1 - Introduction
The distribution of vegetation, its properties and state, is of major importance for a wide range of applications, namely:
- Environmental management;
- Natural Hazards monitoring;
- Agriculture and forestry;
- Climate change studies;
- Numerical weather forecast models.
Changes in the landcover either caused by changes in land use, climate change or natural hazards (like forest fires or droughts, for instance) may have a huge social and economic impact. An example of this was the severe drought that stroke Eastern Africa in 2009, causing crops to shrink and threatening millions of people with starvation ( the-great-drought-in-east-africa).
Remote sensing provides the best means to monitor changes in vegetation over a wide range of temporal scales over large areas. As an example, the decrease in vegetation cover associated to the 2009 Eastern Africa drought, is shown in the comparison between an image of the Fraction of Vegetation Cover in 2007 to the same period during the 2009 (please hit the play button bellow to see the animation):
Figure 1 – Fraction of Vegetation Cover product obtained on the aim of LSA SAF project, from SEVIRI sensor on board Meteosat Second Generation (MSG) satellite.
Several empirical indices have been proposed and used through the years, which allow an easy identification and monitoring of the vegetation conditions from satellite measurements. However such indices have several disadvantages. Because they are ratio based they are nonlinear, have noise effects and are not structural properties of land surface areas. There are several key variables that can be use for a wide range of land biosphere applications that are more directly related to vegetation properties and health than conventional empirical indices. The Satellite Application Facility on Land Surface Analysis (LSA SAF) produces several of those variables, making then available both in near real time and off-line.
The module aims at giving an overall view on the use of satellite data, particularly that provided by Meteorological satellites, for monitoring vegetation cover and properties. It is organized as follows: chapter 2 focus on basic physical concepts; chapter 3 details how vegetation interacts with radiation and describes satellite sensors suitable to retrieve information from vegetation surfaces; chapter 4 shows how the unique spectral signatures of vegetation described in the previous chapter can be used to derive widely used empirical indices and their relation with vegetation properties; chapter 5 presents several vegetation products, which are key variables for the monitoring of dynamic changes in the structure and the functioning of vegetation at the earth surface; and finally chapter 6 provides a list of references where you can find more information about several module subjects.
This module was thought for a broad community, which includes users from: weather forecasting and climate modelling, requiring detailed information on the nature and properties of land; environmental management and land use, needing information on land cover type and land cover changes; users dealing with agricultural and forestry applications, requiring information on incoming/outgoing radiation and vegetation properties; natural hazards managers, requiring frequent observations of terrestrial surfaces.
This module is created within the EUMeTrain project by Carla Barroso (LSASAF - IM, Portuguese Met. Service) and Isabel Monteiro (LSASAF - IM, Portuguese Met.Service). If you think you can contribute to this CAL or you have suggestions or questions please feel free to contact us.
2 - Physical concepts
In this chapter we make a revision of the physical concepts that support the monitoring of vegetation with satellite observations.
2.1 - Radiation principles
The Sun is the main source of energy in the earth-atmosphere system, irradiating electromagnetic energy in a wide range of wavelengths. According to Planck’s law, all bodies with temperature above 0K emit electromagnetic radiation. A black body at Temperature T emits radiation according to equation 2.1:
Bλ (Wm-3) =
energy emitted at a given wavelength λ (m);
c1 and c2 are constants (3.74x10-16 Wm2 and 1.44x10-12 mK, respectively).
The spectral distribution of the emitted radiation of a body (B as a function of λ) can be obtained by applying the Planck function at a given temperature T and considering a range of possible wavelengths.
2.1.1 - Solar radiation
For the case of the Sun, and considering its temperature at approximately 6000K, we can plot its emitted radiance as a function of the wavelength (Figure 2.1):
2.1.2 - Visible light and thermal radiation
The visible region of the electromagnetic spectrum, commonly called visible light, is the portion of the electromagnetic spectrum that is visible to (can be detected by) the human eye. Within this region, the longest wavelength is red and the shortest is violet. Table 2.1 lists the common wavelengths of what humans perceive as colours within visible light:
Up to now we have seen that electromagnetic radiation emitted from a body at a certain wavelength is determined by its temperature. The most relevant features of the solar emitted radiance have been revised.
Having in mind that objects at the surface of the Earth are at lower temperatures than the Sun, should we expect them to emit radiation with characteristics similar to those of solar irradiance?
Figure 2.2 shows the spectral distributions of radiation emitted both by the Sun and the Earth considered to be at a mean temperature of about 300K (close to the temperature of most objects at the surface of the Earth).
Figure 2.2 – Spectral irradiance for the Sun and the Earth
This figure shows significant differences of terrestrial radiation when compared to solar irradiance:
- The peak of terrestrial irradiance is located at about 10μm, in the thermal infrared region;
- The most significant fraction of terrestrial irradiance occurs between [3 - 100μm] (longwave radiation);
So all objects emit radiation (as far as their temperature is above the absolute zero). But emissions from objects at the Earth surface occur mostly at wavelengths at which our eyes are unable to see.
- Earth objects emit radiation at long wavelengths.
- Human eyes are only sensitive to short wavelengths (only emitted by the Sun).
So the question is, how can we see the objects? And why do we distinguish different colours?
To answer these questions we must understand how radiation interacts with matter.
2.2 - Interactions with matter
When electromagnetic energy hits a target three different interactions may occur, as illustrated in Figure 2.3:
Figure 2.3 – illustration of the different possible interactions between radiation and a surface feature.
So from the incident light on a surface, there may be a fraction which is transmitted, another which is absorbed, and a third fraction that is reflected. This portion of reflected radiation is the reason why we can actually see the objects. Moreover, the colour of the object results from the combination of wavelengths of the reflected portion of light bouncing off the object, as illustrated in the figures 2.4 and 2.5.
This principle is applied while making composites of satellite images. An RGB true colour image is a composite of satellite images using red, green and blue spectral channels. The result is an image that has the same appearance as seen from human eyes.
Figure 2.6 is an example of a true colour RGB Image of the earth showing the ongoing eruption of Chile’s Chaitén Volcano, in a region with different types of soil, as well as a variety of bluish tones in sea water.
