TOWARDS ROBUST FOREST LEAF AREA INDEX ... - Semantic Scholar
TOWARDS ROBUST FOREST LEAF AREA INDEX ASSESSMENT USING AN IMAGING SPECTROSCOPY SIMULATION APPROACH Wei Yao, Martin van Leeuwen, Paul Romanczyk, Dave Kelbe, Scott Brown, John Kerekes, and Jan van Aardt Rochester Institute of Technology, Chester F. Carlson Center for Imaging Science, 54 Lomb Memorial Drive, Rochester NY 14623, USA ABSTRACT Few studies have evaluated how per-pixel structural configurations could impact spectral response. This has an impact on how we assess especially large area/global ecosystems. In an effort to understand this impact of sub-pixel structural variation on large-footprint imaging spectroscopy, a simulation approach was used, which provides precise knowledge of target geometry and radiometry. We demonstrated the validity of the proposed simulation in terms of one such structural metric of interest, namely leaf area index (LAI). LAI is a key vegetation structural parameter, which has implications for predicting ecosystems’ foliar spatial distribution, health, photosynthesis, transpiration, and energy transfer. Simulated LAI measurements were validated with field data obtained from AccuPAR measurements (R2 = 0.76) and by comparison to NDVI data obtained from simulated AVIRIS imagery (R2 = 0.92 − 0.65, depending on sampling interval). These data were used to propose an appropriate sampling protocol for LAI data collection, thus providing for efficient data collection, while minimizing variability of individual measurements. These efforts will support preparatory science experiments towards understanding the phenomenology of NASA’s next-generation imaging spectrometer, HyspIRI. Index Terms— HyspIRI, AVIRIS, DIRSIG, Leaf area index, PAR 1. INTRODUCTION As part a NASA’s decadal survey strategy, planning for the Hyperspectral Infrared Imager (HyspIRI) mission is currently underway towards addressing key science questions related to the world’s ecosystems. Preparatory work is ongoing to provide antecedent science data on the anticipated HyspIRI mission and associated science products. In particular, we are investing the impact of sub-pixel structural variation on the assessment of vegetation structure via imaging spectroscopy. Thanks to the NASA HyspIRI Mission for funding (Grant No. NNX12AQ24G).
Vegetation structural parameters are related to the state and dynamics of forest function, and therefore have important implications across ecological domains. Of particular interest is leaf area index (LAI), which is the ratio of one-sided leaf area per corresponding surface area on the ground. LAI dictates the size of the plant-atmosphere interface, and thus is a key input for models predicting ecosystems’ foliar spatial distribution, health, photosynthesis, transpiration, and energy transfer. Typically, LAI is estimated through empirical relations to vegetation indices (VIs), often derived from imaging spectroscopy [1, 2]. Field data, collected using terrestrial instruments, are then used to calibrate and validate large-scale models. The majority of field-based methods for acquiring LAI utilize either the optical analysis of gap size distribution or gap fraction [3]. Both quantities can be obtained from portions of visible sky, measured via hemispherical or line-transect approaches, while gap fractions can additionally be obtained from measurements of above and below-canopy direct and diffuse radiation, which relaxes requirements on weather conditions. One such instrument of the latter type is the AccuPAR LP-80 Ceptometer, which calculates LAI from the ratio of above-canopy and below-canopy photosynthetically active radiation (PAR) [4]. Due to variability among individual measurements, uncertainty exists in terms of the number of measurements that are needed to obtain a reliable mean estimate within a forest plot. Thus, we identified a knowledge gap in terms of the appropriate field sampling protocols necessary to capture the within-pixel variation of LAI structure for a 60 m HyspIRI pixel. To investigate this, we applied the Digital Imaging and Remote Sensing Image Generation (DIRSIG) model, which is a physics-based, first-principles radiometric modeling environment for the creation of synthetic remote sensing imagery that is radiometrically, geometrically, and temporally accurate [5]. The model is designed to generate passive broadband, multispectral, imaging spectroscopy, low-light, polarized, active laser radar, and synthetic aperture radar datasets
