• Nenhum resultado encontrado

Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires in the summer of 2016 in Portugal

N/A
N/A
Protected

Academic year: 2021

Share "Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires in the summer of 2016 in Portugal"

Copied!
64
0
0

Texto

(1)

Normalized Difference

Vegetation Index

(NDVI) for evaluation

of forest fires in the

summer of 2016 in

Portugal

Ana Luísa Castro e Antas Carvalho do Amaral

Mestrado em Engenharia Geográfica

Departamento de Geociências, Ambiente e Ordenamento do Território 2017

Orientador

Ana Cláudia Moreira Teodoro, Professora Auxiliar com Agregação Faculdade de Ciências da Universidade do Porto

(2)

O Presidente do Júri,

(3)

Resumo

Portugal ´e um dos pa´ıses mais afetado por fogos florestais na Europa. Todos os anos, no Ver˜ao, centenas de hectares ardem, destruindo bens e florestas a uma velocidade alarmante. Este trabalho teve como objetivo analisar duas ´areas florestais afetadas por fogos em Portugal no Ver˜ao de 2016 usando dados de diferentes sat´elites com resolu¸c˜oes espaciais distintas (Sentinel-2A e Landsat 8 OLI). Foram escolhidos dados da Primavera de 2016 e 2017 (pr´e-fogo e p´ os-fogo) com o intuito de maximizar os valores do ´Indice de Vegeta¸c˜ao por Diferen¸ca Normalizada (NDVI). Foi utilizado o plugin do Software QGIS – Semi-Automatic Classification Plugin- que permite calcular os valores do NDVI para o sat´elite Landsat 8 OLI e para o sat´elite Sentinel-2A. Os resultados mostraram que o NDVI decresceu consideravelmente em Arouca e Vila Nova de Cerveira depois dos fogos terem ocorrido, havendo assim uma diminui¸c˜ao do n´ıvel de vegeta¸c˜ao. O munic´ıpio de Sintra foi escolhido como ´area de controlo j´a que nesta zona n˜ao ocorreram incˆendios. Como era espect´avel n˜ao houve diminui¸c˜ao do valor de NDVI. Os resultados obtidos com o Landsat 8 OLI e o Sentinel-2A foram semelhantes, apesar disso, com o sat´elite Sentinel-2A obtiveram-se resultados mais precisos relativamente `a ´area ardida visto que este tem uma melhor resolu¸c˜ao espacial que o Landsat 8 OLI. Este estudo pode ajudar na compreens˜ao das causas bem como as consequˆencias dos fogos. Outros ´ındices de vegeta¸c˜ao poderiam ser calculados com o intuito de analisar as ´areas ardidas. Em rela¸c˜ao ao sat´elite Sentinel-2A, a resolu¸c˜ao espacial (10 m) e a resolu¸c˜ao temporal (10 dias) foram recentemente (Mar¸co 2017) otimizados com o lan¸camento do sat´elite Sentinel-2B, pois, a combina¸c˜ao destes dois sat´elites g´emeos faz com que a frequˆencia de recolha de dados para uma determinada zona seja de 5 dias ou at´e menos. Apesar disso, para estudos hist´oricos, o programa Landsat continua a ser uma boa op¸c˜ao.

Palavras-chave: Fogos florestais, Landsat 8 OLI, Sentinel-2A MSI, Semi-Automatic Clas-sification Plugin, QGIS, NDVI.

(4)
(5)

Abstract

Portugal is one of the most affected countries in Europe by forest fires. Every year, in the summer, hundreds of hectares burn, destroying goods and forests at an alarming rate. The objective of this work was to analyze the forest areas burned in Portugal in 2016 (summer) using different satellite data with different spatial resolution (Sentinel-2A MSI and Landsat 8 OLI) in two affected areas. Data from spring from 2016 and 2017 were chosen (pre-fire event and post-fire event) in order to maximize the Normalized Difference Vegetation Index (NDVI) values. The QGIS software’s plugin - Semi-Automatic Classification Plugin- which allowed to obtain NDVI values for the Landsat 8 OLI and Sentinel- 2A was used. The results showed that the NDVI decreased considerably in Arouca and Vila Nova de Cerveira after de fire event, meaning a marked drop in vegetation level. In Sintra municipality this change was not verified because non forest fire was registered in this area during the study period. The results from the Sentinel-2A and Landsat 8 OLI data analysis are in agreement in relation to the burned area, however the Sentinel-2A satellite give results more accurate than Landsat 8 OLI since it has a better spatial resolution. This study could help the experts to understand both the causes and consequences of spatial variability of post-fire effects. Other vegetation spectral indices related with fire and burnt areas could also be calculated in order to discriminate burnt areas. Added to the best spatial resolution of Sentinel-2A (10 m), the temporal resolution of Sentinel-2A (10 days) was increased with the launch of the twin Sentinel–2B (Mar¸co de 2017) and therefore the frequency of the combined constellation revisit will be 5 days or even less. However, for historical studies, the Landsat program remains a good option.

Keywords: Forest fires, Landsat 8 OLI, Sentinel-2A MSI, Semi-Automatic Classification Plugin, QGIS, NDVI.

(6)
(7)

Acknowledgements

First, to Prof. Ana Cl´audia Teodoro,the supervisor of this dissertation, for the availability and guidance provided, the sharing of knowledge and suggestions regarding the problems that arose during the accomplishment of this work.

To my family and friends, especially to my mother, aunt and grandmother for the uncondi-tional love and support demonstrated over the years.

(8)
(9)

Contents

1 Introduction 1

1.1 Forest fires in Portugal . . . 1

1.2 Causes of forest fires . . . 2

1.2.1 Conditions for forest fire propagation . . . 3

1.3 Remote sensing . . . 4

1.3.1 Optical sensors . . . 7

1.3.2 Normalized Difference Vegetation Index . . . 8

1.3.3 Landsat 8 . . . 10

1.3.4 Sentinel-2A . . . 11

1.4 Objectives . . . 13

2 Methodology 15 2.1 Study area . . . 15

2.1.1 Vila Nova de Cerveira . . . 16

2.1.2 Arouca . . . 18

2.2 Selected data . . . 19

2.3 Image processing . . . 21

2.3.1 Conversion to Top of Atmosphere (ToA) Reflectance . . . 21

2.3.2 Surface Reflectance . . . 23

2.3.3 Normalized Difference Vegetation Index (NDVI) . . . 25

3 Results 27 3.1 Statistical analysis . . . 27

3.2 Spatial analysis . . . 28 vii

(10)

3.2.1 Normalized Difference Vegetation Index (NDVI) Maps . . . 28

3.2.2 NDVI differences . . . 41

3.2.3 Factors related to fire propensity . . . 44

3.3 Conclusions and Perspectives . . . 45

3.3.1 Main conclusions . . . 45

(11)

