresponding to the eighth (832.8–832.9 nm) and fourth (664.6–664.9 nm) Sentinel-2 MSI bands, respectively (Xu et al., 2021). In general, NDVI values range from -1.0 to 1.0. The higher the NDVI value, the better the conditions for vegetation development, and extremely low values indicate the presence of water. Furthermore, an NDVI value above 0.4 indicates vegetated surfaces, and those between 0.25 and 0.40 signify soils with the presence of vegetation (Charrua et al., 2021). The vigor of the vegetation increases as the NDVI values come closer to 1.00 (Rouse et al., 1974). Numerous studies have established the NDVI threshold for vegetated land (e.g., Xu et al., 2021; Wong et al., 2019; Liu et al., 2015; Eastman et al., 2013; Yang et al., 2012; Sobrino et al., 2004). Most researchers noted that the NDVI threshold value for vegetation cover typically ranges from 0.15–2.0 (Xu et al., 2021; Eastman et al., 2013; Sobrino et al., 2004). Therefore, the vegetated pixels (e.g., NDVI threshold of > 0.20) present in pre- and post-typhoon NDVI images have been used for vegetation severity analysis. The NDVI threshold is considered to reduce the effect of land cover change from the pre- (1 August 2018) to post-typhoon (15 October 2018) periods.

The degree of vegetation damage has been determined by comparing the NDVI values of the pre- and post-typhoon periods. Various researchers have frequently used the direct difference in NDVI to determine the damage severity caused by typhoons to naturally vegetated land (Wang and Xu, 2018; Konda et al., 2018). It has been calculated on a cell-by-cell basis by subtracting the pre-typhoon NDVI image from the post-typhoon image in ArcGIS using Map Algebra (Zhang et al., 2013; Cakir et al., 2006). The following equation is used to calculate the ΔNDVI (Wang and Xu, 2018): 2ΔNDVI=NDVIpost-typhoon-NDVIpre-typhoon. The difference in NDVI (i.e., ΔNDVI) illustrates the change in natural vegetation, while a negative ΔNDVI value indicates the damage inflicted by a typhoon to the vegetation cover (Xu et al., 2021).

The relative change in NDVI value has been used to investigate the geoecological impact on the forest area (Mishra et al., 2021b). The relative vegetation changes (NDVIr) after Typhoon Soulik have been determined by using the following equation (Kumar et al., 2021): 3NDVIr=ΔNDVINDVIpre-typhoon×100, where the negative NDVIr value indicates vegetation loss caused by typhoons and the positive NDVIr value shows vegetation gain. The NDVIr value has been classified into three categories corresponding to pixels with vegetation cover that has decreased, had no change, or has increased.

On the other hand, we analyze FVC in conjunction with NDVI, which provides additional insights into vegetation conditions and damage severity. Numerous researchers (Wang and Xu, 2018; Song et al., 2017; Bao et al., 2017; Chu et al., 2016; Amiri et al., 2009) used FVC to analyze vegetation damage, restoration, recovery, and inter-annual variability. In the present study, FVC was calculated before and after the typhoon using the derived NDVI data (Wang and Xu, 2018). It is expressed as a percentage and can range from 0 % to 100 %. The formula of FVC is as follows (Wang and Xu, 2018; Amiri et al., 2009; Carlson and Ripley, 1997): 4FVC=NDVI-NDVIm/NDVImax⁡-NDVIm2, where NDVIm and NDVImax⁡ represent the NDVImin⁡ and NDVImax⁡ values calculated using Eq. (1) (Zhang et al., 2021; Ge et al., 2018). The calculated FVC values vary between 0 and 1. After that, the FVC values were converted to percentages to fit the actual FVC classification scheme (Wang and Xu, 2018), which consists of five classes: high (80 %–100 %), medium-high (60 %–80 %), medium (40 %–60 %), medium-low (20 %–40 %), and low (0 %–20 %). Further, the difference in FVC values between the pre- and post-typhoon images was used to calculate the extent of vegetation damage using the following equation: 5ΔFVC=FVCpost-typhoon-FVCpre-typhoon, where ΔFVC denotes the difference between the pre- and post-typhoon FVC. The ΔFVC value represents alterations in vegetation conditions and damage intensity, while a negative value of ΔFVC indicates the extent of damage caused by a typhoon to vegetation cover (Wang and Xu, 2018).

