Introduction

Occurrence of fire is identified as a prominent disturbance regime that influences biodiversity and global climate change (Rudel et al. 2005). Fires are a threat to biodiversity due to the human fire dynamic variability (Fuller et al. 2004). Ecological impact of fires on forest ecosystems, especially across boreal, temperate and tropical biomes, received global attention (Nasi et al. 2002). The increase in temperature and decrease in rainfall in association with land-use change contribute to the spread of human induced fires in Asia (IPCC 2007). The tropical warmer regions are more vulnerable to fires (McKenzie et al. 2004; Chuvieco et al. 2003).

The increase in edge forests caused by fragmentation will lead to greater fire risk, as agricultural residue burning may spread to these forests. The fire incidences and the distance of the forest edge to the interior are found to be inversely proportional (Nepstad et al. 1999; Cochrane et al. 1999; Cochrane and Laurance 2002). The fire behaviour is also influenced by elevation, slope, and aspect (Harmon 1982; Bennett 2010). Another important anthropogenic causative factor for fires is the distance to roads (Nepstad et al. 2001; Arima et al. 2005). The regeneration and recruitment of many species are affected by fires, and it facilitates the encroachment by invasive species which contributes to forest degradation (Reddy et al. 2014).

The rapid spread of fire is mainly caused by the availability of burnable material, inclination of the terrain, high wind velocity, and soaring summer temperature (Rothermel and Richard 1991; Roy et al. 2006). The existence of highly combustible coniferous vegetation and limited moisture content stimulates isolated trees at higher altitudes and scrubs at lower altitudes to easily catch fire (Gupta et al. 2018). The coarse-resolution sensors such as the Moderate Resolution Imaging Spectroradiometer (MODIS) with thermal infrared bands (3.6–12 µm range) are capable of detecting vegetation fires on a real-time basis (Reddy et al. 2017).

Study Area

Myanmar, an Indo-Burma biodiversity hotspot, situated between 9°32′ N and 28°31′ N latitude and 92°10′ E and 101°11′ E longitude. This country consists of seven states: Chin, Kachin, Kayah, Kayin, Mon, Rakhine, and Shan, mainly covering hill regions and seven divisions − Ayeyarwady, Bago, Magway, Mandalay, Sagaing, Tanitharyi, and Yangon, covering the plains. Myanmar is reported to have a large number of fires in Asia (Vadrevu and Justice 2011). The study reported more than 6 million ha of forest being annually affected by fires in Myanmar (FAO 2006). The work had quantified the factors impacting vegetation fires in protected and non-protected areas of Myanmar (Biswas 2015). Even though Myanmar is reported to be having the highest number of forest fires, no studies were done till date to quantify the forest burnt area and identification of hotspots using spatial statistics. The main objective of this work is to generate a first national database on forest burnt area and hotspots of fires in Myanmar.

Materials and Methods

Data Used

  1. 1.

    Landsat 8 OLI data of 2017 (https://earthexplorer.usgs.gov/).

  2. 2.

    MODIS active fire locations from 2003 to 2017 (https://earthdata.nasa.gov/earth-observation-data/near-real-time/firms/active-fire-data).

  3. 3.

    VIIRS active fire locations from 2016 to 2017 (https://earthdata.nasa.gov/earth-observation-data/near-real-time/firms/active-fire-data).

  4. 4.

    Vegetation type map of 2016 prepared as part of the National Carbon Project (Reddy et al. 2019).

Methodology

Forest Burnt Area Mapping

To estimate the burnt area the 64 cloud-free images from Landsat 8 OLI for the dry season of February to May 2017 were used. In this study vegetation burnt area has been extracted using unsupervised ISODATA classification, normalized burn area ratio, and visual interpretation. In the first step, all the primarily non-vegetated areas were masked out in satellite imagery using spatial data of vegetation cover. Burnt areas have lower spectral values than no burnt areas because of the decreased level of photosynthetic activities. A 3 × 3 matrix was used for performing post-classification smoothening. Raster data were converted to vector format to refine the boundaries of forest burnt areas using visual interpretation techniques. The normalized burn ratio (NBR) is an index designed to highlight burnt areas (Miller and Thode 2007). The formula of NBR combines the use of both near-infrared (NIR) and shortwave infrared (SWIR) wavelengths. A high NBR value indicates healthy vegetation, while a low value indicates bare ground and recently burnt areas. Non-burnt areas are normally attributed to values close to zero.

Emerging Hotspot Analysis

An emerging hotspot analysis is conducted to investigate trends of forest fires over space in addition to trends over time. A space–time cube was created using each of the data layers representing a category of fire counts from MODIS active fire product as the input feature. The emerging hotspot analysis tool takes the space–time cube as input and conducts a hotspot analysis using the Getis–Ord Gi* statistic for each bin. The neighbourhood distance and neighbourhood time-step parameters define how many surrounding bins, in both space and time, will be considered when calculating the statistic for a specific bin. Based on the input parameters, the space–time cube took the attributes values and aggregates all points with a distance interval of 1000 m and a time-step interval of 1 year to create a bin. The sum of all the aggregated data attribute values included in the bin is the attribute value for that bin. This tool uses the Gi* statistic as a measure of the degree of association between the attribute value for each bin within the space–time cube to the spatial weight or relationship with the attribute values of its neighbouring bins (Getis and Ord 1992). With statistically significant high (or low) attribute values being surrounded by other bins with high (or low) attribute values, it then compares the attribute value for a bin and its neighbours to the mean attribute value of all bins. The Gi* statistic returned for each bin in the dataset is a z score or standard deviation. After the Gi* statistic has been computed, hotspot/cold spot trends were analysed using the Mann–Kendall trend test to detect temporal trends at each spatial location with data against time (Mann 1945). Using the calculated hotspot z score and p value for each bin and the trend z score and p value, the emerging hotspot analysis tool classifies each location into one of 17 categories (Zhu and Newsam 2016) (Table 1).

