*4.5. Accuracy*

The last but also the most important step for selecting the best fire danger rating system or index is estimating the respective accuracy. For the evaluation process, four regions within the Greek territory were selected, as depicted in Figure 2: (1) Mt. Penteli region in Attica; (2) the Regional Authority of Evros, northeastern Greece; (3) the Region of Kimi-Aliveri in Evoia; and (4) the Regional Authority of Helia-Achaia. As shown in Figure 2, fire incidents for the period 01/06/2022–31/07/2022 were gathered from satellite images provided by the NASA-FIRMS (https://firms.modaps.eosdis.nasa.gov/map/, accessed on 1 August 2022) [234] as well as meteorological data from the local weather stations, provided by the National Weather Service of Greece and the National Observatory of Athens.

**Figure 2.** Fire incidents and weather stations in the areas of interest.

For calculating the values of the indices for every region, a computer program in Python was developed according to the equations presented in this article and the respective SM file. The program uses as input the meteorological data, i.e., temperature, relative humidity, wind speed, and precipitation depth, while it calculates other intermediate parameters needed for the computation of the fire indices, such as the number of days of drought, evapotranspiration, etc. The computational procedure concludes with the output values of every index on daily basis from 1 June 2022 to 31 July 2022. Unfortunately, due to lack of some specialized data, not all presented indices were used. The following indices were excluded from this accuracy experiment: BEHAVE, LASI, FFBT, ICONA, IREPI, PDSI, CWD, Darcy, as well as remote sensing indices, as a different approach would have to be adopted; thus, the comparison would be unrepresentative.

According to the index values in relation to fire incident or no fire occurrence, the following four cases were examined: (1a) no fire occurrence and index "hit"; (1b) no fire occurrence and index "miss"; (2a) fire occurrence and index "hit"; and (2b) fire occurrence and index "miss". In case of no fire incident, an index hit is considered as the outcome value out of the range of extreme fire danger class—according to SM tables. Accordingly, in case of a fire event, an index hit is when the respective outcome value is in the range of the extreme fire danger class. Fire danger classes for some of the indices had to be redefined to be realistic in the Greek environment in order to fit with the outcome value ranges. Cases 1a and 1b were marked with 1 and 0 points, respectively, per day of calculations, while cases 2a and 2b, which were considered more important for fire management, were marked with 2 and −2, respectively. For the final score, all index marks were normalized by dividing with 277 (the maximum mark for all days and all regions). The five most-accurate indices for the tested period and regions in Greece were proven to be the following: (1) NI; (2) KBDI; (3) SFDI; (4) FFDI5; and (5) SDI. The five least-accurate indices were as follows: (1) DW; (2) r (Orieux); (3) IBr; (4) IFI; and (5) PEI. The NI has also been successfully applied in mountainous areas in Greece [235–237], corroborating the findings of the present article.

Eventually, the scores of indices (presented in Table 3) and the accuracy marks were summed, using equal weights (divided with the respective maximum grade), to produce the final evaluation of the included indices, as shown in Table 4.


**Table 4.** Overall performance of environmental fire danger systems and indices based on the five groups of criteria.
