**3.2 Relationships between aboveground biomass and remote sensing indices**

In this study, we investigated the applicability of both optical and radar data as well as their combinations. To obtain the best regression model for estimating aboveground biomass, robust multiple regressions were conducted between field-measured aboveground biomass and various remote sensing-derived variables such as TM spectral reflectance, vegetation

Sparse woodland 57.3 34.0 9.1 – 94.5 8 Low-high shrub 11.1 11.4 2.0 – 47.2 19 Graminoids-dwarf shrub 2.3 2.2 0.4 – 8.3 16

Table 1. Mean value, standard deviation, and range of aboveground biomass measured at sites within and around the Porcupine caribou winter and summer range along the

type Mean Standard

Table 2. Mean value, standard deviation, and range of foliage biomass measured at sites within and around the Porcupine caribou winter and summer ranges along the Dempster Highway, Yukon during the summer of 2006, as well as at sites within and around the Porcupine caribou calving ground and summer range inside the Ivvavik National Park, Yukon during the summer of 2008. Also shown are statistics for the corresponding

Similarly, the foliage biomass values measured at sites within a specific dominant vegetation type also varied significantly (Table 2). For example, foliage biomass ranges from 95.3 to 198.4 g m-2 for low-high shrub sites, 37.9–92.3 g m-2 for mixed graminoids-dwarf shrub-herb sites, 63.1–106.1 g m-2 for coastal plain tussock sites, and 0.0–20.0 g m-2 for hilltop rock lichen sites. Consequently, assigning aboveground or foliage biomass value to a site according to its vegetation type can result in substantial error (Gould et al., 2003; Walker et

**3.2 Relationships between aboveground biomass and remote sensing indices** 

In this study, we investigated the applicability of both optical and radar data as well as their combinations. To obtain the best regression model for estimating aboveground biomass, robust multiple regressions were conducted between field-measured aboveground biomass and various remote sensing-derived variables such as TM spectral reflectance, vegetation

All types 16.4 25.6 0.4 – 94.5 43

Standard deviation (t ha-1)

Low-high shrub 135.4 36.3 95.3 – 198.4 6

shrub 65.3 16.3 37.9 – 92.3 9 Coastal tussock 87.9 22.2 63.1 – 106.1 3 Rock lichen 11.3 10.2 0.0 – 20.0 3 All types 80.8 47.2 0.0 – 198.4 21

Low-high shrub 7.02 7.58 0.36 – 18.16 6

shrub 3.11 2.64 0.32 – 8.55 9 Coastal tussock 0.77 0.53 0.21 – 1.27 3 Rock lichen 0.65 0.57 0.0 – 1.06 3 All types 3.54 4.83 0.0 – 18.16 21

Range (t ha-1)

deviation Range Number

Number of sites

of sites

Mean (t ha-1)

Dempster Highway, Yukon, during the summer of 2004.

Dominant vegetation

Graminoids-dwarf

Graminoids-dwarf

aboveground biomass measurements.

Dominant vegetation type

Foliage biomass (g m-2)

Abovegro und biomass (t h-1)

al., 2003).

indices, and JESR-1/SAR backscatter coefficients. We used 3 × 3 pixels (i.e., 90 m by 90 m) averaged value in place of single pixel value in order to reduce the effect of erroneous spectral features, e.g., features of adjacent pixels may have been assigned to some field plots of the data due to errors in image registration and the location of sample plots. The Landsat images were re-sampled to 100 m resolution for matching the resolution of the North America JERS summer mosaic.

For the sites along the Dempster Highway, we found that strong correlations exist between ln(*Ba*) and remote sensing signals (Table 3). When all types were mixed, the strongest correlation was found against the L-band JERS-1/SAR backscatter, followed by the Landat B4/B5 (Fig. 5). The Landsat bands 3 and 5 show strong negative relationships for the sparse woodlands and all types mixed, but not for shrub and graminoids lands.


Table 3. Correlation coefficient (*r*) and coefficient of determination (*r*2) between ln(*Ba*) and remote sensing indices for mixed graminoids-dwarf shrub-herb, low-high shrub, sparse woodlands, and all types for aboveground biomass measurements along the Dempster Highway in 2004. Remote sensing indices include Landsat red band reflectance (B3), near infrared band reflectance (B4), shortwave infrared band reflectance (B5), ratio of B4/B5, simple ratio (SR = B4/B3), normalized differential vegetation index (NDVI = (B4 - B3)/(B4 + B3)), shortwave vegetation index (SWVI = (B4 - B5)/(B4 + B5)), and L-band JERS-1/SAR backscatter coefficient (JERS).

Fig. 5. Scatter plots between ln(*Ba*) and Landsat B4/B5 as well as JERS-1/SAR backscatter coefficients.

