Forests play a key role in the global carbon cycle. This is because carbon is exchanged naturally between forests and the atmosphere through photosynthesis, respiration, decomposition and combustion . Forest biomass represents the largest terrestrial carbon sink that accounts for approximately 90% of all living terrestrial biomass . Brown defined forest biomass as the total amount of aboveground living organic matter expressed as oven-dry tons per unit area. When forests are cleared or degraded, their stored carbon is released into the atmosphere as carbon dioxide (CO2), acting as a source of greenhouse gases. With an increasing concern about the rising CO2 concentrations in the atmosphere, the role of forest as a carbon sink to mitigate global warming has been realized. As a global warming mitigation effort, the amount of carbon stored in the forest biomass has gained considerable attention under international initiatives such as the Kyoto Protocol . Under this agreement, the signatory countries need to estimate and report the amount carbon dioxide emitted and stored by the forest ecosystems in their countries.
In recent years, mapping and monitoring carbon stocks in forest has gained considerable attention. Reliable estimation of the above ground carbon in forest is essential when studying the role of forest in mitigating global warming. It assumes that 50% of the biomass is made up by carbon . Above ground biomass consists of all living biomass above the soil including stem, stump, branches, bark, seeds and foliage . Apart from estimating biomass for carbon stocks monitoring, it has also been a usual practice to quantify fuel and wood stock for forest management purposes . Different approaches, based on field measurements, remote sensing and GIS, have been applied for above ground biomass estimation . The traditional techniques based on field measurement are often time consuming, labour intensive, and difficult to implement, especially in remote areas. Furthermore, also, they cannot efficiently provide timely information about the spatial distribution of biomass over large areas . The advantage of remotely sensed data, such as repetitive data collection, a synoptic view, a digital format that allows fast processing of large quantities of data, and the high correlations between spectral bands and vegetation parameters, have made it the primary source for large area, especially in areas where access is difficult. Remote sensing based biomass estimation therefore has increasingly attracted scientific interest .
Remote sensing has been shown to be a simple, low cost and less time consuming approach that can be used to estimate biomass over large areas . Some remote sensing methods use spectral indices as a way to estimate above ground biomass at landscape level. In particular, vegetation indices (VIs) and band ratio based models are most commonly used to produce estimates of biomass . A variety of vegetation indices based on the spectral features of green vegetation have been developed using broad-band remotely sensed data .
A vegetation Index (VI) is defined as a dimensionless, radiometric-based measurement computed from some spectral combination of remotely sensed data . A number of vegetation indices reported in literature are computed using data from the red and infrared part of the electromagnetic spectrum . The spectral reflectance of vegetation is characterized by very low reflectance in the red part of the spectrum followed by an abrupt increase in the reflectance at near-infrared (NIR) part (see Figure1). Simple ratio (SR) and the normalized difference vegetation index (NDVI) are the most popular vegetation indices which are using some combination of red and near-infrared (NIR) reflectances .
The first true vegetation index was the simple ratio (SR), which is simply the ratio of red and near infrared reflectances. This provides valuable information about vegetation biomass . The SR index is especially sensitive to biomass variations in high biomass vegetation such as forests (Huete et al., 2002). Many vegetation indices and band ratio have been used in the relation with biomass since past 30 years . The vegetation indices used in this study are i) Normalised Difference Vegetation Index (NDVI), ii) Re-normalised Difference Vegetation Index (RDVI) iii) Difference Vegetation Index (DVI) iv) Simple Ration (SR) and iv) Modified Simple Ration (MSR). These indices are based on the fact that the chlorophyll II pigments in vegetation canopies absorb red light and reflect infrared light and also the low reflectance value of red and high reflectance value of NIR part of wavelength. The ratio between infrared and red radiation is a sensitive indicator of green biomass . By exploiting vegetation indices, forest biomass can potentially be estimated and mapped . Not all vegetation indices are significantly correlated with AGB . Many studies have investigated the ability of remotely sensed data to estimate the biomass of forests .
Figure 1 Typical spectral reflectance curve for vegetation (Jensen, 2000)
The NDVI, which is the difference of the near-infrared (NIR) and red bands divided by their sum, has been the most widely used index in global vegetation studies . Jensen mentioned Rouse et al., (1974) developed the generic Normalized Difference Vegetation Index (NDVI) :
(1)
The "ratio" properties enable NDVI to cancel out a large proportion of the noise caused by changing sun angles, topography, clouds or shadow, and atmospheric conditions . NDVI value for a given pixel ranges from minus one (-1) to plus one (+1) .
