The NAIP NDWI [16] was calculated from two surface reflectance bands using the equation:

$\text{NDWI}=\left(\text{Green}-\text{NIR}\right)/\left(\text{Green}+\text{NIR}\right)$(1)

where, Green is the reflectance band from 0.53 to 0.59 µm, and NIR is the near-infrared reflectance band from 0.83 to 0.92 µm. The difference in land/water area (dNDWI) from one year to another was mapped for every major subbasin and estuary salinity zone inside the study area.

The NDWI values are a ratio that ranges from −1 to 1, with negative values indicating land that includes soil and live vegetation cover, and positive values indicating a majority of water coverage. If NDWI ranges from −1.0 to −0.2, then the pixel is 100% land, and if NDWI ranges from −0.2 to 0.0, then the pixel is a majority land and a minority water (Potter and Amer, 2020). If NDWI ranges from 0.0 to +0.49, then the pixel is a majority water and a minority land, and if NDWI ranges from +0.5 to +1.0, then the pixel is 100% water [16] .

While there has been no spectral or atmospheric preprocessing of NAIP imagery regarding calibration across acquisition dates, the conversion of the RGB-NIR NAIP bands into a NDWI index effectively normalizes any spectral inconsistencies between image acquisition dates. A ratio of bands minimizes the absolute reflectance value differences between image acquisition dates and instead represents the NAIP pixel values in relative terms. Furthermore, atmospheric aerosol scattering effects in the NIR band region are weak, and consequently a time series of NDWI is less sensitive to atmospheric effects than is the normalized difference vegetation index (NDVI) [17] .

The aircraft collecting different NAIP image data for this study would have passed over at different times of the day during the summer and autumn months, subjecting the images to subtle variations in tidal water levels along the lowest-lying areas of the Barataria and Breton Sound basins. The maximum daily tidal amplitude along the southeastern Louisiana coast is normally less than 20 cm [18] , which would potentially introduce a higher level of uncertainty for pixels in the study area that were less than 15 cm elevation above sea level at any given hour and day. Because sea levels are rising in the Gulf of Mexico, the precise elevation relative to the sea level of any NAIP pixel area in the study region is a continuously changing variable over the four-year time period of this study.

Aerial image data were acquired from NAIP (www.fsa.usda.gov) as four-band imagery, composed of red, green, blue (RGB), and near-infrared (NIR) reflectance bands. Each NAIP image tile (DOQQ) covers a 3.7-minute quarter quadrangle.

NAIP image data acquired for the years 2019 (June-July), 2021 (November), and 2023 (June-July) were converted to NDWI raster images using Equation 1 for the Plaquemines Parish sections of the Barataria and Breton Sound basins. These NDWI products were regrouped into 10 equal-interval classes for subsequent statistical summaries. This processing step was taken to bin the NDWI values into broad classes of land/water cover and, as such, assure a high level of consistency between images collected on different dates.

Zonal statistics (including mean and standard deviation) for NDWI in each year were computed for the groups of NDWI class pixels that intersect selected a sub-basin polygon in a vector base layer created for the United States Geological Survey (USGS) Watershed Boundary Dataset (WBD). The hydrologic units (HU) in the WBD are arranged in a nested, 12-digit hierarchical system with each HU in the system identified using a unique code [19] . Any HU in the WBD defines the areal extent of surface water drainage to an outlet point on a dendritic stream network or to outlet points where the stream network is not dendritic.

To characterize the wetland plant species that occurred most frequently in each of the ten equal-interval classes for the NAIP NDWI mapped for the Plaquemines Parish sections of Barataria and Breton Sound basins, the point sampling dataset reported by Nyman et al. (2022) [20] was overlaid on the 2019 NDWI raster layer. Aerial surveys of wetland plant species were conducted in early 2021 from a helicopter in transects oriented in a north-south direction and spaced 3 km apart. Visual survey sites were located at a spacing of 0.8 km along these transects. Plant species present were identified, and their abundance was ranked by percent cover at each survey site. Based on species composition and abundance, each marsh survey station was also assigned a marsh type: fresh, intermediate, brackish, or saline marsh. For the present study, a 1-m polygon buffer zone around each marshland survey point was generated and used to calculate zonal statistics for 2019 NAIP NDWI values at a total of 1154 survey points.

Landscape metrics for the ten equal-interval classes of NDWI were computed using the FRAGSTATS program [21] . Class-level indices from FRAGSTATS were used to quantify the densities and spatial configuration of patches, providing a variety of metrics for the geometry, complexity, and aggregation levels of the wetland- versus water-covered patches across the study area.

