Ecological Indicators

Ecological Indicators 73 (2017) 378–387 Contents lists available at ScienceDirect Ecological Indicators journal homepage: www.elsevier.com/locate/ec...
Author: Delilah Flowers
1 downloads 0 Views 2MB Size
Ecological Indicators 73 (2017) 378–387

Contents lists available at ScienceDirect

Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind

Fusion of airborne LiDAR data and hyperspectral imagery for aboveground and belowground forest biomass estimation Shezhou Luo a,b , Cheng Wang a,∗ , Xiaohuan Xi a , Feifei Pan c , Dailiang Peng a , Jie Zou d , Sheng Nie a , Haiming Qin a a

Key Laboratory of Digital Earth Science, Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100094, China Department of Geography and Program in Planning, University of Toronto, Toronto, ON M5S 3G3, Canada Department of Geography and the Environment, University of North Texas, Denton, TX 76203, USA d Key Laboratory of Data Mining and Information Sharing, Ministry of Education, Spatial Information Research Center of Fujian Province, Fuzhou University, Fuzhou 350002, China b c

a r t i c l e

i n f o

Article history: Received 12 February 2016 Received in revised form 30 July 2016 Accepted 3 October 2016 Keywords: Airborne LiDAR Aboveground biomass Belowground biomass Hyperspectral imagery Fusion Partial least squares regression

a b s t r a c t Vegetation biomass is a key biophysical parameter for many ecological and environmental models. The accurate estimation of biomass is essential for improving the accuracy and applicability of these models. Light Detection and Ranging (LiDAR) data have been extensively used to estimate forest biomass. Recently, there has been an increasing interest in fusing LiDAR with other data sources for directly measuring or estimating vegetation characteristics. In this study, the potential of fused LiDAR and hyperspectral data for biomass estimation was tested in the middle Heihe River Basin, northwest China. A series of LiDAR and hyperspectral metrics were calculated to obtain the optimal biomass estimation model. To assess the prediction ability of the fused data, single and fused LiDAR and hyperspectral metrics were regressed against field-observed belowground biomass (BGB), aboveground biomass (AGB) and total forest biomass (TB). The partial least squares (PLS) regression method was used to reduce the multicollinearity problem associated with the input metrics. It was found that the estimation accuracy of forest biomass was affected by LiDAR plot size, and the optimal plot size in this study had a radius of 22 m. The results showed that LiDAR data alone could estimate biomass with a relative high accuracy, and hyperspectral data had lower prediction ability for forest biomass estimation than LiDAR data. The best estimation model was using a fusion of LiDAR and hyperspectral metrics (R2 = 0.785, 0.893 and 0.882 for BGB, AGB and TB, respectively, with p < 0.0001). Compared with LiDAR metrics alone, the fused LiDAR and hyperspectral data improved R2 by 5.8%, 2.2% and 2.6%, decreased AIC value by 1.9%, 1.1% and 1.2%, and reduced RMSE by 8.6%, 7.9% and 8.3% for BGB, AGB and TB, respectively. These results demonstrated that biomass accuracies could be improved by the use of fused LiDAR and hyperspectral data, although the improvement was slight when compared with LiDAR data alone. This slight improvement could be attributed to the complementary information contained in LiDAR and hyperspectral data. In conclusion, fusion of LiDAR and other remotely sensed data has great potential for improving biomass estimation accuracy. © 2016 Elsevier Ltd. All rights reserved.

1. Introduction Vegetation biomass is a key biophysical parameter, which provides important information about the growth, health and productivity of vegetation ecosystems (Dubayah and Drake, 2000; Ediriweera et al., 2014). As an important input variable for carbon stock estimation, climate change and ecological modeling, vegetation biomass estimation has been increasingly studied (Lucas et al.,

∗ Corresponding author. E-mail address: [email protected] (C. Wang). http://dx.doi.org/10.1016/j.ecolind.2016.10.001 1470-160X/© 2016 Elsevier Ltd. All rights reserved.

2008; Næsset et al., 2013). The output accuracy of these models is closely related to estimation accuracy of vegetation biomass. Therefore, the accurate estimation of vegetation biomass is essential for improving the accuracy and applicability of these models. Vegetation biomass is defined as total dry weight of organic material both aboveground and belowground. The aboveground biomass (AGB) is the sum of the dry weights of flowers, fruits, branches, stems, foliage and bark above the ground surface (Kajimoto et al., 1999). The belowground biomass (BGB) is the dry weight of all living biomass of live roots (Kajimoto et al., 1999). The total biomass (TB) is the sum of AGB and BGB. Direct methods for measuring biomass are the most accurate and reliable; however, a