The saturation problem is one of the major limitations of using vegetation indices for biomass estimation. The VIs , particularly NDVI, asymptotically approach a saturation level after a certain leaf area index (LAI) 7.0 is reached . The red part of wavelength responds to cholophyll and leaf pigments and NIR part responds to spongy leaf mesophyll (see Figure 2). There is a tendency of red part to saturate at a lower level than NIR part, as a result a lower "spectral asymptote" for red . Mutanga and Skidmore explained this problem very well. They explained that when the canopy cover increases to 100 %, the NIR reflection also increase due to increase in leaf number but the red reflection decrease slightly as the amount of red light absorbed by the leaves reaches a peak. This imbalance change of reflectance between NI and red contributes to a change in NDVI value, results poor estimation of biomass.
Figure 2 Diagrammatic cross section of a leaf, showing interaction with incident energy (modified from (Sabins, 1997)).
Soil background conditions exert considerable influence on partial canopy spectra and the calculated vegetation indices. Many studies reported the influence of soil brightness on vegetation indices. Higher value of the ratio vegetation index or the normalized difference vegetation index has been found as an effect of darker soil background for a given amount of vegetation (Elvidge & Lyon, 1985; Huete, et al., 1985).
The topographic effect is another very important factor when the indices are used for biomass estimation in areas of undulating terrain with steep slope . In remote sensing, the topographic effect is defined as the variation in radiance from inclined surfaces compared to radiance from a horizontal surface as a function of the orientation of the surface relative to the light source and sensor position .The definition of topographic effect implies that the spectral response of target (here forest canopy) varies on the position of light source (sun), target and sensor. Due to topographic effect the irradiance of same coverage in inclined plane also differs from horizontal plane. Laungcai & Zingeheng explained that in a mountainous area the total irradiance received by a terrain element can be shown as -
Eλ = Eλs + EλD +, EλR (2)
where Eλs = direct solar irradiance,
EλD= diffused solar irradiance,
EλR=reflected irradiance from adjacent slopes.
The following diagram is showing the geometry of irradiance in a mountainous area-
Figure 3 Components of irradiance received by terrain element in a mountainous area (Adopted from (Laungcai & Zingeheng, 1992)
In mountainous area total irradiance is the sum of direct solar irradiance, diffuse solar irradiance and reflected irradiance from adjacent slopes. Direct solar irradiance depends on sun azimuth and sun elevation angle and reflected irradiance from adjacent slopes depends on the slope position and slope steepness of the terrain. Therefore, in the mountainous or hilly area topographic effect (terrain slope and aspect) need to consider sun, target and sensor geometry (position) (see Figure 4). Holben & Justice measured the topographic effect on remotely sensed data and showed that the effect was most extreme at low solar elevations and greatest for slopes in the principal plane of the sun. One of the earliest attempts to simulate and quantify this topographic effect involved the measurement of differential spectral radiance from a uniform sand surface under various slope and angle combinations (Holben & Justice, 1980). It was found that the magnitude of the topographic effect varied as a function of the solar inclination, azimuth and the terrain's degree of slope and aspect. This effect produces wide variations in radiance. Sadowski and Malila demonstrated that reflectance varies as a function of slope and aspect. Therefore, in estimating biomass in mountainous forest from spectral reflectance particularly from vegetation indices we can assume that there must be some effect of slope and aspect. The topographic effect could be eliminated or weakened when vegetation indices are expressed as band ratios, such as in the NDVI, RVI, etc. . Though some studies mentioned that rationing spectral bands can reduce the topographic effect, there is little quantitative evidence that rationing eliminate the topographic effect. In the literature the rationale behind the use of ratios is hardly discussed . If the effect is similar in visible and NIR part of the spectrum the band ratio of VIs could cancel out the effect.
Figure 4 Sun-Target-Sensor geometry, adopted from (Grodecki & Dial, 2001)
Problem statement
Accurate estimation of biomass is required for carbon stock accounting and monitoring . The most persistent uncertainties concerning global Carbon budgets are the general lack of accurate spatial forest biomass data . Therefore, a high degree of uncertainty prevails in the estimation of the carbon sink. . This uncertainty can be reduced by accurate estimation of biomass.
The possibility of estimating biomass by satellite remote sensing at various spatial scales and environments have been investigated by several studies .Most of the earlier research on above ground biomass (AGB) estimation was conducted for coniferous species since coniferous forests have simple stand structures and species compositions. In the case of tropical forest or mixed natural forest, AGB estimation is a problematic task due to the complex structure and diversified species composition .
