A variation in the fraction of absorbed photosynthetically active
radiation and a comparison with MODIS data in burned black spruce
forests of interior Alaska
Hiroki Iwata
a,*
, Masahito Ueyama
b, Chie Iwama
b, Yoshinobu Harazono
a,b aInternational Arctic Research Center, University of Alaska Fairbanks, Fairbanks, AK 99775-7340, USA
b
Graduate School of Life and Environmental Sciences, Osaka Prefecture University, Sakai, Osaka 599-8531, Japan Received 31 July 2012; revised 31 January 2013; accepted 8 March 2013
Available online 29 March 2013
Abstract
Absorption of photosynthetically active radiation (PAR) by vegetation was observed in two burned black spruce forests, one and seven years after wildfire, in interior Alaska along with several vegetation properties. This study considered PAR absorption by mosses by examining the relationship between PAR transmittance and fractional coverage of green vegetation. Our results suggest that mosses absorbed a considerable fraction of incoming PAR in the burned forests, which cannot be neglected in evaluating the fraction of absorbed PAR (FPAR). The relationships between FPAR and vegetation indices revealed that enhanced vegetation index (EVI) may be suitable for expressing the spatial and temporal variation of FPAR, regardless of stand age after wildfire. The comparison between the observed in situ FPAR and FPAR derived from Moderate Resolution Imaging Spectroradiometer (MODIS FPAR) clearly showed that MODIS FPAR was highly overestimated. The most likely reason for the overestimation was identified as misclassification of land cover type. The current regional estimation of photosynthesis in boreal region based on the light-use efficiency approach and MODIS FPAR is probably overestimated, and an accurate distribution of FPAR is desired for clarifying the regional carbon exchange in boreal forests.
Ó 2013 Elsevier B.V. and NIPR. All rights reserved.
Keywords: Boreal forest; Fire scar; FPAR/LAI; Vegetation index; Vegetation recovery
1. Introduction
Wildfire is a major disturbance in boreal forest ecosystems. Under the current warming climate, an increase of fire frequency and its intensification has
been observed in boreal regions, and it is expected that the increasing trend will continue (e.g.,Flannigan and
Harrington, 1988; Kasischke and Stocks, 2000;
Kasischke and Turetsky, 2006). In the areas after
wildfire, carbon, water, and energy exchange processes are significantly altered from the pre-burn condition
(Chambers and Chapin, 2002; Liu and Randerson,
2008; Amiro et al., 2010; Goulden et al., 2011).
Thus, it is important to explicitly express the exchange processes in burned areas in order to accurately esti-mate the regional exchanges.
* Corresponding author. Present address: Graduate School of Agriculture, Kyoto University, Oiwake-cho, Kita-shirakawa, Sakyo-ku, Kyoto 606-8502, Japan. Tel.:þ1 075 753 6149; fax: þ1 075 753 6149.
E-mail address:hiwata@kais.kyoto-u.ac.jp(H. Iwata).
1873-9652/$ - see front matterÓ 2013 Elsevier B.V. and NIPR. All rights reserved.
http://dx.doi.org/10.1016/j.polar.2013.03.004
Polar Science 7 (2013) 113e124
One method for estimating the regional distribu-tion of photosynthesis e an important component of carbon cycle e is the light-use efficiency approach
(Monteith, 1972), in conjunction with satellite data.
In this approach, the fraction of absorbed photosyn-thetically active radiation (FPAR) is an important biophysical parameter. FPAR data covering the world’s terrestrial area are derived using Moderate Resolution Imaging Spectroradiometer (MODIS) data (hereafter MODIS FPAR;Myneni et al., 2002,2003), and these data have been used in turn to estimate spatial distribution of photosynthesis (Heinsch et al.,
2003; Running et al., 2004; Zhao et al., 2005;
Heinsch et al., 2006; Zhang et al., 2008; Zhao and
Running, 2010). However, validation studies of
MODIS FPAR have been scarce (Steinberg et al., 2006), especially for disturbed areas. The few studies that have conducted validations (Fensholt
et al., 2004; Huemmrich et al., 2005; Steinberg
et al., 2006) have demonstrated that MODIS FPAR
is overestimated compared to in situ observation.
Steinberg et al. (2006) mentioned that the MODIS
FPAR tended to overestimate particularly in more sparsely vegetated burned forests.
In sparsely vegetated areas after wildfire in boreal regions, it is expected that ground cover plants such as mosses largely contribute to photosynthesis (e.g.,
Heijmans et al., 2004). However, the validation
studies mentioned above did not account for the ab-sorption of photosynthetically active radiation (PAR) by ground cover plants, due to the difficulty of direct observation. Since an accurate estimate of PAR ab-sorption by such vegetation is a key to model a re-covery of photosynthesis after wildfire, we collected data regarding the absorption of PAR along with several vegetation properties in burned boreal forests, and estimated the absorption of PAR by ground cover plants.
This study examined how actual FPAR varies seasonally and through the vegetation recovery after wildfires, at the early stage of succession, using in situ observations. The obtained FPAR data can be used for validating MODIS FPAR in burned forests. It also fo-cuses on obtaining an empirical relationship in order to estimate temporal and spatial variations in FPAR from vegetation indices (Goward and Huemmrich, 1992;
Be´gue´, 1993;Hanan et al., 1995;Fensholt et al., 2004)
in the early stage of recovery after wildfire. Such re-lationships are used as a back-up algorithm in MODIS FPAR estimations (Myneni et al., 2003); thus the re-lationships are useful for improving the quality of MODIS FPAR.
2. Observations and data analysis 2.1. Study sites
We conducted observations in two burned black spruce forests, one and seven years after wildfire, in 2011 in interior Alaska.