Typhoons have adversely affected the coastal landform and ecology of the southern and western coasts of the Korean Peninsula every year. Therefore, a GIS-based coastal-change model has been developed to understand the morphodynamics of coastal landforms during typhoons. In the present study, we considered four coastal-landform classes, i.e., wetland, wetland vegetation, land, and water, for the coastal morphodynamic analysis (Maiti and Bhattacharya, 2011). The method consists of two algorithms, i.e., (a) the ISODATA (Iterative Self-Organizing Data Analysis Technique) algorithm used to classify the coastal landform with four main classes, i.e., water, wetland, wetland vegetation, and land, and (b) the change detection technique used to quantify the short- and medium-term coastal changes. In this approach, we accentuate in-depth morphological changes and emphasize minor changes along the Mokpo coast caused by Typhoon Soulik.

The pre- and post-typhoon Sentinel-2 MSI images have been classified using the unsupervised classification technique to distinguish between different coastal landforms of the study region. This approach is used to determine which types of coastal landforms were adversely affected by Typhoon Soulik and which of them have recovered more quickly than others. ERDAS IMAGINE has been used to run the unsupervised classification algorithm (ERDAS, 1997). Based on the k-means algorithm, this technique reduces variability within pixel clusters (Charrua et al., 2021; Aswatha et al., 2020; Bhowmik and Cabral, 2013). Finally, pre- and post-typhoon Sentinel-2 MSI images have been classified into four coastal-landform classes: land, water, wetland, and wetland vegetation.

The accuracy assessment is a commonly used method to determine how closely the classified map matches the reference data (Congalton, 1991). In the present study, the classified data (i.e., coastal-landforms maps) have been derived through an unsupervised classification technique, while 550 random samples collected from different parts of the Sentinel-2 MSI standard false-color image are considered reference data. Thereafter, a confusion matrix was developed based on the reference and classified data to evaluate accuracy statistics (Story and Congalton, 1986). The κ​​​​​​​ coefficient (k) has been used to determine the quantitative accuracy of the classified map (Landis and Koch, 1977). The assessment is quantified using three different statistics: overall accuracy, producer accuracy, and user accuracy (Story and Congalton, 1986). The model's precision is classified into five categories based on the k values: near perfect (k > 0.8), substantial (0.6 < k < 0.8), moderate (0.4 < k < 0.6), fair (0.2 < k < 0.4), and poor (k < 0.2) (Landis and Koch, 1977).

The land transformation model based on mutual spatial replacements has been applied during the post-classification stage, as shown in Fig. 4. The classified coastal-landform classes, such as land, wetland, wetland vegetation, and water, have been spatially replaced in order to create coastal-change units. For example, the coastal-landform class of wetland vegetation in the pre-typhoon period replaced by water in the post-typhoon period indicates the change class of wetland vegetation replaced by water. A total of nine coastal-change classes have been derived, as illustrated in Fig. 4b.

The suspended-sediment concentration (SSC) distribution in coastal regions is a significant indicator of changes in the marine environment caused by typhoon-induced storm surges, strong waves, and subsequent coastal flooding (Min et al., 2012; Gong and Shen, 2009). In a short period, a typhoon may drastically influence the water column structures (Souza et al., 2001), change the transport and deposition of sediment (Li et al., 2015), and affect the distribution of nutrients and biological production in the affected seas (Wang et al., 2016). Extreme storms or typhoons can modify suspended-sediment distribution in coastal regions, which can significantly change marine habitats (Chau et al., 2021; Lu et al., 2018; Li and Li, 2016). Due to strong typhoon wind stress, the concentration of suspended particles in the seawater column and sediment resuspension may increase dozens of times before and after the event (Lu et al., 2018; Bian et al., 2017). Thus, typhoons significantly affect suspended-sediment movement in the coastal region (Zhang et al., 2022; Li and Li, 2016; Goff et al., 2010). The spatiotemporal distribution of the SSC can be impacted by variations in tidal phase, runoff, and wind speed (Tang et al., 2021). Furthermore, the resuspension of sediment can cause numerous problems in ocean engineering and change the region's ecology (Kim, 2010). The amount of material delivered to and adverted across the shelf by typhoons is considerably larger than that of winter storm systems (Dail et al., 2007). The southern and western part of the Korean Peninsula is affected by an average of three typhoons annually passing through the Yellow Sea (KMA, 2018; Altman et al., 2013). Some studies on the SSC distribution impacted by artificial construction along the coastal region of the Yellow Sea have been undertaken by several researchers (i.e., Lee et al., 2020; Eom et al., 2017; Min et al., 2012, 2014; Choi et al., 2014). However, the effects of typhoons on the sedimentary environment in the Mokpo coastal region have not yet been investigated. Therefore, it is imperative to carry out regional-scale SSC mapping and coastal modifications to reveal changes in the marine environment and sediment transport mechanisms over the typhoon period.