Figure 2.6 – True colour image taken on March 6, 2009 with the Advanced Land Imager on the Earth Observing-1 (EO-1) satellite over South America (Image from http://earthobservatory.nasa.gov/NaturalHazards/).
While sensing vegetation from space, we will be interested in the reflectance mechanism. Reflectance is usually expressed as a fraction of the total amount of energy striking an object, taking the values:
- 1 - if all of the incident radiation bounces off and is detected by the sensor;
- 0 - if no light returns from the surface
In most cases, the reflectance value of an object for any region of the electromagnetic spectrum is somewhere between these two extremes.
2.3 - The effect of the atmosphere
So interaction of light with objects allows us to see them. But what affects radiation reaching a sensor at the top-of-atmosphere? What is the contribution from the surface? How does radiation interact with the atmosphere?
Next figure illustrates what happens to solar radiation in its transmission through the atmosphere:
As you can see, solar radiation is attenuated by:
- Reflection (by clouds and atmospheric constituents);
- Absortion (by atmospheric constituents)
What are the atmospheric constituents causing this attenuation of solar radiation? Do they act equally for the all regions of the electromagnetic spectrum?
2.3.1 - Absorption bands and atmospheric windows
Figure 2.8 shows the absorption spectra for the various atmospheric gases between the top of the atmosphere and the earth’s surface, as a function of the wavelength, namely:
- The tropospheric nitrous oxide (N2O) and methane (CH4), which have absorption bands in the infrared. Nevertheless, their concentrations are too low to have a significant impact.
Figure 2.8 – Absorption spectra for various atmospheric gases (adapted from Peixoto and Oort, 1992)
To answer to this question you should have mentally combined the individual effects of the atmospheric gases. This corresponds to Figure 2.9:
Figure 2.9 – Impact of the atmospheric gases on solar radiation (adapted from Peixoto and Oort, 1992)
The spots you selected are called atmospheric windows, where radiation is allowed to pass through the atmosphere with little attenuation (transmissivity is high and absortivity is low). It is through these windows that satellites observe the surface of the Earth on the visible and/or infrared regions of the electromagnetic spectrum.
The areas of the electromagnetic spectrum where radiation is most absorbed are known as absorption bands.
The solar energy that is able to pass through the atmosphere will reach the surface of the earth and interact with the surface features through the same mechanisms of reflection, absorption and transmission.
2.4 - Spectral signature of an object
The reflectance values for different objects, over a range of wavelengths, may be plotted for comparison. Such plots are called “spectral response curves” or “spectral signatures.” Differences among spectral signatures of landscape features are used to help classifying remotely sensed scenes, since the spectral signatures of similar features have similar shapes.
Figure 2.10 shows the spectral signatures of 3 different materials: conifers (in green), water (blue) and soil (red).
Figure 2.10 – Spectral response curves for 3 different types of features on the surface of the Earth.
The previous figure implies that the mentioned feature types are normally spectrally separable. The degree of separation of landscape features depends, however, on the range of wavelengths we are considering. This means that two features may be indistinguishable for a certain spectral range and perfectly individualized for other spectral band.
2.5 - Bidirectional reflection
So far we have seen that the visual perception of colour is related to the differential absorption/reflection of radiation by an object in the different regions of the electromagnetic spectrum. But why do we perceive different intensities of light? What physical explanation is behind the creation of patterns of lightness and darkness?
To answer to this question we must consider the different ways an object may reflect light. This is primarily a function of the object surface’s roughness.
The Sun Glint effect mentioned in the exercise can be seen in next figure taken with MSG-1:
Figure 2.11 - Midday sun glint over the Kongo river (in black) MSG-1, 24
March 2004, 09:00 UTC, Channel 3.9 μm
(Image taken from
http://oiswww.eumetsat.org/WEBOPS/msg_interpretation/PowerPoints/Channels/Channel_IR39.ppt, accessed in 2010)
2.5.1 - BRFDs
The measured surface reflectance depends on the configuration: sun position – observation geometry.
First think of a geostationary satellite, located at fixed position above the Earth. For each individual pixel only the Sun geometry varies from slot to slot.
For the case of polar orbiters, (normally sun-synchronous) it is mainly the satellite observation angles that change between observation slots of the same region.
This means that reflectance values measured with geostationary or polar satellites need to be corrected for directional effects. Such corrections may be easily performed if the Bidirectional Reflectance Distribution Functions (BRDF) are known. The BRDF model describes reflection behavior for all possible combinations of sun-viewing geometries. The characteristics of the BRDF will determine what "type" of material the viewer thinks the displayed object is composed of.
Among the several Bi-directional Reflectance Distribution Functions proposed by different authors, the one by Roujean et al. (1992) has been extensively used for vegetation parameters estimations:
ρ (θs, θv, φ) = K0 + K1 x F1(θs, θv, φ) + K2 x F2(θs, θv, φ)
and F2 are functions, called kernels, which only
depend on the angular configuration of aquisition:
- θs is the sun zenith angle;
- θv is the satellite zenith angle
- φ is the relative azimuth angle between sun and satellite
The K coefficients quantify
the respective contribution of each kernel to the whole BRDF and are
derived by model inversion.
- K0 corresponds to isotropic reflectance, i.e. reflectance values directionally normalised to reference illumination and observation zenith angles of 0°. This geometry leads to a minimum of the shadow proportion (hotspot geometry);
- K1 and K2 represent angular distribution related to geometric (caused by diffuse reflectors) and volumetric (caused by volume scatter reflectors) surface scattering processes.
The isotropic, geometric and volumetric components of surface reflectance are intrinsically linked to the amount and state of vegetation, canopy structure and to the leaf and soil optical properties, as you will see in more detail in chapter 5.
3. Satellite sensors and vegetation
In this chapter we will detail how vegetation interacts with radiation. We will first see how to distinguish between vegetated and non vegetated surfaces using some broad features that are spectrally separable. Then we will emphasize on how spectral resolution is used to distinguish between different species of vegetation, different phases of the life cycle of plants and vegetation stress. Finally we will describe the satellite sensors suitable to retrieve information from vegetation surfaces.