The seven-year burned black spruce forest (6570N, 147260W, 491 m a.s.l.) is located at the Poker Flat Research Range of the University of Alaska Fairbanks, interior Alaska. A wildfire began in the middle of June of 2004 and continued until early August. The remaining soil organic layer was 2 cm deep, underlain by sandy silt with gravels. Major vegetation consisted of sapling of white birch and trembling aspen, Labra-dor tea, bog blueberry, sedge, and mosses. The heights of saplings of white birch and trembling aspen were less than 1.2 m. Many dead trees remained standing at the time of observation. Micrometeorological and flux observations have been conducted continuously since August, 2008 (Iwata et al., 2011).
The other site is a one-year burned black spruce forest (65230N, 148560W, 265 m a.s.l.) located 19 km southwest of Livengood, in interior Alaska. A wildfire began at the end of May of 2010 and remained active until mid-June. The remaining soil organic layer was 13 cm deep (M. Kondo and M. Uchida, personal communication). Major vegetation consisted of Lab-rador tea, cloudberry, horsetail, prickly rose, and sedge. No broadleaf trees were present and mosses covered little of the area. The height of most vegetation was less than 0.5 m. Most dead trees remained standing at the time of observation. At this site, micrometeoro-logical and flux observations began in May, 2011.
Wildfires burned almost all of the vegetation in the vicinity of both observation sites. As a result, the dis-tribution of regrowing vegetation was quite similar, in the range of a few hundred meters. We selected an observation area in the vicinity of a flux and meteo-rology observation mast for each site.
2.2. Field sampling design
At the two burned forest sites, the following vari-ables were observed: transmittance of PAR, fractional coverage of green vegetation, vegetation indices, and leaf area index (LAI). Data were collected at 27 points within a 30 m by 30 m grid, except for LAI, which was measured at 9 points within the grid (Fig. 1).
In the seven-year burned site, data were collected approximately every two weeks, allowing us to examine the seasonal variations of FPAR and vegetation
properties. In the one-year burned site, data were collected at two points in time, when light conditions were suitable for the observation.
2.3. Measurements and analysis
Mostly, all measurements were conducted on the same day. However, when a particular measurement (typically, LAI) was unavailable due to unsuitable sunlight conditions, the measurement was conducted within a week in order to obtain a full set of data. 2.3.1. Transmittance of PAR
The fraction of transmitted PAR, PARt, was observed with a line PAR sensor (LI191SA, Li-Cor, USA), which has a 1-m sensing component for PAR. Measurements were taken above the regrowing vege-tation, at approximately 1.5-m height, and on the ground in order to obtain fraction of transmittance. Leveling of the sensor was confirmed at each observation.
These measurements were conducted in diffuse or weak sunlight conditions. Under strong sunlight, spatial variation of PARt is large in the presence of sunlit area and shadow of dead black spruce. Obser-vation with a line PAR sensor in such conditions results in data with large uncertainty.
2.3.2. Fractional coverage of green vegetation Fractional coverage of green vegetation was ob-tained from digital photographs of the surface, which
were taken from approximately 1.5 m above with a single lens reflex digital camera (Olympus E-620). The resolution of photographs was set to 4032 3024, and the photographs were recorded in JPEG format. The camera’s white balance was fixed using a gray-scale card before taking the photographs, in order to mini-mize the effect of different light conditions on the analysis. Also, the photographs were taken using the flash to reduce shadows on the ground. This mea-surement was conducted in diffuse or weak sunlight conditions, since green vegetation in the dark shadows of dead spruce trees was not identified in the analysis below.
Excess green index (EGI, Woebbecke et al., 1995;
Rasmussen et al., 2007), defined below, was used to
distinguish green plants and other materials in the photographs.
EGI¼ 2G ðR þ BÞ ð1Þ
where R, G, and B represent RGB digital number (0e255). EGI was calculated for each pixel of the digital photographs, and pixels with EGI higher than a threshold value were considered to be green plants. We determined the threshold value at 20 by visually checking the agreement of EGI image green plant dis-tribution with green vegetation in the original photo-graphs. Fractional coverage of green vegetation was calculated by dividing the number of green plant pixels by that of total pixels. This fractional coverage of vegetation is determined solely by colore hence red and yellow leaves in the fall season are not considered as green vegetation. The image analysis above was conducted using GRASS GIS software 6.4.0.
2.3.3. Vegetation indices and leaf area index
Spectral reflectance was observed using a hand-held spectroradiometer (FieldSpec HH, Analytical Spectral Devices, USA) with a spectral range of 325e1075 nm and with 3.5 nm resolution. Before taking measure-ments, the FieldSpec HH was optimized, a dark current measurement was taken, and the Spectralon white panel was used for the reference measurement. This procedure was conducted both in cases when illumi-nation conditions changed, and at least every nine observations. The FieldSpec HH was held at a height of 1.7 m, in order to view the ground at an approximate view zenith angle of 30. The height and zenith angle were chosen in order to sample a large area with one measurement, with the instrument held by the instru-ment operator. The resulting observation area was approximately oval, with a major axis of 1 m. The
Fig. 1. Design of manual field sampling. Vegetation cover, trans-mitted PAR, and vegetation indices were observed at open and solid circles, and leaf area index was observed at solid circles only.
instrument was faced approximately south for obser-vations. This observation was conducted in diffuse light conditions in order to minimize the effect of incidence angle on measured reflectance and hetero-geneity of the incident radiation field within dead spruce trees (Rautiainen et al., 2011).
Normalized difference vegetation index (NDVI:
Rouse et al., 1974; Tucker, 1979;Jones and Vaughan,
2010) and enhanced vegetation index (EVI, Huete,
1988; Huete et al., 2002) were calculated from the
obtained spectral reflectance,r. NDVI¼rNIR rRED
rNIRþ rRED
ð2Þ
EVI¼ 2:5 rNIR rRED
rNIRþ 6rRED 7:5rBLUEþ 1
ð3Þ where the suffixes NIR, RED, and BLUE denote fre-quency range of near-infrared (841e876 nm), red (620e670 nm), and blue (459e479 nm), respectively.
Leaf area index (LAI) was observed using a plant canopy analyzer (LAI-2000, Li-Cor, USA). First, reference signals were taken from above vegetation at a height of 1.8 m, and transmitted signals were then observed on the ground. Observations were conducted only in diffuse sunlight conditions or at dusk.
