Application of gvSIG in a Study Related to Forest Fire Monitoring?

This research aims to compare the vegetation renewal of Karst woodland and pine forest in an area damaged by a fire occurred in July 2003, with the natural evolution of the same vegetation typologies in unburnt areas. This comparison has been made calculating the years needed for the NDVI and NDWI indices, obtained from Landsat satellite multispectral images, to reach the same values found in areas not exposed to fire. The NDVI (Normalized Difference Vegetation Index) quantifies the green biomass while the NDWI (Normalized Difference Water Index) is affected by leaf water content and soil humidity. Although the vegetation structure in the burnt areas is still in an evolutionary stage, after 5 years the two indices almost match the unaltered area values. 

Key words: gvSIG, forest fire, Karst, Landsat, NDVI, NDWI 

A. Altobelli, A. Sgambati, F. Bader, G. Fior, B. Magajna, L. Ferrazzo, R. Braut, P. Urrutia, P. Ganis, S. Orlando


1. Introduction

Vegetation is a fundamental element of ecosystems because it represents the base for the food web and for energy fluxes, it regulates the water cycle and the soil conditions and guarantees the ecosystem's conservation.
Generally severe fires cause almost total disappearance of vegetation cover and a hydrophobic soil condition (MacDonald and Huffman, 2004), with consequent erosion risk due to superficial run-off.
Ecosystem monitoring after a forest fire is based on the study of vegetation dynamics. Remote sensing analysis gives an important contribution in finding quantitative differences in green biomass and soil-plant water amount, allowing to examine the ecosystem's capacity to return to the former conditions (i.e. before the fire), namely its resilience.
NDVI (Normalized Difference Vegetation Index) (Rouse et al., 1974; Tucker, 1979) is the most commonly used green biomass index in remote sensing analysis. It is calculated using the reflectance in the red and in the NIR (Near Infrared) bands (Figure 1). In the spectral vegetation signature the red band is located in the maximal absorbing interval due to chlorophyll, whereas the NIR band is located in the high reflectance plateau due to the structure of spongy mesophyll tissue. Therefore NDVI index is correlated with the green biomass density and with the vegetation health status.
Using the NIR (Near Infrared) and the SWIR (Short Wave Infrared) bands, a second index called NDWI (Normalised Difference Water Index) (Gao, 1996) can be calculated, which is sensitive to leaf water content and soil humidity. SWIR reflectance is negatively related to leaf water content (Tucker, 1980).
Multi-temporal and multi-spectral images, with a special resolution of 30 m and 16 days revisitation time, can be easily obtained by the Landsat7 ETM+ (Enhanced Thematic Mapper Plus) satellite and used for this kind of analysis even though some of these images are affected by striping noise (Liu and Morgan, 2006).

Figure 1. Landsat wavelength bands (Red, NIR, SWIR) superimposed to the reflectance curves of healthy vegetation and bare soil. Adapted from Lillesand and Kiefer (1994).?

2. Study area

The study area is located in Slovenia, a few kilometres from the Italian border, and from the structural point of view it is part of the "anticlinal Karst area".
Karst is positioned in a climatic transition zone, which sits between the Mediterranean and the continental Prealpine region. The zone is characterized by rainy winters, relatively dry summers and extremely short spring and autumn seasons. The two most peculiar physic-geographical characteristics of the Karst are: the discontinuous elevation (between 300-400 m amsl) and the predominant presence of carbonate rocks. Karst area has shallow soils, poor in humus and often moderately productive. Morphologically and lithologically the Karst is characterized by scarce presence of superficial water.
In the study area two forest typologies are prevalent: the Karst woodland and the black (or Austrian) pine (Pinus nigra) planted forest. In the Karst region deciduous woodlands are very common: nowadays the prevailing one is the hornbeam and oak woodland (Ostryo-Quercetum pubescentis) (Poldini, 1989); the following species are also always present: Quercus pubescens (downy oak), Ostrya carpinifolia (hop hornbeam), Quercus petraea (sessile oak) and Fraxinus ornus (flowering ash). Black pine forests are also very significant in the Karst; they all have an anthropic origin, since they have been planted under the Austro-Hungarian dominance from the middle of the 19th century in order to reforest the Karst. The pine forests in the test polygons (Figure 2) were partially planted after World War II and partially originated from natural forest expansion.
A forest fire occurred in this region on the 29th July 2003, burning an area of 10.45 km2 with the following topographical characteristics: altitude 161.42 m, slope 12.48% and aspect 174? (southern exposure) (Figure 2).
In the burnt area the Slovenian forest authority began in 2004 a restoration program: some zones have been planted and other seeded (11200 saplings and 92.3 kg of seeds on a total surface of 69.5 hectares). In the research areas the following tree species were planted: Pinus nigra (black pine), Acer platanoides (Norway maple), Tilia sp. (lime), Prunus avium (wild cherry), Acer monspessulanum (Montpellier maple), Acer campestre (field maple).?