3.1 - Spectral response of different surfaces
As already mentioned in chapter 2.3 different surfaces have different reflectance spectra. Vegetated and non vegetated surfaces can be recognized by some distinctive broad feature types:
Beyond 1.3μm, vegetation essentially absorbs or reflects the incident radiation; transmittance is negligible and the dips in the reflectance spectra correspond to absorption at those wavelengths.
Figure 3.1 – Reflectance spectrum of green grass.
Non vegetated surfaces
The reflectance curve of non vegetated surfaces (in the figure it is exemplified by sandy soil) has smaller dependence on wavelength. Some of the factors affecting soil reflectance are moisture content, soil texture, surface roughness, presence of iron oxide and organic matter.
Figure 3.2 – Reflectance spectrum of sand.
3.2 - Fundamental factors affecting vegetation reflectance
A typical reflectance spectrum of a vegetation canopy can be subdivided into 3 parts, visible (0.40 –m 0.70 μm), near infrared NIR (0.701 – 1.3 μm) and middle-infrared (1.301 – 2.5 μm). Another optical property of the reflectance spectrum of vegetation is the abrupt transition between the strong absorption in the visible red and the strong reflectance in the NIR domain known as Red Edge.
Figure 3.3 – Reflectance spectrum of green grass subdivided into the 4 main optical properties of the reflectance spectrum of vegetation.
NOTE: The classification of the electromagnetic radiation used in this module is the most widely used in the vegetation community. The Visible is within the interval 0.40 μm - 0.70 μm; the Near Infra-red 0.701μm - 1.3μm and the Middle Infra-red 1.301μm - 2.5μm. Thus, in this module, channel 0.6μm is called Visible (VIS); channel 0.8μm - Near Infra-red (NIR) and 1.6μm - Middle Infra-red (MIR), instead of the classification used in the Satellite Meteorology community 0.6μm - Visible (VIS), 0.8μm - Visible (VIS) and 1.6μm - Near Infra-red (NIR)
3.2.1 - Visible region
The visible part of the reflectance spectrum (0.40 μm – 0.70 μm) of vegetation is controlled by the pigments in the green leaf chloroplasts that reside in the outer or Palisade leaf, the Chlorophyll pigments – chlorophyll-a and chlorophyll-b. A pigment is any substance that absorbs light. The colour of the pigment is determined by reflected wavelengths. White pigments/light colours reflect all or almost all of the energy striking them. Black pigments absorb all of the wavelengths that strike them. Chlorophyll is the major absorber of radiation in the visible region and its absorption is dominant in the visible red 0.6 – 0.7 μm – red wavelengths; it is called the green pigment and it is common to all photosynthetic cells. Other leaf pigments also have an important impact on the visible part of the spectrum. The carotene (yellow to orange-red pigment responsible for the colour of some flowers, fruits and leaves without chlorophyll) and xantophyll (responsible for the leaf colour in autumn) have strong absorption in the 0.35 – 0.5 μm, blue wavelengths.
Figure 3.4 – Reflectance spectrum for green grass, for VIS wavelengths.
Figure 3.4. shows a typical reflectance spectrum for green vegetation, for VIS wavelengths. Since absorption occurs in red and blue wavelengths, the predominant reflectance of visible wavelengths is concentrated in the green.
3.2.2 - Near infrared (NIR)
The optical properties in the near infrared spectral domain (0.701 μm 1.3 μm) are explained by leaf structure. The spongy mesophyll cells located in the interior or back of the leaves reflects NIR light, much of which emerges as strong reflection rays. The intensity of NIR reflectance is commonly greater than most inorganic materials, so vegetation appears bright in NIR wavelengths.
Figure 3.5 – Reflectance spectrum for green grass, for NIR wavelengths
3.2.3 - The red edge
The red edge is a region in the red-NIR transition zone of vegetation reflectance spectrum and marks the boundary between absorption by chlorophyll in the red visible region, and scattering due to leaf internal structure in the NIR region. This transition zone is in the basis of several vegetation indices like NDVI which is the normalized difference between the reflectance in the red visible (0.6µm) and the NIR (0.8µm) reflectance. Also the red edge position (REP) is used to estimate the chlorophyll content of leaves or over a canopy.
Figure 3.6 – Reflectance spectrum for green grass, for the transition zone between VIS and NIR wavelengths- the Red Edge
The concept of Red Edge is on the basis of the most widely used vegetation index - the Normalized Difference Vegetation Index (NDVI). This index is the normalized difference between the NIR and the red VIS reflectance.
Where ρ NIR and ρred are the reflectance values for NIR and VIS (red) bands, respectively.
This combination of the strong absorption in the red visible and strong reflectance in the NIR is very specific of vegetation and allow us to distinguished vegetation from bare soil (see slide Figs 3.1. and 3.2.). Calculations of NDVI for a given pixel can always result in a number that ranges from -1 to +1; however, for natural surfaces NDVI values are within the 0 to +1 range. An NDVI close to 0 corresponds to no vegetation, while NDVI close to +1 (0.8 - 0.9) indicates the highest possible density of green leaves. NDVI can be computed at TOA or at the surface, i. e. using atmospherically corrected reflectances for NIR and Red bands.
Figure 3.7 – NDVI daily composite of SEVIRI full disk, June 15, 2008 (extracted from Yu,Y. et al. Development of Vegetation Products for U.S. GOES-R Satellite Mission, Presented on 4th Global Vegetation Workshop Univ. of Montana, Missoula, June 16-19 2009 )
3.2.4 - The middle-infrared region (MIR)
The middle-infrared region (1.301 μm - 2.5 μm) contains information about the absorption of radiation by water, cellulose and lignin and several other biochemical constituents. ( For further information see for ex. Curran,1989).
Figure 3.8 – Reflectance spectrum for green grass, for MIR wavelengths
This region of the vegetation spectrum allows the identification of vegetation stress due to drought.
