4.1. Spatial Pattern of dNBR
The RGB 7-4-3 band combination (Figure 1
) depicts the Las Hurdes fire perimeter in shades of red associated with the low reflectance in the NIR band, a characteristic of zones of scarce vegetation, and high reflectance at 2.1 µm in the SWIR spectral region, typical of areas with a low moisture content. This is the typical spectral response of burned areas [79
] (Figure 2
). Different exposure time and different fire intensity result in the great spatial variability of burn severity in the affected ecosystem. The spatial distribution of burn severity, classified from the original dNBR threshold values, can be seen in Figure 2
. Within the Las Hurdes fire, 32.9% of the burned surface presents HS, 37.4% MHS, 18% MLS and 11.7% LS. On the whole, Las Hurdes was a high severity fire, since more than 70% of the area falls within the MHS and HS categories. However, within the fire perimeter, two wide diagonals of low severity pixels divide the burned area in the north and south (Figure 2
), defining four sectors: two in the north with a large number of high-severity nuclei, a very large one in the center and one of predominantly moderate-low severity in the south. The predominance of the highest burn severity intervals is also related to the initial approach applied to the burn severity assessment, by using an immediate post-fire image and not giving time for the ecosystem to show additional responses to fire [12
4.2. Temporal Dynamics of LST and NDVI Values
This section presents the temporal dynamics of LST and NDVI throughout the study period. Descriptive statistics for LST and NDVI (Tables 2
) refer to data from all of the available images: the pre-fire image (13 July 2009) and 14 post-fire images taken between July 2009 (one day after fire), and September 2011 (two years after fire), while Figure 3
shows data in the form of graphics on four different dates: 13 days before the fire on13 July 2009, and on three midsummer dates corresponding to successive post-fire summer seasons (29 July 2009,16 July 2010, and 4 August 2011). Values are grouped by severity categories. In addition, Figure 4
shows the spatial distribution of LST and NDVI on the same dates as Figure 3
In the pre-fire image, all burn severity categories present similar average LST values (∼30 °C) (Table 2
). The coolest areas associated with greater biomass are those registering the highest severity levels after fire (Figures 3
). The existence of this type of relationship between pre-fire biomass and further burn severity was previously reported by García-Martin et al.
], who demonstrated that knowledge of crown biomass enables the prediction of the burn severity levels.
The immediate effects of the fire on the LST are reflected in the first two post-fire images (29 July and 30 August 2009) closest to the event. For the visual assessment of these effects, Figure 5
presents the spatial distribution of the dLST, where LST values of the post-fire image are subtracted from the pre-fire image. According to this formula and the assigned colors, areas with the greatest increase of LST are highlighted in red, and areas without a change are shown in green. To improve the understanding of pre- and post-fire LST changes, Figure 6
presents the confidence levels of average values for the dLST by burn severity category. The average LST increase is 13°C, reaching 20°C for the HS pixels. The generalized LST increase in the post-fire image (both in the burned and unburned areas) may be due to the fact that this image was acquired on a date closer to the middle of summer than the pre-fire image, and therefore, the air temperature was high. However, thermal differences between HS and UB categories within the post-fire image (>10°C) reveal the influence of burn severity on the spatial distribution of LST (Table 3
). The decrease of aboveground green biomass in the burned zones [12
], especially in those of higher severity, and the appearance of lower emissivity coverage (ash, char and mineral soils) lead to a large increase in the LST.
Elevated LST after fire events is mentioned by several authors (among others Lambin et al.
]; Montes-Helu et al.
]; Wendt et al.
]). Veraverbeke et al.
] studied this increase using MODIS imagery following the major Peloponnese fire in 2007. Until now, few studies have analyzed spatiotemporal patterns of post-fire surface temperature using Landsat data, although the high potential of existing single channel algorithms, such as the mono-window (MW) method by Qin et al.