2.3.4. Uncertainty due to heterogeneity
In burned forests, in general, vegetation is sparse. As a result, the averaged vegetation properties in this study have uncertainty due to heterogeneity in vege-tation distribution within the 30 m by 30 m grid. We evaluated the uncertainty of averaged data by
ε ¼ta;n1ffiffiffis n
p ð4Þ
where n is the number of data, t is Student’s t-value with a confidence level ofa and n 1degrees of freedom, ands is standard deviation (Crockford and Richardson,
1990;Carlyle-Moses et al., 2004;Iida et al., 2005). A
confidence level of 95% was used. 2.3.5. Absorption of PAR
Total absorbed PAR (W m2), PARab, is defined as follows (Hipps et al., 1983; Goward and Huemmrich,
1992;Huemmrich et al., 2005).
PARab¼ PARinþ PARb PARt PARr ð5Þ where PARin, PARb, and PARrare incident PAR above the canopy, PAR reflected at the soil, and reflected PAR observed above the canopy, respectively. At the
observation sites, PARbwas negligibly small on dark ground (PARb/PARin < 0.03), and its contribution of PARbwas omitted.
The fraction of PARtobserved in this study does not include absorption of PAR by mosses. This can lead to underestimation of FPAR. To account for the absorp-tion of PAR by mosses, the relaabsorp-tionship between the fractional coverage for taller plants (excluding mosses) and the fraction of PARtwas examined first, and then the obtained relationship was extrapolated to the value of fractional coverage for all green vegetation to esti-mate the fraction of PARtincluding the effect of PAR absorption by mosses. The assumption made here is that the transmittance of moss is equal to that of taller plants, and transmitted PAR through the moss layer is completely absorbed by the underlying soil. Hereafter, this fraction of PARt is referred to as the corrected fraction of PARt.
The partitioning of fractional coverage of green vegetation was conducted as follows. The fractional coverage of taller plants was calculated as the differ-ence between the observed value and value before the leaf emergence. It was simply assumed that the in-crease in the fractional coverage of green vegetation after leaf emergence was due to an increase in leaf area of taller plants, and superposition of taller plants above mosses was not accounted for. In the one-year burned forest, coverage of mosses was negligible; hence no correction was applied to the observed data.
The fraction of absorbed PAR was obtained by dividing Eq.(5) by PARinand neglecting PARb/PARin as follows:
FPAR ¼PARPARab in¼ 1
PARt PARin
PARr
PARin ð6Þ
where the corrected fraction of PARt, estimated as in the method described above, was substituted for the second term on the right-hand side in order to calculate FPAR. The third term represents the albedo and was calculated from observations at the height of 2.2 m at an obser-vation mast at each site. The obserobser-vation mast was located at a distance of 40 m and 25 m from the center of the observation grid for the seven-year and one-year site, respectively. Albedo was also manually observed at each point on the observation grid a few times, and spatial difference between albedo observed at the mast and that in the observation grid was then adjusted. 2.4. Satellite data
The 8-day composite MODIS FPAR (1-km resolu-tion, MCD15A2 collection 5:Myneni et al., 2003) was
compared with in situ data. The MODIS FPAR algo-rithm uses observed MODIS reflectance product (MOD09 series) and land cover data as inputs. It models canopy reflectance based on bidirectional reflectance distribution functions with sun-view ge-ometries and canopy structure and ground cover con-ditions using a canopy radiation transfer model
(Knyazikhin et al., 1998a). This algorithm compares
observed and modeled canopy reflectance for a suite of canopy and ground conditions, such as canopy archi-tecture, optical properties of canopy elements, and soil and/or understory type (Knyazikhin et al., 1998a), all of which represent a range of expected natural occur-rences. In all cases where the difference between observed and calculated reflectance is less than the uncertainty of the observed reflectance, canopy struc-tural variables used are taken as possible solutions, and FPAR is calculated for each solution (Knyazikhin
et al., 1998b). Then, the FPAR product is calculated
as an average value over all possible solutions. For the collection 5 product, eight-biome land cover data is used e i.e., grasses/cereal crops, shrubs, broadleaf crops, savanna, evergreen broadleaf forest, deciduous broadleaf forest, evergreen needleleaf forest, and de-ciduous needleleaf forest. One pixel encompassing the observation site at both the one-year and the seven-year burned forest was extracted, and only data with “good quality” designation under the quality control description were used.
When comparing in situ data with satellite data, it is important to consider the difference of data in terms of spatial scale. Before the comparison, the heterogeneity of vegetation distribution within the MODIS FPAR pixel was examined using finer Landsat data (30-m resolution, L7 ETMþ) to check whether the observa-tion sites are representative of the larger MODIS FPAR pixel. NDVI was calculated for each Landsat pixel within the MODIS pixel described above. The repre-sentativeness of the observation site within the MODIS pixel was then examined.
For the seven-year site, a cloudless image was captured on DOY 227 in 2011. The mean NDVI and standard deviation were 0.33 and 0.05, respectively. This mean NDVI was almost the same as the mean NDVI averaged for the pixel corresponding to the observation site and surrounding pixels (0.30). This observation site can thus be considered representative of the larger MODIS pixel encompassing the obser-vation site, and data taken at the obserobser-vation sites can be directly compared with MODIS FPAR data. For the one-year site, a cloudless image captured on DOY 186 in 2011 was examined; mean NDVI and standard
deviation were 0.35 and 0.12, respectively. This mean NDVI was substantially larger than the mean NDVI averaged for the pixel corresponding to the observation site and surrounding pixels (0.20). The larger mean NDVI was partly due to the presence of unburned black spruce trees near the boundary of MODIS pixel. This consequence was discussed when comparing in situ data with MODIS data.