List of Figures

1.1 Pedrog˜ao Grande fire consequences. . . 2

1.2 Causes of forest fires in mainland of Portugal between 2000 and 2010 (%.) . . . 3

1.3 Electromagnetic radiation. . . 5

1.4 Electromagnetic spectrum. . . 6

1.5 (a) Specular reflection. (b) Diffuse reflection. . . 7

1.6 Passive Sensor. . . 8

1.7 Healthy leaf interacting with radiation. . . 9

1.8 Landsat 8. . . 10

1.9 Sentinel-2 Processing Levels. . . 12

2.1 Flowchart of the methodology followed in this work. . . 15

2.2 (a)Portugal Mainland district. (b)Municipalities chosen for this study. . . 16

2.3 District of Vila Nova de Cerveira. . . 16

2.4 Average precipitation, temperature and wind velocity values for the study period in Vila Nova de Cerveira district. . . 17

2.5 District of Arouca. . . 18

2.6 Average precipitation, temperature and wind velocity values for the study period in Arouca district. . . 19

2.7 Areas of the different municipalities selected for this study: (a) Arouca; (b) V.N Cerveira; (c) Sintra. . . 20

3.1 2016 NDVI map for all Arouca municipality derived from Landsat 8 image. . . . 29

3.2 2017 NDVI map for all Arouca municipality derived from Landsat 8 image. . . . 29

3.3 2016 NDVI map for all Arouca municipality derived from Sentinel-2A image. . . 30

3.4 2017 NDVI map for all Arouca municipality derived from Sentinel-2A image. . . 30 ix

(12)

3.5 2016 NDVI map for the selected area of Arouca municipality derived from Land-sat 8 image. . . 31 3.6 2017 NDVI map for the selected area of Arouca municipality derived from

Land-sat 8 image. . . 32 3.7 2016 NDVI map for the selected area of Arouca municipality derived from

Sentinel-2A image. . . 32 3.8 2017 NDVI map for the selected area of Arouca municipality derived from

Sentinel-2A image. . . 33 3.9 2016 NDVI map for all Vila Nova de Cerveira municipality derived from Landsat

8 image. . . 34 3.10 2017 NDVI map for all Vila Nova de Cerveira municipality derived from Landsat

8 image. . . 34 3.11 2016 NDVI map for all Vila Nova de Cerveira municipality derived from Landsat

8 image. . . 35 3.12 2017 NDVI map for all Vila Nova de Cerveira municipality derived from Landsat

8 image. . . 35 3.13 2016 NDVI map for all of Vila Nova de Cerveira municipality derived from

Landsat 8 image. . . 36 3.14 2017 NDVI map for all of Vila Nova de Cerveira municipality derived from

Landsat 8 image. . . 37 3.15 2016 NDVI map for the seleted area of Vila Nova de Cerveira municipality derived

from Sentinel-2A image. . . 37 3.16 2017 NDVI map for the seleted area of Vila Nova de Cerveira municipality derived

from Sentinel-2A image. . . 38 3.17 2016 NDVI map for all Sintra municipality derived from Landsat 8 image. . . . 39 3.18 2017 NDVI map for all Sintra municipality derived from Landsat 8 image. . . . 39 3.19 2016 NDVI map for all Sintra municipality derived from Sentinel-2A image. . . . 40 3.20 2017 NDVI map for all Sintra municipality derived from Sentinel-2A image. . . . 40 3.21 dNDVI map for the selected area of Arouca derived from Landsat 8 image. . . . 41

(13)

FCUP xi Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

3.22 dNDVI map for the selected area of Arouca derived from Sentinel-2A image. . . 42 3.23 dNDVI map for the selected area of V.N. Cerveira derived from Landsat 8 image. 42 3.24 dNDVI map for the selected area of V.N. Cerveira derived from Sentinel-2A image. 43 3.25 (a) Slope map for the selected area of Arouca and the NDVI values for the

(14)
(15)

List of Tables

1.1 Essencial bands for remote sensing from the electomagnetic spectrum. . . 6

1.2 Landsat 8 bands settings. . . 11

1.3 Sentinel-2 bands settings. . . 12

2.1 Characteristics of the Landsat 8 satellite data used. . . 20

2.2 Characteristics of the Sentinel-2A satellite data used. . . 20

3.1 NDVI statistical analysis from Arouca selected area. . . 27

3.2 NDVI statistical analysis fromV.N. Cerveira selected area. . . 27

3.3 NDVI statistical analysis from Sintra municipality. . . 28

3.4 Areas calculated from the map of the NDVI differences between 2017 and 2016 (dNDVI) for the selected area of Arouca. . . 43

3.5 Areas calculated from the map of the NDVI differences between 2017 and 2016 (dNDVI) for the selected area of Vila Nova de Cerveira. . . 43

(16)
(17)

1. Introduction

“From a little spark may burst a flame.” Dante Alighieri (1265-1321)

Forest fires have always been a part of the history of the Mediterranean region, in such a way that it molded the native vegetation and favored species adapted to it. Still, in recent years this region has been the target of numerous fires and is here that almost 95 % of the total area burnt in EU occurs1. Compared to other regions of the globe this is where the highest number

of fires and the largest burned area happens [1].

1.1 Forest fires in Portugal

Portugal, being a Mediterranean country, is extremely affected by forest fires. According to Di´ario de Noticias newspaper between 2000 and 2013, there were almost 19 thousand forest fires in the countries of the Mediterranean Basin (Portugal, Spain, France, Italy and Greece) and more than 10,000 happened in mainland of Portugal. The high frequency at which forest fires occur in recent years make stronger the need for a better understanding of forest fire risks, given the real value of sustaining forest resources. Every year, in the summer, hundreds of hectares burn, destroying goods, forests and lives. In the summer of 2016, several districts of Portugal were devastated by forest fires. At least seven major fires were burning out of control in the north of Portugal on August 9, 2016. Also, in Portugal, in Madeira Island, a forest fire started on August 2016 in several places causing damages and casualties never seen before. The local media had reported the death of three people, over 200 people injured, over 950 habitants evacuated, and 50 houses damaged. Very recently, on 18 June 2017, in Pedrog˜ao Grande municipality (Portugal Mainland) a massive forest fire has killed 64 people

1http://copernicus.eu/sites/default/files/documents/Copernicus Briefs/Copernicus Brief Issue54 Fires September2015.pdf

(18)

and over 130 people injured (Figure 1.1). The Portuguese government has declared three days of national mourning following what Prime Minister has called ”the greatest tragedy of human lives” witnessed in the country in years. The high number of forest fires in Portugal is mainly due to its climate, characterized by high precipitation in the winter, which allows the growth of biomass fuel and by a very long and dry summer. The fact that the hot season coincides with the driest period of the year makes it easier to the occurrence of a fire given the state of dryness of the vegetation. Contrary of what was supposed to be expected, or what is normal in other parts of the globe, most fires are not instigated by natural origins but by arson.

