PatrÌ_cia Malico Alexandre1, March 2011
1 Ph.D. student, Wildlife and Forest Ecology Department, College of Agriculture, University of Wisconsin – Madison, USA firstname.lastname@example.org;
Abstract: The objective is to monitor vegetation recovery after the large fires of 2003 in Portugal using a time-series of MODIS Terra Enhanced Vegetation Index (EVI) data. Post-fire vegetation regeneration rate was estimated using Olson’s model. We attempted to model it as a function of fire history, including number of times burned prior to 2003. This study shows that satellite imagery can be very valuable for studying post-fire vegetation response, and this can contribute to a better understanding of wildfires, lead to improved management strategies for prevention, or even improve allocation of firefighting resources.
Key words: Forest fires; EVI; MODIS; Time-series; post-fire regeneration.
The Mediterranean climate is characterized by the coincidence of hot and dry summers, and rainy winters. This particular combination of weather conditions is the reason why this is one of the world’s major fire-prone ecosystems. Here, fire dictates structure and vegetation dynamics (Bond and Keeley, 2005; Naveh, 1975), since plant communities have high resilience to fire and many regenerate by re-sprouting from fire-resistant structures (Arnan et al., 2007). Portugal has a Mediterranean climate and wildfires are a piece of the ecological puzzle. However, these events currently display increased frequency of recurrence. The cycle is altered: It is shorter, with more intense fires, which causes bigger damages to tree bark (some trees re-sprout from subcutaneous buds), and trees don’t have time to reach sexual maturity and produce seeds for regeneration. Besides the ecological effects, there also are social consequences related to loss of property value, burned houses and even direct health effects due to degradation of air quality.
Since 2003, reports from the European Commission’s European Forest Fire Information System refer to Portugal as an example for large forest fires. Portugal is a case study for several reasons, but mainly because of its atypical characteristics when compared to other Southern European countries. It has a large proportion of the territory covered by forest (approximately 68% of the total area), and the highest rates of forest privately owned (approximately 90%). This situation is the opposite of Southeastern and Eastern Europe, where public ownership commonly ranges from 90-100%. Additionally, the property area per owner is very small on average (i.e., very fragmented), and the country lacks an updated cadastre of forest ownership. This causes severe difficulties for an efficient management.
Wildfires affect the production of environmental goods and services through impacts on biodiversity, soils, water resources and air quality. Detailed spatio-temporal monitoring of post-fire vegetation recovery is important to help forecast harmful environmental impacts, such as landslides and floods, and to assist burned area rehabilitation activities. Extended post-fire monitoring of vegetation re-growth is also useful for characterizing fuel hazard dynamics, an important contributor to wildfire risk assessment. In Portugal, the fire season of 2003 was the worst on record, with 430,000 ha (more than 1 million acres) of burned area, about four times the annual average and 1.5 times greater than the previous maximum, recorded in 1985. Between July 30 and Aug. 3, 80 fires consumed more than 220,000 hectares (more than 500,000 acres), coinciding with an extreme heat wave over Europe estimated to have been an event with a 500-year return interval.
Our research goals are to understand post-fire vegetation dynamics in Portugal and evaluate if the recent fire history is influencing vegetation re-growth.
The 2003 fire season in Portugal provides a good case study due its severity and due to existence of a fire perimeter dataset (with information since 1975). That year, 100 wildfires burned more than 500 ha (1,235 acres) each. For this study, we analyze the two largest fires (Figure 1) occurring in mainland Portugal in that year. Fire #93 is located in the central part of the country and the burned area was dominated by pine forest (Figure 2a). This area is managed for wood extraction and sometimes resin. After the fires, these pine trees grow back from the existing seed bank on the soil. Fire #78 is located in the South and the area is characterized by a mix of hardwood, broadleaf, and shrubs (Figure 2b). After a fire, natural regeneration occurs with the seeds that survived the fire.