through the integration of a suite of first-principles-based radiation propagation modules. The objectives of this research were to (i) validate the simulation of the PAR sensor using DIRSIG, and (ii) determine the ideal collection parameters required to obtain a reliable mean estimate of LAI for an 80 × 80 m forest plot. These outcomes will be used to address science questions related to the HyspIRI mission, namely, the assessment of appropriate VIs to estimate LAI. 2. METHODS 2.1. Simulated LAI measurement In order to develop a tool for measuring LAI in the simulation environment, we integrated existing theory to develop a DIRSIG tool, which computes LAI from the observed sensor radiance. Norman and Jarvis proposed a complete radiation penetration model for calculating LAI [6], however it is not suitable for computation due to complexity. A simplified version was presented by Norman et al. [7]: A · (1 − 0.47 · fb ) · LAI Ec (1) = exp 1 Eo (1 − 2·K ) · fb − 1
The first scene was based on an ash tree on the campus of the Rochester Institute of Technology (Fig. 1(a), 43◦ 50 16.900 N, 77◦ 400 49.000 W). Both above-canopy and belowcanopy PAR were simulated in this scene. This scene was used to investigate the LAI of single tree canopy. The second scene was based on the National Ecological Observatory Network (NEON)’s Domain D17 (Pacific Southwest) site 116. Site 116 is an oak savanna located in San Joaquin Experimental Range (SJER), California, USA (Fig. 1(c), 37◦ 60 43.7700 N, 119◦ 440 11.8500 W). We used this scene to determine the proper number of below-canopy PAR readings (Fig. 1(d)). Results were verified using the field and airborne data collected in the area.
(a)
(b)
where Eo is the incident (above-canopy) irradiance, Ec is the below-canopy irradiance (i.e., a combination of radiation transmitted and scattered by the vegetation canopy), LAI is the LAI of the canopy, fb is the fraction of incident PAR from the sun vs. other sources, K is the canopy extinction coefficient and A is a constant equal to 0.283 + 0.785a − 0.159a2 , where a is the leaf absorbtivity in the PAR band. The irradiance (E) can be calculated by: X E' Li cos θi Ωi (2) i
(c)
where Li is the radiance from the solid angle, Ωi , received by the detector cell i, while θi is the angle between radiance and the normal vector of the detector surface. The output of DIRSIG is Li for a simulated scene and viewing conditions. Therefore, the LAI is determined by working backwards from the observed radiance, i.e., first substituting Eq. 2 into Eq. 1, and then solving for LAI. 2.2. Study area and virtual scene To validate this simulated LAI measurement tool, we developed two virtual scenes with appropriate radiometric and geometric characteristics. Modeled trees in the virtual scenes match the height, crown size, and leaf and bark optical properties of the field-measured trees (Fig. 1(b)). While the actual number of branches and leaves may be different than the virtual modeled trees, these scenes provide generative examples of real scene structure. Moreover, the virtual scene provides for full knowledge of vegetation-structural attributes for which AccuPAR readings can be derived and analyzed.
(d) Fig. 1. Study areas and the virtual scenes. (a) An ash tree on RIT’s campus; (b) 3D model of the ash tree; (c) The site 116 in NEON’s D17 Domain; and (d) Side-view image of site 116 scene generated by DIRSIG.
2.3. Field and airborne data These virtual scenes were supplemented with corresponding real data obtained as follows. Reference LAI measurements were collected using AccuPAR instruments at site 116 in sum-
In order to assess the simulated LAI measurement tool, PAR measurements were taken both above and below the tree canopy, using AccuPAR instruments, for an ash tree (Fig. 1(a)). These were then compared to simulated data obtained from the corresponding virtual tree (Fig. 1(b)). Moreover, we extracted the normalized difference vegetation index (NDVI) from synthetic imaging data obtained from an AVIRIS-like virtual sensor, and compared it to LAI obtained from the field data collection. Although NDVI is known to have limitations, it provides an opportunity to assess correlations between an established narrow-band index and our virtual LAI. For forested environments, it is common to obtain LAI measurement along several parallel transects. The 80 × 80 m site (116) was divided into 25: 15 × 15 m pixels. For each pixel, LAI was measured in the field according to three sampling protocols: (i) 450 PAR readings along 5 minterval transects, (ii) 300 PAR readings along 10 m-interval transects, and (iii) 150 PAR readings along 15 m-interval transects. These were compared to simulated data and provided opportunity to assess the optimal sample spacing for field data collection, in order to achieve adequate mean estimates, given the variability within an 80 × 80 m forest plot.