S. Luo et al. / Ecological Indicators 73 (2017) 378–387

large amount of field-observed data is acquired, including destructive sampling and non-destructive field measurements (Englhart et al., 2011). Therefore, direct observation methods are only practical in a relatively small area because, for large study areas, they are expensive, time-consuming and laborious (García et al., 2010; Houghton, 2005). However, remote sensing technologies provide a promising approach to rapidly and periodically estimate vegetation biomass over large areas (Sun et al., 2011). Numerous studies have been conducted to estimate vegetation biomass using optical remotely sensed data (Chen et al., 2009; Foody et al., 2003) and radar data (e.g., Gao et al., 2013; Mitchard et al., 2009). Vegetation biomass cannot be directly acquired using remote sensing techniques; it is estimated through the regression relationships between vegetation biomass and vegetation parameters derived from remotely sensed data. However, one problem for traditional optical remote sensing is saturation in the estimation, a common phenomenon in study areas with high biomass or high canopy density (when the leaf area index reaches a certain level); this can lower the estimation accuracy of vegetation biomass. The saturation problem is still common in radar data, although radar data are more sensitive to biomass at high biomass density (Mitchard et al., 2009; Nelson et al., 2007). However, Light Detection and Ranging (LiDAR) is an efficient remote sensing technology (Lin and West, 2016) and has been widely used for estimating vegetation biomass, because LiDAR is able to overcome the saturation problem and improve the estimation accuracy of biomass (Clark et al., 2011; He et al., 2013; Lu et al., 2012; Swatantran et al., 2011). LiDAR is an active remote sensing technology and can quickly and accurately acquire three-dimensional information concerning the earth’s surface (Lefsky et al., 2002; Qin et al., 2015). Laser pulses from LiDAR systems can penetrate through a forest canopy to the ground, and can therefore reliably measure forest canopy and subcanopy height (Listopad et al., 2015; Thomas et al., 2006). LiDAR data have been extensively used for estimating vegetation height (e.g., Glenn et al., 2011; Hopkinson et al., 2006; Wang et al., 2009) and leaf area index (LAI) (e.g., Luo et al., 2015; Nie et al., 2016; Olsoy ˜ et al., 2004; Richardson et al., 2009; Zhao and et al., 2016; Riano Popescu, 2009). Previous studies have shown that LiDAR systems are a promising technology for accurately estimating vegetation biomass, and many studies have obtained reliable estimates of vegetation biomass using LiDAR data (Ahmed et al., 2013; Chen et al., 2016; Kronseder et al., 2012; Li et al., 2015; Lu et al., 2012; Popescu et al., 2011; Zhao et al., 2009). In recent years, fusion of LiDAR and other data has been widely used for vegetation biomass research (Zolkos et al., 2013). Moreover, the results have demonstrated that estimation accuracy of biomass could be improved by fusing LiDAR and other data sources (Ediriweera et al., 2014; Sarrazin et al., 2011; Tsui et al., 2012; Vaglio Laurin et al., 2014). These improvements are mainly due to complementary information of LiDAR and other data (Koch, 2010). Hyperspectral sensors can provide fine spectral resolution bands and abundant spectral information about the Earth’s surface (Chen et al., 2009). Hyperspectral data have been widely used for vegetation species classification as well as biophysical and biochemical parameters estimation (Chen et al., 2009; Dalponte et al., 2012; Thomas et al., 2006). In particular, the use of narrow band vegetation indices derived from hyperspectral data can reduce the effects of background soil reflectance, atmospheric and water absorption and the saturation problem of broadband vegetation indices (Chen et al., 2009; Swatantran et al., 2011). Thus, hyperspectral data can provide complementary spectral information of vegetation characterization to the canopy structural information derived from LiDAR data (Swatantran et al., 2011). Most previous studies using fused LiDAR and hyperspectral data have primarily focused on land cover and forest species classification (Geerling et al., 2007; Ghamisi et al., 2015; Koetz et al., 2008; Luo et al.,