Table 1 Classification scheme used in this study for portraying different categories of statistically significant hotspots
Full size table

Results and Discussion

Extent of Forest Burnt Area

About 15.2% of vegetation area was affected by fires in 2017. Among the four vegetation types, forests had experienced a high extent of burnt area (Fig. 1). The spatial analysis indicates the highest percentage of burnt area in grassland, followed by scrub and plantations (Table 2). Analysis of burnt area at state level indicates 58.4% of forest in Kayah affected by fires in 2017 (Table 3). Analysis of burnt area across the forest types shows deciduous forests were affected by more fires (Table 4). Error matrix confirms that majority of the classified fire pixels are correct. Overall classification accuracy was achieved at 92.02%, and kappa statistic was above 0.87.

Fig. 1
figure 1

Forest burnt area map of Myanmar for 2017

Full size image
Table 2 Burnt area distribution across the vegetation types of Myanmar (area in km2)
Full size table
Table 3 Areal extent of forest burnt area in Myanmar—2017 (area in km2)
Full size table
Table 4 Burnt area distribution across the forest types of Myanmar—2017 (area in km2)
Full size table

Trend in Fire Occurrences from 2003 to 2017 Based on MODIS

In all the years from 2003 to 2017 a significant number of fires were found in all vegetation classes (Table 5). The number of fire occurrences in forests is found to be the highest than the other vegetation types. The year 2007 is showing the highest number of forest fire occurrences followed by 2010, 2004, and 2012. Overall 15 years of analysis shows declining trend of fires in Myanmar indicates increased conservation measures.

Table 5 Number of vegetation fire occurrences from 2003 to 2017
Full size table

Of the total vegetation fires in 2003–2017, forests accounted for the highest percentage (44.7%) of fire occurrences followed by scrub (19%), plantation (12.8%), agriculture (12.2%), and grassland (11.4%). About 39.7% of fire occurrences were recorded in forests in 2017, followed by agriculture (18%), scrub (17.8%), grassland (13.1%), and plantation (11.4%). The number of fire occurrences is found to be higher in the months of February, March, and April, collectively contributing to more than 90% of fires, while 2.2% of forest fire events were recorded alone in the month of May based on earth observations from 2003 to 2017 (Table 6).

Table 6 Total cumulative monthly fire occurrences from 2003 to 2007
Full size table

Fire Occurrences Recorded by MODIS and VIIRS (2016–2017)

The comparison of fire incidences recorded by MODIS and VIIRS shows that VIIRS had recorded a highest number of fire incidences. The year 2016 had a greater number of fires in all vegetation classes. Two-year data on fires by VIIRS also show decreasing pattern of vegetation fires in Myanmar (Table 7).

Table 7 Fire occurrences in 2016 and 2017
Full size table

On analysing the monthly fire occurrences again, it is found that the number of fire occurrences are more in February, March, and April (Table 8). The monthly fire incidences recorded for 2016 and 2017 had shown a significant change in the number of fires by VIIRS and MODIS. This change may be attributed to the difference in the spatial resolution of their thermal bands. The thermal band of MODIS has 1000-m resolution per pixel, whereas VIIRS has a 375-m resolution per pixel. This higher resolution enables VIIRS to detect fires that MODIS overlooks. However, VIIRS is a more sensitive instrument when it comes to detecting fires. MODIS provides crisper background images because MODIS has 250-m spatial resolution for other than thermal bands and can produce more detailed land surface images.

Table 8 Comparative evaluation of monthly forest fire occurrences for VIIRS and MODIS (2016–2017)
Full size table

Emerging Hotspot Analysis

The emerging hotspot analysis had shown the highest spatial extent of persistent hotspots followed by oscillating hotspots (Table 9). A total of 161 intensifying hotspots and 450 new hotspots were found. About 1545 hotspots out of a total of 17,639 hotspots were found to be diminishing, and 1424 hotspots were found to be historical hotspots. Forest fire hotspots are concentrated mostly in the states of Kayah, Shan, Bago, Nayi Pyi Taw, Magway, Mandalay, Chin, and Kayin (Fig. 2). New hotspots are found only in Tanitharyi. The coastal states of Ayeyarwady, Yangon and snow-covered states of Sagaing and Kachin were least affected by fire (Table 10). The highest area under diminishing hotspots is recorded in Magway.

Table 9 Distribution of forest fire hotspots in Myanmar
Full size table
Fig. 2
figure 2

Forest fire emerging hotspot map in Myanmar

Full size image
Table 10 State-wise distribution of hotspot categories (area in %)
Full size table

Conclusions

This work had assessed the annual forest burnt area in Myanmar using Landsat 8 OLI satellite data for 2017. The deciduous forests were found to be more vulnerable to fires. Emerging hotspot analysis had done using MODIS active fire locations to study the spatiotemporal trends of forest fires. Analysis of 15 years of fire occurrences from 2003 to 2017 indicates a declining trend of fires in Myanmar. In a biodiversity-rich country like Myanmar more scientific studies are required to devise fire management plans to minimize the ecological impact of fire events on forest ecosystems.