2000
PAR (µmol m−2 s−1)
2.4. Validation of simulated LAI
R2 = 0.76). We found that the LAI readings were more uniform in the center of the shadow and that there were some locations with very low LAI because of the gaps in the canopy. The size of the largest gap is about 1 m, which is larger than the length of the AccuPAR probe. Therefore, for a single tree, multiple field-based PAR measurements along a transect within the shadow of a tree will be required.
1500
1000
Simulated PAR Measured PAR
500 6
8
10
12 Time
14
16
18
(a) 2500
PAR (µmol m−2 s−1)
mer 2013 and fall 2014. During the summer 2013 collection, terrestrial laser scanner (TLS) data were also collected to support virtual scene construction. The 2014 PAR data were collected according to the collection protocol proposed in this paper for validation. The airborne data were collected by Airborne Visible / Infrared Imaging Spectrometer (AVIRIS) on June 12, 2013 and October 6, 2014 with 15 m spatial resolution. The PAR data of the ash tree were collected using AccuPAR instruments on RIT’s campus on May 31, 2014.
Simulated PAR Measured PAR
2000 1500 1000 500 0 −4
−2
0
2
4
6
Relative distance in South−North direction (m)
(b)
3. RESULTS AND DISCUSSION
Fig. 2. Simulation results on site 116. (a) The measured and simulated above-canopy PAR (RMSE = 23.94µmol m−2 s−1 and R2 = 0.99); and (b) The measured and simulated below-canopy PAR (RMSE = 319.2µmol m−2 s−1 and R2 = 0.76).
The first experiment assessed the validity of the proposed LAI measurement tool in the simulation environment. Fig. 2(a) shows the simulated and actual above-canopy PAR measurements obtained for site 1, while Fig. 2(b) shows the simulated and actual below-canopy PAR measurements along the South-North transect for the ash tree on RIT’s campus. Note that in both cases, PAR peaks at 12h00 (noon) due to the angular projection effects of the sun’s flux on the horizontal detector element. Spikes in the plot are due to sun flecks, i.e., where sunlight reaches the detector directly through gaps in the canopy. Note that these PAR features are not directly overlapping between simulated and actual data, due to small differences in tree structure between the modeled virtual and real tree. The simulation tool was found to be a good estimator of true above-canopy PAR (RMSE = 23.94µmol m−2 s−1 and R2 = 0.99) and below-canopy PAR (RMSE = 319.2µmol m−2 s−1 and
Fig. 3 shows the simulated NDVI vs. LAI for 450 measurements of 5 m transect spacing (R2 = 0.92), 300 PAR readings of 10 m transect spacing (R2 = 0.77), and 150 measurements of 15 m transect spacing (R2 = 0.66). A 15 m-interval provides insufficient predictive ability due to the variation of the fewer available measurements. The 5 m transect, on the other hand, provided good results but is timeconsuming for collection in-field. Comparison of the 10 minterval to the 5 m-interval, R2 dropped by 15% while only requiring half the number of measurements. Therefore, the 10 m-interval may provide an appropriate sampling protocol, which balances efficiency and precision. This 10 m interval was subsequently used in real data collection as follows. Our field team collected LAI with 10 m-interval transects in three 80m plots of SJER on October 5-7, 2014 and verified results with NDVI extracted from AVIRIS data (Fig. 4). The R2 was 0.61, which is slightly higher than regression models
6
6
6
Data Model
Data Model
Data Model
4
4
LAI
LAI
LAI
4
0
2
2
2
0
0.1
0.2
0.3
0.4
0.5
0.6
0
0.7
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0
0
0.1
0.2
0.3
NDVI
NDVI
(a)
(b)
0.4
0.5
0.6
0.7
NDVI
(c)
Fig. 3. NDVI extracted from a synthetic AVIRIS image was used to verify the simulated forest LAI. (a) 450 PAR readings were simulated along three transects in each 15 × 15 m area (R2 = 0.92); (b) 300 PAR readings were simulated along two transects in each 15 × 15 m area (R2 = 0.77); and (c) 150 PAR readings were simulated along one transect in each 15 × 15 m area (R2 = 0.66). 2.5
Schaaf (UMB), Dr. Alan H. Strahler (BU) and their students. The authors want to thank Mr. Chris DeAngelis for building virtual scenes. The authors also thank the NEON airborne observation platform (AOP) team for collecting data.