Remote sensing has long contributed to the advancement of water quality studies (Hossain et al., 2021). In the present study, we attempted to calculate both the qualitative and quantitative SSC in the inner-shelf region of the Mokpo coast using Sentinel-2 MSI data. The relative suspended-sediment concentration has been calculated from pre- and post-typhoon Sentinel-2 MSI images using NDSSI, which has been used in various water quality research (Kavan et al., 2022; Hossain et al., 2010). Further, many studies (Shahzad et al., 2018; Arisanty and Saputra, 2017) have successfully used Landsat and Sentinel-2 data to calculate NDSSI. This index determines the relative concentration of suspended sediment, with values ranging from -1 to 1, where -1 indicates the highest concentration and +1 indicates the lowest (Hossain et al., 2010). The NDSSI value has been calculated by using the following equation: 6NDSSI=ρBlue-ρNIRρBlue+ρNIR,

where ρBlue and ρNIR represent the surface reflectances of band 2 (492.1–492.4 nm) and band 8 (832.8–833.0 nm) of Sentinel-2 MSI data, respectively. The NDSSI value is based on the observation that turbid waters reflect more in the NIR band but less in the visible band. The negative NDSSI value represents the reflectance of water in the NIR band being greater than that in the blue band (Shahzad et al., 2018; Hossain et al., 2010). Therefore, the positive values of NDSSI represent a lower SSC or more transparent water, while a negative value indicates a higher SSC. The spatial patterns of the relative SSC during the typhoon period have been determined using NDSSI.

On the other hand, the empirical model has also been used to quantify the suspended-sediment concentration before and after Typhoon Soulik. This method is widely used for SSC mapping and monitoring around the world (Eom et al., 2017; Hwang et al., 2016; Son et al., 2014; Min et al., 2012; Lee et al., 2011; Choi et al., 2014). For this purpose, we reviewed the existing relations between the in situ SSC (SSC, g m-3) and remote sensing reflectance (Rr) developed by various researchers for the southern and western coasts of South Korea, as illustrated in Table 2. In the present study, the SSC algorithm developed by Choi et al. (2014) for the Mokpo coastal region based on the in situ SSC and a spectral ratio of water reflectance around 660 nm has been used to quantify the SSC distribution. The atmospherically corrected Sentinel-2 MSI image (red band) has been used to calculate the SSC.

The shorelines (i.e., land and water boundary) of the Mokpo coast for short and medium periods have been extracted using a semi-automatic technique (Maiti and Bhattacharya, 2009). Here, we used the normalized difference water index (NDWI) and manual digitization approach to separate the land and water boundary. The technique is widely used for dividing the land and water boundary (Santos et al., 2021; Dai et al., 2019). By using Sentinel-2 MSI imagery, NDWI can be achieved with the following formula (McFeeters, 1996): 7NDWI=ρGreen-ρNIRρGreen+ρNIR, where ρGreen is the green band and ρNIR is the near-infrared band of Sentinel-2 MSI data.

The extracted land and water boundary of the Mokpo region are then converted into polygons, and the shoreline has been determined using ArcGIS software. The shoreline change statistics have been calculated using the DSAS program (Thieler et al., 2009). The extracted shoreline for pre- and post-typhoon periods has been merged, and a 10 m interval transect perpendicular to a baseline has been created (Santos et al., 2021). After that, the net shoreline movement (NSM) method was used to calculate the total shoreline movement (in meters) between the pre- and post-typhoon shoreline positions of each transect (Kermani et al., 2016): 8NSM=shpost-shpre, where shpost and shpre represent the post- and pre-typhoon shoreline positions, respectively.

On the other hand, the backshore surface area changes due to shoreline movement (retreat/advance) over the typhoon period has also been calculated using the Geostatistical Analyst toolbox. Several researchers (Awad and El-Sayed, 2021; Deabes, 2017; Kermani et al., 2016) have also previously mapped the surface changes in the backshore region. To create the surface area change map, we first generated two polygon layers based on the extracted shoreline, one for the pre- and one for the post-typhoon periods. Next, we utilized the Symmetrical Difference tool in ArcGIS to compute the difference between these polygon layers during the period affected by the typhoon. Finally, two feature classes have been derived, one for erosion and another for accretion. In addition, the attribute table contained in each zone illustrates the magnitude of spatial changes (amount of erosion and accretion) during the typhoon period.

The VDSM shows the degree of vegetation damage due to typhoons. The comparison of pre- and post-typhoon NDVI and FVC distribution shows a significant loss of vegetated land as the number of no-productivity and low-productivity pixels increases in the post-typhoon NDVI and FVC image.