] or the single-channel (SC) method by Jiménez-Muñoz and Sobrino [22
], has already been demonstrated [89
]. The greater surface heterogeneity of the burned areas due to the incorporation of combustion products, changes to lighter-colored soil and ash, char and scorched, then blackened, vegetation [12
] results in an increase in post-fire thermal variability (SD values ∼5, Table 2
Some interesting ideas arise regarding the influence of burn severity on the LST distribution and the contrast between areas of different burn severity categories. The general decrease in LST observed in the first post-fire autumn (September and October data of 2009) is probably associated with lower solar illumination angles. When the sun is directly above the observation location and the sunlight is perpendicular to the land surface, the amount of solar radiation received by the surface is at its maximum. However, as the angle between the sun and a surface is continually changing, the surface gets only part of the incident sunlight. Topography (slope) and sun azimuth also affect the incidence angle of sunrays and the time the area is illuminated by the sun. However, burn severity remains the main factor influencing the spatial distribution of LST: higher LST values correspond to higher burn severity and vice versa
). Likewise, post-fire thermal variability within burn severity categories maintains the level observed in the immediate post-fire image.
The same patterns of LST changes are observed in the images from different years: (i) same season LST values (spring, summer, autumn) become lower from year to year; (ii) the spatial distribution of LST values is qualitatively in agreement with the burn severity categories; and (iii) differences between extreme severity categories in 2010 are slightly lower compared to 2009 (the year of fire) and even lower in 2011 (Table 2
). This smoothing of contrast among categories can be explained by both the effects of time on the combustion products and, most of all, the effects of vegetation regeneration, reflected in NDVI values registered in all of the temporal series in the various burn severity categories (Table 3
) and a visual comparison of three dNDVI images (Figure 7
): each image is calculated as the subtraction of the NDVI raster of one of the post-fire summer seasons from the pre-fire NDVI (dNDVI2009
). The images show areas with higher dNDVI in red and those with lower dNDVI (similar to the pre-fire situation) in green. The progress of vegetation recovery is quite evident: while the highest differences are characteristic to the first post-fire summer, the contrast between burned and unburned areas is smoothed in 2010 and especially in 2011 data, with much lower dNDVI values in the corresponding images.
This successful regeneration process is explained by the efficient recovery mechanisms of the vegetation species dominant in this area. Pinus pinaster
, the main species affected by the Las Hurdes fire, is highly adapted to fire-prone environments through the massive release of seeds from serotinous cones after fire forgermination [92
]. In the same way, shrubland species observed in fire affected areas near the study site (Erica arborea
, Erica lusitanica
, Cistus ladanifer
, Phillies angustifolia
, Cytisus scoparius
, Calluna vulgaris
]also apply efficient post-fire reproduction strategies in recolonizing burned areas.
Our results demonstrate that the LST increase in fire-affected areas was evident in the analyzed series of images, which cover all of the seasons of the two post-fire years, except winter. This is similar to the results reported by the previous research [5
], although the range and the size of the differences between severity categories of the same date is much larger than that detected in the earlier studies. This is probably due to the different response of the analyzed vegetation: much more homogeneous in this study (predominantly conifer forests) than analyzed in the study by Veraverbeke et al.
] (shrublands, olive groves, coniferous and deciduous forests). Immediate post-fire increment in LST calculated from Landsat is much more pronounced than that registered for similar vegetation cover at the same phenological stage registered by MODIS, because of the difference in spatial resolution between the two sensors, i.e.
, the smoothing of contrasts in the lower resolution images.
4.3. Analysis of fsdLST and fsdNDVI
The results of the date-by-date LST-NDVI comparison by severity categories are shown below. It can be seen that the differences between burned/unburned areas increase with burn severity in terms of LST from around 3 °C to almost 7 °C (Figure 8a
) and from 0.09 to 0.21 in terms of the NDVI (Figure 8b
). Mean fsdLST between the successive severity categories is 1.42°C. In fact, significant statistical differences (p
< 0.05) were registered in all of the pairs, with the only exception being HS-MHS and MHS-MLS in the images from August and September of 2011.
A detailed date-by-date analysis of the differences between severity categories allows for the identification of common features. Each pair in Figure 8
shows the distribution of the fsdLST and fsdNDVI by date. The size of the circle reflects the between-category distance for each pair (i.e.