Data Model
2
LAI
1.5 1 0.5 0
6. REFERENCES 0
0.1
0.2
0.3
0.4
0.5
NDVI
Fig. 4. NDVI extracted from a real AVIRIS image was used to verify the in situ forest LAI, where R2 = 0.61. proposed in other papers[1, 2]. 4. CONCLUSIONS In an effort to understand the impact of sub-pixel structural variation on large-footprint imaging spectroscopy, e.g., as obtained from HyspIRI, a simulation approach could be used, which provides absolute knowledge of target geometry and radiometry. We presented a simulation approach for measuring one structural metric of interest, LAI, using the fractional PAR ratio. This approach was validated with field data obtained from AccuPAR measurements (R2 = 0.76) and by comparison to NDVI data obtained from simulated AVIRIS imagery. Finally, an appropriate sampling protocol for LAI data collection was proposed, which provides for efficient data collection, while minimizing the variability of individual measurements. These efforts will support preparatory science experiments towards understanding the phenomenology of NASA’s next-generation imaging spectrometer, HyspIRI. 5. ACKNOWLEDGEMENTS The authors would like to thank the field team members: Ms. Ashley Miller, Mr. Terence Nicholson, Ms. Claudia Paris, and Mr. Alexander Fafard, and our collaborators: Dr. Crystal B.
[1] D. Turner et al., “Relationships between leaf area index and landsat tm spectral vegetation indices across three temperate zone sites,” Remote Sensing of Environment, vol. 70, no. 1, pp. 52–68, 1999. [2] P. Gong et al., “Estimation of forest leaf area index using vegetation indices derived from hyperion hyperspectral data,” Geoscience and Remote Sensing, IEEE Transactions on, vol. 41, no. 6, pp. 1355–1362, 2003. [3] I. Jonckheere et al., “Review of methods for in situ leaf area index determination: Part i. theories, sensors and hemispherical photography,” Agricultural and Forest Meteorology, vol. 121, no. 1, pp. 19–35, 2004. [4] “AccuPAR PAR/LAI ceptometer model LP-80 operator’s manual,” http://www.decagon.com/ education/lp-80-manual/, June 2013. [5] J. Schott et al., “An advanced synthetic image generation model and its application to multi/hyperspectral algorithm development,” Canadian Journal of Remote Sensing, vol. 25, no. 2, pp. 99–111, 1999. [6] J. Norman and P. Jarvis, “Photosynthesis in sitka spruce (picea sitchensis (bong.) carr.): V. radiation penetration theory and a test case,” Journal of Applied Ecology, pp. 839–878, 1975. [7] J. Norman and G. Campbell, Plant Physiological Ecology: Field methods and instrumentation, chapter Canopy structure, pp. 301–325, Springer Netherlands, 1989.