3.3 - The importance of spectral resolution
In the previous section we saw that the spectral characteristics of the vegetation spectrum are related with the leaf pigments, leaves internal structure and water content within leaves. A close analysis of the vegetation spectral reflectance provides information on the phase of vegetation life cycle, health, and even the identification of individual vegetation cover types.
3.3.1 - Distinguishing among different life cycle phases
When a plant undergoes senescence, chlorophyll is no longer produced, and other pigments present in plants become visible.
3.3.2 - Vegetation stress
Following changes in vegetation, and therefore in foliar chemistry and membrane structure, the vegetation spectral signatures may be substantially modified. Stress is indicated by a progressive decrease in near-Infrared reflectance accompanied by an increase in middle-Infrared reflectance. In the visible domain, changes are limited to variation in colour, as shown in the previous subchapter.
Figure 3.10 – Reflectance spectrum of healthy and stress sugar beets (Kyllo, 2003).
figure extracted from http://rst.gsfc.nasa.gov/, accessed in 2010.
3.4 - Satellite sensors sensitive to vegetation
Given the interaction between vegetation and radiation analysed throughout the previous sections, the monitoring of vegetation requires sensors with bands/channels within following domains: VIS, NIR, MIR.
Figure 3.11 Bands/channels suitable to retrieve vegetation information.
For detailed information about the different sensors used to retrieve vegetation parmeters please consult Satellite Operator- spectral channels.doc
3.5 - RGB images
RGB images can be used to extract information about vegetation cover. RGB techniques work by associating a colour to a particular channel.To extract maximum information from a single image each colour (Red, Green or Blue) should be associate to a different channel with distinct physical properties.
To extract information about vegetation cover, the RGB composite 1.6; 0.8; 0.6 μm, often referred as "Day natural colours" composite, can be used.
In this composite Red is associated to NIR 1.6 μm; Green to VIS 0.8 μm and Blue to VIS 0.6 μm. So as you can see in the image:
Figure 3.12 – Meteosat-9 colour composite image of 1.6μm; 0.8μm and 0.6 μm
- Reddish/pink is dominant for bare soil and arid regions.
- Green is the dominant component for vegetation.
- Cyan is evident in ice clouds.
A complete description of how to apply RGB techniques in satellite images and forecasting, can be found in www.zamg.ac.at/eumetrain/CAL_Modules/CALRGB/
4 - Empirical Vegetation Indices
Vegetation indices (VI’s) have a long history of use over a wide range of applications, such as vegetation monitoring; climate and hydrologic modelling; agricultural activities; drought studies and public health issues. As shown in chapter 3, vegetation has unique spectral signatures, which evolve with the plant life cycle. VI’s are dimensionless radiometric measures, which combine information from different channels, particularly in the red and NIR portions of the spectrum, to enhance the 'vegetation signal'. Such indices allow reliable spatial and temporal inter-comparisons of terrestrial photosynthetic activity and canopy structural variations. They are generally computed for all pixels in time and space, regardless of biome type, land cover condition and soil type, and thus represent true surface measurements. Due to their simplicity, ease of application, vegetation indices have a wide range of usage within the user community.
4.1 - NDVI
As detailed in chapter 3, the relationship of red and NIR reflected energy is clearly related to the amount of vegetation present on the ground (Huete et al. 1999). Reflected red energy decreases with plant development due to the chlorophyll absorption within actively photosynthetic leaves. Reflected NIR energy, on the other hand, will increase with plant development through scattering processes in healthy leaves.
The most widely used VI is the Normalized Difference Vegetation Index, it consists of a normalized ratio of the NIR and red bands
ρred and ρNIR(either top of atmosphere or surface bidirectional) reflectance
measurements for VIS red and NIR bands, respectively.
For land targets the index ranges from values close to 0 for arid or barren areas to ~ 1 for densely vegetated areas. Negative values of NDVI usually correspond to urban areas. The NDVI over water surfaces is very close to -1 due to their very low reflectance in the NIR band.
Figure 4.1. MODIS/Terra Vegetation NDVI 16-Day L3 Global 250m SIN Grid.
(extracted from https://lpdaac.usgs.gov/lpdaac/products/modis_products_table/, accessed in 2010)
NDVI main advantages and disadvantages.
- As a simple transformation of spectral bands, NDVI is easily computed without assumptions regarding land cover classes, soil type or climatic conditions
- Long time series (more than 20 years) available.
- Inherent nonlinearity because it is a ratio based index
- Additive noise effects
- Asymptotic (saturated) signals over high biomass conditions
- Very sensible to canopy background brightness (Huete, 2002)
- Is not a structural property of a land surface areas
4.2 - EVI
The Enhanced Vegetation Index was developed by the MODIS Science Team to take full advantages of the sensor capabilities. In order to increase thesensitivity to the vegetation signal, the index makes use of measurements in the red and near infrared bands (as in the case od NDVI), and also in the visible blue band, which allows for an extra correction of aerosol scattering. EVI also performs better than NDVI over high biomass areas, since it does not saturate as easily.
Where ρ are atmospherically corrected or partially corrected (Rayleigh and ozone absorption) reflectances, L is the canopy background adjustment,C1 and C2 are coefficients related to aerosol correction and G is a gain factor. The blue band is used to remove residual atmosphere contamination caused by smoke and sub-pixel thin clouds
Figure 4.2.MODIS/Terra Vegetation EVI 16-Day L3 Global 250m SIN Grid.
(extracted from https://lpdaac.usgs.gov/lpdaac/products/modis_products_table/, accessed in 2010)
EVI main advantages and disadvantages.
- Found to perform well under high aerosol loads , biomass burning conditions (Huete et al., 2002)
- Inherent nonlinearity because it is a ratio based index
- To compensate for the effects of NDVI saturation over high biomass areas, EVI tends to present relatively low values in all biomes and also lower ranges over semiarid sites.
- The correction for aerosol impact on the final index makes use of reflectances measurements within VIS blue, not always available (the case of AVHRR or SEVIRI sensors).