3. Results and discussion
3.1. Seasonal variations of vegetation properties At the seven-year site, complete snowmelt occurred at the beginning of May (DOY 126) in 2011. Saplings of white birch and trembling aspen and low bush plants such as blueberry started to grow their leaves in early June (Fig. 2b). The deciduous trees had fully expanded their leaves by the middle of July (Fig. 2f). Leaves started to turn red and yellow in the middle of August
(Fig. 2h), and defoliation had completed by the end of
September (Fig. 2l). Snow then began to accumulate on the ground in October (Fig. 2m).
Fig. 3 shows a seasonal variation of fractional
coverage of whole green vegetation at the seven-year burned forest, where 10 m by 10 m grid averages and an average of all points are indicated. Seasonal variations of other data are similarly shown below, unless otherwise specified. Fractional coverage of whole green vegetation showed a clear seasonal pattern at the seven-year burned forest, increasing with leaf growth of deciduous vegetation and decreasing with leaf senescence and defoliation. The vegetation cover reached up to 0.50 at DOY 199. Photographs were also taken at DOY 296 after snow cover began, and the vegetation cover was calculated as zero.
Fractional coverage of green vegetation was observed before leaf emergence and after complete defoliation; with a value of about 0.17. This fractional coverage nearly corresponds to moss coverage, since there were no deciduous leaves present and coverage of evergreen plants such as cranberry remained small in the observation area.
Spatial variation of fractional coverage was quite large: data observed at individual points ranged from 0.32 to 0.77 for DOY 199, for example (data not shown). However, the uncertainty due to heterogeneity calculated with Eq.(4)was about 12% of the average throughout the growing season, and thus the averaged data reasonably represent the observation area.
NDVI and EVI observed by a hand-held spectror-adiometer also showed clear seasonal variations
(Fig. 4), reflecting seasonal increase of vegetation cover and vegetation activity. NDVI and EVI showed maximum values of 0.66 (DOY 186) and 0.42 (DOY 217), respectively. Observation was taken after snow cover began, and both NDVI and EVI were close to zero. The uncertainties due to heterogeneous distribu-tion of vegetadistribu-tion were 7% and 11% of the average during the growing season for NDVI and EVI, respectively.
Fig. 5shows a seasonal variation of LAI, in which
individual point data and an average of all data are indicated. LAI showed a clear seasonal variation. Un-certainty, however, was larger than other properties. Typical uncertainty during the growing season was 30% of the average. This may be due to a smaller
sampling number of LAI. LAI was also observed after snow cover at about 0.22. LAI observed with LAI-2000 inherently corresponds to plant area index, since the device cannot distinguish light interception by leaves and by non-photosynthetic parts such as trunks and twigs. Since there were no green leaves
Fig. 2. Photographs taken at a certain point in the seven-year burned forest to show the seasonal change in vegetation. The photographs were taken on a) May/18, b) Jun/3, c) Jun/16, d) Jun/23, e) Jul/5, f) Jul/18, g) Aug/5, h) Aug/17, i) Sep/1, j) Sep/9, k) Sep/15, l) Sep/30, and m) Oct/23 in 2011. 0.0 0.2 0.4 0.6 0.8 1.0 100 150 200 250 300
Fractional coverage of green
vegetation
Day of Year
Fig. 3. Seasonal variation of vegetation cover. The broken lines show each 10 m by 10 m grid average, and the thick black line shows the average of all points. The error bar shows a 95% confidence interval.
0.0 0.2 0.4 0.6 0.8 1.0 100 150 200 250 300 NDVI Day of Year
(a)
0.0 0.2 0.4 0.6 0.8 1.0 100 150 200 250 300 EVI Day of Year(b)
Fig. 4. Seasonal variations of (a) NDVI and (b) EVI. The broken lines show each 10 m by 10 m grid average, and the thick black line shows the average of all points. The error bar shows a 95% confi-dence interval.
during snow cover, this value is due to light intercep-tion by non-photosynthetic parts.
Observed fraction of PARt showed a noticeable seasonal variation responding to the change in vege-tation cover (Fig. 6). This value was about 0.87 before the emergence of leaves (DOY 138), after which it decreased to 0.66 in the middle of the growing season. Along with leaf senescence and defoliation, the observed fraction of PARt increased again, showing a value of 0.86 after the complete defoliation (DOY 296). As with LAI, this observation includes light interception by non-photosynthetic parts. Hence, the observation after the complete defoliation represents absorption of PAR by non-photosynthetic parts. Un-certainty due to heterogeneous distribution of vegeta-tion was typically 7% of the average during the growing season.
3.2. PAR absorption by vegetation
Fig. 7shows the relationship between the fractional
coverage of taller plants and the observed fraction of
PARt. As expected, the observed fraction of PARt decreased as the fractional coverage increased. The relationship was nearly linear. Two entries are avail-able for the one-year burned forest, and these data were plotted inFig. 7. The relationship at each site appears to be different, perhaps related to a difference in vegetation structure. In the one-year burned forest, most vegetation were low herbaceous plants, whereas woody plants had begun to dominate in the seven-year burned forest. The structure of the vegetation layer is simple in the one-year burned forest compared to the seven-year burned forest, and thus more light can penetrate to the ground in the one-year burned forest for certain vegetation cover. In contrast, light can be absorbed more efficiently by multiple vegetation layers e i.e., upper wood plants and lower herbaceous plants e in the seven-year burned forest.
Extrapolating this relationship to the value of frac-tional coverage of whole green vegetation, we esti-mated the fraction of PARtincluding the effect of PAR absorption by mosses. This contribution to PAR ab-sorption by mosses was large, since FPAR increased by 0.10 on average including the effect of mosses (Fig. 8). FPAR was calculated using the corrected fraction of PARtand then used in the following analysis.
The FPAR observed in situ at the seven-year burned forest was 0.25 before the emergence of leaves (DOY 138) and increased up to 0.45 in the middle of the growing season (Fig. 8). Then, it decreased with defoliation. The maximum FPAR observed at the one-year burned forest was 0.16. FPAR obviously increased with vegetation recovery. The FPAR at the seven-year burned forest in the mid-growing season (0.45) was lower compared to that at an eight-year burned black spruce forest (0.6) reported in Steinberg et al. (2006). Their value is probably even larger when PAR ab-sorption by mosses is taken into consideration. This 0.0 0.5 1.0 1.5 2.0 2.5 100 150 200 250 300 LAI Day of Year
Fig. 5. Seasonal variation of LAI. The broken lines show each point data, and the thick black line shows the average of all points. The error bar shows a 95% confidence interval.