Vegetation structural parameters are related to the state and dynamics of forest function, and therefore have important implications across ecological domains. Of particular interest is leaf area index (LAI), which is the ratio of one-sided leaf area per corresponding surface area on the ground. LAI dictates the size of the plant-atmosphere interface, and thus is a key input for models predicting ecosystems’ foliar spatial distribution, health, photosynthesis, transpiration, and energy transfer. Typically, LAI is estimated through empirical relations to vegetation indices (VIs), often derived from imaging spectroscopy [1, 2]. Field data, collected using terrestrial instruments, are then used to calibrate and validate large-scale models. The majority of field-based methods for acquiring LAI utilize either the optical analysis of gap size distribution or gap fraction [3]. Both quantities can be obtained from portions of visible sky, measured via hemispherical or line-transect approaches, while gap fractions can additionally be obtained from measurements of above and below-canopy direct and diffuse radiation, which relaxes requirements on weather conditions. One such instrument of the latter type is the AccuPAR LP-80 Ceptometer, which calculates LAI from the ratio of above-canopy and below-canopy photosynthetically active radiation (PAR) [4]. Due to variability among individual measurements, uncertainty exists in terms of the number of measurements that are needed to obtain a reliable mean estimate within a forest plot. Thus, we identified a knowledge gap in terms of the appropriate field sampling protocols necessary to capture the within-pixel variation of LAI structure for a 60 m HyspIRI pixel. To investigate this, we applied the Digital Imaging and Remote Sensing Image Generation (DIRSIG) model, which is a physics-based, first-principles radiometric modeling environment for the creation of synthetic remote sensing imagery that is radiometrically, geometrically, and temporally accurate [5]. The model is designed to generate passive broadband, multispectral, imaging spectroscopy, low-light, polarized, active laser radar, and synthetic aperture radar datasets
through the integration of a suite of first-principles-based radiation propagation modules. The objectives of this research were to (i) validate the simulation of the PAR sensor using DIRSIG, and (ii) determine the ideal collection parameters required to obtain a reliable mean estimate of LAI for an 80 × 80 m forest plot. These outcomes will be used to address science questions related to the HyspIRI mission, namely, the assessment of appropriate VIs to estimate LAI. 2. METHODS 2.1. Simulated LAI measurement In order to develop a tool for measuring LAI in the simulation environment, we integrated existing theory to develop a DIRSIG tool, which computes LAI from the observed sensor radiance. Norman and Jarvis proposed a complete radiation penetration model for calculating LAI [6], however it is not suitable for computation due to complexity. A simplified version was presented by Norman et al. [7]: A · (1 − 0.47 · fb ) · LAI Ec (1) = exp 1 Eo (1 − 2·K ) · fb − 1
The first scene was based on an ash tree on the campus of the Rochester Institute of Technology (Fig. 1(a), 43◦ 50 16.900 N, 77◦ 400 49.000 W). Both above-canopy and belowcanopy PAR were simulated in this scene. This scene was used to investigate the LAI of single tree canopy. The second scene was based on the National Ecological Observatory Network (NEON)’s Domain D17 (Pacific Southwest) site 116. Site 116 is an oak savanna located in San Joaquin Experimental Range (SJER), California, USA (Fig. 1(c), 37◦ 60 43.7700 N, 119◦ 440 11.8500 W). We used this scene to determine the proper number of below-canopy PAR readings (Fig. 1(d)). Results were verified using the field and airborne data collected in the area.
(a)
(b)
where Eo is the incident (above-canopy) irradiance, Ec is the below-canopy irradiance (i.e., a combination of radiation transmitted and scattered by the vegetation canopy), LAI is the LAI of the canopy, fb is the fraction of incident PAR from the sun vs. other sources, K is the canopy extinction coefficient and A is a constant equal to 0.283 + 0.785a − 0.159a2 , where a is the leaf absorbtivity in the PAR band. The irradiance (E) can be calculated by: X E' Li cos θi Ωi (2) i
(c)
where Li is the radiance from the solid angle, Ωi , received by the detector cell i, while θi is the angle between radiance and the normal vector of the detector surface. The output of DIRSIG is Li for a simulated scene and viewing conditions. Therefore, the LAI is determined by working backwards from the observed radiance, i.e., first substituting Eq. 2 into Eq. 1, and then solving for LAI. 2.2. Study area and virtual scene To validate this simulated LAI measurement tool, we developed two virtual scenes with appropriate radiometric and geometric characteristics. Modeled trees in the virtual scenes match the height, crown size, and leaf and bark optical properties of the field-measured trees (Fig. 1(b)). While the actual number of branches and leaves may be different than the virtual modeled trees, these scenes provide generative examples of real scene structure. Moreover, the virtual scene provides for full knowledge of vegetation-structural attributes for which AccuPAR readings can be derived and analyzed.
(d) Fig. 1. Study areas and the virtual scenes. (a) An ash tree on RIT’s campus; (b) 3D model of the ash tree; (c) The site 116 in NEON’s D17 Domain; and (d) Side-view image of site 116 scene generated by DIRSIG.
