**3. Results**

The mapped land cover classes of the Qeshm Island mangrove ecosystem are given in Figure 6.

**Figure 6.** Classification map of the Worldview-2 image (**A**), classification map of the Sentinel-2 image with the same extent as the Worldview-2 image (**B**), and the final map based on the upscaling approach (**C**). The visible differences between (I) and (II) are related to misclassified shallow water. This error happened two reasons: First, the spectral profiles of the shallow water and tidal zone classes were similar in the Sentinel-2 image. Second, the date of the images differed, and the relative sea level rise had acted as an important factor in converting the tidal zone class (**D**) to shallow water (**E**). However, we were able to show more details of mangrove ecosystems with this approach, such as individual trees and aerial roots (**F**).

Model accuracies of the RF classification were assessed in two steps. In the first step, random reference data based on the segmentation of Worldview-2 images was used to validate the subset of Sentinel-2 imagery. An overall accuracy of 88% and a kappa coefficient of 0.85 were obtained. The validation revealed the producer's accuracy of the four classes shallow water (96.5%), deep water (94.8%), closed canopy mangrove (89.2%), and mudflat (83.1%) (Figure 7). In the second step of the validation, the overall accuracy of the upscaling approach was calculated at 65.5% and the kappa coefficient was 0.63. Whereas the user's accuracy for the two classes deep water (100%) and closed canopy mangrove (75.1%) was high, the producer's accuracy for the class mudflat with aerial roots (66.1%) and without aerial roots (73.3%) were lower (Figure 8). These two classes included a corollary omission error of 33.9% and 26.7%, respectively. The results of the confusion matrix of the different classification extents are given in Tables 3 and 4.

**Figure 7.** Accuracy statistics of the classification map of Sentinel-2 over the same extent as for Worldview-2.

**Figure 8.** Accuracy statistics of the classification map of the upscaling approach.

**Table 3.** Confusion matrix for the classification map of Sentinel-2 over the same extent as for Worldview-2. Bold-faced numbers indicate the agreemen<sup>t</sup> between a class.



**Table 4.** Confusion matrix for classification map of the upscaling approach. Bold-faced numbers indicate the agreemen<sup>t</sup> between a class.

The classification revealed that the largest area (27,678 ha) belongs to the class deep water and smallest (62 ha) to the class individual mangrove trees. The classes closed canopy mangrove, open canopy mangrove, mudflat, aerial root and tidal zone cover an area of 4857, 3474, 13,099, 2296, and 2026 ha, respectively. The *p*-value of the Wilcoxon test for di fferences in the user's and producer's accuracies between the two classification maps were 0.11 and 0.32, respectively, which is greater than the significance level alpha = 0.05. We can conclude that the accuracy assessments did not di ffer significantly between the two classification maps.