Mapping Aboveground and Foliage Biomass Over the Porcupine Caribou

Calibration

estimates using a single data type alone.

1000

0.1

1

10

**Measured aboveground biomass (t ha-1)**

100

Habitat in Northern Yukon and Alaska Using Landsat and JERS-1/SAR Data 241

For equation (7), the *CF* = 1.35. The 1:1 comparison of estimated aboveground biomass using *CF*-corrected equation (7) and the measured values for the sites along the Dempster Highway is shown in Fig. 6, with the slope = 0.95, *r*2 = 0.72, and *SEE* = 13.5 t h-1 over measured aboveground biomass range from 0.4 to 94.5 t h-1. These results agreed well with the finding of Moghaddam et al. (2002) that the estimation accuracy of forest biomass was significantly improved when radar and optical data were used in combination, compared to

> 0.1 1 10 100 1000 **Estimated aboveground biomass (t ha-1)**

Fig. 6. A 1:1 comparison between estimated and measured aboveground biomass values for the calibration sites along the Dempster Highway transect measured during the summer of 2004. For clarity over the low biomass range, the results are shown on the log-log scale.

In this study, we further validated the relationship using measurements collected over the same study area at later dates. We used aboveground biomass measured at 21 sites along the Dempster Highway during the summer of 2006 in the Ivvavik National Park during the summer of 2008, where foliage biomass was the main measurement target, to validate the above relationship. Fig. 7 shows the validation result, with the slope = 0.91, *r*2 = 0.90, and

These validation results suggest that the aboveground biomass can be reliably estimated

*SEE* = 1.5 t h-1 over measured aboveground biomass range from 0 to 18.16 t h-1.

**3.3 Validation of aboveground biomass relationship with remote sensing indices**  In a previous study (Chen et al., 2009a), we validated equation (7) using aboveground biomass measurements at 33 sites around Yellowknife, Northwest Territories, and Lupin Gold Mine, Nunavut Territories. The 1:1 comparison of estimated aboveground biomass and the measurements for the Yellowknife and Lupin Gold Mine study area indicates that the equation holds up very well, with *r*2 = 0.81, slope =1.17, and *SEE* = 9.67 t h-1 over

measured aboveground biomass range from 0.9 to 103.3 t h-1.

using Landsat B4/B5 and JERS-1/SAR backscatter coefficient.

<sup>2</sup> *SEE* /2 *CF e* (8)

Many vegetation indices have been developed and applied for estimating aboveground biomass (Anderson & Hanson, 1992; Anderson et al., 1993; Mutanga & Skidmore, 2004; Lu 2005). In Table 3, we examined 4 vegetation indices: ratio of Landsat B4/B5, simple ratio (SR = B4/B3), normalized differential vegetation index (NDVI = (B4 - B3)/(B4 + B3)), shortwave vegetation index (SWVI = (B4 - B5)/(B4 + B5)). In general, vegetation indices can partially reduce the impacts on reflectance caused by environmental conditions and shadows, thus improve correlation between AGB and vegetation indices, especially in those sites with complex vegetation stand structures (Lu et al., 2004). However as shown in Table 3, not all vegetation indices are significantly correlated with aboveground biomass. Consequently, various degrees of success had been obtained in estimating aboveground biomass using Landsat vegetation indices (Sader et al., 1989; Lee & Nakane, 1996; Nelson et al., 2000; Steininger, 2000; Foody et al., 2003; Phua & Saito, 2003; Zheng et al., 2004). For example, Nelson et al. (2000) found that aboveground biomass cannot be reliably estimated using Landsat data without the inclusion of secondary forest age. Steininger (2000) explored the ability of Landsat data for estimating aboveground biomass of tropical secondary forests and found that saturation was a problem for advanced successional forests.

Similarly, different degrees of success had been obtained in previous studies of using radar data for estimating aboveground biomass, with longer-wavelength L-band and Pband SAR data proven to be more valuable (Sader, 1987; Le Toan et al., 1992; Rauste et al., 1994; Ranson et al., 1997; Luckman et al., 1997; Kurvonen et al., 1999; Kuplich et al., 2000; Tsolmon et al., 2002; Sun et al., 2002; Castel et al., 2002; Santos et al., 2002). For example, Kuplich et al. (2000) used JERS-1/SAR data for aboveground biomass estimation of regenerating forests. Sun et al. (2002) found that multi-polarization L-band SAR data were useful for estimating aboveground biomass of forest stands in mountainous areas. Castel et al. (2002) identified the significant relationships between the backscatter coefficient of JERS-1/SAR data and the stand biomass of a pine plantation. Santos et al. (2002) used JERS-1/SAR data to analyse the relationships between backscatter signals and biomass of forest and savanna formations. The significant correlation between aboveground biomass and JERS-1/SAR backscatter coefficient, as shown in Table 3 and Fig. 4, indicates that longer-wavelength L-band SAR data are also valuable in the Arctic. Nevertheless, the saturation problem is also common in estimating aboveground biomass using radar data (Luckman et al., 1997; Balzter, 2001, Lucas et al., 2004; Kasischke et al., 2004). For example, Luckman et al. (1997) found that the longer-wavelength L-band SAR image was more suitable to discriminate different levels of forest biomass up to a certain threshold than shorter-wavelength C-band SAR data.