The simplest index of landscape pattern is patch size. Patch size distribution can be summarized at the class level by a variety of metrics (e.g., mean, median, max and variance). The area-weighted mean patch size is also known as correlation length and indicates the distance that one might expect to traverse the map while staying in a particular patch, from a random starting point and moving in a random direction [22] . The boundaries between patches (or edges) represent another fundamental spatial attribute of a class mosaic. The most common measures of patch shape complexity are based fundamentally on the perimeter-to-area ratio (PARA).

The following landscape metrics for class area coverage and edge characteristics were compared for NDWI classes from 2019, 2021, and 2023 NAIP imagery: Percentage of landscape (PLAND), Area-weighted mean patch area (AREA_AM), Coefficient of variation in patch area (AREA_CV), Total edge (TE), Edge Density (ED). PLAND is a measure of how much of the landscape is comprised of a particular class. AREA_AM (m) provides a patch size metric as a function of the number of patches in the class and total class area. TE (m) is an absolute measure of total edge length of a particular patch type. ED (m ha-1) equals the sum of the lengths (m) of all edge segments in the landscape, divided by the total landscape area.

The following landscape metrics for patch shape characteristics were compared for NDWI classes from 2019, 2021, and 2023 NAIP imagery: GYRATE_AM is the area-weighted mean patch radius of gyration (m), which provides a measure of landscape continuity (also known as correlation length). This metric represents the average traversability of the landscape; specifically, the average distance one can move from a random starting point and traveling in a random direction without leaving the patch. CIRCLE is a shape metric that is computed as one minus patch area (m²) divided by the area of the smallest circumscribing circle; this metric approaches 0 for circular patches and approaches 1 for elongated patches. SHAPE is the shape index metric and equals the patch perimeter (m) divided by the square root of patch area. SHAPE corrects for the size problem of the PARA index by adjusting for a square standard and, as a result, is the simplest and perhaps most straightforward measure of shape complexity. SHAPE approaches 1 when the patch is square and increases without limit as patch shape becomes more irregular.

Four representative sub-basin areas (7 km² each) were selected for FRAGSTATS analysis and visualization of wetland patch dynamics, two in probable impact areas of the Mid-Barataria sediment diversion and two in probable impact areas of the Mid-Breton sediment diversion ( Figure 2 ; geographic coordinates provided in Table 1 ). The central locations of these sub-basins were selected on the basis of the presence of relatively continuous coverage of wetland vegetation in 2019 adjacent to named bayous, ponds, or lakes.

The state of Louisiana’s Coastwide Reference Monitoring System (CRMS) has been designed to monitor the effectiveness of coastal restoration projects [23] . CRMS includes around 400 sites encompassing a range of fresh, intermediate, brackish, and salt marshes, monitored using standardized data collection techniques and fixed sampling schedules. The CRMS datasets for water level and land surface elevation were accessed and plotted for this study to compare to results for recent NAIP water/land class mapping.

The Rod Surface Elevation Table (RSET) method implemented within the CRMS network is used to estimate elevation change rates. One RSET benchmark is located at each CRMS site (with the exception of floating marshes and perpetually flooded, flocculent swamps). From the RSET benchmark, surface elevation is measured at nine points in four directions annually.

Surface elevation change is defined as cumulative elevation change since station establishment. For each sampling event, mean cumulative elevation change is calculated for each of the four directions, and an elevation change rate is estimated using linear regression of elevation change against time. The lines in this figure are linear regressions of elevation change (n = 4 per sample date) over time. Rates reported for elevation change are the slopes of the linear regression (cm yr⁻¹).

In a recent survey report, Sharp (2021) [10] documented Hurricane Ida’s impacts in August 2021 on the monitoring equipment and the marshlands at around 120 CRMS monitoring sites in coastal Louisiana. The photographic information on damage to marshlands assembled for these sites was compared to changes in NAIP NDWI from the 2019 and (post-Ida) 2021 image acquisitions in 100-m radius circular areas (0.3 km²) centered on each reported station geo-location.

