Evaluation of Fire Danger and Fire Potential Indices for South Africa : case studies in Mpumalanga and the Western Cape

Abstract

Wildfires are a common phenomenon on earth and can have disastrous effects on the environment, infrastructure and surrounding communities. At the same time, many ecosystems are fire prone and require burning at regular intervals, in order to maintain the health of the ecosystems. It is necessary to minimise the negative effects of fires where possible. Information needs to be provided to fire management officials to facilitate efficient planning and mitigation in order to minimise the negative effects. Wildfires are influenced by many variables including vegetation type, fuel load, fuel moisture, proximity to roads, proximity to settlements, elevation, slope, aspect, temperature, precipitation, wind and relative humidity. These variables can be used to build a fire potential index that determines the probability of a fire occurrence and the possibility of the fire to become an out of control fire. Fire potential indices provide information on where fire potential is high so fire management officials can plan resources accordingly and thus minimise negative impacts of wildfires. Many fire potential indices have been developed but their usefulness in South Africa has not been verified. The aim of the research was to implement and evaluate different fire potential indices utilising geographic information, including remote sensing products, to predict fire potential in South Africa. The Mpumalanga and the Western Cape provinces were used as case studies. The time periods included February to December 2015 for Mpumalanga and August 2014 to June 2015 for the Western Cape. A number of candidate fire potential indices were implemented in the Python scripting language. A variety of data sources were used to implement the fire potential indices. The fire potential indices were evaluated along with a few fire danger indices. The performance evaluation compared satellite detected active fire events to the fire potential indices in the study areas based on statistical metrics including Pseudo R2, C-Index, Eastaugh’s Two-Part Parametric, Bhattacharyya Coefficient and Percentile Shift. The evaluation was performed per pixel for the entire date range. A performance ranking was then calculated for all the indices based on the pixel performance and a final ranking was assigned to each index. The Fire Potential Index performed best amongst the implemented candidate fire potential indices. The Canadian Fire Weather Index performed well in Mpumalanga and the Fine Fuel Moisture Code performed well in the Western Cape. The overall performance of the indices was not very high. This is due to the fact that even though fire potential is high in an area, an ignition source might not be present to cause an actual fire event. The performance of fire potential indices and fire danger indices were different in the two provinces. Future work can be done to develop an index based on South African conditions or calibrate the indices implemented in this research for an area.