Χώρο-χρονική ανάλυση σημείων έναρξης δασικών πυρκαγιών με τη χρήση των γεωγραφικών συστημάτων πληροφοριών. Για την χρονική περίοδο 2020-2024
Spatio-temporal analysis of wildfires ignitions using geographic information systems
View/
Keywords
Δασικές πυρκαγιές ; Γεωγραφικά συστήματα πληροφοριών ; Χωρική ανάλυση ; Διαχείριση κινδύνουAbstract
This study examines spatial and temporal patterns of wildfire ignition events in Greece for the period 2000-2024, utilizing Geographic Information Systems tools. The aim is to understand the factors associated with wildfire occurrence, identify trends and patterns in both space and time, and develop a predictive model of ignition risk. The analysis of these patterns contributes to a better understanding of wildfire dynamics and supports preventive planning.
Initially, a literature review was conducted, focusing on wildfires and the key factors influencing ignition events. At the same time, the role of GIS as tool for analysis, management and decision support in natural hazard environments was examined, as well as its importance in understanding complex phenomena related to climate change. In addition, the role of machine learning models and their integration with geospatial data for predicting potential future events was investigated.
Next, corresponding to the practical component of the study, data were collected, processed and transformed into an appropriate format for analysis. Due to limitations in the spatial accuracy of the data prior to 2020, the methodological approach was distinguished between the two different time periods. Specifically, for the period up to 2019, a more generalized spatial analysis was conducted at the level of operational service areas, aiming to capture the distribution of ignition events and burned areas. In the other hand, for the period 2020-2024, where accurate spatial coordinates were available, ignition points were enriched with additional geospatial and climate variables, including elevation, slope, land use, population density, drought index and mean temperature.
Subsequently, spatial and temporal analyses were carried out, along with the development of applications and interactive maps for improved visualization of the data and the results. Finally, a Random Forest machine learning model was implemented, trained on the available data and evaluated to predict the probability of wildfire ignition incidents.
The results revealed spatial and temporal patterns, with ignition events primarily associated with human activities and predominantly occurring in agriculture and mixed areas. Strong seasonality was also observed. The predictive model demonstrated satisfactory performance and identified areas of increased risk, confirming the contribution of geospatial technologies and machine learning to the study and management of wildfire related risk.