, a size of four corresponds to combinations of the extreme burn severity categories UB-HS). The color in these figures represents the type of categories paired: green when one of the categories in the pair is the UB; red and orange when the HS category is involved and blue for the combinations of the intermediate categories. Pairs combining high burn severity levels (HS and MHS) and the UB class register the most pronounced differences (between 7 °C and 10 °C). A seasonal pattern is observed during 2009 and 2010, as well as a stronger decrease and temporal stabilization in 2011, two years after the fire. Pairs formed by consecutive categories (HS-MHS and MHS-MLS) (size = 1) show lower differences (<3 °C), without any specific temporal pattern.Differences below 1°C are almost exclusively observed in combinations of high severity levels (HS-MHS) on all of the dates, except in the image taken just after the fire, where they are slightly above 1 °C.Between these two groups of high and low differences, differences for the HS-LS (orange), UB-MLS (light green) and MHS-LS (blue) (size = 2–3) pairs show a large range of variation (from 2 °C to 7 °C).
The greatest fsdNDVI between 0.25 and 0.35 (Figure 8b
) are always related to the comparison between UB and all of the other severity categories (green).They are mainly observed in the images of the first post-fire summer, when the effects of fire are more obvious, and especially in March and April 2010.Lower fsdNDVI (0.25–0.15) are characteristic of the pairs formed by the UB and HS (size = 4) in the last images of 2010, MHS and MLS (light blue, size = 1) and also between HS-LS (orange, size = 3) until the spring of 2010.Many pairs register differences between 0.15 and 0.05. The majority of images included in this group are from 2011. Differences below 0.05 correspond to the HS-MHS pair on all of the dates and the HS-MLS pair on the dates after June 2010.
Analysis of the fsdLST and fsdNDVI in 2011 reveals: (1) lower fsdLST compared to 2010; and (2) progressive smoothing of contrast between severity categories. fsdNDVI are below 0.10 (Figure 8b
), and fsdLST values are less than 5°C (Figure 8a
), except in May, when they are slightly higher. A general downward trend is observed in both the fsdLST and fsdNDVI throughout the time series, especially significant in 2011. Thus, the scatterplot in Figure 9
highlights the strong relationship between these two variables (r
= 0.84): most of the bigger circles in the graphic are located in the upper part of the scatterplot, except those corresponding to the 2011 dates (in green), which are located in the lower part of the plot, always below the reference line, due to the minimizing effects of vegetation regeneration on fsdLST. Unusually low fsdLST values observed in March of 2010 are due to particularly low air temperature on this date.
However, the spatial distribution of LST depends not only on burn severity and its interaction with local-scale variables, such as surface emissivity (vegetation regeneration).Factors explaining intra- and inter-annual LST changes also include illumination geometry controlled by solar azimuth and elevation angles and topography reflected in slope and aspect. The role of aspect in the spatial distribution of LST is shown in Figure 10
. The figure presents mean LST values for eightcategories of aspects (at 45-degree intervals) grouped by burn severity levels from the pre-fire image up to the image taken 27 months after the fire in September of 2011.
At first glance, some influence of aspect on the spatial distribution of LST and its relationship with severity levels, cover type and day of the year is observed. High LST values are systematically registered on SE-facing slopes (between 90°and 180°), with slight variations depending on the image date. Conversely, values corresponding to pixels in NNE- and NW-facing slopes (between 270° and 360° and between 0°and 45°, respectively) always register lower LST. Differences between hot and cold orientations deepen in the spring and autumn images, due to the lower elevation angles of the sun. For example, in the image from July2010, the differences between the aspect intervals described above can exceed 6°C, and in October2009, they can be higher than 15°C, because of the deeply shaded areas. However, thermal contrast between pixels of different aspects is not as pronounced in the pre-fire image, where it does not exceed 4°C. Therefore, the fire and the consequent vegetation removal lead to greater thermal heterogeneity, increasing the role of aspect in the spatial distribution of LST. In later images (those from 2011), however, vegetation regeneration reduces the differences in LST between burn severity categories and aspect intervals, as can be appreciated in the 2011 images.