To take advantages of the ability of Landsat vegetation indices and JERS-1/SAR backscatter coefficient, we used both data for estimating aboveground biomass. The best fit relationship between aboveground biomass and JERS-1/SAR backscatter coefficient *σ* as well as Landsat B4/B5 for all types mixed in the Dempster Highway study area is given by:

$$
\ln(B\_a) = 2.3759(\text{B4/B5}) + 0.5542\sigma + 4.0948\,\text{A}\,\tag{7}
$$

with a coefficient of determination *r*2 = 0.72, and standard estimation error (*SEE*) = 0.78. Because the logarithmic equations could introduce a systematic bias when used for back calculating biomass, it has now become fairly widely recognized that a correction factor is necessary to counteract this bias (Sprugel, 1983). The correction factor (*CF*) can be calculated by using the formula:

Many vegetation indices have been developed and applied for estimating aboveground biomass (Anderson & Hanson, 1992; Anderson et al., 1993; Mutanga & Skidmore, 2004; Lu 2005). In Table 3, we examined 4 vegetation indices: ratio of Landsat B4/B5, simple ratio (SR = B4/B3), normalized differential vegetation index (NDVI = (B4 - B3)/(B4 + B3)), shortwave vegetation index (SWVI = (B4 - B5)/(B4 + B5)). In general, vegetation indices can partially reduce the impacts on reflectance caused by environmental conditions and shadows, thus improve correlation between AGB and vegetation indices, especially in those sites with complex vegetation stand structures (Lu et al., 2004). However as shown in Table 3, not all vegetation indices are significantly correlated with aboveground biomass. Consequently, various degrees of success had been obtained in estimating aboveground biomass using Landsat vegetation indices (Sader et al., 1989; Lee & Nakane, 1996; Nelson et al., 2000; Steininger, 2000; Foody et al., 2003; Phua & Saito, 2003; Zheng et al., 2004). For example, Nelson et al. (2000) found that aboveground biomass cannot be reliably estimated using Landsat data without the inclusion of secondary forest age. Steininger (2000) explored the ability of Landsat data for estimating aboveground biomass of tropical secondary forests

Similarly, different degrees of success had been obtained in previous studies of using radar data for estimating aboveground biomass, with longer-wavelength L-band and Pband SAR data proven to be more valuable (Sader, 1987; Le Toan et al., 1992; Rauste et al., 1994; Ranson et al., 1997; Luckman et al., 1997; Kurvonen et al., 1999; Kuplich et al., 2000; Tsolmon et al., 2002; Sun et al., 2002; Castel et al., 2002; Santos et al., 2002). For example, Kuplich et al. (2000) used JERS-1/SAR data for aboveground biomass estimation of regenerating forests. Sun et al. (2002) found that multi-polarization L-band SAR data were useful for estimating aboveground biomass of forest stands in mountainous areas. Castel et al. (2002) identified the significant relationships between the backscatter coefficient of JERS-1/SAR data and the stand biomass of a pine plantation. Santos et al. (2002) used JERS-1/SAR data to analyse the relationships between backscatter signals and biomass of forest and savanna formations. The significant correlation between aboveground biomass and JERS-1/SAR backscatter coefficient, as shown in Table 3 and Fig. 4, indicates that longer-wavelength L-band SAR data are also valuable in the Arctic. Nevertheless, the saturation problem is also common in estimating aboveground biomass using radar data (Luckman et al., 1997; Balzter, 2001, Lucas et al., 2004; Kasischke et al., 2004). For example, Luckman et al. (1997) found that the longer-wavelength L-band SAR image was more suitable to discriminate different levels of forest biomass up to a certain threshold than

To take advantages of the ability of Landsat vegetation indices and JERS-1/SAR backscatter coefficient, we used both data for estimating aboveground biomass. The best fit relationship between aboveground biomass and JERS-1/SAR backscatter coefficient *σ* as well as Landsat

ln( ) 2.3759(B4/B5)+0.5542 +4.0948 *Ba*

with a coefficient of determination *r*2 = 0.72, and standard estimation error (*SEE*) = 0.78. Because the logarithmic equations could introduce a systematic bias when used for back calculating biomass, it has now become fairly widely recognized that a correction factor is necessary to counteract this bias (Sprugel, 1983). The correction factor (*CF*) can be calculated

