Navigation mit Access Keys

Hauptinhalt

 
 

Fire, weather and land cover interactions

 

Understanding fire, weather and land cover interactions from long-term terrestrial observations and satellite data in a north to south transect in Europe and North Africa (GRADIENT)

MARIE SKŁODOWSKA-CURIE ACTIONS
Marie Curie Individual Fellowship project - H2020-MSCA-IF-2015

Researcher: Nikos Koutsias

  

Project overview

Long-term historical time series records of fire activity (number of fires and total area burned) extending back to the late 1800s, that are very rare worldwide, were found and used within the GRADIENT project that correspond to (i) Switzerland, central Europe (1900-2014), (ii) Greece, south Europe (1897-2014), (iii) Algeria, north Africa (1870-2014) and (iv) Tunisia (1902-2015), north Africa which together with the spatial-explicit reconstruction of recent fire history from Landsat satellite images (1984-2016), gave a unique and excellent opportunity to understand fire, weather and land use/land cover (LULC) interactions in a north to south transect. The Tunisia study case was added during the implementation period of the project since in the original proposal only the first three study cases were proposed.

Differences in bio-geographical characteristics provided by the four selected study areas, located on a large geographical gradient covering two continents gave the opportunity to document the role of fire in different biomes, to explore cross-scale issues and assess how fire-weather-LULC interactions vary across different scales, especially under a climate change context. The GRADIENT project consisted of three topics, that corresponding mainly to three different scales. The specific objectives were: (i) the identification of trends, patterns and relationships between forest fires, weather, land cover and socio-economic variables from long-term observations, (ii) the reconstruction of recent fire history and the assessment of burning patterns and fire selectivity on an annual basis from satellite images, and (iii) the exploration of post-fire vegetation dynamics and recovery for selected large fire events using time series satellite images.

Some remarks and outcomes for the specific objectives of the project are the fact that there are similarities and non-similarities among the four study area that compose the gradient from north North to Ssouth. Specifically, for each one of the three objectives the main overall conclusions are:

 

  1. in principle there is a characteristic fire activity in all four study areas defined by the general pyro-environment with certain peaks occurringed at specific years, associated to physical and social factors. The role of precipitation is different in the gradient from the wet to dry areas. Moisture is more evident as an underlying explanation mechanism in the wet study area while temperature is more evident in the dry study areas. It was evident that there is a clear and different role of precipitation from promoting to discouraging fire activity across the north to south gradient together with the social aspects and the role of human dimension, 
  2. remote sensing can be used with semi-automatic methods to reconstruct the recent fire history though certain difficulties exist especially in the North study area where cloudiness is a potential problem. The huge number of satellite images neededs the development of automatic or semi-automatic techniques to avoid the time-consuming manual methods and techniques. This is a critical issue in cases where many satellite images are required, such as the spatially explicit reconstruction of recent fire history, where hundreds of images might be used in the processing chain. For all study areas selective burning, which is evident that also depends on the available to burn landscape classes available to burn, was investigated. Frequent fires were also observed tohat burn mainly grasslands and shrublands. 
  3. remote sensing either by using low resolution or medium-high resolution satellite data provides critical information of the phenology of the fire affected areas that can be used efficiently to post-fire studies. Satellite time series data need special attention because of atmospheric and other sources of radiometric problems. The consideration of the full phenological cycle of landscape enhances the interpretation power of the spectral signal. The synergy and co-processing of low spatial resolution high temporal resolution time series data (e.g. MODIS) with high spatial resolution low temporal resolution time series data (e.g. Landsat, Sentinel-2) needs further understanding. Within the GRADIENT project vegetation phenology and time series statistics proved very useful not only to study vegetation recovery in fire affected areas but also to identify the time period where the fire or fires occurred and to define also the vegetation phenology before the fire. This is very useful first for Iintegrating this concept into a burned land mapping approach was very useful and second forallowed to identifying what the type of vegetation is burned.

The results and outcomes of the GRADIENT project can contribute further to the a better understanding of fire, weather and land cover interactions, and therefore provides knowledge for fire and land cover management practices, especially under a climate change context. Understanding of post-fire vegetation dynamics and recovery can be useful for the mitigation of short and long-term consequences of fire occurrence. The knowledge acquired from the past can be used to understand current processes and project them to future. 

 

Overall remarks and conclusions

WP 4

Some remarks and outcomes for the specific objective of the WP6 WP4 are the fact that there are similarities and non-similarities among the four study area that compose the gradient from north to south. In principle there is a characteristic fire activity in all four study areas defined by the general pyro-environment with certain peaks occurred at specific years associated to physical and social factors (Fig. 1). The role of precipitation is different in the gradient from the wet to dry areas. Moisture is more evident as an underlying explanation mechanism in the wet study area while temperature is more evident in the dry study areas. It was evident that there is a clear and different role of precipitation from promoting to discouraging fire activity across the north to south gradient together with the social aspects and the role of human dimension.

From the autocorrelation analysis of Total Burned Area at yearly basis time series data for all four study areas seems that there is no severe autocorrelation to create statistical issues; however, there are interesting structures in the data. Periodical occurrence of fires can be observed especially in Switzerland and Greece.