Figure 1.1: Pedrog˜ao Grande fire consequences Source:Associated Press Photo

1.2 Causes of forest fires

A fire can start for many reasons and it is very important to know its origin whether for legal, statistical or preventive purposes 2 (Figure 1.2). For a fire to start and spread, there

must be fuel, appropriate atmospheric conditions and a source of ignition. Nowadays, human action is a determining factor in the source of forest fires. It’s necessary to reach temperatures

(19)

FCUP 3 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

of about 180◦C for ignition to happen so it is very difficult to reach this naturally. Therefore,

much of forest fires are caused by humans, through negligence or voluntarily.

Figure 1.2: Causes of forest fires in mainland of Portugal between 2000 and 2010 (%).

Source: ICNB,2012

1.2.1 Conditions for forest fire propagation

In the forest, there are three mainly factors that determine the fire behavior, they are: (i) quantity and type of fuel present; (ii) meteorological conditions at the time of the fire occurrence and (iii) topography of the landscape.

• Fuel material

The combustible materials in the forest, are essentially, wood, leaves, non-volatile mate-rials that burn above 65◦C and volatile organic substances, which vegetation releases into

the atmosphere. Thus, is created a continuous fuel medium that extends in altitude wich are capable of being ingnited above room temperature with a minor source of heating [2]. The flammability of a specie is determined by its structure, amount of dead biomass, moisture content and presence of volatile substances [3].

(20)

• Weather and climate conditions

Climatic and meteorological parameters play an important role in the proliferation and behaviour of forest fires. High temperatures and low humidity values causes the drying of fuels, facilitating their ignition. Summers with high temperatures, low to no precipita-tion, strong evaporation and highly flammable vegetation are the ideal conditions for the proliferation of a fire [4].

On the other side, wind introduces oxygen into the reaction zone, increasing the heat transfer, the speed of progression, the intensity and the possibility of occurrence of other fires by projection of inflamed material. The randomness of the wind behavior makes difficult the prediction of the fire attitude [4].

• Topography

There are several topographic characteristics that define a terrain some of which are fundamental to the behaviour of a fire. The altitude and the sun exposure of a hill influences the climate in general and, therefore, the availability of fuel. Precipitation tends to increase and temperature decreases with altitude. The slope is also one of the essential characteristics for the propagation of fires, the bigger it is the inclination the faster the fire spreads. The fire progression is larger and faster in ascending than descending directions, the upward flame will preheat the above fuels increasing the fire progression. High slopes also originate intense upward winds [5].

1.3 Remote sensing

Remote sensing is the science that allows us to analyze elements on Earth from electromag-netic sensors placed in aerial or space devices without being in direct contact with the object under study. This technique is in constant development and there is a great diversity of ar-eas where it can be applied. Examples of these are: agriculture, meteorological monitoring, management of natural resources and forecasting of natural disasters.

(21)

FCUP 5 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

the object of study. This energy comes in the form of an electromagnetic wave. All electromag-netic radiation has fundamental properties and behaves in a predictable way according to the basic principles of wave theory. Electromagnetic radiation consists in the propagation of two fields: electric field (E) and magnetic field (M). Those are orthogonal to each other, oriented at right angles and they move at the speed of light (Figure 1.3).

Figure 1.3: Electromagnetic radiation Source: CCRS, 2015a

An electromagnetic wave is defined by three elementary parameters: wavelength, frequency and propagation speed. The wavelength corresponds to the distance between two successive wave crests. The frequency shows the number of complete oscillations that a wave performs in each unit of time. The speed of propagation depends on the medium in which the radiation propagates. Theoretically, the velocity of an electromagnetic wave in the vacuum is 3.0 x 108

m/s but this velocity is always lower in any other medium [6]. These physical quantities are related through 1.1:

c = λf, (1.1)

Where,

c = speed of light in vacuum≈3,0 x 108 m/s;

λ = wavelength (m); f = frequency (s−1 or Hz).

(22)

This sequence is called the ”electromagnetic spectrum” and each strip of this spectrum have a different designation, according to its properties. It ranges from very high frequencies, cosmic radiations, to low-frequency, radio waves. The wavelength is inversely proportional to frequency [7] (Figure 1.4).

Figure 1.4: Electromagnetic spectrum. Source:Fundamentals of UV-Visible Spectroscopy

In this spectrum, there are bands essential to remote sensing [8] (Table 1.1).

Table 1.1: Essencial bands for remote sensing from the electomagnetic spectrum.

When radiation interacts with an object on the Earth’s surface, this energy can be absorbed, reflected and/or transmitted in different proportions. In remote sensing, the importance is in the radiation reflected by the targets. This happens when the radiation that hits an object

(23)

FCUP 7 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

bounces off to the atmosphere. There are two kinds of reflection: specular and diffuse (Figure 1.5 (a) and 1.5 (b)). The specular reflection happens when the radiation hits a smooth surface and is reflected in a single direction. When the surface is irregular, the energy reflects in different directions and occurs the diffuse radiation.

Figure 1.5: (a) Specular reflection. (b) Diffuse reflection.

In some cases, a mixture of specular and diffusive reflectance can be obtained. It is usual to classify a surface as being diffusive when 25% or more of the radiation is diffusely reflected. In nature, most objects have diffuse behaviour [9].

1.3.1 Optical sensors

To obtain and record information about an object on the Earth’s surface, a sensor must be installed on a consistent platform with a considerable distance from the target. Although there are several terrestrial and space platforms that can be used for remote sensing data collection, nowadays, is more frequent to use satellite imagery. The optical sensors installed in satellites capture the electromagnetic energy reflected by the target in the visible and infrared region. These types of sensors are known as passive because they capture the reflected energy of an object that had been illuminated by external radiation, usually the Sun.

(24)

Figure 1.6: Passive Sensor. Source: CCRS,2015

1.3.2 Normalized Difference Vegetation Index

One of the main objectives of remote sensing is to identify the composition of different materials on the Earth’s surface, as vegetation, land use patterns, rocks among others. (Crosta, 1992)

The radiation emitted by the Earth’s surface is collected by satellite sensors and is often com-bined to produce vegetation indices (IV). Wavelengths in the Red Band (0.6-0.7 µm) and Near Infrared (0.75-1.75 µm) are recognized as being the most useful in remote sensing vegetation studies [10].

The leaf of a plant is the organ specialized in the capture of light and gas exchanges with the atmosphere to carry out photosynthesis, gutting, transpiration and breathing. Those are composed of a chemical called chlorophyll which absorbs red and blue wavelengths but reflect the green ones and it is this reflection that gives plants their green colour.

The composition of chlorophyll changes throughout the seasons of the year so the percentage of absorption of the red wavelengths also vary. The internal structure of a healthy leaf acts as an excellent diffuse reflector of the infrared wavelengths 3.

Currently there are different vegetation indices possible to analyze, but the Normalized Dif-ference Vegetation Index (NDVI) remains the most significant. The NDVI is one of the most applied indices for the monitoring of seasonal vegetation changes and depends on vegetation

3

(25)

FCUP 9 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

reflectance properties. Several studies have been reported in the literature to investigate the impact of forest fires events through NDVI [11]. NDVI is computed as 1.2:

N DV I = ρN IR − ρRED

ρN IR + ρRED (1.2)

Where,

ρN IR = Near-infrared (NIR) reflectance; ρRED = = Red reflectance.