2.3. Field and airborne data These virtual scenes were supplemented with corresponding real data obtained as follows. Reference LAI measurements were collected using AccuPAR instruments at site 116 in sum-
In order to assess the simulated LAI measurement tool, PAR measurements were taken both above and below the tree canopy, using AccuPAR instruments, for an ash tree (Fig. 1(a)). These were then compared to simulated data obtained from the corresponding virtual tree (Fig. 1(b)). Moreover, we extracted the normalized difference vegetation index (NDVI) from synthetic imaging data obtained from an AVIRIS-like virtual sensor, and compared it to LAI obtained from the field data collection. Although NDVI is known to have limitations, it provides an opportunity to assess correlations between an established narrow-band index and our virtual LAI. For forested environments, it is common to obtain LAI measurement along several parallel transects. The 80 × 80 m site (116) was divided into 25: 15 × 15 m pixels. For each pixel, LAI was measured in the field according to three sampling protocols: (i) 450 PAR readings along 5 minterval transects, (ii) 300 PAR readings along 10 m-interval transects, and (iii) 150 PAR readings along 15 m-interval transects. These were compared to simulated data and provided opportunity to assess the optimal sample spacing for field data collection, in order to achieve adequate mean estimates, given the variability within an 80 × 80 m forest plot.
2000
PAR (µmol m−2 s−1)
2.4. Validation of simulated LAI
R2 = 0.76). We found that the LAI readings were more uniform in the center of the shadow and that there were some locations with very low LAI because of the gaps in the canopy. The size of the largest gap is about 1 m, which is larger than the length of the AccuPAR probe. Therefore, for a single tree, multiple field-based PAR measurements along a transect within the shadow of a tree will be required.
1500
1000
Simulated PAR Measured PAR
500 6
8
10
12 Time
14
16
18
(a) 2500
PAR (µmol m−2 s−1)
mer 2013 and fall 2014. During the summer 2013 collection, terrestrial laser scanner (TLS) data were also collected to support virtual scene construction. The 2014 PAR data were collected according to the collection protocol proposed in this paper for validation. The airborne data were collected by Airborne Visible / Infrared Imaging Spectrometer (AVIRIS) on June 12, 2013 and October 6, 2014 with 15 m spatial resolution. The PAR data of the ash tree were collected using AccuPAR instruments on RIT’s campus on May 31, 2014.
Simulated PAR Measured PAR
2000 1500 1000 500 0 −4
−2
0
2
4
6
Relative distance in South−North direction (m)
(b)
3. RESULTS AND DISCUSSION
Fig. 2. Simulation results on site 116. (a) The measured and simulated above-canopy PAR (RMSE = 23.94µmol m−2 s−1 and R2 = 0.99); and (b) The measured and simulated below-canopy PAR (RMSE = 319.2µmol m−2 s−1 and R2 = 0.76).
The first experiment assessed the validity of the proposed LAI measurement tool in the simulation environment. Fig. 2(a) shows the simulated and actual above-canopy PAR measurements obtained for site 1, while Fig. 2(b) shows the simulated and actual below-canopy PAR measurements along the South-North transect for the ash tree on RIT’s campus. Note that in both cases, PAR peaks at 12h00 (noon) due to the angular projection effects of the sun’s flux on the horizontal detector element. Spikes in the plot are due to sun flecks, i.e., where sunlight reaches the detector directly through gaps in the canopy. Note that these PAR features are not directly overlapping between simulated and actual data, due to small differences in tree structure between the modeled virtual and real tree. The simulation tool was found to be a good estimator of true above-canopy PAR (RMSE = 23.94µmol m−2 s−1 and R2 = 0.99) and below-canopy PAR (RMSE = 319.2µmol m−2 s−1 and
