*3.3. Accuracy Evaluation*

The accuracy of open-water flood mapping from S1 was evaluated against water mapped from S2 by estimating the overall accuracy and kappa hat coefficient. For accuracy assessment, the Semi-Automatic Classification Plugin in QGIS was used [56]. The kappa hat coefficient (*κc*) was estimated for the evaluation of classification accuracy between water mapped from S1 and S2 datasets as follows:

$$\kappa\_{\varepsilon} = \frac{p\_a - p\_{\varepsilon}}{1 - p\_{\varepsilon}} \tag{2}$$

where *pa* represents the overall percentage of agreements between S1 and S2 data in raster and *pe* represents the percentage of chance agreement of S1 and S2 data.

The evaluation of accuracy in the case of water mapped under the vegetation is a complicated task, as water under vegetation cannot be directly mapped from the optical satellite images. The evaluation of water mapped under the vegetation was performed using drone photos and observations (44 observation points) collected at the Soomaa test site on 22 March 2019.

#### **4. Results**

#### *4.1. Mapping Accuracy*

The open water mapping accuracy from EW HV polarization data was evaluated against the MNDWI index estimated from S2 imagery at three test sites (Table 6, Figure 4). The accuracy of open water mapped from S1 EW HV polarization data at the Matsalu test site was 97.8% with a kappa hat coefficient of 0.94 (Figure 4a, Table 6). The accuracy and kappa hat coefficient of S1 IW VH data from the Alam-Pedja (96.70% and 0.84, respectively; Figure 4c) and Matsalu (95.90% and 0.86, respectively; Figure 4b) test sites were very high, while at the Soomaa test site, the corresponding numbers were lower, 93.60% and 0.62, respectively (Figure 4d).

**Table 6.** Validation results. Accuracy and *κc* values for different test sites and imaging modes.


**Figure 4.** Validation results. (**a**) Open water mapped from S1 EW HV data vs. S2 MNDWI (4 April 2019), (**b**) Open water mapped from S1 IW VH data vs. S2 MNDWI (23 March 2019), (**c**) Open water mapped from S1 IW VH data vs. S2 MNDWI (5 April 2019), (**d**) Open water mapped from S1 IW VH data vs. S2 MNDWI (16 November 2017) and (**e**) Comparison of in situ validation points (22 March 2019) with open water mapped from S1 VH data (24 March 2019) and water mapped under vegetation from S1 EW HH data (23 March 2019).

> The evaluation of the water mapped under vegetation was performed using drone photos and observations collected at the Soomaa test site on 22 March 2019. Figure 4e shows the water mapped from S1 images and observation points of the water. As seen in Figure 4e, the validation points coincide with mapped water under vegetation from the S1 EW mode HH data. However, it must be noted that the applied methodology for mapping FUV is less accurate (incidence angle normalization is not applicable) (Figure 2b). Still, we see that statistical analysis enables the identification of areas where water accumulates in forested areas.