- Is not a structural property of a land surface areas
4.3 - EVI vs NDVI performance and common drawbacks
Figure 4.3.NDVI vs EVI performance
A problem common to all VI’s is their empirical nature. These two figures show the steep gradient in vegetation cover over South America, including: the hyperarid Atacama desert (no vegetation), the semi-arid and sub-humid portions of the Brazilian Cerrado (savannah biome), and the humid and perhumid portions of the Amazon tropical rainforest.
4.4 - Other vegetation Indices.
Several other vegetation indices have been developed for different sensors and with different purposes:
- Perpendicular Vegetation
- Soil Brightness
- Green Vegetation
- Vegetation Condition
this index is derived from NDVI, NDVImin and NDVImax are NDVI multiyear absolute maximum, and minimum, respectively.
VIs are widely used to assess how environmental changes affect the distribution and dynamics of vegetation, particullarly at large temporal and spatial scales, and/or in areas of limited in situ data. In this chapter we will present to examples of NDVI use to assess environmental consequences of DROUGHT:
- Vegetation stress
- Crop monitoring
Drought and vegetation stress in Portugal 2004
The strong dependence of vegetation dynamics on water availability has been for long recognized in semi-arid regions (Gouveia et al. 2009). Under such conditions VI's can be used to identify areas prone to drought. However VI's must be used carefully. For example low values of NDVI indicate bare/low vegetated areas, but NDVI by itself cannot be considered a drought index since the concept of drought implies an extreme condition and a deviation from normal status. In terms of vegetation dynamics, a drought is an event that hampers the normal vegetation growth. In order to capture this effect, and to compare drought conditions in areas with different land cover, NDVI information has to be complemented with climatic information.
Next figure presents monthly NDVI anomalies for September 2004 - to - August 2005, defined as departures from the respective monthly medians. The latter statistics were obtained from NDVI time-series encompassing the September 1998 - to - July 2006 period, and derived from VEGETATION instrument onboard SPOT 4 and SPOT 5 (Gouveia, C. et al. 2009).
Please hit the play button bellow to see the animation.
Figure 4.4. NDVI monthly anomalies between September 2004 and August 2005 derived from images acquired by the VEGETATION instrument onboard SPOT 4 and SPOT 5 (Gouveia et al. 2009).(*) - During the summer season of 2005, Portugal was also hit by large wild fires. You can learn further details about this Fire event at the Forest Fires EUMETRAIN CAL Module
Drought and Crop production decrease in Kenya 2009
The amount of precipitation of Kenyas 2009 "long rains" season was well below the climatological average and also under the 2008 "long rains", which was also a dry year. The figure shows the MODIS NDVI 16-day composite for the period between Sep 14 and Sep 29 in 2007 (left panel) and in 2009 (right panel). The figure shows the MODIS NDVI 16-day composite from Sept 14 until Sept 29 for 2007 on the left and for 2009, on the right.
Figure 4.5. NDVI 16-day composite for Sep 14 - Sep 29 2007(left) and the same period in 2009 (right) derived from MODIS 250m spatial resolution.
The 2009 NDVI shows a dramatic decrease on of vegetation dynamics when compared with 2007. Because NDVI correlates well with relative grain yields for most agro-climates in Kenya (USDA FAS - Commodity Intelligency Report, Dec 2009), the comparison of 2007 and 2009 NDVI suggests that the severe 2009 drought will reduce the Kenya's 2009 "long rain" corn yields. In fact USDA's Foreign Agricultural Service forecasted for Kenyas 2009/10 corn production 1.8 million tons, down 0.3 million tons compared to the previous year already poor crop and considerably less than the 5-year average of 2.6 million tons. We will return to this subject latter on this module section 5.9.3. Other perspective of Kenya’s drought can be found in http://oiswww.eumetsat.org/iotm/20091001_drought/ in which the authors present the day microphysics RGB monthly averages as an indicator of cloud and rainfall amounts.
5 - Vegetation Products
Satellite observations are a valuable tool for the monitoring of dynamic processes occurring at the earth surface. The changes in the structure and the functioning of vegetation are part of such processes, that can be detected through remote sensing. In particular, FVC (Fractional Vegetation Cover), LAI (Leaf Area Index) and fAPAR (Fraction of Absorbed Photosynthetically Active Radiation) are key variables for a wide range of land biosphere applications ( Product User Manual (PUM) of LSA SAF Vegetation Products ).
In this chapter you will learn:
- how this variables can be retrieved from satellite observations;
- the main applications of this vegetation products;
- the main characteristics of FVC, LAI and fAPAR estimated on the aim of the LSA SAF project;
- the advantages of using geostationary satellite derived products in comparison to polar orbiters;
5.1 - Definitions
Although the names of these variables might be self-descriptive, we should pay attention to their physical definitions:
5.1.1 - Fraction of Vegetation Cover (FVC)
The FVC accounts for the amount of vegetation distributed on a flat background. For a pixel such as that on next figure (hit the play button bellow to see the animation)
FVC would tell us the fraction of the total pixel that is covered by the trees.
FVC is relevant for agriculture and forestry, environmental management and land use, hydrology, natural hazards monitoring and management, vegetation-soil dynamics monitoring, drought conditions and fire scar extent. (Fraction of Vegetation Cover Product User Manual (PUM)).
5.1.2 - Leaf Area Index (LAI)
LAI [m2/m2] is geometrically defined as the total one-sided area of photosynthetic tissue per unit of ground surface area. It represents the amount of leaf material in ecosystems and controls the links between biosphere and atmosphere through various processes such as photosynthesis, respiration, transpiration and rain interception.
Monitoring the distribution and changes of LAI is important
for assessing growth and vigour of vegetation on the planet. It is
fundamentally important as a parameter in land-surface processes and
parameterizations in climate models.
LAI is a key parameter in Numerical Weather Prediction (NWP) models, regional and global climate modelling, weather forecasting and global change monitoring. Monitoring the distribution and changes of LAI is important for assessing growth and vigour of vegetation on the planet. (Leaf Area Index Product User Manual (PUM)).
FVC and LAI are important structural properties of land surface areas occupied by plant canopies, which yield complementary information to describe the three-dimensional structure of the vegetation attributes.
