In this issue we are presenting the 3^rd and last BEACON Business Case that represents the Greek Case, conducted in partnership with KARAVIAS Underwriting Agency based in Athens, Greece. Karavias Underwriting Agency was founded in May 2014, as the continuation of Karavias & Associates who has been handling Binder Agreements since 1992 for a variety of risks. The Agency provides risk assessment, risk underwriting, policy issuance and claims management services.

This Business Case focused on flood insurance. The Greek Business Case has been implemented in order to approve crop damages caused by flood and validate the accuracy of the BEACON Damage Assessment Calculator component.

Karavias provided crop damages originated from Thessaly region. Thessaly is a lowland area located in Central Greece.  It is a closed geographic region with mountainous borders on three sides. Karditsa Basin in the west and Larisa Basin in the east cover most of Thessaly. Notably, it is estimated that almost 14.2% of the primary products of Greece are being produced on the Thessalian plain.

Karavia’s input included flood damages mostly related to winter cereals, as well as summer crops, such as cotton and maize. The damage severity caused by flood in the samples ranged between 40% and 100%. The provided parcel size area ranged between 0.1ha and 55ha, with an average area of 3ha.

These parcels were insured during the growing season of 2019-2020. The total number of provided damaged parcels from flood was 165. For flood damage assessment, BEACON employs optical and C-Band SAR satellite data for delineating and mapping a flooded area, for determining the beginning and the duration of the flood, as well as for the damage assessment.

Natural hazards included in this case were heavy rains, long periods of rain, and snowmelt flash-floods and floods that could be induced when river overflow their banks and water inundates the surrounding agricultural land. The actual flood mapping took place after this preliminary processing of SAR GRD images, with the use of Sigma VV timeseries.

In particular, the method was based on multi- temporal image series analysis and the computation of two indices: the Normalized Difference Flood Index (NDFI) for highlighting flooded areas, and the Normalized Difference Flood in Vegetated areas Index (NDFVI) for highlighting shallow water in short vegetation.

Initially a statistical analysis of SAR time-series was performed to better characterise the behaviour of each pixel in a specified area, and then normalized indices were calculated which help to effectively compress information on long time series. The calculated temporal statistics were used to compute the NDFI, which aimed at highlighting temporary open water bodies. NDFI easily allowed to categorize as “flooded” those areas solely temporarily covered by water with respect to permanent water bodies and non-water land cover classes. Normalizing the difference, NDFI assisted in defining a threshold for values between 0 and 1 to mask the inundated areas and calculate the flood extent.

On the other hand, NDFVI was computed to detect shallow water in short vegetation. NDFVI was used for detecting and delineating flood events in well-developed crops, which is particularly important in BEACON.

After the computation of the two indices, a suitable threshold value was applied to extract flooded areas. The maps obtained after the thresholding were then filtered to reduce three sources of errors. Firstly, to reduce the effect of speckle, the classified image was segmented and filtered using two morphological filters: 1. dilate filter which gradually enlarged the boundaries of regions of foreground pixels, and 2. closing filter which removed small holes while preserving boundaries. Both filters used a window of 3 by 3 pixels, which was identified as a suitable solution based on empirical analysis. Secondly, all clusters smaller than 10pixels were excluded to reduce spurious flooded areas. Thirdly, the pixels falling in a slope of>5^o, where a flood would be unlikely, were excluded. The two morphological filters allowed the inclusion of flooded pixels that had slightly lower NDFI values but still with high probability to be part of the flood.