The NDVI assumes values between -1 and +1. The closer to one, the bigger is the vegetative activity. Negative values or close to zero indicate areas of water, buildings, soil and others where there is few to no chlorophyll activity. The basic principle of this index takes advantage of the fact that the more active the vegetation is, the greater is the absorption in the red region and the reflection in the near infrared region of the sunlight. Because of this, the digital information that arrives at the satellite, have low values in the red band and high values in the near infrared band (Figure 1.7).

Figure 1.7: Healthy leaf interacting with radiation.

As a ratio, the NDVI has the advantage to minimize certain types of band-correlated noises and influences attributed to variations in direct and diffuse irradiance, cloud shadows, sun and view angles, topography, and atmospheric attenuation [12].

(26)

1.3.3 Landsat 8

The Landsat Data Continuity Mission (LDCM) have the purpose to obtain data and images for use in agriculture, education, science and by government. This mission allows the repetitive acquisition of high-resolution multispectral data from the Earth’s surface in a global way. The data acquired by Landsat, constitute the largest register of the Earth’s surface and have an enormous detail, quality and coverage 4.

Landsat 8 (Figure 1.8) was launched on February 11, 2013 and is the eight satellite of the Landsat mission. This was developed in a collaboration between NASA and the U.S. Geological Survey (USGS) in order to obtain images with higher quality than those obtained with the satellites previously launched by this mission. It has a heliosynchronous circular orbit with an inclination of 98.2o, a period of 99 minutes, an altitude of 705 km, a bandwidth of 185

km and takes 16 days to pass in the same region.

Figure 1.8: Landsat 8. Source: NASA

According to NASA, the Landsat 8 system is robust, have a high-performance and provides high-quality data. This system:

• Provides systematic and global measurements of multispectral data with medium resolu-tion;

• Provides large amount of data;

• Uses an automatic system to detect shadows and clouds in images, avoiding the acquisition of data without useful information;

(27)

FCUP 11 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

• Ensures that all obtained images are stored by a ground station.

Landsat 8 has two types of sensors: an Operational Land Imager (OLI) Spectrometer and a Thermal Infrared Sensor (TIRS). Is comprised of 11 multispectral bands with different resolu-tions (Table 1.2).

Table 1.2: Landsat 8 bands settings.

* in micrometers; ** TIRS bands are acquired at 100-meter resolution, but are resampled to 30 meter in delivered data product

The OLI spectral sensor, similar to the sensor in Landsat 7, has the innovation of being able to measure in two new bands: Band 1 (Coastal / aerosol) and Band 9 (Cirrus). These new bands are useful for coastal, aerosol studies and to detect clouds. In total, this spectral sensor has nine bands and has a resolution of 30 meters from bands 1 to 7 and in band 9, while the band 8, is a panchromatic band, and has a resolution of 15 meters.

The TIRS thermal sensor measures the temperature of the Earth’s surface in two different thermal bands and has a resolution of 100 meters.

1.3.4 Sentinel-2A

The Sentinel is composed by a set of satellites being developed by ESA in collaboration with the European Commission for the specific requirements of the Copernicus programme. Sentinel missions have a diversity of technologies, such as RADAR and multispectral images to land, ocean and atmospheric monitoring.

The Sentinel-2 mission is a constellation which comprises two satellites, Sentinel-2A (launched on June of 2015) and Sentinel-2B (launched in March of 2017), with the same polar-orbiting

(28)

and phase difference of 180◦. This mission operates in the infrared and visible bands of the

electromagnetic spectrum, essentially dedicated to the monitorization of the Earth and offers an unprecedented combination of capabilities. First, makes systematic global coverage of the Earth’s surface (from 84◦ South and 84◦ North) in a five-day time, using the two satellites. The

Sentinel-2 mission is equipped with a MSI (MultiSpectral Instrument) sensor with 13 spectral bands (Table 1.8) with high spatial resolution (10, 20 or 60 m, depending on the band) constituting the possibility of a new perspective on soil and vegetation and have a bandwidth of 290 Km.

Table 1.3: Sentinel-2 bands settings.

Sentinel provides different levels of quality for the same product (Figure 1.9).

Figure 1.9: Sentinel-2 Processing Levels. Source: ESA

(29)

FCUP 13 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

accessible to common users. Level-1B processing uses the Level-1A product and applies the essential radiometric corrections. Level-1C processing uses the Level-1B product and applies geometric and radiometric corrections. All acquired data from the MSI sensor are processed to Level-1C by the Payload Data Ground Segment (PDGS).

1.4 Objectives

The main objectives of this work are:

1. Evaluate the utility of the Normalized Difference Vegetation Index (NDVI) to evaluate the burnt areas;

2. Explore the potentials of the Semi-Automatic Classification Plugin;

3. Compare the results obtained with the Landsat 8 and Sentinel-2A satellites for this type of work.

(30)
(31)

2. Methodology

This chapter is intended to report the characterization of the data used, as well as, different stages of the methodology applied in this work. A flowchart with the methodology employed is showed in 2.1.

Figure 2.1: Flowchart of the methodology followed in this work.

2.1 Study area

The selected area for this study consisted in two municipalities of the north part of Portugal Mainland: Arouca (from Aveiro district) and Vila Nova de Cerveira (from Viana do Castelo district), which were greatly affected by the 2016 summer fires. Another municipality (Sintra) in Lisboa district was also considered as control case, since no fire as occur there. Also, the Sintra municipality is mainly an urban area, unlike Arouca and Vila Nova de Cerveira (V. N Cerveira), which are mainly rural/forest areas. In 2.2 are presented the three areas selected

(32)

for this study.

Figure 2.2: (a)Portugal Mainland district. (b)Municipalities chosen for this study.

2.1.1 Vila Nova de Cerveira

The municipality of Vila Nova de Cerveira is inserted in the district of Viana do Castelo. It is placed in the extreme north of Portugal and borders with Spain. This district has an area of 108.47 km2 and is constituted by 11 parishes (Figure 2.3). Vila Nova de Cerveira has a forest

area that occupies near 38.6% (4,189 ha) of its territory and an area of uncultivated land and bushes of around 2,986 ha.

Figure 2.3: District of Vila Nova de Cerveira.

(33)

FCUP 17 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

occupies 3,334 ha (80% of the county’s forest area). There are other species like Eucalyptus (Eucalyptus globulus), Oak (Quercus robur ), Chestnut (Castanea sativa), and Alder (Alnus glutinosa) but in a much smaller occupied area.

Vila Nova de Cerveira is a humid region, with abundant precipitation, medium cloudiness, short autumns and summers with soft meteorological conditions, especially in the riverside. The presence of the Minho River reduces the thermal amplitudes, providing a mild climate. The annual average temperature is above 12.5◦C.

