In this issue we are presenting BEACON’s 2nd Business case that represents the Serbian Business Case conducted together with the Agricultural Insurance company Triglav. Triblav Osiguranje is a part of the Adria region – Triglav Group, one of the leading insurance groups in SouthEast Europe. Triglav Osiguranje is a very active insurance company on the Serbian market (among the 5 biggest insurance companies) and active in provision of agri-insurance services for last couple of years.

BEACON services were tested in this case focusing on the insurance against hail as 90% of damages in Serbia are caused by this weather event. This business case was implemented in the frames of approving crop damages caused by hail and validated the accuracy of the BEACON Damage Assessment calculator component. TRIGLAV provided crop damages originated from Vojvodina the Serbian flatland. Vojvodina is a region located in the northern part of Serbia and has the best quality of herbal land in the whole of Serbia with high standards of yield.

The three most dominant crop types in the region are maize, wheat and soya bean. Therefore, Triglav provided input on hail damages mostly related to these crops. The damage severity caused by hail in the samples ranged between less than 5% and 100%, for the three selected crops for parcels that were insured during the growing season of 2015-20106 to the growing season of 2019-2020.

The general framework under which the Damage Assessment Calculator estimates hail damage on insured parcels, is the implementation of a change detection index. The DPI (Difference Percentage Index) serves as a measure and can be applied on any vegetation index or canopy biophysical paramater of choice, to detect significant change index or canopy biophysical parameter of choice, to detect significant change between pre- and post-hazard crop status. Optical satellite data was used for the calculation of NDVI, GNDVI, MCARI, REIP vegetation indices (with 10m spatial resolution) and the biophysical parameters Leaf Area Index, FCOVER and FAPAR (with 20m spatial resolution). The indices were then used to calculate the DPI index per applying Equation.

In the Damage Assessment Calculator of BEACON, the MPDI was calculated and tested in change detection. Other SAR derived indices tested in the DPI, were the PRI (Polarization Ratio Index), which is the VH/VV ratio, as well as the VH and the VV backscatter signals. To tackle the irregularity in pre- and post-damage image acquisition dates due to cloud cover, the pre-event and post event sampled images were created, corresponding to the Vegetation Index feature of 6days before the event and 20days after the hail event.

The logic behind the machine learning approach tested in BEACON, is described and illustrated in the following workflow for damage classification. Firstly, the insurer imports the date of the hail event. Then, the pre-event composite image is produced by the 36-day pre-event image stack by the temporal interpolation tool. Also, the post-event composite image is produced by the 45-day post-event image stack by the temporal interpolation tool. The Difference Percentage Index is produced by the pre- and post-damage composite image.  The zonal statistics are extracted across the parcel (zonal minimum, maximum and average). The machine learning model takes into account the crop features such as growth stage and parcel-based Difference Percentage Index to predict and classify the damage. In the end the output is damage classification (0-10, 10-30, 30-70 and 70-100%) or a single damage percentage.