379

2016b). However, only a few studies have been conducted using fused LiDAR and hyperspectral data for estimating forest biomass (e.g., Anderson et al., 2008; Clark et al., 2011; Swatantran et al., 2011; Vaglio Laurin et al., 2014). Anderson et al. (2008) estimated forest AGB using fused LiDAR (LVIS) and AVIRIS hyperspectral data, and the results showed improvements of 8–9% for AGB estimation across all forest conditions and 25% or more for the unmanaged forest. Similarly, Vaglio Laurin et al. (2014) found that the estimation accuracies of forest biomass were improved (6%) using integrated LiDAR and hyperspectral data compared with LiDAR data alone. Their results indicated that forest biomass estimation could be improved using integrated LiDAR and hyperspectral data. However, further research efforts are needed to explore the potential of fused LiDAR and hyperspectral data for estimating biomass over different geographical environments, vegetation types and LiDAR data. Previous studies showed that airborne LiDAR data could reliably estimate belowground biomass (Cao et al., 2014; Næsset and Gobakken, 2008); however, no study has been performed on belowground biomass estimation using fused LiDAR and hyperspectral data. Scale is a fundamental and crucial issue in remote sensing studies and image analysis (Weng, 2014), especially in the application of passive optical remote sensing (Woodcock and Strahler, 1987; Wu and Li, 2009). Similarly, use of LiDAR data is also affected by scale. Plot sizes of LiDAR data and field observations have a strong effect on the estimation accuracy of biomass. However, most previous studies have concentrated on the effect of variable field plot size on the biomass estimation accuracy (Frazer et al., 2011; HernándezStefanoni et al., 2014; Mauya et al., 2015; Ruiz et al., 2014), and only a few studies have focused on the effect of variable LiDAR plot size on biomass estimation accuracy (Estornell et al., 2011; Hayashi et al., 2015; Strunk et al., 2012). The use of variable plot size for LiDAR metrics extraction can be beneficial for obtaining accurate biomass estimations. Therefore, it is essential to explore the effect of LiDAR plot size on biomass estimation accuracy. The main goal of this study was to explore the potential of the fused LiDAR and hyperspectral data to estimate forest BGB, AGB and TB. The specific objectives of this study were to: 1) establish biomass prediction models using LiDAR- and/or hyperspectralderived metrics and field-observed biomass; 2) determine the optimal LiDAR plot size for estimating forest biomass in our study; and 3) assess the potential of fused LiDAR and hyperspectral data for improving biomass estimation accuracy. 2. Materials and methods 2.1. Study area The study area was located in the middle Heihe River Basin in Zhangye City of Gansu Province northwest China (38◦ 56 15 N to 38◦ 59 46 N, 100◦ 22 36 E to 100◦ 28 10 E). This study was one research component of the Heihe Watershed Allied Telemetry Experimental Research (HiWATER) (Li et al., 2013). The main tree species in Zhangye City are poplar (Populus spp.), willow (Salix spp.), spruce (Picea asperata Mast.) and birch (Betula platyphylla Suk. L.). In this study, we mainly estimated the biomass of poplar. 2.2. Field observations Field data collection was carried out on July 10th and 17th, 2012. The field plot was a circular area with a radius of 15 m. In each plot, the tree height (H, m) and diameter at breast height (DBH, cm) of all trees were measured. The coordinates of all plot centers were measured using a Trimble real time kinematic (RTK) GPS (Trimble Navigation Ltd.). A total of 33 plots were measured, and three

380

S. Luo et al. / Ecological Indicators 73 (2017) 378–387

Table 1 Summary statistics of the field-measured forest parameters at the plot level. H represents tree height, and DBH represents the diameter at breast height. BGB, AGB and TB represent belowground, aboveground and total biomass, respectively. Biophysical parameter

Minimum

Maximum

Mean

Standard deviation

Mean H (m) Mean DBH (cm) BGB (Mg/ha) AGB (Mg/ha) TB (Mg/ha)

4.71 2.62 6.009 16.367 23.103

19.58 52.76 35.131 199.746 231.150

12.48 20.96 19.027 100.534 119.561

4.23 10.93 8.025 45.785 52.966

The horizontal accuracy of LiDAR data was 0.14 m, which was provided by the data provider. To assess the vertical accuracy of laser point cloud elevations, 36 GPS coordinates were measured using a 1 cm level RTK-GPS in the relatively flat non-vegetated areas. GPS elevations were compared with the raw laser point cloud elevations (see Luo et al. (2015) for more details). The mean error (ME) (i.e., vertical bias) and root mean squared error (RMSE) were calculated using Eqs. (1) and (2), respectively.

Table 2 Hyperspectral CASI (Compact Airborne Spectrographic Imager) data acquisition parameters in this study. Parameter

Specification

Flying altitude Swath width Number of spectral bands Spatial resolution Spectral resolution Field of view Spectral range

2000 m 1500 m 48 1.0 m 7.2 nm 40◦ 380–1050 nm

of them contained dense young trees. Allometric equations (e.g., Basuki et al., 2009; Parresol, 2001), which have been widely used to estimate biomass of a single tree, are often developed by regression analysis between H, DBH, DBH2 or DBH2 × H and actual biomass harvested from forest. In this study, the AGB and BGB of each tree were calculated using the allometric equations developed by Cheng (2007). The biomass density was calculated using the total biomass in each plot (sum of single tree biomass in the field plot) divided by the area of field plot. The summary statistics of field-measured forest parameters for the 33 plots are listed in Table 1.

1 (ZLiDARi − ZGPSi ) n n

ME =

(1)

i=1

  n 1 RMSE =  (ZLiDARi − ZGPSi )2 n

(2)

i=1

where ZLiDARi is laser point elevation of ith GPS location, ZGPSi is the ith elevation GPS-measured, n is the number of test points. 2.5. Hyperspectral metrics for biomass estimation Many vegetation indices (VIs) have been developed and applied in vegetation research. Vegetation indices have been widely used to estimate biomass using empirical relationships with biomass (Foody et al., 2003). With the development of hyperspectral remote sensing, narrow band vegetation indices have been increasingly used to estimate biomass. Compared with broadband vegetation indices, narrow band vegetation indices can reduce the effect of saturation problem and improve the estimate of biomass (Chen et al., 2009; Mutanga and Skidmore, 2004). In this study, ten narrow band vegetation indices calculated from CASI data (see Table 3) were used to estimate biomass. The  in Table 3 represents the reflectance at a specific wavelength in nm.