The Figure 2.4 shows a plot of the average monthly precipitation, temperature and wind velocity for the study period (Summer of 2016), these data were obtained from information recorded in the meteorological station of Vila Nova de Cerveira and made available by SNIRH

1. The average monthly precipitation assume values between 0.6 and 23.6 mm, it is possible

to assume that in this region the rain was scarce in this time of the year. Is perceptible an accentuate decrease of the values from June to July and in August the precipitation value increases again. In relation to the average monthly temperatures, this assumes values between 18.1◦C and 21.6◦C, with the hottest month being July. Despite this, the hottest day of 2016 in

this region was on 8 of August when it reached an average air temperature of 29.7oC. For the

three months the average monthly wind speed had a constant value close to 0.6 m/s.

Figure 2.4: Average precipitation, temperature and wind velocity values for the study period in Vila Nova de Cerveira district.

(34)

2.1.2 Arouca

Arouca covers an approximate area of 328 km2, is located in the extreme northeast of the

Aveiro district and is composed of 16 parishes (Figure 2.5). The municipality of Arouca borders with the counties of Gondomar, Castelo de Paiva, Cinf˜aes, Castro Daire, S. Pedro do Sul, Vale de Cambra, Oliveira de Azem´eis and Santa Maria da Feira. The forest area of Arouca is around 25 000 ha. The forest area of Arouca is mainly constituted by creeping plants that grow naturally like Heather (Calluna vulgaris) and Carqueja (Baccharis trimera) [13]. In the hillside areas, the predominant vegetation are Pines, Oaks and Arbutus [14].

Figure 2.5: District of Arouca.

Arouca have a very mild climate, since the annual average temperature is around15oC. The

coldest months are December, January and February with typical temperatures near 10◦C and

the hottest months are June, July and August, with temperatures that often exceeds 30oC. The

climate of this region is defined as Mediterranean, presenting a hot season with several hours of sun and a cold season characterized by some rain. The Figure 2.6 displays a plot of the average monthly precipitation, temperature and wind velocity for the study period (Summer of 2016), this graphic was made considering the data obtained in the meteorological station of Castelo de Burg˜aes Dam (nearest station of Arouca) accessible in SNIRH database 2. The

average monthly temperature assumes values between 18.2◦C and 21.7◦C with the hottest day

of 2016 being on 8th of August with 29.9◦C. The average precipitation have the lowest value in

(35)

FCUP 19 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

July with 0.016 mm and the highest value in June with 1.514 mm of rain. The wind velocity mean takes values between 0.77 and 0.87 m/s.

Figure 2.6: Average precipitation, temperature and wind velocity values for the study period in Arouca district.

2.2 Selected data

The main fire events for the selected study areas were registered on 11 August 2016 in Arouca and in 16 August 2016 in V. N. de Cerveira. Considering the study period (Summer of 2016) were selected pre and post fire Landsat-8 images from Arouca, Vila Nova de Cerveira and Sintra (control case) (Table 2.1 and 2.2). To minimize the effects of seasonality, the images were obtained on dates corresponding to the same season of the year, so the determinants factors of vegetation spectral responses did not suffer interference from natural phenomena, thus having equivalent spectral response parameters, in order to not interfering with the analyzes. During the selection of the images the percentage of clouds was taken into account and did not exceed 5% of cloud cover. These images are available for free from the USGS (United States Geological Survey)3 in WGS84 (World Geodetic System 1984) coordinate system and are geometrically

corrected. Its characteristics are listed in Chapter 1.3.3. From all the information obtained, the OLI sensor bands used were: 4th band (RED) and 5th (NIR) band. These bands have a

resolution of 30 meters.

(36)

Satellite Sensor Acquisition date (Arouca) Acquisition date (V. N. de Cerveira) Acquisition date (Sintra) Observation Landsat 8 OLI July 14, 2016 March 14, 2016 June 12, 2016 Pre fire event Landsat 8 OLI April 12, 2017 March 19, 2017 April 28, 2017 Post fire event

Table 2.1: Characteristics of the Landsat 8 satellite data used.

Satellite Sensor Acquisition date (Arouca) Acquisition date (V. N. de Cerveira) Acquisition date (Sintra) Observation Sentinel-2A MSI July 19, 2016 March 24, 2016 June 16, 2016 Pre fire event Sentinel-2A MSI April 15, 2017 March 19, 2017 May 3, 2017 Post fire event

Table 2.2: Characteristics of the Sentinel-2A satellite data used.

For the development of the work it was necessary to download the shapefiles of Corine Land Cover (CLC) and the Official Administrative Map of Portugal (CAOP) from DGT 4, as well

as, the DEM (Digital Elevation Models) generated considering the Shuttle Radar Topographic Mission (SRTM).

In order to analyses the real effects of the fire events in the NDVI, a specific area was selected in each municipality, in order to perform a deep analysis. The selected areas are marked in the images of the Figure 3. These areas were chosen taking into account the most burned areas in Vila Nova de Cerveira and Arouca using the data published in the Fire Report of 2016 provided by the ICNF 5.

Figure 2.7: Areas of the different municipalities selected for this study: (a) Arouca; (b) V.N Cerveira; (c) Sintra.

4http://www.dgterritorio.pt/

(37)

FCUP 21 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

2.3 Image processing

All the data were processed in an open source GIS environment. The image processing tasks were computed considering the QGIS plugin “Semi-Automatic Classification Plugin” developed by Luca Congedo [15].

2.3.1 Conversion to Top of Atmosphere (ToA) Reflectance

The spectral reflectance corresponds to the ratio of the reflected radiant flux (total radiated energy in all directions per unit of time) and the incident.

• Sentinel-2A

In Sentinel-2A images (Level-1C) the pixel radiometric measurements are available in scaled Top-Of-Atmosphere (TOA) reflectances6. This came in DN (digital numbers) and

is necessary to apply the following formula (Equation 2.1) to obtain physical values of the reflectance 7:

Ref letance(f loat) = DN

Quantif icationV alue (2.1)

Where,

DN = Digital Numbers;

Quantif icationV alue = value defined in .xml metadata file.

• Landsat 8

According to Mishra [16], in Landsat 8 images, digital numbers are converted to spectral radiance at the sensor’s aperture, Lλ, using the equation 2.2:

6https://earth.esa.int/web/sentinel/user-guides/sentinel-2-msi/product-types/level-1c 7https://sentinel.esa.int/documents/247904/349490/S2 MSI Product Specification.pdf

(38)

Lλ = Mρ× Qcal+ Aρ (2.2)

Where,

Lλ = Spectral radiance at the sensor’s aperture (W/(m2.µm));

Mρ = multiplicative factor of a given band;

Aρ = additive factor of a given band;

Qcal = value of the calibrated pixel.

The planetary reflectance at the top of the atmosphere, ρ0λ, can be determined from the

equation 2.3 8: ρ0λ = π × Lλ× d2 ESU Nλ× cos θ (2.3) Where,

ρ0λ = TOA planetary reflectance, without correction for solar angle;

Lλ = Spectral radiance at the sensor’s aperture;

d = Earth-Sun distance in astronomical units; ESU Nλ = Mean solar exo-atmospheric irradiances;

θ = Solar zenith angle in degrees.