5.1.3 - Fraction of photosynthetically active radiation (fAPAR)
FAPAR represents the fraction of incoming solar radiation in the photosynthetically active radiation (PAR) spectral range (0.4 - 0.7 μm) that is absorbed by the green parts of the canopy.
fAPAR has been recognized as one of the fundamental terrestrial state variables in the context of the global change sciences (Steering Committee for GCOS, 2003; Gobron et al., 2006). It is a key variable in models of vegetation primary productivity and, more generally, in carbon cycle models implementing up-to-date land surfaces process schemes (e.g., Sellers et al., 1997). Besides, it is an indicator of the health of vegetation (Fraction of Absorbed Photosynthetic Active Radiation Product User Manual (PUM)).
Seasonal variations in LAI and fAPAR are vital to determine landscape water, energy and carbon balances, as well as in the detection of long-term climate change (Potter et al, 1993, Churkina and Running 1998, cited by Huemmrich et al, 2005).
VEGETATION PRODUCTS FROM OTHER SATELLITE PROGRAMS:
The following programs have also implemented algorithms in their operational lines to provide advanced biophysical products:
- POLDER (Leroy et al. 1997, Roujean and Lacaze 2002);
- MODIS and MISR (Knyazikhin et al. 1999);
- MERIS (Gobron et al. 1999, Bacour et al. 2006, Baret et al. 2007);
- SEAWIFS (Gobron et al. 2001);
- VEGETATION (Baret et al. 2007, Bartholomé et al. 2006);
- GLOBCARBON (Plummer et al. 2006, Deng et al. 2006)
5.2 Spatial Coverage of the LSA SAF Vegetation Products
The Land SAF vegetation products are estimated at the full spatial resolution of the MSG/SEVIRI instrument over 4 geographical areas, covering the full MSG disk: (hit the play button to see the areas).
5.3 Temporal Resolution & Data Availability
The LSA SAF vegetation products are produced on a daily (and 10-day) basis. To access LSA SAF products a simple registration is required on the following address:http://landsaf.meteo.pt/
The data files can be obtained by different ways:
- In Offline mode:
- Direct download from the webpage;
- In Near Real Time:
- ftp transfer;
- via EUMETCAST
5.4 Product Content
The FVC, LAI and FAPAR products contain 3 datasets each, comprising the following fields:
- a vegetation field;
- an error estimate field;
- a quality control information field.
The data is coded in HDF5 format. The HDF5 files in Land SAF system have the following structure:
- A common set of attributes for all kind of data, containing general information about the data;
- A dataset for the parameter values;
- Additional datasets for metadata (e.g., quality flags, error field).
The retrieval of FVC, LAI and fAPAR in the LSA SAF rely on Spectral Mixture Analysis (SMA); the signal from a single pixel (surface reflectance) is assumed to be the contribution of different vegetation and bare soil components within the scene (next figure).
As detailed in Chapter 2, the surface BRDF may be written as:
ρ (θs, θv, φ) = K0 + K1 x F1(θs, θv, φ) + K2 x F2(θs, θv, φ)
K0 - represents isotropic reflectance;
K1 - represents geometric reflectance;
K2 - represents volumetric reflectance;
Parameters K0 in the SEVIRI channels VIS 0.6 μm, VIS 0.8 μm and NIR 1.6 μm are used to determine FVC and LAI, while fAPAR uses K0, K1 and K2.
5.5.1 - LSA SAF FVC
The estimation of FVC is based on the analysis of K0 for
VIS 0.6 μm, VIS 0.8 μm and NIR 1.6 μm channels.
The K0 component of surface reflectance is the
less affected by shadows and, therefore is able to provide a physically correct estimation of FVC.
Taking into account the spectral signature of pure types of vegetation (V1,V2,V3) and soil
(S1,S2,S3), the SMA process allows to estimate the fraction of the vegetation types present in the pixel.
Next figure shows an example, considering a 2 dimensional space [0.8 μm K0 and 0.6 μm K0 ]
Therefore FVC is estimated by the following expression:
Pure vegetation types consider species prevalent in:
- V1 - Crops
- V2 - Herbaceous ecosystems
- V3 - Forest
Soil types include:
- S1 - Bare soil
- S2 - Rock
- S3 - Human-built surfaces
Next figure shows an example of FVC over the Full SEVIRI disk (figure on the left), and the corresponding Error (figure on the right) for the 15th April 2007.
The FVC image adequately captures the spatial patterns of vegetation cover at continental level. Large spatial variation gradients are observed over Africa. This figure also highlights the spatial homogeneity and the absence of large gaps in the FVC field.
5.5.2 - LSA SAF LAI
LAI in the LSA SAF is retrieved using a semi-empirical method proposed by Roujean and Lacaze (2002), in which LAI is related to FVC trough the following expression:
- a0 - is a coefficient within the range 1.04 to 1.07;
- b = 0.945;
- G = 0.5;
- Ω is the so-called clumping index (Nilson, 1971)
Both coefficients a0 and Ω depend on a Land Cover Classification Map (GLC2000) GLC2000_EUR.pdf
The next figure shows an example of LAI over the Full SEVIRI disk (figure on the left), and the corresponding Error (figure on the right) for the 15th April 2007.
Large spatial gradients of LAI are also found over Africa:
- LAI ~ 0 on the Sahara desert;
- LAI gradually increases in the Sahel from areas with sparse vegetation to woody Savannas, reaching the highest values in the equatorial forests;
- In the southern hemisphere, LAI decreases again from tropical forest through the woody savannas to the desert of Namibia;
- The higher values within the Meteosat disk (LAI= 6.5) are found in Amazonian forest.
5.5.3 - LSA SAF fAPAR
The fAPAR is estimated from a NDVI-like vegetation index, the RDVI (Renormalized Difference Vegetation Index).