The frequency of plant species present across the entire study area in 2021, derived from the aerial survey sampling reported by Nyman et al. (2022) [20] overlaid on the 2019 NDWI raster layer (example in Figure 2 ), showed that the NDWI class 4 (range of −0.2 to −0.4) had about 30% water cover ( Table 2 ) and the highest frequency of freshwater-favoring marsh plants such as Baccharis halimifolia, Colocasia esculenta, Phragmites australis, Sagittaria lancifolia, and Typha domingensis. NDWI class 5 (range of 0.to −0.2) showed 63% water cover and was dominated by Phragmites australis, Spartina alterniflora, and Spartina patens in land cover areas. NDWI classes 6 and 7 (ranges of 0 to 0.2 and 0.2 to 0.4, respectively) both showed about 80% water cover and were dominated by brackish-favoring Spartina alterniflora and Spartina patens in land cover areas. NDWI

Key to plant species in Table 2.

classes 8, 9, and 10 (ranging from 0.4 to 1.0) showed greater than 90% water cover and commonly less than 5% coverage by Spartina sp. or any other wetland plants.

The percent cover of pixels in NAIP images acquired in 2019, 2021, and 2023 for the Barataria Basin portion of the study area was compiled within NDWI equal-interval classes ( Table 3 ) and showed that the greatest magnitude of change

Table 3. Percent cover of pixels in NAIP images from 2019, 2021, and 2023 within NDWI classes of the Barataria Basin of the study area. Coverage classes with the greatest magnitude of change are shaded in gray.

between 2019 and late 2021 (before and after the passage of both Hurricanes Zeta and Ida, respectively) was within NDWI class 5 (range of 0 to −0.2), which showed a 12% increase in marshland area dominated by Phragmites australis and Juncus roemerianus, and also within NDWI class 6 (range of 0 to 0.2), which showed a 16% decrease of marshland area dominated by Spartina sp. by late autumn of 2021. Within NDWI classes 8 and 9, which were relatively shallow water-dominated areas, these zones lost around 6% coverage between 2019 and 2021 and evidently contributed to a 27% increase in deeper water depth (NDWI class 10) coverage by late autumn of 2021, following Hurricanes Zeta (2020) and Ida (2021).

Changes in the percent cover of pixels in NAIP images of the Barataria Basin portion of the study area between late 2021 and mid-2023 showed that the greatest magnitude of change was within NDWI class 7 (range 0.2 to 0.4, 80% water cover) with sparse wetland cover dominated by brackish-favoring Spartina sp. This class 7 coverage increased by nearly 40% over the period since late 2021 (post-Hurricane Ida), whereas NDWI classes 8 - 10 each decreased to 1% or lower cover area in 2023. NDWI classes 5 class, co-dominated by Phragmites australis and Spartina sp., increased by 6% in area over the period since late 2021.

Changes in NAIP NDWI from the 2019 (pre-Ida) and 2021 (post-Ida) image acquisitions at the (total of 85) CRMS sites photographed by Sharp (2021) [10] were summarized in zonal statistics for 100-m radius circular areas (0.3 km²) centered at each reported site (geo-location map in Figure 3 ). Comparison of frequency distribution plots ( Figure 4 ) of the mean NDWI values for these 85 0.3 km² areas showed that the majority of sites shifted from mean values to the range NDWI of −0.2 to 0.2 in 2019 to a NDWI range of 0 to 0.4 in 2021, indicating a higher frequency of post-Ida water coverage. Comparison of frequency distribution plots ( Figure 4 ) of the standard deviation of NDWI values for these 85 0.3 km² areas showed that the majority of sites shifted from values in the range of 0.25 to 0.35 in 2019 to a range of 0.35 to 0.45 in 2021, indicating a higher variability in post-Ida land versus water coverage.

The four CRMS station sites that showed the greatest change in mean NDWI from 2019 to 2021 NAIP image acquisitions (labeled with station numbers in map Figure 3 ) were selected for comparisons to the photographic evidence of marshland change provided by Sharp (2021) [10] . The pre-Ida photos of these locations showed dense coverage of marshland vegetation surrounding the boardwalks and station pole markers at all four CRMS sites ( Figure 5 and Figure 6 ), whereas post-Ida photos from Sharp (2021) [10] showed widespread removal of plant coverage and

replacement by open water surrounding the station markers still remaining after the 2021 storm surge. The disturbance patterns revealed in these before- and after-Ida CRMS photographs verified that NAIP NDWI changes accurately quantified a higher frequency of post-Ida water coverage.