The approach is based on the use of a vegetation index derived from satellite imagery to characterize the evolution of vegetation greenness and response to fires in each of the study areas, in the analyzed period. Using a product provided by NASA, the MODIS vegetation index, a time series was created spanning the years from 2000 until 2009 for the selected burned areas of the 2003 fire season. The abrupt decrease of EVI index values (EVI, Enhanced Vegetation Index) allows the identification of the moment of fire occurrence and how the greenness values change after that.
We used data from NASA MODIS/Terra satellite, which has a 16-day return interval. We relied on the MODIS Enhanced Vegetation Index (EVI) with 250 m spatial resolution, in the form of a product named ÛÏVegetation Indices 16-Day L3 Global 250mÛ (MOD13Q1), covering the period from 2000 to 2009. The images were processed with MODIS Reprojection Tool (MRT) in order to mosaic and clip the study area. Then, these data were screened with the Time Series Generator (TiSeG) software (Colditz, Conrad, Wehrmann, Schmidt, & Dech, 2008), to detect and replace bad or missing data. Interpolation was performed for each pixel, using cubic splines. The starting date of February 2000 allows the characterization of undisturbed (recently unaffected by fire) vegetation dynamics in the fire-affected areas.
The post-fire length of the time-series should allow for detecting substantial vegetation recovery, up to fuel loads capable of sustaining new fires. In order to extract the EVI values from each pixel, a sample grid of 500 m was created (Figure 3). The EVI values inside the 2003 burned areas will provide information allowing characterizing the pre- and post-fire situation. Next, time series were created for each burned area using an IDL script to extract the EVI values. For each point, we then analyzed its fire history based on a dataset that contains the fire perimeters since 1975, which were mapped using LANDSAT imagery.
CORINE Land Cover code for 1990, 2000 and 2006). We then separated the points according to their fire history, i.e.: Û÷never burned since 1975′, Û÷burned once’, and Û÷burned twice since 1975′). Within each of these groups we averaged the EVI values.For this work, sample points from the two largest fires of that year were chosen, having the fire IDs #78 and #93. Inside those fire perimeters. we selected sample points in areas of forest cover (using sample points that had the same
We used Olson’s model (Olson, 1963) to run a quantile regression using the quantreg package from R, and adjust a curve to the post-fire data. Olson’s model has an asymptote which can be used as an empirical approximation to the pre-fire average of EVI for comparison between the pre- and post-fire periods.
Results and Discussion
If we assume that the asymptote is an approximation of the average EVI value that a specific vegetated place will display once it has reached its equilibrium growth or it has a constant growth rate, then we can compare that value with the pre-fire average EVI values (Figure 4).
The EVI values extracted from the MODIS images provide a time series for each area. These series allow us to observe seasonal variations and the moment in time where the fire occurred. The following charts illustrate the forest’s seasonality and the pre-fire average EVI values as well as the Olson’s curve adjusted with the quantile regression (Figures 5a-c).
A preliminary, essentially visual analysis of the EVI time series of the two sample fires from the summer of 2003 reveals distinct post-fire vegetation responses, namely in the magnitude of the fire-induced drop in EVI values and in the time required for the burned area to recover. Areas that burned mostly pine forest (Fire #93) display larger drops in EVI values, take longer to recover, never seem to achieve pre-fire values in the six-year, post-fire period, and show no apparent differences related to fire history.
Fire #78, occurring in a more southern location, presents decreasing EVI values when the number of fires before 2003 increases. This confirms an expected result, if we consider the time that a forest needs to recover from a fire; if another fire occurs, the regeneration process starts all over again with probably a smaller bank seed in the soil, since there was not enough time for the maturation of trees. Some species may never recover and there also could be a species shift due to the fire, thus explaining the different EVI values.