Fig. 3 shows the simulated NDVI vs. LAI for 450 measurements of 5 m transect spacing (R2 = 0.92), 300 PAR readings of 10 m transect spacing (R2 = 0.77), and 150 measurements of 15 m transect spacing (R2 = 0.66). A 15 m-interval provides insufficient predictive ability due to the variation of the fewer available measurements. The 5 m transect, on the other hand, provided good results but is timeconsuming for collection in-field. Comparison of the 10 minterval to the 5 m-interval, R2 dropped by 15% while only requiring half the number of measurements. Therefore, the 10 m-interval may provide an appropriate sampling protocol, which balances efficiency and precision. This 10 m interval was subsequently used in real data collection as follows. Our field team collected LAI with 10 m-interval transects in three 80m plots of SJER on October 5-7, 2014 and verified results with NDVI extracted from AVIRIS data (Fig. 4). The R2 was 0.61, which is slightly higher than regression models
6
6
6
Data Model
Data Model
Data Model
4
4
LAI
LAI
LAI
4
0
2
2
2
0
0.1
0.2
0.3
0.4
0.5
0.6
0
0.7
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0
0
0.1
0.2
0.3
NDVI
NDVI
(a)
(b)
0.4
0.5
0.6
0.7
NDVI
(c)
Fig. 3. NDVI extracted from a synthetic AVIRIS image was used to verify the simulated forest LAI. (a) 450 PAR readings were simulated along three transects in each 15 × 15 m area (R2 = 0.92); (b) 300 PAR readings were simulated along two transects in each 15 × 15 m area (R2 = 0.77); and (c) 150 PAR readings were simulated along one transect in each 15 × 15 m area (R2 = 0.66). 2.5
Schaaf (UMB), Dr. Alan H. Strahler (BU) and their students. The authors want to thank Mr. Chris DeAngelis for building virtual scenes. The authors also thank the NEON airborne observation platform (AOP) team for collecting data.
Data Model
2
LAI
1.5 1 0.5 0
6. REFERENCES 0
0.1
0.2
0.3
0.4
0.5
NDVI
Fig. 4. NDVI extracted from a real AVIRIS image was used to verify the in situ forest LAI, where R2 = 0.61. proposed in other papers[1, 2]. 4. CONCLUSIONS In an effort to understand the impact of sub-pixel structural variation on large-footprint imaging spectroscopy, e.g., as obtained from HyspIRI, a simulation approach could be used, which provides absolute knowledge of target geometry and radiometry. We presented a simulation approach for measuring one structural metric of interest, LAI, using the fractional PAR ratio. This approach was validated with field data obtained from AccuPAR measurements (R2 = 0.76) and by comparison to NDVI data obtained from simulated AVIRIS imagery. Finally, an appropriate sampling protocol for LAI data collection was proposed, which provides for efficient data collection, while minimizing the variability of individual measurements. These efforts will support preparatory science experiments towards understanding the phenomenology of NASA’s next-generation imaging spectrometer, HyspIRI. 5. ACKNOWLEDGEMENTS The authors would like to thank the field team members: Ms. Ashley Miller, Mr. Terence Nicholson, Ms. Claudia Paris, and Mr. Alexander Fafard, and our collaborators: Dr. Crystal B.
[1] D. Turner et al., “Relationships between leaf area index and landsat tm spectral vegetation indices across three temperate zone sites,” Remote Sensing of Environment, vol. 70, no. 1, pp. 52–68, 1999. [2] P. Gong et al., “Estimation of forest leaf area index using vegetation indices derived from hyperion hyperspectral data,” Geoscience and Remote Sensing, IEEE Transactions on, vol. 41, no. 6, pp. 1355–1362, 2003. [3] I. Jonckheere et al., “Review of methods for in situ leaf area index determination: Part i. theories, sensors and hemispherical photography,” Agricultural and Forest Meteorology, vol. 121, no. 1, pp. 19–35, 2004. [4] “AccuPAR PAR/LAI ceptometer model LP-80 operator’s manual,” http://www.decagon.com/ education/lp-80-manual/, June 2013. [5] J. Schott et al., “An advanced synthetic image generation model and its application to multi/hyperspectral algorithm development,” Canadian Journal of Remote Sensing, vol. 25, no. 2, pp. 99–111, 1999. [6] J. Norman and P. Jarvis, “Photosynthesis in sitka spruce (picea sitchensis (bong.) carr.): V. radiation penetration theory and a test case,” Journal of Applied Ecology, pp. 839–878, 1975. [7] J. Norman and G. Campbell, Plant Physiological Ecology: Field methods and instrumentation, chapter Canopy structure, pp. 301–325, Springer Netherlands, 1989.
