NBR Index-Based Fire Detection Using Sentinel-2 Images and GIS: A Case Study in Mosul Park, Iraq

DOI: https://doi.org/10.52939/ijg.v19i3.2607

Forest fires are the second factor leading to ecosystem degradation after pollution [1], and are natural and man-made disasters that can endanger human life [2]. Forecasting fire hazards is important for preventing fires and allocating resources, and mapping fire risks is an effective way to measure regional risk [3]. Fire-risk controlling starts by assessing the most inflammable sites [2]. Information about the spatial pattern of fire hazards in a region plays a major role in forest succession[4], and environmental impacts [5]. Every year hundreds of hectares are damaged because of forest fires thus its necessary to apply the appropriate management for controlling and assessments [6]. With the advent of geographic information systems (GIS) and Remote Sensing (RS) [3] and [7], fire risk forecast has become popular in sophisticated fire-controlling scenarios [8]. GIS offers massive benefits in evaluating many geographical features and analyzing the events that develop according to these geographical features [9]. GIS identifies and prioritizes the geographical site of fire events and hotspots and enables map development to visualize the scenario results [6]. Besides, GIS develops forest fire hazard maps that attach potential fire regions to environmental factors [10]. Evaluating different map layers related to large regions in a short time [11]. Creating wildfire maps and obtaining spatial data information related to wildfires [12], helps manage fire risk [13]. Moreover, satellite images and RS technology are widely used to monitor forest fire behavior [14].

Recently, severe forest fires have occurred due to climatic influences, including dry periods [15]. Forest areas close to road networks and settlements are more exposed to fires as the possibility of man-made or accidental fires increases [11]. 70% of some forest fires are caused near main roads, with less than 500 m distance [16]. Forest fires can be responsible for severe air quality periods [17]. Where exposure to particulate matter [18] from wildfire smoke is associated with negative health effects [19] and may cause death [17]. Many factors are taken into consideration to detect the forest-fire risk potentially such as; temperature, wind speed, distance from roads, rainfall, slope, altitude, land use, etc [15]. Many studies showed that topography, vegetation, fire history, and climatic factors are significant variables in forest-fire susceptibility modeling [10].

Several techniques are adopted to evaluate the probability of forest fires, such as analytical hierarchy process, support vector machines, fuzzy logic models, and system dynamics [8]. Also, statistical approaches such as Pearson correlation coefficient are used to analyze the relationship between forest fire events and other variables [20]. Besides, regression models are useful in predicting wildfire risks for instance geographically weighted regression, linear regression, and logistic regression [10]. The NBR index is applied to measure spectral changes post-fire events [21]. The ability of the Normalized Burn Ratio index to detect fire-events was investigated [22]. Since 2022 Incidents have become frequent in Iraq, including the outbreak of fires in Mosul Park, which covers an area of about 90 hectares. The effects of the fires have spread to large areas, and large parts of them have turned into burnt areas and ashes.

1.1. Research Objective and Design

Exposure to forest fire smoke is associated with adverse health effects on the human respiratory system. Thus this study aimed to create a burn area map in Mosul Park, Iraq from Sentinel 2A satellite imageries and determine the correlation between fire events and AQI in Mosul Park. Linear regression and Pearson correlation were adopted based on the results of Sentinel-2 image analysis and GIS tools. Moreover, two indices were used to detect the burnt areas. The NDVI was used to map the characteristics of Mosul Park and the NBR index was applied for mapping fire-events and post-fire during the year 2022 and discovering the affected areas.

2. Study Area

2.1 Methodology

The study area is Mosul Park which lies in Mosul city, Iraq (Figure 1). The study area occupies 90 ha and is bonded between latitudes (36° 21′ 40″ to 36° 23′ 20″) N, and longitudes (43° 06′ 06″ to 43° 07′ 40″) E. Mosul city is located in the north of Iraq and is characterized by temperatures of 20 °C-40 °C in the summer and 5 °C -15 °C in the winter [23]. Since 2017, Iraq has recorded multiple forest fire events, including the recent Mosul Park fire in June 2022. Thus it is important to embark on research on fire detection, and mapping fire events for evaluating affected areas, and risk assessment.