It is possible to correct the error that the angle of the sun produces in the reflectance at the top of atmosphere using the following equation 2.4:

(39)

FCUP 23 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

ρλ =

ρλ0 cos(θSE)

(2.4)

Where,

ρλ = TOA planetary reflectance, with the correction of solar angle;

ρ0λ = TOA planetary reflectance, without correction for solar angle;

θSE = Local sun elevation angle. This information comes in degrees and is available in

the metadata.

2.3.2 Surface Reflectance

According to Mather [17], atmospheric correction is essential for computed index calculation, such as NDVI (Normalized Difference Vegetation Index).

Dark Object Subtraction is a method of image-based atmospheric correction to calculate the surface reflectance. Atmospheric interference is estimated only with the digital numbers of the image, which makes it simpler and possible to be applied at any scene, as it does not require data of atmospheric conditions at the time of acquisition of the images (for example, visibility). In this method it is assumed that there is a high probability that there are dark areas (pixels) in the scene, such as shadows caused by topography or clouds, which should have a very low DN in the image, equivalent to about 1% of reflectance [18] and therefore these dark pixels serve as reference for the correction of atmospheric scattering.

For the use of this method the following equation 2.5 is applied:

ρ = π × (Lλ− Lρ) × d

2

ESU Nλ× cos(θs)

(40)

Where,

ρ = land surface reflectance;

Lλ = value of spectral radiance (µm) in the sensor;

Lρ = Path Radiance;

d = Distance between the Sun and the planet Earth, given in astronomical units; θs = Solar zenith angle;

ESU Nλ = Mean solar exo-atmospheric irradiances (W/(m2*µm)).

To find the values of Lρ is applied the following equation 2.6:

Lρ =

ML× DNmin+ AL− 0.01 ∗ ESU Nλ× cos(θs)

π × d2 (2.6)

Where,

ML and AL = these values are in the file of the image metadata, and vary by the band;

DNmin = corresponds to the lowest value found in the raw product (digital number) of the

band that is going to be corrected.

Since the Landsat 8 and Sentinel 2A images are in 16 bits, the minimum value of the DN is between 0 and 65535, without units of measurement.

The value of ESU Nλ for Sentinel-2 sensor is provided in the metadata file and is different

for each band. In relation to landsat 8 it is necessary to calculate it using the following formula 2.7:

ESU Nλ =

π × d2× RADIAN CE

M AXIM U M

REF LECT AN CEM AXIM U M

(41)

FCUP 25 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

Where,

RADIAN CEM AXIM U M and REF LECT AN CEM AXIM U M = information provided in the

file of the image metadata, and vary through band.

2.3.3 Normalized Difference Vegetation Index (NDVI)

For this step the NDVI was calculated using the post-processing tool BandCalc of the SCP plugin. The characteristics as well as the formula to calculate this index is described in the Chapter 1.3.2.

(42)
(43)

3. Results

3.1 Statistical analysis

A statistical analysis was carried out in order to have a better understanding of the data obtained. These statistics resulted from the NDVI computation performed in the three study areas. They are presented in the Tables 3.1- 3.3, respectively for Arouca, V.N. Cerveira and Sintra. Landsat 8 Sentinel-2A NDVI 2016 2017 2016 2017 Minimum 0.125 0.036 -0.038 -0.126 Maximum 0.912 0.872 0.834 0.845 Mean 0.666 0.391 0.546 0.323 Standard dev. 0.115 0.140 0.112 0.142 Median 0.687 0.366 0.564 0.296

Table 3.1: NDVI statistical analysis from Arouca selected area.

Landsat 8 Sentinel-2A NDVI 2016 2017 2016 2017 Minimum 0.257 0.282 0.052 0.029 Maximum 0.775 0.820 0.724 0.736 Mean 0.610 0.521 0.517 0.397 Standard dev. 0.126 0.156 0.144 0.174 Median 0.659 0.511 0.578 0.390

Table 3.2: NDVI statistical analysis fromV.N. Cerveira selected area.

(44)

Landsat 8 Sentinel-2A NDVI 2016 2017 2016 2017 Minimum -0.169 -0.193 -0.276 -0.339 Maximum 0.912 0.904 0.873 0.883 Mean 0.653 0.684 0.611 0.619 Standard dev. 0.179 0.161 0.183 0.171 Median 0.705 0.739 0.671 0.676

Table 3.3: NDVI statistical analysis from Sintra municipality.

From the statistical analysis (Tables 3.1- 3.3) it is possible to conclude that the standard deviation and median values are identical for the results obtained with the two sensors analysed (OLI and MSI). In this analysis was given more focus to the median than the mean because for a large quantity of data, the median, unlike the mean, is not affected by outliers.

For Arouca selected area, the median of the NDVI values of 2017 decreased to almost half compared to the values of 2016. Considering the two sensors analysed, the NDVI values are higher (in both years) for the Landsat-8 than Sentinel-2A.

For V.N. Cerveira selected area, the mean and median values are identical, but the results considering the two sensors are different. Considering the Landsat 8 sensor, is noted a decrease in the median value of 0.659 in 2016 to 0.511 in 2017, while this decrease is much more pro-nounced in the NDVI values estimated considering the Sentinel-2A sensor (0.578 in 2016 and 0.390 in 2017).

In Sintra municipality, the median values are in general higher than the mean values; however, in terms of variability we can conclude that the NDVI values did not change between 2016 and 2017, since there were no records of forest fire in this zone.

3.2 Spatial analysis

3.2.1 Normalized Difference Vegetation Index (NDVI) Maps

For all the selected areas were generated NDVI maps before (2016) and after (2017) the fire events. These results are shown in Figures 3.1- 3.20.

(45)

FCUP 29 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

Figure 3.1: 2016 NDVI map for all Arouca municipality derived from Landsat 8 image.

Figure 3.2: 2017 NDVI map for all Arouca municipality derived from Landsat 8 image.

(46)

Figure 3.3: 2016 NDVI map for all Arouca municipality derived from Sentinel-2A image.

Figure 3.4: 2017 NDVI map for all Arouca municipality derived from Sentinel-2A image.

(47)

FCUP 31 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

Analyzing the maps of the municipality of Arouca (Figure 3.1 and 3.4) created from the NDVI values is clearly perceptible a considerable variation in the vegetation cover from 2016 to 2017. It is possible to notice that there was a decrease of the vegetation in the totality of the county being more notorious in the east zone of Arouca.

Figure 3.5: 2016 NDVI map for the selected area of Arouca municipality derived from Landsat 8 image.

(48)

Figure 3.6: 2017 NDVI map for the selected area of Arouca municipality derived from Landsat 8 image.

Figure 3.7: 2016 NDVI map for the selected area of Arouca municipality derived from Sentinel-2A image.

