University of Twente Student Theses

Login

Forest fire detection for near real-time monitoring using geostationary satellites

Manyangadze, Tawanda (2009) Forest fire detection for near real-time monitoring using geostationary satellites.

[img] PDF
1MB
Abstract:Forest fires are an important component of the savanna, tundra and boreal forest ecosystems. The increasing rate of the occurrence of fires however has increased the concern over their impacts on climate change and fragile ecosystems. This requires efficient and effective methods in forest fire detection for near real-time monitoring so as to minimize these impacts. Remote sensing has been widely used in active forest fire detection; however there are some limitations in contextual algorithms which are used in forest fire detection. These contextual algorithms are greatly affected by clouds and different land cover types such as land and water with inherent temperatures included in the N x N matrix and this brings errors. As a step towards minimizing these problems an automated multi-temporal threshold algorithm was developed in this study using MSG satellite and ground fire data from Portugal. The algorithm is based on temperature anomalies detected in IR3.9 channel and the difference between IR3.9 and IR10.8 channels as well as the solar zenith angles for day, night and twilight conditions. Thresholds were set to determine actual fires and possible fires depending on how far the temperature of a particular point or pixel deviates from the normal background temperature which is estimated using the images directly prior to the actual image. The accuracy of the algorithm was compared with that of the MSG FIR-G product. The McNemar’s test was used for significance test of the difference between the multi-temporal threshold algorithm and the MSG FIR-G product which uses a contextual algorithm. This study shows that the multi-temporal threshold algorithm has higher fire detection rate (50%) as compared to MSG FIR-G (3.7%) when ground data from Portugal was used for validation. There is a significant difference between these methods (McNemar’s test ( x2) = 5.45, df = 1, p-value = 0.0196). The superiority of the multitemporal threshold algorithm over the contextual algorithm and significant difference between these methods was also confirmed in Southern Africa when MODIS fire product was used for validation. The automated procedure takes less than 15 minutes to produce the fire map, so it can cope with MSG satellite 15 minutes temporal resolution. Therefore the multi-temporal threshold algorithm performs better than the contextual algorithms in forest fire detection however there are some outstanding problems such as the transparent clouds that are not easily detected which may increase the errors in fire detection. Although this method was developed based on Portugal data it has been shown that it can be applied to other areas in the view of MSG satellite. This method can be easily adapted to other geostationary satellites and only the solar zenith angles have to be specific to the particular satellite.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/92711
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page