3. Materials and methods

Four areas have been selected for this research (Figure 2, Table 1). Two of these are covered with Karst woodland, the other two with planted pine forest. One of the Karst woodlands and one of the pine forests were exposed to the fire. 

Table 1. Topographical characteristics of the 4 test areas. 
?

Landsat images

To follow the evolution of the vegetation in the burnt areas, we have used a series of multi-temporal Landsat images (2003-2009) from Landsat 5 Thematic Mapper (TM) and Landsat 7 Enhanced Thematic Mapper Plus (ETM+) downloaded from the Glovis website (http://glovis.usgs.gov/). The images chosen were the following: 16/07/2003 (TM), 17/08/2003 (TM), 08/06/2004 (ETM+), 29/07/2005 (ETM+), 14/06/2006 (ETM+), 19/07/2007 (ETM+), 19/06/2008 (ETM+), 24/07/2009 (ETM+).?

gvSIG

From the wide range of free and open source GIS software available, gvSIG 1.9 version (produced by Generalitat Valenciana) has been chosen, due to its remote sensing extension. Sextante, a set of 239 free geospatial analysis tools included in gvSIG and distributed under GPL license, has also been repeatedly used.

The geographic data were handled in the following way:

• importation of the fire area polygon and land cover data;
• individuation of 4 polygons: a burnt and analogous non burnt area, both for pine forest and Karst woodland;
• destriping of Landsat 7 ETM+ images with striping noise:?these images have been treated replacing missing values with a mean local value calculated by a moving window of 5x5 cell dimension;
• importation of Aster DEM image (15 m resolution) and calculation, with Sextante, of aspect and slope;
• calculation with Sextante of vegetation indices NDVI, NDWI with bands 3, 4 and 5 of the Landsat images, from 2003 to 2009;
• calculation, with Sextante, of grid statistics (mean, minimum, maximum, variance) of NDVI and NDWI indices in the 4 polygons.?

Vegetation indices

Once corrected the striping noise on the downloaded Landsat images, the data could be analysed to obtain the green biomass index (NDVI) and the leaf-soil water content (NDWI).?

Bands B3, B4 and B5 from the spectral reflectance curve were used to obtain these indices values and their images. The NDVI values were obtained by combining bands 3 and 4 while the NDWI values by combining bands 4 and 5:

NDVI = (B4-B3)/ (B4+B3)

NDWI = (B4-B5)/ (B4+B5) 

Where: B3 = band 3 (Red), B4 = band 4 (NIR), B5 = band 5 (SWIR)?

Statistical analysis

The non parametric U-Mann Whitney test was used to test the differences between indices of NDVI and NDWI in burnt and unburnt areas on a random sample for each year and for each vegetation type. The non parametric test has been selected because the indices were not normally distributed.

Statistical analysis has been carried out by the free software R (http://www.r-project.org/).?

4. Results and discussion

With the gvSIG software the Landsat data has been analysed and 16 images were obtained, 8 for NDVI index values (Figures 3a, 3b) and 8 for NDWI index values (Figures 3c, 3d) (two for 2003, before and after the fire, and one per year from 2004 to 2009).
The comparison of the reflectance values before and after the fire broke out, pointed out these aspects: a) the damage caused by fire on both vegetation typologies leads to a strong signal loss in the near infrared (NIR) and a decrease of NDVI value, due to leaf tissue degradation;?b) leaf tissue dehydration and the development of a hydrophobic soil layer (MacDonald and Huffman, 2004) are the main causes of reflectance increase in short wave infrared (SWIR) and of the consequent decrease of NDWI value.
Signs of recovery for both vegetation typologies start appearing since spring 2005 (two years after the fire).?

Figure 3. a) NDVI 16th July 2003 before the fire, b) NDVI 17th August after the fire 2003, c) NDWI 16th July 2003, d) NDWI 17th August 2003 

NDVI differences between the burnt and the unburnt areas consistently decrease in 2008, 5 years after the fire (Figure 4). So the current photosynthetic activity and CO2 balance could be considered similar to those before the fire, although the vegetation structure, verified during a field trip (May 2009), still differs.