(49)

FCUP 33 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

Figure 3.8: 2017 NDVI map for the selected area of Arouca municipality derived from Sentinel-2A image.

Analysing the selected area of the municipality of Arouca (Figure 3.5- 3.8), it is notable that all the area had a very sharp decrease of the vegetation level. Both maps, created from Landsat 8 and Sentinel-2A images, show that the NDVI’s maximum and minimum values have decreased. It would be expected a more pronounced reduction in the maximum value but there may have been forest development in some part of the study area, which made this decrease not so perceptible.

(50)

Figure 3.9: 2016 NDVI map for all Vila Nova de Cerveira municipality derived from Landsat 8 image.

Figure 3.10: 2017 NDVI map for all Vila Nova de Cerveira municipality derived from Landsat 8 image.

(51)

FCUP 35 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

Figure 3.11: 2016 NDVI map for all Vila Nova de Cerveira municipality derived from Landsat 8 image.

Figure 3.12: 2017 NDVI map for all Vila Nova de Cerveira municipality derived from Landsat 8 image.

(52)

In relation to the municipality of Vila Nova de Cerveira (Figure 3.9 - 3.12), analyzing the the maps of the NDVI created by Landsat 8 and Sentinel-2A images, overall there is no perception of a decrease in vegetation index. In some areas it is even noticeable the development of vegetation. A more detailed analysis was then made and is shown in the Figures 3.13- 3.16.

Figure 3.13: 2016 NDVI map for all of Vila Nova de Cerveira municipality derived from Landsat 8 image.

(53)

FCUP 37 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

Figure 3.14: 2017 NDVI map for all of Vila Nova de Cerveira municipality derived from Landsat 8 image.

Figure 3.15: 2016 NDVI map for the seleted area of Vila Nova de Cerveira municipality derived from Sentinel-2A image.

(54)

Figure 3.16: 2017 NDVI map for the seleted area of Vila Nova de Cerveira municipality derived from Sentinel-2A image.

Considering the selected area in the municipality of Vila Nova de Cerveira (Figure 3.13 -3.16), it is more noticeable the changes in vegetation. The western zone of this area was largely affected by the decrease of vegetation and it is also visible that in the Southeast area there was an intensification of flora. This means that this extension was not affected by forest fires and that from one year to the other occurred forest development. The growth of vegetation would be expected in most of the land if there were no phenomena that destroyed it. The part of Vila Nova de Cerveira chosen was a relatively small area, making the images created with Landsat-8 very pixelated since this sensor has a lower resolution compared to the resolution of Sentinel-2A.

(55)

FCUP 39 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

Figure 3.17: 2016 NDVI map for all Sintra municipality derived from Landsat 8 image.

Figure 3.18: 2017 NDVI map for all Sintra municipality derived from Landsat 8 image.

(56)

Figure 3.19: 2016 NDVI map for all Sintra municipality derived from Sentinel-2A image.

Figure 3.20: 2017 NDVI map for all Sintra municipality derived from Sentinel-2A image.

(57)

FCUP 41 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

For Sintra municipality was not presented a NDVI image for a specific selected area because this area was only selected to be used as control case, since no fire as occur there in the period of analysis. In Figure 3.17- 3.20 (and analyzed in Table 3.3) neither an increase nor a decrease in the vegetation index is perceptible as expected.

3.2.2 NDVI differences

For the two selected areas affected by forest fires in Arouca and V.N Cerveira, a map of the NDVI differences between 2017 and 2016 (dNDVI) were computed. This analysis is showed in Figures 3.21- 3.24. From the dNDVI maps it is possible to conclude that the results obtained consider the Landsat and Sentinel-2A images are identical. However, the best spatial resolution of Sentinel-2A sensor allows to generate a more accurate map.

Figure 3.21: dNDVI map for the selected area of Arouca derived from Landsat 8 image.

(58)

Figure 3.22: dNDVI map for the selected area of Arouca derived from Sentinel-2A image.

Figure 3.23: dNDVI map for the selected area of V.N. Cerveira derived from Landsat 8 image.

(59)

FCUP 43 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

Figure 3.24: dNDVI map for the selected area of V.N. Cerveira derived from Sentinel-2A image.

In order to analyse the difference between the results obtain with Landsat-8 and Sentinel-2A was applied a method to realize if exist a very large variance between the total values obtained with the two sensors. To the raster with the differences between NDVI’s was made a reclassification that assumed that values less than zero represent a decrease in the value of NDVI and it was considered that for the rest of the values there was no change or there was an increase of vegetation index. After this reclassification, the areas less than zero and the areas equal to or greater than zero were calculated. The results are shown in the Table 3.4 and 3.5.

area km2 (>1) area km2 (<1) Landsat-8 1,9962 27,7407

Sentinel-2A 3,3003 26,4431

Table 3.4: Areas calculated from the map of the NDVI differences between 2017 and 2016 (dNDVI) for the selected area of Arouca.

area km2 (>1) area km2 (<1)

Landsat-8 0,9827 9,8792 Sentinel-2A 1,2794 9,5825

Table 3.5: Areas calculated from the map of the NDVI differences between 2017 and 2016 (dNDVI) for the selected area of Vila Nova de Cerveira.

(60)

It is not possible to conclude that these areas are close to the areas burned in reality but it is perceptible that there is no great difference between the values obtained by the two sensors. There are many factors that can lead to a decrease in NDVI value besides fire deforestation.

3.2.3 Factors related to fire propensity

There are numerous other factors beyond vegetation, such as ignition agents, topography, landscape fragmentation, and fire management activities that could influence the fire activity in a region. For instance, slope can significantly influence fire behavior. In order to show how the approach developed in this work can also be used to investigate the relationship between the slope and other factors we relate the NDVI values with slope. The slope map presented in Figure 3.25 (a) was generated consider the SRTM 90 m Digital Elevation Model (DEM) with an original resolution of 90m at the Equator [19], and resampling for 25 m. Analyzing the maps presented in Figure 3.25 for the Arouca selected area, more specifically for the areas marked in the images, it is possible to check that the areas with the highest slope values were the areas with the highest decrease of NDVI values (2017); which leads us to conclude that these areas are associated with fires occurrences. A more detailed analysis should be performed in order to take more robust conclusions, but it is proved the usefulness of this type of analysis.

Figure 3.25: (a) Slope map for the selected area of Arouca and the NDVI values for the Sentinel-2A image for (b) 2016 and; (c) 2017.

In Chapter 2.1 the weather conditions of the time under study are shown. The existing conditions, high temperatures and low perception, at the time maximized the hypothesis of fire propagation. The hottest days of the year of 2016 for the study zones coincided with the time in which the fires occurred with greater magnitude. The high temperatures provoke

(61)

FCUP 45 Normalized Difference Vegetation Index (NDVI) analysis for evaluation of forest fires

in the summer of 2016 in Portugal

the evaporation of the existing water in the vegetation and this propitiates the ignition of this vegetation. Also, the areas affected by the fires are extents characterized by fire-prone forest, meaning, combustive flora. This type of vegetation creates the perfect medium for a proliferation of fire.