The 2019 to 2021 (post-Ida) NAIP NDWI map comparison at CRMS station 115 ( Figure 7 ), located in the Spanish Lake sub-basin near the Pipeline Canal of inner Breton Sound Basin, showed extensive erosion of the vegetated marshland shoreline and transformation of NDWI class 5 (range of 0 to −0.2) dominated by Phragmites australis and Spartina sp. to NDWI class 8 open water coverage. Likewise, the 2019 to 2021 (post-Ida) NAIP NDWI map comparison at CRMS station 161 ( Figure 7 ), located near the Raphael Pass sub-basin in the transition from Breton Sound Basin to the Chandaleur Sound, showed loss vegetated marshland, widening of several canals, large pond formations, and extensive transformation of NDWI classes 4 and 5 dominated by Spartina sp. to NDWI classes 8 to 10 open water coverage.

The 2019 to 2021 (post-Ida) NAIP NDWI map comparison at CRMS station 4103 ( Figure 8 ), north of the Bayou Dupont sub-basin near a pipeline canal of inner Barataria Basin, showed deep erosion of all vegetated bayou shorelines, extensive pond formation and fragmentation of marshlands, and extensive transformation of most NDWI classes 4 and 5 coverage dominated by Phragmites australis and Spartina sp. to NDWI classes 8 and 9 open water coverage. The 2019 to 2021 (post-Ida) NAIP NDWI map comparison at CRMS stations 387 (sub-stations 1 and 2; Figure 8 ), located in the Grand Bayou Blue sub-basin of the Terrebonne Basin, showed nearly complete loss of vegetated marshland coverage, shoreline fragmentation, large pond formations, and extensive transformation of most NDWI classes 5 and 6 dominated by Spartina sp. to NDWI classes 9 to 10 open water coverage. These are now “plucked” marshlands that show numerous closely spaced, irregular scours and ponds [24] .

Analysis of the area cover and edge complexity of the wetland versus water-covered patches in probable impact areas of the Mid-Barataria and Mid-Breton sediment diversions was carried out for four circular 7 km² sub-basin areas, each with a diameter of 3 km ( Figure 9 ). FRAGSTATS results ( Table 4 ) showed that the NDWI class 4 (range of −0.2 to −0.4), dominated by freshwater-favoring marsh plants, had a PLAND of just 2.5% in 2019 and a relatively low AREA_AM of 1.4 m². In contrast, NDWI class 5 (range of 0 to −0.2), dominated by brackish-favoring marsh plants like Spartina sp., covered 15% of these sub-basin areas in 2019 and showed higher ED values along with an AREA_AM of 11 m². NDWI classes 6 - 8 (ranging from 0 to 0.6) and dominated by 80% - 90% water cover, covered 51% of these sub-basin areas in 2019 and showed the highest edge length (TE and ED) index values of all the NDWI classes. The AREA_AM coverage of these three classes ranged from 14 to 50 m² in 2019. NDWI class 9 covered 32% of these sub-basins in 2019 with an AREA_AM coverage of 253 m², whereas the (deepest) water class 10 covered less than 1% of these sub-basins in 2019.

FRAGSTATS analysis of the shape and connectivity levels of the wetland versus water-covered patches in probable impact areas of the Mid-Barataria and Mid-Breton sediment diversions ( Table 4 ) showed that the GYRATE and SHAPE metrics increased with increasing water cover across the ten NDWI classes in 2019. A longer average traversability length and a more irregular overall shape in NDWI classes 7 - 9 patches were discovered, compared to classes 4 - 6 patches. The CIRCLE metric showed that NDWI classes 9 and 10 (deepest water cover) were those with the most circular (least irregular) patch shapes in 2019.

Between 2019 and 2021 NAIP image acquisitions, several notable changes were detected in the landscape metrics for NDWI classes in these same four 7 km² sub-basins. The most striking shift was in the percent cover (PLAND) and AREA_AM

of shallow water-dominated NDWI classes 6 - 8 that transitioned into relatively large patches of deeper water-dominated classes 9 and 10 ( Table 4 ). In addition, the TE and ED values of NDWI classes 9 and 10 increased by several-fold between 2019 and 2021, suggesting that longer, more complex shorelines and marshland edge boundaries were developed with classes 5 and 6. The reduction of the GYRATE and SHAPE metric values across all NDWI classes 4 - 9 between 2019 and 2021 further suggested that more compact, less irregular patch shapes of both wetland and water cover were formed during the years 2020 and 2021 in the four sub-basins areas. The CIRCLE metric increased noticeably for all NDWI classes implying more circular (and less irregular) patch shapes were formed in these sub-basins between 2019 and 2021.