The analysis of the fire occurring in central Portugal shows very different and almost opposite results. The points that burned twice before 2003 present higher EVI values than the ones that burned only once or that have never burned (since 1975). This result can have many different explanations, such as new planting of trees, thus causing larger EVI values than just natural regeneration. Points that never burned show higher post-fire average EVI values than the pre-fire average, maybe due to the fact that after the fire the soil was quickly covered by pine trees, which is a pioneer species. However, if there was a fire before (once since 1975) we can see a slightly slower recovery, although a statistical analysis is required to verify if these values are significantly different.
It should be noticed that we are comparing two values that were obtained with two different methods, therefore with distinct biological significance. This comparison may lead to inaccurate conclusions since a simple pre-fire average is not the same as an asymptote obtained from a regression model, i.e., the latter is an estimate and not an observed value. These are preliminary results that need further statistical support.
One cannot be 100% sure if there is or is not a significant fire-history effect. More studies and more variables are needed to further formulate and confirm hypotheses. Nevertheless, the results show that these forests take up to 4-5 years to recover and to be ready to burn again. There could be long-term consequences if the natural fire regime is being disrupting by shortening the fire return interval. Unfortunately, those consequences may not be clear yet because of the short time period under analysis. Any effects on the resilience of forest systems may require a longer observation period before they become apparent. Besides, more explanatory variables can/should be considered to eliminate the influence of other factors that also interfere with plant growth. Examples of such variables are temperature, rainfall, soil type, slope and aspect, previous land cover and even distance to the nearest unburned area because it can be an important source of seeds. These results need to be verified with statistical assessment and more explanatory variables could be added to the analysis.
The use of satellite imagery is a useful and effective way to access substantial amounts of information for large areas in a consistent way, or for locations that are inaccessible, remote, or too dispersed in space. Satellite imagery allows the observation of phenomena that might be invisible from the ground or that have a frequency that would make in situ analysis too expensive, as in this case of vegetation re-growth after fires. This resource also allows for temporal comparisons or the creation of time-series, like it was done in the present work.
This study shows that data such as the MODIS vegetation index product are valuable to evaluate post-fire vegetation response. Having information on how fast the soil is covered with vegetation is important for study of soil losses and desertification purposes. Knowing how long it takes for a specific area to recover to pre-fire greenness values provides information about the capacity of that area to adjust to the disturbance. Having the knowledge that the seed bank in the soil is limited and that this has a big impact on its recovery can contribute to better understanding wildfire and vegetation dynamics, and contribute to improve management strategies for fire prevention. It is also economically important to understand these processes and the time frame in which they happen in order to improve allocation of firefighting resources.
With future statistical analysis to support these first results and with the addition of more explanatory variables, it will be possible to produce a good model of post-fire regeneration dynamics.
Arnan X., Rodrigo A., Retana J. and Collins B. 2007. Post-fire regeneration of Mediterranean plant communities at a regional scale is dependent on vegetation type and dryness. Journal of Vegetation Science 18: 111-122.
Bond W.J. and Keeley J.E. 2005. Fire as a global Û÷herbivore’: the ecology and evolution of flammable ecosystems. Trends in Ecology and Evolution 20: 387-394.
Colditz, R. R., Conrad, C., Wehrmann, T., Schmidt, M., & Dech, S. (2008). TiSeG: A flexible software tool for time-series generation of MODIS data utilizing the quality assessment science data set, IEEE Transactions on Geoscience and Remote Sensing, pp 3296-3308.
Naveh Z. 1975. The evolutionary significance of fire in the mediterranean region Vegetatio 21: 199-208.
Olson J.S. 1963. Energy storage and the balance of producers and decomposers in ecological systems. Ecology 44: 322-331.
PatrÌ_cia Alexandre holds a Master’s degree in Forestry and is currently a research assistant at the Department of Forest and Wildlife Ecology of University of Wisconsin-Madison. Her PhD research is funded by the Fulbright exchange program and the U.S. Forest Service and focuses on forest fires and the influence of human disturbances at the Wildland-Urban Interface (WUI).