3.3 Conclusions and Perspectives

3.3.1 Main conclusions

This work had the objective to analyze the forest fires that occurred in the year of 2016 in Portugal using the Normalized Difference Vegetation Index (NDVI).

With the development of this work it was possible to perceive that remote sensing is an excellent resource for data collection providing reliable and rapid information, making it possible to analyze and effective monitoring the landscape changes through Geographic Information Systems. The common user easily have access to information with great precision that allows to study different ways of analyzing the Earth globe and the changes that occur in it. Conventional inventories allow us to obtain local information, while remote sensing permits us to have a more global data of a given parameter under study.

Through the analysis of the NDVI it was possible to identify areas where there was a decrease in the vegetation level. This decrease in some areas corroborates the information obtained for example in the Forest Fire Report of the ICNF.

Given the results obtained, it can be stated that the higher spatial resolution of Sentinel-2A sensor allows a higher sensibility in the estimation of NDVI, which lets determine the areas affected by the fires with high accuracy. Therefore, this sensor should be considered in this type of analyses. Added to the best spatial resolution of Sentinel-2A, the temporal resolution of Sentinel-2A (10 days) was increased with the launch of the Sentinel-2B (very recently) and therefore the frequency of the combined constellation revisit will be 5 days. However, for historical studies, the Landsat program remains the best option. The NDVI is an excellent indicator in order to be used to analyze the impact of forest fires, allowed in addition to

(62)

performing a robust statistical analysis, a very clear visual interpretation.

3.3.2 Perspectives

The facility of free satellite imagery and open-source imaging applications by ESA and NASA will allow further studies in this area and in other areas of RS. The arrival of Sentinel-3 may be an advantage in this type of studies and possibly an integration of this sensor with Sentinel-2A would be an advantage in the analysis of forest areas. Not only for the NDVI but as far as the development of other indices.

Forest fires are going to continue to be a very worrying case of study as more and more fires destroy everything, putting entire ecosystems at risk. For this reason other methods of analyzing this problem will be created in order to understand the causes, risks and damages caused.

(63)

References

[1] R. Velez. A defesa da floresta contra incˆendios florestais: estrat´egias, recursos, orga-niza¸c˜ao. ISAPress, 2011.

[2] M. F. Cam˜oes. Porque ardem as florestas. eng. 2nd. Folhas de Qu´ımica. Sociedade Por-tuguesa de Qu´ımica, 2006.

[3] J. P. Fernandes and A. R. M. Freitas. Introdu¸c˜ao `a Engenharia Natural. Volume II. EPAL - Empresa Portuguesa das ´Aguas Livres, S.A., 2012. isbn: 978-989-97459-5-7.

[4] J. Ventura and M. Vasconcelos. O fogo como processo f´ısico-quimico e ecol´ogico, Incˆendios Florestais em Portugal, Caracteriza¸c˜ao, Impactes e Preven¸c˜ao. ISAPress, Lisboa, 2016. [5] J. Obano and G. Makokha. AGE 301: Physical Geography. Department of Geography.

Astrophys. J. 796, 132. 2005.

[6] D. Halliday, R. Resnick, and J. Walker. Fundamentals of physics. Volume 1. 1993. [7] T. Owen. Fundamentals of Modern UV-Visible Spectroscopy : A Primer. Hewlett-Packard,

1996.

[8] E. Chuvieco. Fundamentos de teledetecci´on espacial. 2th. Ediciones Rialp.S.A., 1995. isbn: 84-321-2680-2.

[9] P. Meneses and T. Almeida. Introdu¸c˜ao ao processamento de imagens de sensoriamento remoto. 1st. 2002.

[10] R. Myneni, F. Hall, P. Sellers, and A. Marshak. Themeaning of spectral vegetation indices. IEEE Trans. Geosci. Remote Sensing., 1995.

[11] G. Navarro, T. Chu, and X. Guo. Remote sensing techniques in monitoring post-fire effects and patterns of forest recovery in boreal forest regions: a review. Remote Sensing., 2013. doi: 10.3390/rs6010470.

(64)

[12] A. R. Huete1, K. Didan1, and W. V. Leeuwen. Modis Vegetation Index(MOD 13), Algo-rithm theoretical basis document. Vegetation Index and Phenology Lab, 1999.

[13] M. Caixinhas, P. Forte, and M. Sousa. Trevos, Anafes e Luzernas de Portugal. Verbo. 2016. isbn: 9789722231664.

[14] G. Mendes. Carateriza¸c˜ao de proveniˆencias de Pinus elliottii e Pinus taeda para instala¸c˜ao de ensaios de proveniˆencias. eng. 2017. doi: 7.

[15] L. Congedo. Semi-Automatic Classication Plugin Documentation. eng. 2016. doi: 10 . 13140/RG.2.2.29474.02242/1.

[16] N. Mishra et al. Radiometric Cross Calibration of Landsat 8 Operational Land Imager (OLI) and Landsat 7 Enhanced Thematic Mapper Plus (ETM +). Volume 6. Remote Sensing. 2014. doi: 10.3390/rs61212619.

[17] P. Mather and M. Koch. Computer Processing of Remotely-Sensed Images: An Intro-duction. eng. 4th edition. John Wiley and Sons, Ltd, 2011. isbn: 9780470742396. doi: 10.1002/9780470666517.

[18] P. Chavez. Radiometric calibration of Landsat thematic mapper multispectral images. vol-ume 55. Photogrammetric Engineering and Remote Sensing. 1989.

[19] E. Rodriguez, C.Morris, and J. Belz. A global assessment of the SRTM performance. Photogramm. Eng. Rem. Sens. 2006.

Referências

Documentos relacionados

Alguns ensaios desse tipo de modelos têm sido tentados, tendo conduzido lentamente à compreensão das alterações mentais (ou psicológicas) experienciadas pelos doentes

Este relatório relata as vivências experimentadas durante o estágio curricular, realizado na Farmácia S.Miguel, bem como todas as atividades/formações realizadas

The probability of attending school four our group of interest in this region increased by 6.5 percentage points after the expansion of the Bolsa Família program in 2007 and

Para tanto foi realizada uma pesquisa descritiva, utilizando-se da pesquisa documental, na Secretaria Nacional de Esporte de Alto Rendimento do Ministério do Esporte

Ao Dr Oliver Duenisch pelos contatos feitos e orientação de língua estrangeira Ao Dr Agenor Maccari pela ajuda na viabilização da área do experimento de campo Ao Dr Rudi Arno

Neste trabalho o objetivo central foi a ampliação e adequação do procedimento e programa computacional baseado no programa comercial MSC.PATRAN, para a geração automática de modelos

The present study aimed to evaluate the accuracy of the Normalized Difference Vegetation Index (NDVI) and Inverse Ratio Vegetation Index (IRVI) in the prediction of grain yield

If, on the contrary, our teaching becomes a political positioning on a certain content and not the event that has been recorded – evidently even with partiality, since the