Figure 5 depicts the spatial distribution of pre- and post-typhoon NDVI images. Further, to determine the severity of vegetation damage, the pre- and post-typhoon NDVI images have been classified into six categories, namely non-vegetation (-1.0–0.0), low vegetation (0.0–0.2), medium-low vegetation (0.2–0.4), medium vegetation (0.4–0.6), medium-high vegetation (0.6–0.8), and high vegetation (0.8–1.0). The pre- and post-typhoon mean NDVI values were observed to be 0.159 and 0.143, respectively, indicating a mean NDVI value decline of 0.016 after the typhoon.

Table 3 depicts the area changes for each NDVI category over the typhoon period. It has been observed that the high NDVI values (> 0.8) have changed drastically after Typhoon Soulik. The area changes in the low- and non-vegetation categories along the Mokpo coastal region revealed that the wetland (mudflat) had accreted after the typhoon. On the other hand, the post-typhoon image was acquired 2 months after Typhoon Soulik, which suggests that the grasses and crops have recovered well. This recovery is reflected in Table 3 from medium-low to medium-high NDVI levels.

On the other hand, the physical presence of vegetation has also been measured using FVC analysis. In general, NDVI provides information on the health and productivity of vegetation, while FVC provides information on the physical presence and distribution of vegetation. Figure 6 depicts the pre- and post-typhoon FVC map of the Mokpo coast. The area of each FVC category is illustrated in Table 4. The results reveal that the typhoon caused a substantial decrease in FVC in the area, with the average FVC reducing significantly from 33.43 % to 23.64 % after the typhoon. It was observed that the area of medium-high to high FVC decreased from 485.5 to 212.2 km2, while the area of medium to low FVC increased from 1360.1 to 1633.5 km2. The vegetation category of high FVC was more severely affected and decreased considerably after the typhoon. These results indicate that the typhoon significantly impacted the wetland vegetation in the region.

In order to determine the damaged vegetation areas along the Mokpo coast, we compared pre- and post-typhoon NDVI images. A decrease in ΔNDVI is one of the most distinctive features of abrupt canopy modifications detectable by optical remote sensing (Xu et al., 2021). Thus, we can only determine vegetation deterioration from the two NDVI images. Subsequently, an NDVI threshold of 0.2 has been used to extract only vegetation features from the pre- and post-typhoon NDVI images. The threshold value has been manually adjusted to achieve the highest accuracy of vegetation pixels. The extracted vegetated pixels have been compared with reference samples randomly collected from the original high-spatial-resolution images to determine the accuracy (Schneider, 2012; Xu et al., 2021). The two extracted vegetation images obtained within 6 or 7 weeks of Typhoon Soulik (i.e., before the damaged vegetation had recovered) exhibits an overall accuracy of 95.7 % for the pre- and 94.5 % for the post-typhoon periods.

Figure 7a depicts the spatial distribution of ΔNDVI, where the negative ΔNDVI indicates a region with highly impacted vegetation areas. The negative ΔNDVI is attributed to about 26.7 % of the total area (1845.60 km2), which suggests that Typhoon Soulik affected approximately 493.98 km2 of vegetated land. The lowest ΔNDVI value is -0.89, which indicates either tree wind throws or a change in land surface cover from vegetation to built-up land or other non-vegetation covers (Zhang et al., 2013). The results showed that wetland vegetation and agricultural land experienced the most significant NDVI changes, with ΔNDVI values below -0.3. This suggests that these two types of land cover were severely affected by Typhoon Soulik.

On the other hand, Fig. 7b displays the change map obtained from the difference in FVC (ΔFVC), which reveals areas of altered vegetation after the typhoon. The negative ΔFVC is attributed to about 32.07 % of the total area, which suggests that Typhoon Soulik affected approximately 591.89 km2 of vegetated land. It has also been observed that the pure vegetation pixels (i.e., NDVI > 0.6 and FVC > 60 %) were drastically changed over the typhoon period. The changed area determined for NDVI and FVC is -153.43 and -273.40 km2, respectively (Tables 3 and 4). The results obtained from both techniques indicate a significant decrease in vegetation cover after the typhoon. The probable reason for the change is that Typhoon Soulik made landfall close to the Mokpo coastal region.