0.0 0.2 0.4 0.6 0.8 1.0 100 150 200 250 300
Observed fraction of PAR
t
Day of Year
Fig. 6. Seasonal variation of the observed fraction of transmitted PAR. The broken lines show each 10 m by 10 m grid average, and the thick black line shows the average of all points. The error bar shows a 95% confidence interval. 0.0 0.2 0.4 0.6 0.8 1.0 0 0.1 0.2 0.3 0.4 0.5
Observed fraction of PAR
t
Fractional coverage of taller plants =-0.58 +0.80 y x 2 =0.75 R =0.001 p 7-year 1-year
Fig. 7. Relationship between the observed fraction of transmitted PAR and vegetation cover for taller plants. The regression line was fitted to the 7-year site only.
difference is attributable to the difference in vegetation recovery; tree heights in their site ranged from 1 to 3 m, whereas most trees were less than 1.2 m in our seven-year site.
In this study, fractional coverage of green vegetation was calculated based on a visually determined threshold (20). We conducted sensitivity analysis, in which FPAR was calculated in the same way but with different thresholds. We found that FPAR varied from 0.42 to 0.47 for threshold of 15e25 in mid-summer (DOY 199). Beyond this range, fractional coverage of green vegetation was obviously overestimated or underestimated. Similar uncertainty of FPAR was ob-tained for other period.
3.3. Relationships among FPAR and vegetation properties
As seen in Fig. 7, the observed fraction of PARt approximately linearly decreased with fractional coverage of taller plants. Since the variation in the albedo of PAR was small in these burned forests (0.07e0.09 at the seven-year site, and 0.03e0.04 at the one-year site) e only weakly dependent on the frac-tional coverage of vegetation e FPAR was mainly controlled by this variation in PARt. Hence, FPAR increased approximately linearly with the fractional coverage of vegetation for the seven-year burned forest (data not shown). The linear dependence is a reason-able result in the sparsely vegetated area, since further incoming PAR is intercepted as the fractional coverage of vegetation increases to nearly the complete ground cover. Besides, the fractional coverage of green vege-tation was linearly dependent on NDVI and EVI
(Fig. 9) for the range of fractional coverage of green
vegetation observed in this study.
The two linear relationships above suggest that FPAR also depends linearly on vegetation indices. This was confirmed by the observations, and there was a linear relationship between FPAR and NDVI for the seven-year burned forest (Fig. 10). Similarly, linear relationships were also reported in Fensholt et al.
(2004), for a dry grassland, and Huemmrich et al.
(2005), for a woodland. In addition, FPAR also
line-arly depended on EVI, although the correlation was lower for EVI.
Data for the one-year burned forest deviated from the FPAR-NDVI relationship for the seven-year burned forest. This suggests that NDVI may not be the single parameter for explaining spatial and temporal distri-bution of FPAR over boreal landscapes where patches with different ages after wildfire are distributed in forest ecosystems. Different relationships according to the stand age after wildfire suggest that the vegetation recovery may need to be considered when estimating FPAR distribution from NDVI. In contrast, data for the one-year burned forest seems to distribute around the relationship for the seven-year burned forest in the FPAR-EVI relationship, suggesting that EVI may be more suitable for expressing spatial and temporal variation of FPAR during the early stages of vegetation recovery. 0.0 0.2 0.4 0.6 0.8 1.0 100 150 200 250 300 FPAR Day of year
in situ (seven-year with moss) in situ (seven-year without moss) in situ (one-year) MODIS (seven-year) MODIS (one-year)
Fig. 8. Seasonal variations of in situ FPAR and MODIS FPAR: in situ FPAR considering PAR absorption by mosses at the seven-year site (black solid circle); in situ FPAR without PAR absorption by mosses at the seven-year site (open circle); in situ FPAR at the one-year site (black solid square); MODIS FPAR at the seven-one-year site (gray solid circle); and MODIS FPAR at the one-year site (gray solid square). 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Fractional coverage of geen vegetation NDVI
(a)
=1.53 -0.55 y x 2 =0.94 R <0.001 p 7-year 1-year 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Fractional coverage of geen vegetation EVI(b)
=1.76 -0.27 y x 2 =0.80 R =0.001 p 7-year 1-yearFig. 9. Relationship between the fractional coverage of green vegetation and (a) NDVI and (b) EVI. Regression lines were fitted to the seven-year forest only. Broken lines indicate 95% confidence intervals.
The advantage of EVI when describing spatial variation of LAI was also reported in Rocha and
Shaver (2009). They argued that the contribution of
dark soil artificially increased NDVI by dispropor-tionately decreasing the denominator of Eq.(2), though EVI is less sensitive to dark soil due to additional weight on the red reflectance. Thus, EVI can express the gradient of vegetation recovery, suppressing the influence of soil reflectance. The same reason applies to our result that EVI is more suitable for expressing the difference in FPAR over vegetation recovery.
Although the vegetation indices were obtained from off-nadir observation, the effects on vegetation indices
are probably small. According to Goward and
Huemmrich (1992), their simulation results of the
SAIL (Scattering by Arbitrarily Inclined Leaves) model (Verhoef, 1984) showed that the effect of a 30-degree view zenith angle in forwarding scattering upon NDVI was less than a few percent for any LAI. In addition, diffuse light condition also minimized the directional effect on measured vegetation indices. 3.4. Comparison with remote sensing product
As examined in Section2.4, the seven-year site may be considered representative of the MODIS FPAR pixel encompassing the observation site; however, the one-year site had lower NDVI compared to the mean
NDVI from the scale of the MODIS FPAR pixel. Hence, we will focus comparison of FPAR at the seven-year site in this section.
