*3.1. OBIA Classification*

The 2015 Worldview-2/Worldview-3 image classification achieved an overall accuracy of 92.75% (Table 2). Visual inspections revealed an appealing result with an appropriate gradient from *S. alterniflora*, high marsh, Phragmites, and upland vegetation (Figure 3). Errors were mostly between the two types of *S. alterniflora* due to their spectral similarities and early seasonal acquisition date of the imagery. There was no confusion between mudflat/water and other classes making this classification an ideal baseline for change analysis.

The 2015 Worldview-2/Worldview-3 imagery classification was compared with a 1997 classification conducted with aerial imagery. When comparing salt marsh vegetation between the two periods a reduction from 531.27 ha to 505.95 ha was observed or 1.41 ha y<sup>−</sup>1.


**Table 2.** An error matrix for the 2015 Worldview-2/Worldview-3 classification of Fire Island National Seashore.

**Figure 3.** (**a**) The 2015 Worldview-2/Worldview-3 image classification of salt marshes in FIISs. (**b**) A section of FIIS directly to the west of the old inlet breach. (**c**) The Floyd Bennet estate, a section of FIIS on the mainland.

#### *3.2. Tidal Stage E*ff*ect*

The 2017 NAIP imageries were acquired at an approximate tidal stage of 35.66 cm above MLLW at U.S. Geological Survey (USGS) 01305575 at Watch Hill [43]. The 2014 DEM derived from topobathymetric LiDAR was used to determine how much inundation of *S. alternilflora* would be expected at this elevation. The analysis found 7.39% of the 2015 classification's *S. alterniflora* classes were inundated. The indundation could be subcanopy and have little impact on the 2017 classification. The areas of modeled inundation were most prevalent in mosquito ditches, sandbars, and interior mudflats (Figure 4).

**Figure 4.** (**a**) The 2015 Worldview-2 and Worldview-3 Classification for a section of Fire Island east of the breach. (**b**) The modeled tidal inundation of the 2015 classification at a tidal stage of 14.3 cm above NAVD 88, which corresponded to the highest tidal stage of the imagery used in the study.

#### *3.3. Change Analysis (1994–2017)*

The panne change analysis achieved overall accuracy > 85% for all years (Table 3). However since these classifications were being used in tandem it is important to note that propogated percent error calculated by the square root of the sum of squares was 16.7, 10.2, 11.8, and 13.1 for 1994–2011, 2011–2013, 2013–2015, and 2015–2017, respectively. Models with low overall accuracies were evaluated and manually digitized when necessary. The analysis included 475 pannes mapped in 2015 that were > 10 square meters. In 2015, the mean panne size was 441.6 m compared to the 1994 mean size of 121.4 m. In 1994, 257 of the 475 pannes were vegetated. The mean yearly rates of panne change for each of the periods were 25.00, 5.34, 27.98, and 15.91 m<sup>2</sup> y<sup>−</sup><sup>1</sup> for 2015–2017, 2013–2015, 2011–2013, and 1994–2011, respectively (Figure 5). There were statistical differences between the yearly change rates (H (3) = 30.097, *p* < 0.001) were compared with the Wilcoxon rank sum test (Table 4). There were significant differences between edge erosion rates between 1994 and 2011 and 2011 and 2017 (F1, 1597 = 206.06, *p* <0.001). There were no significant differences between panne change in 1994–2011 and 2011–2017 (F 1,948 = 0.13, *p* = 0.72). The edge erosion rates for the mainland, bayside islands, and barrier island locations were compared before and after the Hurricane Sandy breach with least-square means (Figure 6). Pannes, in general, became larger from 1994 to 2017, the temporal resolution from 2011–2017 shows fluctuation in these increases (Figure 7). In 2015–2017, pannes were more likely to revegetate if they had a hydrological connection to a mosquito ditch (χ2=28.049, *p* < 0.001). Edge erosion to the east of the breach from 2011 to 2017 had a significant linear trend (F1, 27 = 28.2, *p* < 0.001) and an R<sup>2</sup> of 0.51. Edge erosion to the west of the breach from 2011–2017 had no trend (F1,94 = 1.5, *p* = 0.22) and an R<sup>2</sup> of 0.02.


**Table 3.** Panne and edge overall classification accuracies, 1994–2015.

**Figure 5.** Interior pannes total area and change rates from (1994–2017) both the average yearly change rates of a time period and total area of pannes and pools.

**Table 4.** Wilcoxon rank sum test between annual panne change rates.

**Years 2017–2015 2015–2013 2013–2011 2015–2013** *p* < 0.05 **2013–2011** *p* = 0.72 *p* < 0.72 **2011–1994** *p* < 0.001 *p* < 0.72 *p* < 0.001

**Figure 6.** Edge erosion rates compared by time period (1994–2011, 2011–2017) and location (barrier island, mainland, or back bay island) with least square means with Bonferroni *p*-value adjustment. Location/dates that share letters did not demonstrate significant differences (*p* > 0.05).

**Figure 7.** Panne classification from 1994 to 2017 for an area to the west of the old inlet breach. Each inset has the corresponding years NIR in panchromatic or red band in 1994. The locus map is a Sentinel-2 image from 5/21/2016.
