A Practical Approach to Assess the Wildfire Ignition and Spreading Capacities of Vegetated Areas at Landscape-scale


 We bring a practical and comprehensive GIS-based framework to utilize freely available remote sensed datasets to assess wildfire ignition probability and spreading capacities of vegetated landscapes. The study area consists of the country-level scale of the Romanian territory, characterized by a diversity of vegetated landscapes threatened by the consequences of climate change. We utilize the Wildfire Ignition Probability/ Wildfire Spreading Capacity Index (WIPI/ WSCI). WIPI/ WSCI models rely on a multi-criteria data mining procedure assessing the social, environmental, geophysical, and fuel properties of the study area based on open access remote sensed data. We utilized the Receiver Operating Characteristic (ROC) analysis to weigh each indexing criterion's impact factor and assess the model's overall sensitivity. Introducing ROC analysis at an earlier stage of the workflow elevated the final Area Under the Curve (AUC) of WIPI from 0.705 to 0.778 and WSCI from 0.586 to 0.802. The modeling results enable discussion on the vulnerability of protected areas and the exposure of man-made structures to wildfire risk. Our study shows that within the wildland-urban interface of Bucharest's metropolitan area, there is a remarkable building stock like healthcare, residential and educational that are significantly exposed to wildfire spreading the risk.

36 Chapin et al. 4 define wildfire as a wicked problem. According to Levin et al. 5 , wildfires are multi-layered 37 phenomena that implicate diverse interacting cycles between causes and effects acting in certain territories.

38
Identifying the relevant factors that significantly correlate with the wildfire regimes remains a critical 39 challenge to scientists 6-8 .

40
In the classical wildfire assessment approach, the interaction of favorable weather conditions with the 41 study area's geophysical and fuel properties is considered the core prerequisite of the fire environment triangle

52
A more recent study presents the impact of COVID-19 lockdown on the wildfire regimes in a wildfire-53 prone region like the Mediterranean 3,17 . The authors report a significant decrease in the total burned area 54 during this period compared to the estimations that counted for similar drought-related circumstances to 55 previous years. The decrease in social activities has resulted in a significant reduction of wildfire events. Thus, 56 3 the integration of anthropogenic factors within the wildfire risk assessment tools has become indispensable to 57 increase the models' sensitivity 18 .

58
Although the anthropogenic factors that impact wildfire events have gained considerable attention in the 59 literature, their combined usage alongside hydro-meteorological and biophysical factors in wildfire spreading 60 capacities models is not spread enough. In this study, we shortlisted sixteen criteria about the anthropogenic 61 (S-social), hydro-meteorological (E-environmental), geophysical (P-physical), and fuel (F) properties of the 62 study area (Romania) following our earlier GIS-based method 19

80
The study area is represented by Romania's territory, a country located in the central-eastern part of 81 Europe with an area of 238391 km 2 (Fig. 1

130
The third stage consists of subgrouping the criteria into two sets according to their relationship with either 131 wildfire ignition or spreading (see Fig.2, third stage). This division is based on a literature review shown in 132 our earlier work 19 . Moreover, a relevancy indicator is given to each criterion according to their direct or 133 indirect relationship with wildfire regimes. This is explained in detail in Table 1

201
According to Fig.3, the RCM that cross the Romanian territory in the central region from north to south-202 west direction stand as a determinant of the spatial distribution of the relative risk for the majority of the 203 criteria. First, the hydro-meteorological criteria visually correlate with the topography of the study area.

207
Wind speed (E4) is the only environmental criterion that is not tightly correlated with altitude (Fig. 3d).

208
The slopes of RCM face south and south-east and remain an exposed area to winds flowing from the Black 209 Sea and the Mediterranean. Similarly, the remaining plain territories in Romania's south-eastern region facing 210 the Black sea are exposed to considerable average wind speeds compared to the north-western plains (    inventory value and the weighted factor via the ROC/AUC method (Fig. 2). The revised weights, as presented 267 in Table 1, led to improved model sensitivity. Fig. 6  improvement is more visible in the case of the WSCI model (Fig. 6b)

354
We performed an exposure assessment of built structures to the wildfire spreading capacity of vegetated 355 surfaces within the Romanian capital city's metropolitan area (Fig. 9a). The wildfire exposure analysis relies 356 on the juxtaposition between the WSCI_ROC results and existing building stock, as shown in Fig. 9. We bring 357 a demonstrative example from Ilfov metropolitan area, which includes the capital city, Bucharest. The hazard 358 map of wildfire spreading capacity relies on the results reported in this article (see Fig. 7

378
According to the box plot, just one hospital is located within the critical wildfire spreading capacity 379 heatmap. Nevertheless, it has a WSCI_ROC heatmap value of 2.37, which indicates a significant exposure of 380 its users to wildfire spreading risk. According to the outlier values (dots) shown in Fig. 9c, there are four 381 industrial, one house, and one school building highly exposed to wildfire spreading the risk.

403
The method we presented in this study is reproducible in other wildfire-prone geographies. It is also 404 flexible enough to integrate the most up-to-date and the most reliable remotely sensed geospatial data.