The use of remote sensing to estimate tree variables is based on the relationship between remote sensing data and tree parameters data measured in the field. Vegetation indices are often used as remote sensing approach to estimate forest parameters. Generally, vegetation indices are used to enhance the spectral contribution which is minimized by the background of vegetation . Several factors like canopy closure, understory vegetation and background reflectance effects this relationship . Canopy geometry, soil background, sun view angles, and atmospheric condition effect the spectral response of vegetation which is removed by the VIs .To develop a model for estimating above ground biomass from remote sensing data requires good understanding of the relationship between remote sensing data and forest stand parameters . It is difficult to select a suitable wave length or combination of wave length to extract information about specific biophysical parameter of a given study location . The relationship between vegetation indices and above ground biomass is not a settled issue yet. A number of studies reported significant relationship between VIs and biomass . Heiskanen found NDVI had the highest R2 value (0.85) for biomass while estimating above ground tree biomass in a mountain birch forest using ASTER satellite data Hall, et al. (2006) found strong relationship between red and near-infrared bands with above ground biomass. However, some results have shown poor relationships . Foody found a weak relationship between NDVI and biomass in tropical forests. Investigations by Huete, et al., revealed that NDVI was spectrally saturated over forested areas and was sensitive to canopy background. However, Matsushita, et al., demonstrated that the topographic effect is an important factor, especially when the indices (Enhanced Vegetation Index (EVI) and NDVI) are used in areas of undulating terrain. Topography plays an important role influencing vegetation indices . Hence, estimating biomass from spectral reflectance particularly from vegetation indices is likely to be affected by slope and aspect. Since a few studies carried out to estimate above ground tree biomass in mountainous mixed forest, this study will contribute to scientific research in biomass estimation.
We need to find out which vegetation indices are more sensitive to the topographic effect particularly slope, compared to the other vegetation indices while estimating above ground biomass from VIs in a mountainous forest. The innovative part of this research is that this will help to understand the relationship between different vegetation indices with above ground tree biomass and in addition whether slopes affect the relationship between vegetation indices and biomass estimation. This can help to formulate a new idea of slope adjusted vegetation indices.
Research Objectives and Research Questions
To address the research problem outlined in the previous section, this research has the following two objectives:
1. To determine if and how slope effects the estimation of above ground tree biomass from vegetation indices.
2. To determine the most suitable vegetation index for estimating aboveground tree biomass of mixed natural mountain forest in Belasitsa Mountain forest, Bulgaria.
To achieve these two objectives, two questions need to be answered:
1. What is the effect of slope on the accuracy of above ground tree biomass estimation from vegetation index?
2. Which vegetation index relates the best to above ground tree biomass?
Hypothesis
For research question 1:
Steeper slopes significantly reduce the accuracy of biomass estimation with vegetation indices (NDVI, RDVI, DVI, SR and MSR)
For research question 2:
Compared to RDVI, DVI, SR and MSR, NDVI is the best vegetation index related to above ground tree biomass.
Research Approach
In order to answer the research questions the integration of remote sensing data and above ground tree biomass estimation was a pre-requisite. A slope map (see Figure 6) was prepared from ASTER DEM to make stratification of sampling design. Field measurements were conducted to collect forest stand parameters such as diameter at breast height (DBH), tree height and tree species for estimating above ground tree biomass (AGTB). Slope, aspects and altitude data were also recorded in the plots. DBH and Tree Height were converted into above ground tree biomass per tree using allometric equations. Then above ground biomass at stand level was calculated using tree density values that were estimated using a method described in the method part. Vegetation indices were calculated from an ASTER satellite image by using appropriate formula in ArcGIS Desktop version 9.3.1 software. There are numerous approaches to estimate above ground biomass from satellite data . Regression analysis is the most common modelling approach . Simple linear regression analyses were used to find the best vegetation index to estimate above ground tree biomass and multiple linear regression analyses were used to explore the effect of slope, altitude and forest types in estimating above ground tree biomass from SVIs. To compare the accuracy of above ground biomass estimation in different slope classes the root mean square error (RMSE) were considered. In addition, in this research the forest of the study area was categorized based on percentage of species composition as i)Pure Chest nut forest ii)Pure Beech forest iii)Pure Oak forest and iv) Mixed forest for multiple regression model. The sampling plots having equalled or more than 75% of individual species was considered as pure forest. In the field data set there was two plantation plots of Pinus nigra which was discarded from the regression analysis as rest of plots were natural forest. Finally, the findings were integrated into the results followed by the discussions, conclusions and recommendations. The followed method is described in the Figure 5
Figure 5 Flow chart of research approach