Comparison of in situ FPAR data with MODIS FPAR revealed that MODIS FPAR are overestimated compared to the in situ data: the maximum MODIS FPAR was approximately 1.7 times as large as the in situ data for the seven-year burned forest (Fig. 8). The magnitude of overestimation was smaller in the early and late growing season. These results were essentially consistent with those ofSteinberg et al. (2006).
Several reasons for the inconsistency between in situ FPAR and MODIS FPAR have been discussed in
Steinberg et al. (2006): (1) underestimation of in situ
FPAR by neglecting the contribution of mosses, (2) spatial inconsistency between in situ observation and MODIS FPAR pixel, and (3) land cover misclassifi-cation. Among these possible reasons, we can rule out underestimation of in situ FPAR by neglecting the contribution of mosses. Spatial inconsistency between in situ observation and MODIS FPAR pixel was examined, and we found that the observation sites fairly represented the larger MODIS FPAR pixel as shown in Section2.4. In addition, uncertainty related to heterogeneous distribution of vegetation in the 30 m by 30 m grid is probably about 7%, judging from the uncertainty of the observed fraction of PARt(Fig. 6). Hence, we believe that the difference in spatial scale was not the primary source of inconsistency between in situ FPAR and MODIS FPAR. The most likely reason, then, for this inconsistency is land cover misclassifi-cation in the calculation of MODIS FPAR (Tian et al.,
2000;Steinberg et al., 2006). MODIS FPAR is derived
with the three-dimensional radiative transfer model using specific parameters for determined land covers. We obtained the base land cover map, from which MODIS FPAR was estimated, from scientists at Boston University (D. Sulla-Menashe and M. Friedl, personal communication); this data indicated that the seven-year site was classified as “evergreen needleleaf forest”. Thus, PAR absorption by bare soil in the real situation may be misinterpreted as PAR absorption by under-story vegetation in the estimation of MODIS FPAR. This explains why the overestimation of MODIS FPAR was mitigated in the early and late growing season
(Fig. 8), during which it is probably simulated that
understory vegetation does not fully expand leaf cover in evergreen needleleaf forest e i.e., the canopy ar-chitecture and leaf distribution within the canopy are more or less similar (Tian et al., 2000) between burned forest and evergreen needleleaf forest. If this PAR absorption by the soil is correctly treated in the 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 FPAR NDVI
(a)
=0.90 -0.17 y x 2 =0.91 R <0.001 p 7-year 1-year 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 FPAR EVI(b)
=1.03 -0.01 y x 2 =0.77 R =0.002 p 7-year 1-yearFig. 10. Relationship between the fraction of absorbed PAR and (a) NDVI and (b) EVI. Regression lines were fitted to the seven-year forest only. Broken lines indicate 95% confidence intervals.
estimation, the MODIS FPAR accuracy in burned areas will be improved. Currently, the MODIS FPAR product is not designed to consider disturbed areas, and in-clusion of disturbed area in the biome classes, in addition to more frequent update of the biome class, is necessary to express the variation of FPAR in frequently disturbed areas.
A similar comparison between in situ LAI and MODIS LAI revealed that MODIS LAI also over-estimated e the maximum in situ LAI was approxi-mately 1.0 in the seven-year burned forest, whereas maximum MODIS LAI was approximately 2.0. This overestimation also probably arises from the misclas-sification of land cover type; low reflectance is mis-interpreted as absorption of PAR by extended leaves.
For the one-year site, the difference between in situ FPAR and MODIS FPAR is, in part, due to scale dif-ference. As examined in Section 2.4, Landsat NDVI for the observation site was smaller than average Landsat NDVI for MODIS pixel by a factor of about two. When comparing the in situ observation with MODIS data, it is necessary to consider this difference. However, taking the scale difference into account probably does not suffice to explain the difference in FPAR. For example, MODIS FPAR data at the one-year site divided by two as a first-order estimation was still larger than the in situ FPAR. The land cover type for the one-year site was also misclassified as evergreen needleleaf forest, and the same explanation above can be applied to this site.
It should be noted that our transmittance observa-tion was conducted under diffuse light condiobserva-tions. It is expected that diffuse light is more efficiently absorbed within the canopy (Hollinger et al., 1998; Roderick
et al., 2001;Gu et al., 2003), and therefore FPAR
be-comes larger in such condition. Goward and
Huemmrich (1992) showed that FPAR increased by
10% as the fraction of diffuse light increased from 0 to 100% in the SAIL model simulation. Taking this into account, however, results in larger discrepancy of FPAR between the in situ and MODIS data.
4. Conclusions
The obtained FPAR in this study considered ab-sorption of PAR by mosses, and the result suggested that mosses absorbed considerable fraction of incoming PAR in the burned forests. This absorbed PAR cannot be neglected in evaluating FPAR from in situ observations.
The in situ FPAR data obtained in this study revealed that MODIS FPAR was overestimated in
burned forests. The most likely reason for the over-estimation is the misclassification of land cover in deriving the MODIS FPAR, and PAR absorption by the soil was probably misinterpreted as PAR absorption by ground cover plants. The relationship between FPAR and vegetation indices suggested that EVI has potential to be applicable for estimating spatial and temporal variation of FPAR, regardless of stand age after wildfire during the early stage of vegetation succession. Applying an empirical relationship between FPAR and NDVI derived from one site to other sites with different age probably introduces errors in FPAR, which propa-gates to errors in estimation of photosynthesis.