Figure 1: The study area map; (a) Iraq, (b) Mosul city north Iraq, and (c) Mosul Park

Table 1: The attributes of Sentinel-2 images

Figure 2: The methods used for fire detection and correlations

2.2 Data and Methods

The used data in this study was based on a satellite image of Mosul Park and AQI data in burned areas. Sentinel-2 images have been downloaded from https://apps.sentinel-hub.com/eo-browser/". The image attributes of the used data are described in Table 1. Furthermore, historical AQI data for Mosul city was downloaded from Air Matters through the website https://air-quality.com/". The methods used for fire detection and correlation of AQI with burned areas in this study are represented in Figure 2. Where two indices have been used to map the study area during fire events and post-fire. NDVI mapping was applied to identify the characteristics of the study area. Moreover, the NBR index was used to map the study area and detect burned areas based on three bands (B12, B11, and B8A) of Sentinel-2 images. NBR was applied at fire events and post-fire. Comparison has been done between burned areas and non-affected areas based on the NBR map. Besides, statistical analyses have been implemented; Pearson Correlation, and Linear regression. The applied method was to investigate the AQI relationship with the burned areas at fire-events.

ArcGIS10.3 is used to perform Mosul Park fire damage and detection of the affected area to analyze the severity of fire events using Sentinel-2 images. The equation that specifies the Normalized Burn Ratio (NBR) can be written as:

Equation 1

where here for Sentinel-2 image NIR: is band 8, and SWIR; is band 12. NBR is applied to present the affected areas in fire-events on a single map [21] and [24]. For NDVI, the equation can be specified as:

Equation 2

where, RED: is band 4, and NIR: is band 8 [21] and [24]. The NDVI is a simple and effective index that can quantify green vegetation. It is based on the reflected light from the vegetation of a certain wavelength. NDVI was designed by Sentinel-2 image in the study area. Accordingly, spatial changes in vegetation and the distribution variation of vegetation due to fire can be distinguished.

The NDVI first advanced in 1974 which can be used for the prediction of vegetation and biomass changes fire-events. It’s based on red and NIR that are reflected by vegetation [21].

Moreover, the Pearson correlation equation between γ and burned areas (B) in hectares for npairs can be specified as;

Equation 3

Pearson correlation coefficient (γ) is a common method for evaluating the linear relationship [24]. It results in a number between (-1 to 1 ) which describes the power and direction of the two variables' correlation. It reports the change of the variable based on the other one in the same direction. To apply the Pearson correlation coefficient we mapped the affected areas based on the Sintenel-2 image of fire-events. In order to evaluate air quality at different locations the affected areas were then divided into three sites; 5 ha, 5 ha, and 6.7 ha which were selected randomly. In each zone, we defined the AQI value with suitable positions in the study area of fired palaces. The value of AQI in each position was determined based on their spatial reliance on surrounding stations around the study area [25]. The equation was applied based on AQI data in each zone with the covered burned area.

Moreover, the collinearity of AQI with the burned areas was analyzed and the linear coefficient correlation was determined.

3. Results and Discussion

3.1 Resultant fire Detection Maps

To detect the fire-affected sites in Mosul Park we used a fire-detecting spectral index, the Normalized Burn Ratio index (NBR) which is organized based on the equation (1). Figure 3 represents the NBR Index map based on Sentinel-2 fire-events. NBR index map shows the affected areas in fire-events on a single map. Based on ArcGIS analysis results, the affected areas by the fire were about 16.7 ha. It can be easily noted by the shaded part that appears on the map. Furthermore, Figure 4 represents NBR Index map based on Sentinel-2 post-fire. By comparison between maps can be visually seen the converted affected area to barren land.

Figure 3: NBR index map based on Sentinel-2 fire-events

Figure 4: NBR index map based on Sentinel-2 post-fire

Figure 5: NDVI map of Mosul Park

3.2 Resultant NDVI Map

NDVI map was derived from Sentinel-2 image acquired on 30-06-2022 in the study area Mosul Park. Consequently, the vegetation spatial changes and the distribution variation caused by fire can be noted. NDVI for a given pixel results in (-1 to +1). NDVI less than 0.1 points to sand, rock, and barren lands. NDVI of (0.1 to 0.2) indicates to the soil, a shallow value. NDVI of (0.2 to 0.5) refers to shrub, grassland, and sparse vegetation spaces. More than 0.5 a high NDVI represents forests [21]. Figure 5 represents the NDVI map of Mosul Park.

3.3 Statistical Analysis Results

Statistical analysis involved linear regression and Pearson correlation. Polynomial linear fitting is the base of the validation [26] and [27]. Table 2 represents the statistical outputs. Moreover, Figure 6 represents a GIS-based correlation processes map. Results of the correlations and regression process have shown significant relationships by (r = 0.92, and R² = 0.84) as shown in Table 2. Indicating a positive correlation between AQI and burned areas. Based on Figure 6, the recorded AQI values ranged between (360-402) which indicates very unhealthy air quality [18]. Correlations of AQI with burned areas are in a positive direction significantly increasing AQI which indicates a negative impact in fire-events.

Table 2: Statistical outputs

Figure 6:GIS-based correlation processes map

4. Conclusion