In Figure 4 the NDWI trend suggest that the humidity content of leaves and soil returns to "normal conditions" when the vegetation cover is regenerated, although the structure of the forest is still different. After a forest fire, the damage to the ecosystem is such that its whole structure is changed or destroyed. This is shown by red lines in the graphs, where it is well displayed as a steep drop of the NDVI and NDWI values. It is also important to highlight that just after the fire (August 2003) and one year later (2004) the soil water absorption was very low. This could be due to the lack of vegetation and to the formation of a hydrophobic sheet above the superficial soil layer. This "crust" blocks the water infiltration and furthermore causes soil erosion due to water run-off (MacDonald and Huffman, 2004).

Figure 5. Total rainfall: above, Karst woodland NDVI and NDWI (2003a is before fire and 2003b is after fire.) Below, pine forest NDVI and NDWI (2003a is before fire and 2003b is after fire.)

The NDWI index falls immediately after the fire and it starts rising again two years later (Figure 5). As mentioned before, this suggests that the soil in the burnt areas has difficulties in absorbing rain water (hydrophobic layer); this is particularly evident for the pine forest. It must be noted that meteorological conditions in 2003 were extreme: summer temperatures were especially high, with scarce precipitations.

Table 2 shows the values of U-Mann Whitney test applied to the NDVI and NDWI indices of two random sampled areas for each year (2003-2009), comparing burnt and unburnt areas for both vegetation typologies.

Table 2. U-Mann Whitney indices (U) with two-tailed probability (p) applied to random pixels of the test areas for each year. Non significant values are in bold (2003a before the fire, 2003b after the fire). 

Before the fire (2003a) the NDVI difference between the two woodlands is not significant, confirming the expected similarity of their green biomass. The NDVI values of the burnt and unburnt Karst woodlands are always significantly different after the fire (from 2003b to 2009), although their differences considerably diminish in 2008 and 2009.
For the pine forests the NDVI values result significantly different already before the fire (2003a). This could be due to the slope (29.76%) of the unburnt pine forest that does not allow an abundant growth of the vegetation cover. The differences become not significant in 2008 and 2009, meaning that the quantity of green biomass in the burnt pine forest has reached the value of the unburnt one. The observed trends (Figure 4, Table 2) suggest that in few years after 2009 the NDVI index will probably return to its original values.
NDWI statistical results are very similar to those of NDVI, except for the pine forest in 2009. In this year the NDWI difference is significant, probably because the amount of rainfall was low (Figure 5) and the vegetation cover is still in a developmental stage. 

5. Conclusions

The consequences of the forest fire occurred in the study area in 2003 have been studied through free multi-temporal satellite images . Using gvSIG, an open source software, it has been possible to analyse satellite images and rapidly obtain useful results for ecosystem monitoring.
NDVI and NDWI indices have been calculated to detect the time needed for the vegetation to recover after the fire . According to the statistical analysis the whole study period (six years) is insufficient to reach the original NDVI and NDWI values, however from the values trend it can be assumed that this will happen in the next few years .
In ecological terms, although the vegetation has not achieved its original structure, 5 years after the fire it can already accomplish some of its biological roles such as photosynthetic activity, CO2 and water absorption .?

6. References

Campbell, J.B. (2002), Introduction to remote sensing, Taylor and Francis, London - New York, pp.157-179.

Gao, B.C. (1996), "NDWI - A normalized difference water index for remote sensing of vegetation liquid water from space", Remote Sensing of Environment, 58, pp. 257-266.

Lillesand, T.M, R.W. Kiefer (1994), Remote sensing and image interpretation, John Wiley & Sons, Inc., New York, p.18.

Liu, J. G., G.L.K. Morgan (2006), "FFT selective and adaptive filtering for removal of systematic noise in ETM plus imageodesy images", IEEE Transactions on Geoscience and Remote Sensing, 44 (12), pp. 3716-3724.

MacDonald, L.H., E.L. Huffman (2004), "Post-fire soil water repellency: persistence and soil moisture threshold", Soil Science Society of America Journal, 68, pp. 1729-1734.

Rouse J.W., R.H. Haas, J.A. Schell, D.W. Deering (1974), "Monitoring vegetation systems in the Great Plains with ERTS", Proceedings of the Third Earth Resources Technology Satellite-1 Symposium, pp. 301-317.

Poldini L. (1989), La vegetazione del Carso Isontino e Triestino, Lint, Trieste.

Tucker C.J. (1979), "Red and photographic infra red linear combinations for monitoring vegetation", Remote Sensing of Environment, 8, pp.127-150.

Tucker, C. J. (1980), "Remote sensing of leaf water content in the near infra red", Remote Sensing of Environment, 10, pp. 23-32. 

Acknowledgements 

Prof. Franco Stravisi, Dipartimento di Scienze della Terra - Stazione Meteorologica di Trieste, Università degli Studi di Trieste for meteorological data.

?