FPAR is an important variable in water and carbon dioxide exchange modeling. Overestimation of FPAR directly results in an overestimation of photosynthesis in the light-use efficiency approach, for example. The current regional estimation of photosynthesis in boreal region based on the light-use efficiency approach and MODIS FPAR is probably overestimated. Boreal for-ests are composed of patches of different age after wildfire. Thus, an accurate distribution of FPAR is desired in order to clarify the current carbon exchange in regional scale and its relation with climate changes. Acknowledgments
This research was funded through grants to the University of Alaska Fairbanks, International Arctic Research Center from NSF under the Carbon Cycle Program of IARC/NSF and the Japan Aerospace
Exploration Agency (JAXA) under the “Arctic
Research Plan Utilizing the IARC-JAXA Information System (IJIS) and Satellite Imagery”, and the Japan Agency for Marine-Earth Science and Technology
(JAMSTEC) under the “JAMSTEC and IARC
Collaboration Studies”. We are grateful to Dr. Nagano for his help in the field work and Drs. Sulla-Menashe and Friedl at Boston University for providing the land cover map. Soil data were provided by Drs. Kondo and Uchida at National Institute of Environ-mental Studies. We also would like to thank Dr. Bauer for editing the manuscript and two anonymous re-viewers for the constructive comments.
References
Amiro, B.D., Barr, A.G., Barr, J.G., Black, T.A., Bracho, R., Brown, M., Chen, J., Clark, K.L., Davis, K.J., Desai, A.R., Dore, S., Engel, V., Fuentes, J.D., Goldstein, A.H., Goulden, M.L., Kolb, T.E., Lavigne, M.B., Law, B.E., Margolis, H.A., Martin, T., McCaughey, J.H., Misson, L., Montes-Helu, M., Noormets, A., Randerson, J.T., Starr, G.,
Xiao, J., 2010. Ecosystem carbon dioxide fluxes after disturbance in forests of North America. J. Geophys. Res. 115, G00K02. Be´gue´, A., 1993. Leaf area index, intercepted photosynthetically
active radiation, and spectral vegetation indices: a sensitivity analysis for regular-clumped canopies. Remote Sens. Environ. 46, 45e59.
Carlyle-Moses, D.E., Laureano, J.S.F., Price, A.G., 2004. Through-fall and throughThrough-fall spatial variability in Madrean oak forest communities of northeastern Mexico. J. Hydrol. 297, 124e135. Chambers, S.D., Chapin III, F.S., 2002. Fire effects on
surface-atmosphere energy exchange in Alaskan black spruce ecosys-tems: implications for feedbacks to regional climate. J. Geophys. Res. 107 (D1), 8145 [printed 108(D1), 2003].
Crockford, R.H., Richardson, D.P., 1990. Partitioning of rainfall in a eucalypt forest and pine plantation in southeastern Australia: I throughfall measurement in a eucalypt forest: effect of method and species composition. Hydrol. Process. 4, 131e144. Fensholt, R., Sandholt, I., Rasmussen, M.S., 2004. Evaluation of
MODIS LAI, fAPAR and the relation between fAPAR and NDVI in a semi-arid environment using in situ measurements. Remote Sens. Environ. 91, 490e507.
Flannigan, M.D., Harrington, J.B., 1988. A study of the relation of meteorological variables to monthly provincial area burned by wildfire in Canada (1953e80). J. Appl. Meteorol. 27, 441e452. Goulden, M.L.., McMillan, A.M.S., Winston, G.C., Rocha, A.V., Manies, K.L.., Harden, J.W., Bond-Lamberty, B.P., 2011. Patterns of NPP, GPP, respiration, and NEP during boreal forest succes-sion. Glob. Change Biol. 17, 855e871.
Goward, S.N., Huemmrich, K.F., 1992. Vegetation canopy PAR absorptance and the normalized difference vegetation index: an assessment using the SAIL model. Remote Sens. Environ. 39, 119e140.
Gu, L., Baldocchi, D.D., Wofsy, S.C., Munger, J.W., Michalsky, J.J., Urbanski, S.P., Boden, T.A., 2003. Response of a deciduous forest to the Mount Pinatubo eruption: enhanced photosynthesis. Science 299, 2035e2038.
Hanan, N.P., Prince, S.D., Be´gue´’, A., 1995. Estimation of absorbed photosynthetically active radiation and vegetation net production efficiency using satellite data. Agric. For. Meteorol. 76, 259e276.
Heijmans, M.M.P.D., Arp, W.J., Chapin III, F.S., 2004. Carbon di-oxide and water vapour exchange from understory species in boreal forest. Agric. For. Meteorol. 123, 135e147.
Heinsch, F.A., Reeves, M., Votava, P., Kang, S., Milesi, C., Zhao, M., Glassy, J., Jolly, W.M., Loehman, R., Bowker, C.F., Kimball, J.S., Nemani, R.R., Running, S.W., 2003. User’s Guide: GPP and NPP (MOD17A2/A3) Products NASA MODIS Land Algorithm. Heinsch, F.A., Zhao, M., Running, S.W., Kimball, J.S.,
Nemani, R.R., Davis, K.J., Bolstad, P.V., Cook, B.D., Desai, A.R., Ricciuto, D.M., Law, B.E., Oechel, W.C., Kwon, H., Luo, H., Wofsy, S.C., Dunn, A.L., Munger, J.W., Baldocchi, D.D., Xu, L., Hollinger, D.Y., Richardson, A.D., Stoy, P.C., Siqueira, M.B.S., Monson, R.K., Burns, S.P., Flanagan, L.B., 2006. Evaluation of remote sensing based terrestrial productivity from MODIS using regional tower eddy flux network observations. IEEE Trans. Geosci. Remote Sens. 44, 1908e1925.
Hipps, L.E., Asrar, G., Kanemasu, E.T., 1983. Assessing the inter-ception of photosynthetically active radiation in winter wheat. Agric. Meteorol. 28, 253e259.
Hollinger, D.Y., Kelliher, F.M., Schulze, E.-D., Bauer, G., Arneth, A., Byers, J.N., Hunt, J.E., McSeveny, T.M., Kobak, K.I.,
Milukova, I., Sogatchev, A., Tatarinov, F., Varlargin, A., Ziegler, W., Vygodskaya, N., 1998. Foresteatmosphere carbon dioxide exchange in eastern Siberia. Agric. For. Meteorol. 90, 291e306.
Huemmrich, K.F., Privette, J.L., Mukelabai, M., Myneni, R.B., Knyazikhin, Y., 2005. Time-series validation of MODIS land biophysical products in a Kalahari woodland, Africa. Int. J. Remote Sens. 26, 4381e4398.
Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 83, 195e213.
Huete, A.R., 1988. A soil-adjusted vegeation index (SAVI). Remote Sens. Environ. 25, 295e309.
Iida, S., Tanaka, T., Sugita, M., 2005. Change of interception process due to the succession from Japanese red pine to evergreen oak. J. Hydrol 315, 154e166.
Iwata, H., Ueyama, M., Harazono, Y., Tsuyuzaki, S., Kondo, M., Uchida, M., 2011. Quick recovery of carbon dioxide exchanges in a burned black spruce forest in Interior Alaska. SOLA 7, 105e108.
Jones, H.G., Vaughan, R.A., 2010. Remote Sensing of Vegetation: Principles, Techniques, and Applications. Oxford University Press, Oxford, p. 353.
Kasischke, E.S., Stocks, B.J., 2000. Introduction. In: Kasischke, E.S., Stocks, B.J. (Eds.), Fire, Climate Change, and Carbon Cycling in the Boreal Forest. Springer, pp. 1e5. (Chapter 1).
Kasischke, E.S., Turetsky, M.R., 2006. Recent changes in the fire regime across the North American boreal regioneS patial and temporal patterns of burning across Canada and Alaska. Geo-phys. Res. Lett. 33, L09703.
Knyazikhin, Y., Martonchik, J.V.., Diner, D.J., Myneni, R.B., Verstraete, M., Pinty, B., Gobron, N., 1998a. Estimation of vegetation canopy leaf area index and fraction of absorbed photosynthetically active radiation from atmosphere-corrected MISR data. J. Geophys. Res. 103, 32239e32256.
Knyazikhin, Y., Martonchik, J.V., Myneni, R.B., Diner, D.J., Running, S.W., 1998b. Synergistic algorithm for estimating vegetation canopy leaf area index and fraction of absorbed photosynthetically active radiation from MODIS and MISR data. J. Geophys. Res. 103, 32257e32275.
Liu, H., Randerson, J.T., 2008. Interannual variability of surface energy exchange depends on stand age in a boreal forest fire chronosequence. J. Geophys. Res. 113, G01006.
Monteith, J.L., 1972. Solar radiation and productivity in tropical ecosystems. J. Appl. Ecol. 9, 747e766.
Myneni, R., Knyazikhin, Y., Glassy, J., Votava, P., Shabanov, N., 2003. User’s Guide: FPAR, LAI (ESDT: MOD15A2) 8-day Composite NASA MODIS Land Algorithm.
Myneni, R.B., Hoffman, S., Knyazikhin, Y., Privette, J.L., Glassy, J., Tian, Y., Wang, Y., Song, X., Zhang, Y., Smith, G.R., Lotsch, S., Friedl, M., Morisette, J.T., Votava, P., Nemani, R.R., Running, S.W., 2002. Global products of vegeation leaf area and fraction absorbed par from year one of modis data. Remote Sens. Environ. 83, 214e231.
Rasmussen, J., Nørremark, M., Bibby, B.M., 2007. Assessment of leaf cover and crop soil cover in weed harrowing research using digital images. Weed Res. 47, 299e310.
Rautiainen, M., ottus, M.M., Heiskanen, J., Akuja¨rvi, A., Majasalmi, T., Stenberg, P., 2011. Seasonal reflectance dynamics of common understory types in a northern European boreal for-est. Remote Sens. Environ. 115, 3020e3028.
Rocha, A.V., Shaver, G.R., 2009. Advantages of a two band EVI calculated from solar and photosynthetically active radiation fluxes. Agric. For. Meteorol. 149, 1560e1563.
Roderick, M.L., Farquhar, G.D., Berry, S.L., Noble, I.R., 2001. On the direct effect of clouds and atmospheric particles on the pro-ductivity and structure of vegetation. Oecologia 129, 21e30. Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., Harlan, J.C.,
1974. Monitoring the Vernal Advancement and Retrogradation Ggreenwave Effect) of Natural Vegetation. Tech. rep.. NASA/ GSFC Final report, Greenbelt, MD, USA.
Running, S.W., Nemani, R.R., Heinsch, F.A., Zhao, M., Reeves, M., Hashimoto, H., 2004. A continuous satellite-derived measure of global terrestrial primary production. Bioscience 54, 547e560. Steinberg, D.C., Goetz, S.J., Hyer, E.J., 2006. Validation of MODIS
FPARproducts in boreal forests of Alaska. IEEE Trans. Geosci.
Remote Sens 44, 1818e1828.
Tian, Y., Zhang, Y., Knyazikhin, Y., Myneni, R.B., Glassy, J.M., Dedieu, G., Running, S.W., 2000. Prototyping of MODIS LAI and FPAR algorithm with LASUR and LANDSAT data. IEEE Trans. Geosci. Remote Sens. 38, 2387e2401.
Tucker, C.J., 1979. Red and photographic infrared linear combina-tions for monitoring vegeation. Remote Sens. Environ. 8, 127e150.
Verhoef, W., 1984. Light scattering by leaf layers with application to canopy reflectance modeling: the SAIL model. Remote Sens. Environ. 16, 125e141.
Woebbecke, D.M., Meyer, G.E., Von Bargen, K., Mortensen, D.A., 1995. Color indices for weed identification under various soil, residue, and lighting conditions. Trans. Ameri. Soc. Agric. Eng. 38, 259e269.
Zhang, K., Kimball, J.S., Hogg, E.H., Zhao, M., Oechel, W.C., Cassano, J.J., Running, S.W., 2008. Satellite-based model detection of recent climate-driven changes in northern high-latitude vegetation productivity. J. Geophys. Res. 113, G03033. Zhao, M., Heinsch, F.A., Nemani, R.R., Running, S.W., 2005.
Im-provements of the MODIS terrestrial gross and net primary production global data set. Remote Sens. Environ. 95, 164e176. Zhao, M., Running, S.W., 2010. Drought-induced reduction in global terrestrial net primary production from 2000 through 2009. Sci-ence 329, 940e943.