Between NAIP image acquisitions in 2021 (post-Hurricane Ida) and in 2023, several noteworthy changes were detected in the landscape metrics for NDWI classes in these same four sub-basins. A distinct shift occurred in the percent cover (PLAND) and AREA_AM of deep water-dominated NDWI classes 9 and 10, transitioning into large patches of relatively shallow water-dominated classes 5 - 7 ( Table 4 ). In addition, the TE and ED values of NDWI classes 9 and 10 decreased considerably between 2021 and 2023, transitioning to the longest, most complex shorelines and marshland edge boundaries in NDWI classes 5 - 7. Associated increases in the GYRATE and SHAPE metric values cross NDWI classes 5 - 7 further suggested that less compact, more irregular patch shapes of wetland cover were formed in the four sub-basins areas between 2021 and 2023. The CIRCLE metric decreased markedly for NDWI classes 8 - 10 (predominantly water cover), implying less circular (and more irregular) patch shapes by 2023 for the deepest water cover zones.

Mapping of NAIP NDWI classes in these four 7 km² sub-basins over time revealed numerous locations where shoreline erosion and/or wetland drowning have occurred ( Figure 10 ). Extensive shoreline erosion was detected between 2019 and 2021 on the southern margins of the Bayou McCutchen sub-basin and in the center of the Bayou Tambour sub-basin. In the Breton Sound sub-basin areas of Little Oak Pond and Lake Batola, replacement of NDWI class 6 (partially vegetated) marshland patches detected in 2019 by NDWI class 9 and 10 open water cover was extensive by 2021.

Patterns of change between 2019 and 2021 in NDWI classes 7 – 10 could be seen in these four 7 km² sub-basins, most prominently in the shift from a predominance of shallow water depths to deeper water depths in all sub-basin examples. The absence of smooth gradations from NDWI class 7 to class 8 that were evident in 2019 NAIP images, instead showing deeper NDWI class 9 and 10 areas in 2021 NAIP imagery in these buffer zones adjacent to patches of marshland plants.

The changes in patch geometry between 2021 (post-Hurricanes Zeta and Ida) and 2023 in NDWI classes could be seen most prominently in the shift from a predominance of relatively deep-water depths to shallower water depths and a widespread recovery of NDWI class 7 (80% water cover) with sparse wetland cover dominated by brackish-favoring Spartina sp. Patches of NDWI classes 4 and 5, dominated by Phragmites australis and Juncus roemerianus in 2019 NAIP imagery, have failed to recover and regrow extensively since 2021 in these four sub-basins ( Figure 10 ).

Spatial variability in patch metrics, as measured by the coefficient of variation (CV; Table 4 ), increased for the AREA_AM, SHAPE_AM, and CIRCLE_AM metrics of NDWI classes 9 and 10 between 2019 and 2021. Variability in the patch metrics AREA_AM, GYRATE_AM, and SHAPE_AM of NDWI classes 5 - 7 increased by several-fold between 2021 and 2023.

Plots of water elevation change ( Figure 11 from the CRMS data viewer) from the closest CRMS station locations to the four sub-basins in the Barataria and Breton Sound Basins (shown in Figure 9 ) all showed a substantial rise in water levels immediately following both Hurricane Zeta (2020) and Hurricane Ida (2021).

In general, the water elevation at all four stations reached peak measured change levels of 5 to 15 cm over the time period between 2019 and 2023 immediately following these intense storm surges. Water elevation dropped by around 5 cm over the years 2022 - 2023 at all four stations. This difference in water levels after the two hurricanes in 2020 and 2021 verified that the shift from NDWI class 6 and 7 to classes 9 - 10 was associated with deeper water depths detected by NAIP images. It is worth noting that the monitoring equipment nearest the Lake Batola sub-basin was evidently disabled for around 16 months after the passage of Hurricane Ida.

Plots of land surface elevation change ( Figure 12 from the CRMS data viewer) from the closest CRMS station locations to the four sub-basins in the Barataria and Breton Sound Basins (shown in Figure 9 ) displayed more diverse patterns. The CRMS stations closest to the Bayou McCutchen and Lake Batola sub-basins both showed steep negative trends in land elevation levels (and apparent wetland drowning) following Hurricane Zeta, whereas land elevation trends at the stations closest to the Bayou Tambour and Little Oak Pond sub-basins showed minor declines in land elevation following Hurricane Zeta and Hurricane Ida.