2.3. Hyperspectral data Airborne hyperspectral data were acquired by the Compact Airborne Spectrographic Imager (CASI) on June 29th, 2012. The CASI data covered the spectral range from 380 nm to 1050 nm with a 1 m spatial resolution. The specific parameters of the CASI data are listed in Table 2. The CASI images were atmospherically corrected using the FLAASH (Fast-Line-of-sight Atmospheric Analysis of Spectral Hypercube) module of the ENVI software (www.exelisvis.com) to reduce atmospheric effects (Dalponte et al., 2012). The manual co-registration of CASI data was performed using the ENVI software and the co-registration error was less than 1 pixel. 2.4. LiDAR data LiDAR data were acquired on July 19th, 2012, using a Leica Airborne Laser Scanner (ALS70) (Xiao and Wen, 2014). The maximum scan angle was 18◦ with a 60% flight-line overlap. The average point density of the LiDAR data over the study area was 6.7 points/m2 with an average post spacing of 0.39 m. The laser point clouds were preprocessed and then classified into ground and non-ground points using LiDAR data post processing software (TerraScan, TerraSolid, Ltd., Finland). In this study, the ground point elevations of the LiDAR data were interpolated into a digital terrain model (DTM) with a 1 m spatial resolution using a triangulated irregular network (TIN) interpolation method. The relative heights of laser points were calculated by subtracting corresponding elevations of the DTM from all laser points (Næsset and Gobakken, 2008). This procedure removed the effect of topography and produced vegetation heights relative to the ground surface (Kulawardhana et al., 2014). All LiDAR metrics were calculated based on these relative heights of laser points.

2.6. LiDAR metrics for biomass estimation Commonly used LiDAR metrics for estimating vegetation biomass include the percentiles, maximums, means and standard deviations of LiDAR heights and the laser intercept index (LII) (e.g., Ahmed et al., 2013; Hudak et al., 2012). We tested a range of LiDAR metrics (Table 4) in this study to obtain an optimal biomass estimation model. The simplest method to extract point cloud for calculating LiDAR metrics is to use the same size as the field plot. However, this method may not yield the best estimations because the plot size of LiDAR data has a significant effect on the estimation accuracy of vegetation parameters (Estornell et al., 2011; Jochem et al., 2011; Morsdorf et al., 2006). The estimation accuracies vary with different LiDAR plot sizes. To identify the optimal LiDAR plot size leading to the best estimation accuracy, different LiDAR plot sizes need to be tested. Some studies have investigated the effect of variable sampling size of LiDAR data on vegetation parameter estimation accuracy (e.g., Hayashi et al., 2015; Luo et al., 2016a; Richardson et al., 2009; Strunk et al., 2012). The results showed that using variable plot sizes for LiDAR metrics extraction could be beneficial to accurate estimations of vegetation parameters. Therefore, variableradius plot sizes (from 11.0 m to 30.0 m incremented by 1.0 m) were used to extract the LiDAR metrics in this study. Moreover, a height threshold for separating canopy returns from ground returns was used to reduce the influence of undergrowth (Mauya et al., 2015). The LiDAR variable LII (see Table 4) for describing canopy cover was calculated using Eq. (3). LII =

Ncanopy (h) Nall

(3)

S. Luo et al. / Ecological Indicators 73 (2017) 378–387

381

Table 3 Narrow-band vegetation indices and equations used in this study. Variables

Equations

Normalized difference vegetation index (NDVI) Modified Normalized difference vegetation index (MNDVI) Simple ratio vegetation index (SRVI1) Simple ratio vegetation index (SRVI2) Atmospherically resistant vegetation index (ARVI) Soil-adjusted vegetation index (SAVI) Optimization of soil-adjusted vegetation index (OSAVI) Modified soil-adjusted vegetation index (MSAVI) Enhanced vegetation index (EVI)

797.7 797.7 754.9 754.9 797.7 669.1 754.9 712 797.7 797.7

References

− 669.1 + 669.1 − 740.6 + 740.6

Rouse et al. (1973) and Tucker (1979) Mutanga and Skidmore (2004) Jordan (1969) Mutanga and Skidmore (2004)

− RB RB = 669.1 − (454.4 − 669.1 )( = 1.0) + RB 797.7 − 669.1 (1 + L) (L = 0.5) 797.7 + 669.1 + L 797.7 − 669.1 (1 + L) (L = 0.16) 797.7 + 669.1 + L  2797.7 + 1 −

(2797.7 + 1) − 8 (797.7 − 669.1 )

2 797.7 − 669.1 (G = 2.5,C1 = 6,C2 = 7.5,L = 1.0) 797.7 + C1 669.1 − C2 454.4 + L i − 697.7 783.4 + 669.1 697.7 + × 42.9, i = 740.6 − 697.7 2

Table 4 LiDAR-derived metrics for estimating biophysical parameters. LiDAR metrics