From the trend analysis based on the Mann-Kendall trend test and the non-parametric Theil-Sen approach, properly modified to account for autocorrelation of the time series data, seems that there are no severe trends although for certain variables in specific study areas there are some trends or periods with different fire activity identified also by the change point estimation methods.

 

Concerning the extremes, two clear patterns were observed according to the two discrete roles of the explanatory variables recognized previously; a first pattern, where the explanatory variable promotes fire activity - as the example of Switzerland and the effect of precipitation or dry periods promoting fires (e.g. drought) - and a second pattern, where the role of the explanatory variable is to discourage fire activity - as the example of Greece or Algeria with the effect of precipitation or dry periods discouraging fires (e.g. wet conditions). We recognize a gradient from north (Switzerland) to south (Algeria) where the role of explanatory variables to fire activity is changing from promotion to discourage.

From the analysis based on the Generalized Linear Models (GLM) - that is a flexible generalization of ordinary linear regression which allows the response variables to have error distribution models other than a normal distribution - we found that the Negative Binomial GLM were the best performing among the four models applied (Poisson, Negative Binomial and Gamma), and it was the only one which was not over-dispersed (Fig. 2). The explanatory variables that presented a statistical significance were precipitation and temperature, with the  above mentioned underlying mechanisms .

 

WP 5

Burned land mapping can be in general achieved by (i) field surveys using on site human-made observations, (ii) the use of b/w, colour or infrared aerial photography, and (iii) remotely sensed data acquired by various satellite systems. The first, field survey is considered a time consuming, expensive but highly accurate method, although it poses serious limitations for burned land mapping since it only provides general statistics due to time and cost limitations (Koutsias et al., 1999). The second, aerial photography, covers larger geographical areas than field surveys and processes the data with less costs. However its usage is minimal in burned land mapping mainly due to cost constraints. The third, satellite remote sensing, appears to be a suitable approach to map and monitor burned areas compared to other more traditional methods, but this method is not error free. Various land cover classes present similar spectral responses to the burned surfaces, thus resulting in spectral confusions and errors such as with water bodies and cloud shadows as observed in NOAA AVHRR data. Also, cloudiness can minimize the potential use of remote sensing to reconstruct the recent fire history as for example in Switzerland within the GRADIENT project.

USGS and ESA archived Landsat scenes processed to Standard Terrain Correction (Level 1T) where possible, are available to the public at no charge from U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center (http://glovis.usgs.gov/) and European Space Agency. These historical archives cover large spatial and temporal extents at continental scale and provide a unique opportunity to overcome cost constraints when reconstructing fire history at local scales. The Landsat satellites series has a long history of data capture that started with the launch of Landsat 1 on 23/07/1972 with Landsat Multispectral Scanner (MSS) onboard (originally known as Earth Resources Technology Satellite [ERTS]). Since then Landsat satellites have been taking repetitive images of the Earth’s surface at continental scale, thus creating a huge historical archive that can be used to reconstruct the past. To understand dynamic processes like wildfires and assess their effects on landscape dynamics by understanding the recovery processes of the affected ecosystems and explaining the observed restoration patterns, we require observations in temporal and spatial extents that allow these processes to be captured. The Landsat archive is a valuable data source that allows reconstruction of short-term fire occurrences by mapping past-fire events in cases where no data or only rough estimates are available. In this line Sentinel-2 data which provide enhanced satellite observation can continue and improve such efforts. The reconstructed burnt area patches are showed in Figure 3.

The interactions of landscape components and fire were analyzed by comparing the relative proportions of what the fires burnt during the period 1984-2015 against what is available to burn across the landscape (e.g. CORINE, or other available global land cover data, e.g. ESA global land cover data), considering a random model that accounts for spatial autocorrelation. To determine whether the wildfires burn significantly different proportions of LULC classes than what is available to burn we applied a Monte Carlo randomization test considering spatial autocorrelation on the basis of randomizing the fire events using however the exact fire shape to account for the spatial autocorrelation. For all study area selective burning is evident that also depends on the available to burn landscape. Frequent fires were also observed that burn mainly grassland and shrublands.

 

WP6

Vegetation phenology is an important element of vegetation characteristics that can be useful in vegetation monitoring especially when satellite remote sensing observations are used. In that sense temporal profiles extracted from spectral signal of time series MODIS and LANDSAT satellite images can be used to characterize vegetation phenology and thus to be helpful for monitoring vegetation recovery in fire-affected areas.

Satellite remote sensing data from MODIS and LANDSAT satellites in the period from 1984 to 2016 were acquired and processed to extract the temporal profiles of the spectral signal for selected areas within the fire-affected areas. This dataset and time period analyzed together with the time that these fires occurred gave the opportunity to create temporal profiles for almost half years before and half years after the fire (Figures 4, 5 and 6). The different scale of the data used gave us the chance to understand how vegetation phenology and therefore the recovery patterns are influenced by the spatial resolution of the satellite data used.

Within the GRADIENT project vegetation phenology and time series statistics proved very useful not only to study vegetation recovery in fire affected areas but also to identify the time period where the fire or fires occurred and define also the vegetation phenology before the fire. This is very useful first for integrating this concept into a burned land mapping approach and second for identifying what type of vegetation is burned.