As you have seen, in chapter 4, NDVI is given by the normalized difference between reflectances in channels 0.6 μm and 0.8 μm. RDVI, is a very similar parameter given by:
RDVI has a behaviour close to that of NDVI for large LAI, but tends to be more sensitive to changes in vegetation coverage under low LAI conditions. Since the view-illumination geometry may have a large impact on reflectance observations, RDVI is computed using 0.6 µm and 0.8 µm surface reflectances corrected to an optimal geometry, which maximizes the correlation with fAPAR. Such corrected reflectance values are estimated, for each channel, from K0, K1 and K2 parameters (Roujean et al. 1992):
corresponding to solar and view zenith angles of 45ş and 60ş, respectively. fAPAR estimation explores its linear relationship to RDVI (obtained for the optimal geometry):
Next figure shows an example of fAPAR over the Full SEVIRI disk (figure on the left), and the corresponding Error (figure on the right) for the 15th April 2007.
5.6 Overall quality of LSA SAF Vegetation Products
The overall quality of the LSA SAF vegetation products depends on the pixels location:
- regions with large view zenith angles (e.g., North of Europe, South America) are expected to have larger errors (next Figure).
This is mainly the case of Europe during wintertime as a combination of multiple effects, such as:
- low illumination and high observation angles;
- higher anisotropy;
- higher cloud occurrence;
- larger shadows or traces of snow cover;
- the African continent and the Mediterranean Basin are consolidated areas with optimal geometry of acquisition.
The spatial distribution of the mean FVC uncertainty along the year 2007 is depicted in next Figure:
The fraction of valid pixels (i.e. processed with reliability) for FVC over SAfr and NAfr zones is nearly 100% through the whole year. This is partly because products benefit from the high temporal sampling of the SEVIRI data, guaranteeing a high rate of cloud-free radiances per pixel.
The Europe and SAme continental zones present generally a decreased accuracy.
5.7 Scientific validation of LSA SAF products
Validation techniques are required to provide the confidence intervals that are mandatory for the users in a number of applications. LSA SAF products are validated through (LSA SAF VEGA Validation Report ):
- 1) Intercomparison with vegetation products derived from other satellite sensors (e.g., MERIS on ENVISAT, MODIS on EOS and VEGETATION on SPOT)
- 2) Comparison with in situ measurements
1) Intercomparison with vegetation products derived from other satellite sensors
Next figures show composites of actual products for 2 distinctive months, July and November. As observed there is a high spatial consistency between LAI from SEVIRI/MSG, MERIS/ENVISAT and MODIS/TERRA. The latter overestimates LAI at equatorial Africa, when compared to those obtained from the remaining sensors. SEVIRI/MSG LAI presents less gaps in vegetated areas.
2) Comparison with in situ measurements
Next Figure shows the comparison between FVC, LAI and fAPAR products estimated from MSG measurements with in-situ based maps obtained within the framework of the VALERI (Validation of Land European Remote sensing Instruments) project.
As shown in the figures above, LSA SAF products compare well with globally distributed VALERI* in-situ observations.
Further results on scientific validation of the LSA SAF vegetation products can be found on the LSA SAF VEGA Validation Report Document.
5.8 Comparison of LSA SAF to Polar-Orbiter Derived Products
Since LSA SAF Vegetation Products are derived from geostationary satellite observations (MSG), they benifit from a high temporal sampling and therefore from a relatively high rate of cloud-free radiances per pixel. As a consequence these products have important advantages when compared with similar parameters derived from polar-orbiters, namely:
- Spatial Continuity;
- Temporal Continuity;
- Better monitor the vegetation dynamics;
- Robustness against double-seasons false alarms;
- Extracting phenological parameters;
- Spatial Continuity - As seen in 5.6 - Overall quality of LSA SAF Vegetation Products there are almost no gaps in Africa; On the other hand, if we look at the percentage of gaps of polar-orbiter derived products over Africa, we can conclude that the occurrence of missing data is highly correlated with the cloud coverage, as demonstrated by next two figures:
- First figure(*) presents the spatial distribution of the percentage of gaps of LAI derived from MODIS over Africa:
- the higher % gaps in South Africa occurs in nov-dec & jan-feb, which coincides with the position of the The Intertropical Convergence Zone (ITCZ);
- during July-Aug the ITCZ moves Northwards, wich results in an increase of clouds, and therefore in the percentage of gaps in North Africa region.
- Next figure(**) shows equivalent results but for LAI-VEGETATION (VGT4Africa) product:
- Temporal Continuity - In contrast to polar-orbiter derived products, LSA SAF vegetation products provide continuously-updated information about the vegetation, enabling to:
- better monitor the vegetation dynamics;
- For some regions with persistent cloud coverage as in western and Central Africa, polar orbiter estimates are missing for long periods, particularly during the growing season, limiting the ability to capture underlying modes of seasonality.
- LAI from MSG (LSA SAF) is the product showing less temporal gaps;
- LAI from MSG (LSA SAF) follows the seasonality of the vegetation activity during 2006 and 2007;
- Robustness against double-seasons false alarms:
- Extracting phenological parameters;
- the amplitude;
- the maximum and minimum values;
- the timing of phenological stages (onset of greenness, maximum development, senescence);
- the growing season length.
- determining the length of the growing period;
- an early assessment of the crop production.
(*) and (**): Images on the left are taken from LSA SAF Biophysical Products FVC, LAI, FAPAR. F. Camacho, J.García-Haro , A. Verger, J. Meliá: Workshop on product availability for users in Africa, EUMETSAT, 27-28 August 2009. The images on the right have been adapted from http://en.wikipedia.org/wiki/File:ITCZ_january-july.png , accessed in 2010
Next figure (source: LSA SAF VEGA Validation Report ) shows the temporal evolution of LAI over 2006 and 2007 for the Dahra (Senegal), estimated from MSG , MODIS, MERIS and measured in-situ:
From this figure:
In some regions a double season may occur (two vegetation peaks per year) and there is no a priory method to distinguish a real double season from a fake second season. Next Figure (please hit the play button bellow) illustrates the occurrence of repeated false alarms of double seasonality on time series of the FVC product from VEGETATION (VGT-JRC), after a period of frequent cloud occurrence:
One important feature of the LSA SAF vegetation products is the ability to generate spatially coherent images of phenological parameters, such as:
In particular, the date of the start of growing season (SOS) is a critical parameter f or food security monitoring. For deciduous plants the SOS is the time when new leaves start their development. Determining the SOS accurately using satellite remote sensing is essential to:
The overall pattern through the year observed here appears to be in agreement with the climatic
patterns in this region.