Description

H max H mean H sd H cv H p (30, 40, 50, 60, 70, 80, 90, 95, 99) LII

Maximum of LiDAR height Mean of LiDAR heigh Standard deviation of LiDAR height Coefficient of variation of LiDAR height Percentile of LiDAR height Laser intercept index (canopy returns/total returns), a description of fractional canopy cover

where LII is the laser intercept index, Ncanopy is the number of canopy returns in which LiDAR heights are greater than h m, Nall is the total number of returns, and h is height threshold (in this study, h = 1.0 m). 2.7. Statistical analyses and modeling In this study, three types of datasets were used to estimate forest biomass, i.e., LiDAR data, VIs, and LiDAR + VIs. Moreover, log-transformed biomass values and predictors were tested. The biomass estimation results from single metrics and fused metrics were compared and analyzed to assess the potential of the combination of LiDAR and hyperspectral data for estimating biomass. Moreover, the optimal LiDAR plot size was obtained through comparison of all biomass prediction models with different LiDAR plot sizes. The LiDAR and hyperspectral metrics have multicollinearity problems because of high correlations among the metrics (Næsset et al., 2005; Vaglio Laurin et al., 2014). Partial least squares (PLS) regression is a multivariate statistical method, closely related to principal components regression (PCR) (Næsset and Gobakken, 2008). PLS regression can effectively resolve small samples and multicollinearity problems that are often faced in multiple linear regression (Chen et al., 2009). It has been increasingly used to estimate vegetation biomass (e.g., Chen et al., 2009; Cho et al., 2007; Duncanson et al., 2015). For PLS regression, overfitting may occur if too many latent variables are used in the prediction model (Duncanson et al., 2015). To avoid overfitting of PLS regressions, a cross-validation method is commonly used to determine the optimal number of latent variables (Næsset et al., 2005). In this study, we selected the optimal number of latent variables using the leaveone-out cross-validation (LOOCV) method. Moreover, LOOCV was also used as an accuracy indicator of prediction model in our study. The Variable Importance in the Projection (VIP) shows the contribution of each predictor variable (x-variable) in fitting the PLS model,

Huete et al. (1994) Rondeaux et al. (1996)

2

G

Red edge inflection point (REIP)

Kaufman and Tanre (1992)

Qi et al. (1994) Huete et al. (1994) Guyot et al. (1988)

and variables with VIP values of more than 0.8 were considered (Luedeling and Gassner, 2012 Wold, 1995). For all biomass estimation models, the R2 , adjusted R2 (adj.R2 ), RMSE and Akaike information criterion (AIC) were calculated to assessed the established models. The AIC provides a simple, effective, and objective method for selecting an estimated model (Burnham and Anderson, 2002). The AIC value was calculated using Eq. (4). The AIC with lower value indicates a more parsimonious and better fitting model (Kelloway, 2014). In addition, the relative RMSE (RMSEr) was calculated using Eq. (5), which accounted for the difference in biomass magnitude (Zolkos et al., 2013). Because there were no additional data for validating the predictive power of the models developed in this study, the LOOCV method was also used to assess the biomass prediction models. LOOCV is an efficient method for assessing estimation models if independent validation data are not available (Duncanson et al., 2015). The RMSE values based on the LOOCV (RMSEcv ) were calculated using Eq. (6). Low RMSEcv values indicated that the model had good prediction ability (Jensen et al., 2008). The estimation accuracies were assessed based on these accuracy indicators. AIC = 2n − 2 ln(maximum likelihood)

(4)

where n is the number of parameters. RMSEr =

RMSE y¯

(5)

where y¯ represents the mean of the field observed biomass.

  n  2  (ˆyi − yi ) 

RMSEcv =

i=1

n

(6)

where yi represents the field-observed biomass of sample i, y ˆi represents the predicted biomass of sample i, and n represents the total number of samples. 3. Results Vertical accuracy of the LiDAR data was validated using the RTK-GPS measured elevation. The negative bias (ME = −0.01 m) indicated that on the flat bare ground LiDAR measured elevations were on average slightly lower than the GPS-measured elevations. The low RMSE value (0.09 m) showed that LiDAR points could accurately measure ground elevations. The horizontal accuracy of LiDAR data was 0.14 m, which was provided by the data provider. Therefore, LiDAR data in this study were reliable. The center coordinates of each field plot were measured using a centimeter-level RTK-GPS

Ten VIs derived from CASI were used to estimate biomass by the PLS regression. Similarly, all VIs with VIP values greater than 0.8 were considered as the inputs in the PLS model. However, CASI-metrics only explained 51.3% (RMSE = 5.691 Mg/ha, AIC = 224.3), 54.5% (RMSE = 31.361 Mg/ha, AIC = 336.9) and 55.2% (RMSE = 36.016 Mg/ha, AIC = 346.1) of the variance in BGB, AGB and TB estimations, respectively (Table 5), and all p-values were less than 0.0001. Fig. 4 shows the scatterplots of field-observed versus predicted biomass using CASI data alone. Compared with LiDAR data, CASI data had a significantly weaker relationship with the

RMSEcv

309.8 346.1 306.0 20.695 36.590 18.877 16.86 30.12 15.46

RMSEr RMSE

20.154 36.016 18.479 0.855 0.538 0.878

adj.R2 R2

0.860 0.552 0.882 296.7 336.9 293.3

Total biomass

AIC RMSEcv

16.946 31.882 15.555 16.46 31.19 15.16 16.545 31.361 15.245 0.869 0.531 0.889 0.874 0.545 0.893 205.3 224.3 201.4 4.261 5.824 3.886 21.76 29.91 19.89 4.141 5.691 3.784 0.734 0.497 0.778 0.742 0.513 0.785