, (7)

B4/B5 for all types mixed in the Dempster Highway study area is given by:

and found that saturation was a problem for advanced successional forests.

shorter-wavelength C-band SAR data.

by using the formula:

$$CF = e^{SEE^2/2} \tag{8}$$

For equation (7), the *CF* = 1.35. The 1:1 comparison of estimated aboveground biomass using *CF*-corrected equation (7) and the measured values for the sites along the Dempster Highway is shown in Fig. 6, with the slope = 0.95, *r*2 = 0.72, and *SEE* = 13.5 t h-1 over measured aboveground biomass range from 0.4 to 94.5 t h-1. These results agreed well with the finding of Moghaddam et al. (2002) that the estimation accuracy of forest biomass was significantly improved when radar and optical data were used in combination, compared to estimates using a single data type alone.

Fig. 6. A 1:1 comparison between estimated and measured aboveground biomass values for the calibration sites along the Dempster Highway transect measured during the summer of 2004. For clarity over the low biomass range, the results are shown on the log-log scale.

#### **3.3 Validation of aboveground biomass relationship with remote sensing indices**

In a previous study (Chen et al., 2009a), we validated equation (7) using aboveground biomass measurements at 33 sites around Yellowknife, Northwest Territories, and Lupin Gold Mine, Nunavut Territories. The 1:1 comparison of estimated aboveground biomass and the measurements for the Yellowknife and Lupin Gold Mine study area indicates that the equation holds up very well, with *r*2 = 0.81, slope =1.17, and *SEE* = 9.67 t h-1 over measured aboveground biomass range from 0.9 to 103.3 t h-1.

In this study, we further validated the relationship using measurements collected over the same study area at later dates. We used aboveground biomass measured at 21 sites along the Dempster Highway during the summer of 2006 in the Ivvavik National Park during the summer of 2008, where foliage biomass was the main measurement target, to validate the above relationship. Fig. 7 shows the validation result, with the slope = 0.91, *r*2 = 0.90, and *SEE* = 1.5 t h-1 over measured aboveground biomass range from 0 to 18.16 t h-1.

These validation results suggest that the aboveground biomass can be reliably estimated using Landsat B4/B5 and JERS-1/SAR backscatter coefficient.

Mapping Aboveground and Foliage Biomass Over the Porcupine Caribou

with *r*2 = 0.81, *SEE* = 20.6 g m-2, *F*= 81, *P* = 2.7×10-8, and *n* = 21 (Fig. 8).

follows

0

during the summer of 2008.

g m-2, *F*= 158, *P* = 2.6×10-12, and *n* = 27.

**3.5 Baseline maps of aboveground and foliage biomass** 

in circa 2000 over the Porcupine caribou habitat.

50

100

150

**Foliage biomass (g m-2)**

200

250

Habitat in Northern Yukon and Alaska Using Landsat and JERS-1/SAR Data 243

previous studies (Franklin & Hiernaux, 1991; Hall et al., 1995; Chen & Cihlar 1996; Turner et al., 1999; Brown et al., 2000; Chen et al., 2002; Wylie et al., 2002; Phua & Saito, 2003; Laidler & Treitz, 2003; Lu, 2004). The best fit relationship between foliage biomass (*Bf*, g m-2) and Landsat-based simple ratio (*SR*) over the Porcupine caribou habitat is given as

> 0 5 10 15 **Landsat-based simple ratio**

Fig. 8. Relationship between Landsat simple ratio and foliage biomass measured at sites along the Dempster Highway during the summer of 2006 and in the Ivvavik National Park

Because we had only a relative small sample size of foliage biomass over the Porcupine caribou habitat, we didn't leave a fraction of the foliage biomass measurement points as validation. Nevertheless, we did find a similar relationship between Landsat-based simple ration and foliage biomass for the Bathurst caribou habitat located in Northwest Territory, Nunavut Territory, and northern Saskatchewan (Chen et al., 2011), with *r*2 = 0.86, *SEE* = 26.3

Applying equations (7) and (8) to the co-registered and geo-referenced Landsat and JERS-1/SAR mosaics data over the Porcupine caribou habitat, we produced aboveground biomass for the Porcupine caribou habitat. Fig. 9 shows aboveground biomass distribution

16.62 - 1.1906 *B SR <sup>f</sup>* , (9)

Fig. 7. A 1:1 comparison between estimated and measured aboveground biomass values for the validation sites along the Dempster Highway transect measured during the summer of 2006 and in the Ivvavik National Park during the summer of 2008. For clarity over the low biomass range, the results are shown on the log-log scale.