Similar outcomes for SOS based on NOAA rainfall estimate were obtained during year 2007 in this same region (Brown and de Beurs, 2008).
5.9 - Further Applications of LSA SAF Vegetation Products
In this section we will show how vegetation products can be used for a wide range of land biosphere applications that are directly related to vegetation properties and health.
5.9.1 - Estimation of Vegetation productivity
Vegetation productivity along the Kalahari transect:
The Kalahari Transect (KT), is one of the
International Geosphere-Biosphere Programme (IGBP)
(accessed in 2010) proposed Transects.
It spans a strong climatic gradient in southern Africa, from the arid south to the humid north, while remaining on a single broad soil type, the deep sands of the Kalahari basin.The vegetation ranges over the length of the transect from shrubland through savannas and woodlands to closed evergreen tropical forest, with land uses ranging from migratory wildlife systems, through pastoralism, subsistence cropping to forestry. (Information from: http://medias.obs-mip.fr/www/Reseau/Lettre/10/en/Dossiers/pa2021/pa2021.htm, accessed in 2010).
The aim to define this transect was to:
- integrate ongoing and future biophysical and relevant socio-economic work in the Kalahari eco-region;
- to enhance understanding of the entire system within the context of human and climatically induced environmental change.
5.9.2 - Dry Matter Productivity
Dry Matter Productivity (DMP) is the increase in dry matter biomass. Its estimation is useful for:
- Crop monitoring;
- Yield forecasting;
- Carbon sequestration / Net Primary Productivity (NPP)
The Joint Research Center (JRC) is computing this parameter following the approach proposed by Monteith (1972) (LSA SAF 3rd User Training Workshop ), in which DMP is a function of the:
- Incoming solar radiation (0.2-3.0 μm);
- Surface Temperature;
Next figure shows the comparision between DMP as estimated by JRC, based on fAPAR from SPOT and MSG , respectively, (Example for 2008, February, Dekad 3 over Africa). As observed there is a very good spatial consistency between the two estimated fields.
A North-South transect through Africa along the 20° Meridian (Lybia to Cape Town) also highlights the good correspondence between SPOT and MSG:
5.9.3 - Drought Monitoring
SEVERE DROUGHT OVER EAST AFRICA
In 2009, the long rainy season over East Africa did not start until late March and produced insubstantial rain, resulting in severe drought. As a consequence, at the end of the first rainy season (March to June) of 2009, satellite observations revealed that plant growth - both crops and natural vegetation - across Kenya was significantly lower than normal (earthobservatory.nasa.gov). This is quite evident when we compare a LSA SAF LAI image from September 2009 with the same day from 2007 (please hit the play button to see the animation):
You might have noticed that this case was already presented in chapter 4.5, where the production decrease in Kenya, as a consequence of the severe drought in 2009 was shown with a MODIS NDVI composite of 16 days. Here, we can see the same effect just by comparing 2 daily images, since due to the temporal resolution of MSG satellite, LSA SAF products show very high spatial and temporal continuity.
This severe drought affected 10 million people, a third of the population. As a consequence the government declared a state of national emergency.( Reuters)
6 - References
ASTER - Spectral Library 1998: ASTER Spectral Library - Version 1.2 California Institute of Technology
Brown, M.E. and K.M. de Beurs 2008: Evaluation of multi-sensor semi-arid crop season parameters based on NDVI and rainfall. Remote Sensing of Environment, 112 (2008), 2261–2271.3
Curran, P. J. 1989: Remote sensing of foliar chemestry. Remote Sensing of Environment, 30, 271-278.
Foreign Agricultural Service (FAS) 2009: Kenya's Grain Basket Experience Drought and Lowers "Long Rains" Corn Output Commodity Inteligence Report (December 2009) - United States Department of Agriculture
Filella, I. and J. Panuelas 1994: The red edge position and shape as indicators of plant chlorophyll content, biomass and hydric status. International Journal of Remote Sensing, 15, 1459 -1470
Gouveia C., R. M. Trigo and C.C. DaCamara 2009: 2009 Drought and vegetation stress monitoring in Portugal using satellite data. Natural Hazards and Earth System Sciences, 9, 185-195
Horler, D. N. H., M. Dockray, J. Barrer 1986: The red edge of a plant leaf reflectance. International Journal of Remote Sensing.4, 273-288
Huete, A.,K. Didan, E. P. Rodriguez X. Gao and L. G. Ferreira 2002: Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83, 195-213
Jacquemoud, S. and F. Baret 1990: PROSPECT: A Model of Leaf Optical Properties Spectra. Remote Sensing of Environment 34:75-91
Kyllo, K. P. 2003: NASA funded research on agricultural remote sensing, Department of Space Studies, University of North Dakota.
LSASAF 2008: Product User Manual (PUM) Vegetation Parameters( FVC, LAI, FAPAR )SAF/LAND/UV/PUM_VEGA/2.1 (February 2008)
LSASAF 2008: Validation Report (VR) Vegetation Parameters( FVC, LAI, FAPAR )SAF/LAND/UV/VR_VEGA/2.1 (January 2008)
MARS Bulletin 2009: Crop Monitoring in Kenya June 2009. Joint Research Center of the European Commission (June 2009)
Pettorelli, N., J. O. Vik, A. Mysterud, J. Gaillard,C. J. Tucker and N. C. Stenseth 2005: Using the satellite-derived NDVI to assess ecological responses to environmental change Trends in Ecology & Evolution vol.20, 9, 503-510
Peixoto, J. P. and A. H. Oort 1992: Physics of Climate American Institute of Physics
Roujean, J-L., M. Leroy and P.-Y. Deschamps, 1992: A bidirectional reflectance model of the Earth's surface for the correction of remote sensing data J. Geophys. Res. , 97(D18), 20455-20468.
Trewartha, Glenn T. and Lyle H. HornAn Introduction to Climate McGraw-Hill Internacional