3.2. Biomass estimates from CASI data

LiDAR-metrics CASI-metrics LiDAR- and CASI-metrics

Biomasses were estimated by the PLS regression method using the LiDAR-derived metrics alone. The VIP was calculated using the PLS regression, and Fig. 1 shows the VIP values of all LiDAR metrics. In our study, all biomass estimations were performed based on LiDAR metrics with VIP values greater than 0.8. These metrics included H max, H p50, H p60, H p70, H p80, H p90, H p95, H p99 and LII. The effects of 20 different LiDAR plot sizes (from 11.0 m to 30.0 m incremented by 1.0 m) on estimation accuracy were studied. Fig. 2 illustrates the estimation accuracies of BGB, AGB and TB with different LiDAR plot radii. The results showed that the estimation accuracy of biomass was affected by the LiDAR plot size, and the optimal LiDAR plot size for estimating biomass in this study had a 22 m radius. The biggest differences in R2 values for BGB, AGB and TB estimation in 20 different LiDAR plot sizes were 14.6%, 8.5% and 9.2%, respectively. The highest estimation accuracies of BGB, AGB and TB using LiDAR-metrics, CASI-metrics, and a combination of LiDAR- and CASI-metrics are listed in Table 5. For LiDAR data alone, the highest R2 values for BGB, AGB and TB were 0.742 (RMSE = 4.141 Mg/ha, AIC = 205.3, p < 0.0001), 0.874 (RMSE = 16.545 Mg/ha, AIC = 296.7, p < 0.0001) and 0.860 (RMSE = 20.154 Mg/ha, AIC = 309.8, p < 0.0001), respectively. Fig. 3 shows the scatterplots of field-observed versus predicted biomass, with the highest estimation accuracy using the LiDAR data alone for estimation. The results showed a strong relationship between the field-observed biomass and LiDAR-metrics. Therefore, the LiDAR-metrics alone can be used to reliably estimate forest biomass by PLS regression.

RMSEr

3.1. Biomass estimates from LiDAR data

RMSE

with a horizontal accuracy of 0.012 m and a vertical accuracy of 0.01 m (Luo et al., 2014). Therefore, the field measurements and the LiDAR data matched well.

adj.R2

Fig. 1. VIP values of LiDAR-derived metrics for biomass estimation.

R2

LiDAR-derived metrics

Aboveground biomass

0.0

AIC

0.2

RMSEcv

0.4

RMSEr

0.6

RMSE

0.8

adj.R2

VIP

1.0

R2

1.2

Belowground biomass

1.4

Input variables

1.6

AIC

S. Luo et al. / Ecological Indicators 73 (2017) 378–387 Table 5 Estimation accuracies of biomass using LiDAR-metrics alone, CASI-metrics alone, and fused LiDAR- and CASI-metrics. R2 and RMSE (Mg/ha) represent the coefficient of determination and root mean square error, respectively. RMSEr (%) and adj.R2 represent relative RMSE and adjusted R2 . RMSEcv (Mg/ha) represents the RMSE obtained using the LOOCV method. AIC represents the value of Akaike Information Criterion.

382

S. Luo et al. / Ecological Indicators 73 (2017) 378–387

0.76

5.2

0.64

4.6 4.4 R² RMSE

0.56 11

13

15

17

19 21 23 Plot radius (m)

25

27

4.0

29 22 (b)

0.86

20

0.84

19

0.82

18

0.80

17 R²

0.78

RMSE

0.76 11

13

15

17

19 21 23 Plot radius (m)

25

27

16 15

27

0.86

26 25

0.84

24

0.82

23 22

0.80 R²

0.78

RMSE

0.76 11

13

15

17

19 21 23 Plot radius (m)

25

27

21

RMEScv (Mg/ha)

(c)

30 25 20 15 10

0

29

0.88

(a)

5

21

0

RMEScv (Mg/ha)

0.88

R2

4.2

5

10

15

20

25

30

35

40

Predicted biomass (Mg/ha) 210

R² = 0.874 RMSE = 16.545 (Mg/ha) RMSEr = 16.46%

180

(b)

150 120 90 60 30 0

20

0

30

60

90

120

150

180

210

Predicted biomass (Mg/ha)

19

29

Fig. 2. Estimation accuracies of biomass using LiDAR-metrics alone, with different LiDAR plot radii (from 11 m to 30 m), (a) BGB, (b) AGB, and (c) TB. R2 and RMSE represent the coefficient of determination and root mean square error, respectively.

field-observed biomass in this study. Thus, the VIs derived from CASI had limited prediction ability in estimating biomass.