Figure 8 compares vegetation damage based on the number and percentage of the decreased pixels for ΔNDVI and ΔFVC. It exhibits decreased pixels in different categories of vegetation damage, ranging from low damage to extensive damage. The pixels showing the most significant vegetation damage (i.e., ΔNDVI of -0.2 to -0.5 and ΔFVC of -20 % to -50 %) account for about 30.9 % and 61.5 % of the total pixels, respectively. On the other hand, the pixels showing extensive vegetation damage (i.e., ΔNDVI < -0.5 and ΔFVC < -50 %) account for only 8.31 % and 10.76 % of the total pixels. It was observed that the dominant vegetation in the region is wetland vegetation, which is mainly due to the prevalence of wetlands or mudflats in the area. Therefore, the significant vegetation damage implies that wetland vegetation was most severely impacted during typhoons.

The pre- and post-typhoon Sentinel-2 false-color images and the corresponding relative change in NDVIr and ΔFVC values are presented in Fig. 9. The standard false-color composite (FCC) imagery (left panel of Fig. 9) for pre- and post-typhoon shows that NDVIr is more effective in detecting areas of damaged vegetation compared to ΔFVC (right panel, Fig. 9). It was observed that the typhoon-induced damaged vegetation area (i.e., pixels with NDVIr and ΔFVC of < -50 %) detected by NDVIr (106.5 km2) was greater than that detected by ΔFVC (51.3 km2). The dissimilarity in the ability of NDVIr and ΔFVC to detect the destruction of vegetation caused by the typhoon can be ascribed to the alteration in the color of the post-typhoon vegetation. This change can be detected more accurately by NDVI compared to FVC because the vegetation in the affected areas still existed, and there was not a significant reduction in vegetation coverage after the event (Wang and Xu, 2018). Thus, NDVI is highly sensitive to the health status of vegetation and a more appropriate approach for assessing the damage to vegetation induced by the typhoon, while FVC is more representative of vegetation coverage status (Wang and Xu, 2018; Jing et al., 2011). Consequently, the dramatic vegetation loss (< -80 %) that occurred in mostly wetland vegetation is detected mainly in NDVIr. In addition, moderate greenness loss has been identified in natural forests. Furthermore, the decrease in NDVIr values from higher classes to lower classes indicates that the typhoon has severely damaged the low-lying coastal regions and the wetland vegetation.

The affected area's topography can influence typhoons' impact on vegetation. The interaction between topography and typhoon-generated wind and rain can result in complex and varied patterns of damage across different landscapes (Abbas et al., 2020; Lu et al., 2020; Zhang et al., 2013). This can affect the severity and spatial patterns of vegetation damage. Therefore, the relationship between topography and damaged vegetation has also been established in the present study. For this purpose, high-resolution (5 m × 5 m) DEM data provided by the NGII are used to calculate the region's topographic slope and explore the relationship between topography and typhoon-induced vegetation damage.

The Mokpo coastal region showed an elevation range between 0 and 403 m, as shown in Fig. 1b. It was observed that the number of trees damaged by Typhoon Soulik decreased as the elevation increased, as illustrated in Fig. 10a. The highest number of damaged trees was observed in areas with an elevation of 50 m or less. This is likely due to the fact that these areas are predominantly covered by wetlands, which can be more vulnerable to strong winds associated with Typhoon Soulik. In general, low-lying areas may not have the same natural windbreaks and barriers as higher elevations, which can exacerbate the impact of the wind. In addition, lowly elevated vegetation may have shallower root systems due to the less stable soil conditions, making them more vulnerable to uprooting during heavy rainfall or strong winds (Zhang et al., 2013; Lugo et al., 1983). A significant difference in the number of decreased ΔNDVI and ΔFVC pixels was observed among different elevation ranges, and a correlation analysis between the number of damaged pixels and elevations showed a negative correlation (i.e., damaged pixels decreased with increasing elevation). The majority of damaged pixels (76.37 %) were observed at elevations between 0 and 50 m, with a decrease to 13.5 % between 51 and 100 m. The vegetation exhibited a sharp decline at higher elevations, as shown in Fig. 10a, with the proportion of pixels displaying negative ΔNDVI and ΔFVC, decreasing to 6.1 % between 100 and 150 m and decreasing to 0.02 % between 350 and 403 m.

On the other hand, Fig. 10b illustrates the extent of damaged vegetation across different slope ranges. It has been noted that there is a negative correlation between the slope and the percentage of damaged vegetation pixels, indicating that the amount of vegetation damage decreases with a higher slope. For instance, when the slope was between 0–5∘, approximately 47.63 % of vegetation pixels were damaged. As the slope increased, the percentage of damaged vegetation pixels decreased accordingly, with values of 18.15 %, 15.01 %, 10.71 %, 7.74 %, 0.73 %, and 0.009 % observed for slope ranges of 5–10, 10–15, 15–20, 20–30, 30–40, and greater than 40∘, respectively.