3.3. Biomass estimates using a fusion of LiDAR and CASI data According to Table 5, a fusion of LiDAR and CASI metrics produced the best estimation accuracies among three types of input variables (LiDAR, CASI, and LiDAR + CASI metrics), i.e., R2 values were 0.785 (RMSE = 3.784 Mg/ha, AIC = 201.4), 0.893 (RMSE = 15.245 Mg/ha, AIC = 293.3) and 0.882 (RMSE = 18.479 Mg/ha, AIC = 306.0) for BGB, AGB and TB, respectively, with p < 0.0001. Compared with the results from single LiDAR data and CASI data, the fused data produced the highest R2 values and the lowest AIC values. The scatterplot of field-observed versus predicted biomass from the fusion of LiDAR and CASI metrics is shown in Fig. 5. The results showed that estimation accuracies of BGB, AGB and TB could be improved using fused LiDAR and hyperspectral data. In this study, we found that log-transformed data could not improve forest biomass estimation accuracy.

R² = 0.860 RMSE = 20.154 (Mg/ha) RMSEr = 16.86%

240

Field-observed biomas (Mg/ha)

0.60

Field-observed biomas (Mg/ha)

R2

4.8

Field-observed biomas (Mg/ha)

5.0 0.68

R² = 0.742 RMSE = 4.141 (Mg/ha) RMSEr = 21.76%

35

RMEScv (Mg/ha)

0.72

R2

40

5.4 (a)

383

210

(c)

180 150 120 90 60 30 0 0

30

60

90

120

150

180

210

240

Predicted biomass (Mg/ha) Fig. 3. Scatterplots of field-observed versus predicted biomass. The biomass was predicted using LiDAR data alone by PLS regression (plot radius = 22 m, n = 33 samples), (a) BGB, (b) AGB, and (c) TB. The dotted lines represent the 1:1 line.

384

S. Luo et al. / Ecological Indicators 73 (2017) 378–387

40 R² = 0.513 RMSE = 5.691 (Mg/ha) RMSEr = 29.91%

30

24

18

12

6

35

6

12

18

24

30

25 20 15 10 5 0

36

0

5

10

Predicted biomass (Mg/ha) 210

(b)

Field-observed biomass (Mg/ha)

Field-observed biomass (Mg/ha)

180 150 120 90 60 30

25

30

R² = 0.893 RMSE = 15.245 (Mg/ha) RMSEr = 15.16%

180

35

40

(b)

150 120 90 60 30

30

0

60

90

120

150

180

0

210

30

60

240 R² = 0.552 RMSE = 36.016 (Mg/ha) RMSEr = 30.12%

(c)

120

150

R² = 0.882 RMSE = 18.479 (Mg/ha) RMSEr = 15.46%

240

Field-observed biomass (Mg/ha)

210

90

180

210

Predicted biomas (Mg/ha)

Predicted biomass (Mg/ha)

Field-observed biomass (Mg/ha)

20

0

0

180 150 120 90 60 30 0

15

Predicted biomas (Mg/ha)

210 R² = 0.545 RMSE = 31.361 (Mg/ha) RMSEr = 31.19%

(a)

30

0 0

R² = 0.785 RMSE = 3.784 (Mg/ha) RMSEr = 19.89%

(a)

Field-observed biomass (Mg/ha)

Field-observed biomass (Mg/ha)

36

210

(c)

180 150 120 90 60 30 0

0

30

60

90

120

150

180

210

240

Predicted biomass (Mg/ha) Fig. 4. Scatterplots of field-observed versus predicted biomass. The biomass was predicted using hyperspectral CASI data by PLS regression (n = 33 samples), (a) BGB, (b) AGB, and (c) TB. The dotted lines represent the 1:1 line.

0

30

60

90

120

150

180

210

240

Predicted biomas (Mg/ha) Fig. 5. Scatterplots of field-observed versus predicted biomass. The biomass was predicted using the fusion of LiDAR and hyperspectral CASI metrics by PLS regression (n = 33 samples), (a) BGB, (b) AGB, and (c) TB. The dotted lines represent the 1:1 line.

S. Luo et al. / Ecological Indicators 73 (2017) 378–387

4. Discussion To obtain the optimal LiDAR plot size for estimating biomass in this study, a range of plot sizes or different radii were tested. The results showed that the estimation accuracies of biomass varied with the LiDAR plot sizes (Fig. 2). Therefore, the LiDAR plot size was an important factor affecting LiDAR-forest biomass estimation. Similar findings were reported by Estornell et al. (2011). In our study, the optimal LiDAR plot size for estimating BGB, AGB and TB had a radius of 22 m. However, the optimal LiDAR plot size for estimating biomass was different for different study areas and vegetation types (Estornell et al., 2011; Hayashi et al., 2015). Thus, the optimal LiDAR plot size in this study might not be applicable to other areas and the optimal LiDAR plot size should be determined based on the specific data of different study areas. We also found that the optimal plot size was inconsistent with field plot size (15 m radius). If the field plot size was directly used as LiDAR plot size, we could not obtain the best estimation results of forest biomass. Therefore, to improve estimation accuracy of biomass, it is essential to determine the optimal LiDAR plot size. Many previous studies have successfully estimated forest biomass using airborne discrete-return LiDAR data (Asner et al., 2012; d’Oliveira et al., 2012; Gobakken et al., 2012; Hansen et al., 2015; Lim and Treitz, 2004; Næsset and Gobakken, 2008). However, the estimation accuracies of these studies showed significant differences. Gobakken et al. (2012) estimated forest biomass using airborne LiDAR data, and the results showed that metrics derived from LiDAR data were strongly related to biomass (R2 = 0.95, RMSE = 19.02 Mg/ha, RMSEr = 19.67%). However, Hansen et al. (2015) estimated forest biomass based on empirical relationships between field-observed biomass and variables derived from LiDAR data, and a relatively lower accuracy was obtained (R2 = 0.71, RMSE = 158.02 Mg/ha, RMSEr = 34.4%). Our biomass estimation accuracies for BGB, AGB and TB were all within the ranges reported in these previous studies. The differences in biomass estimation accuracy were mainly due to different types of vegetation, sample sizes, canopy densities, LiDAR point densities and statistical methods. Compared with LiDAR metrics, narrow band VIs from hyperspectral CASI data obtained lower estimation accuracy of biomass, and the R2 values for BGB, AGB and TB were reduced by 0.229, 0.329 and 0.308, respectively. Similar findings were reported by Vaglio Laurin et al. (2014); they estimated forest biomass using hyperspectral data and found that hyperspectral data alone had limited prediction ability (R2 = 0.36). Although narrow band VIs were more sensitive to biomass than broad band VIs, the estimation accuracy over the high biomass area was still affected by the saturation problem of VIs. Thus, VIs derived from spectral signatures were less effective than the LiDAR metrics for estimating forest biomass (Kulawardhana et al., 2014). LiDAR is considered as one of the most promising and effective methods for biomass estimation (Lu et al., 2012; Næsset et al., 2013). A study by Anderson et al. (2008) showed that fused LiDAR and hyperspectral data could improve estimation accuracy of biomass compared with either data source alone. The results from our study confirmed this finding, i.e., fused LiDAR and hyperspectral data yielded the highest estimation accuracy of forest biomass using the PLS regression. The main reason is that LiDAR data can provide the three-dimensional structural infrmation of the vegetation canopy, while hyperspectral CASI data can provide complementary spectral characteristics of the vegetation. However, CASI data did not provide significant improvements in estimation accuracy of forest biomass in our study. Compared with LiDAR data alone, the additional hyperspectral CASI data improved R2 by 5.8%, 2.2% and 2.6% for BGB, AGB and TB, respectively, and reduced RMSE by 8.6%, 7.9% and 8.3% for BGB, AGB and TB, respectively. Simi-

385

larly, previous studies showed that fused LiDAR and hyperspectral data only marginally improved the estimation accuracy of forest biomass (Clark et al., 2011; Latifi et al., 2012; Swatantran et al., 2011; Vaglio Laurin et al., 2014). Because LiDAR metrics alone yielded a strong correlation with biomass, additional hyperspectral data made a small contribution to improving estimation accuracy of biomass. However, fused metrics from LiDAR and other sensors have the potential for improving estimation accuracy of forest biomass (Zolkos et al., 2013). We assessed the prediction models using the LOOCV cross-validation method. Results shown in Table 5 indicate that the prediction models derived from fused LiDAR and hyperspectral data produced the lowest RMSEcv values. Moreover, the RMSEcv values obtained in this study were in agreement with their RMSE values, which showed that the biomass prediction models using fused LiDAR and hyperspectral data were not overfitting and had a strong predictive power. As a key ecological indicator, forest biomass was estimated using the fused airborne LiDAR and hyperspectral data in this study. Our findings showed that the fused LiDAR and hyperspectral data improved estimation accuracy of BGB, AGB and TB. The methods developed in this study could be applied to produce accurate biomass data for forest health monitoring, management, biodiversity, carbon stock estimation and ecological modeling. However, accurate biomass estimation is a challenging task, especially in a high LAI or biomass vegetation areas, in which passive optical remote sensing data easily reach saturation. LiDAR data can help overcome the saturation problem and have the potential to obtain highly accurate biomass estimates. A fusion of LiDAR data and other data source for estimating forest biomass has great potential for improving biomass estimation accuracy, which is worthy of further exploration and research. Our study could provide valuable guidance for accurate biomass estimates using the fused LiDAR and other multi-source data and improve the application quality and performance of ecological indicator of biomass. 5. Conclusions In this study, we explored the potential of fusion LiDAR and hyperspectral data for estimating BGB, AGB and TB. The primary conclusions drawn from this study are: (1) the airborne discretereturn LiDAR data could estimate BGB, AGB and TB reliably; (2) the estimation accuracy of forest biomass was affected by the LiDAR plot size; (3) the optimal LiDAR plot size for biomass estimation in this study had a radius of 22 m; (4) LiDAR data were more effective and robust in estimating biomass than hyperspectral data; and (5) the estimation accuracies of BGB, AGB and TB could be improved by using fused LiDAR and hyperspectral data. In summary, the results showed that the fusion of LiDAR and other remotely sensed data has the potential for improving biomass estimation accuracy. The methods developed and tested in this study could be useful for accurately estimating biomass using fused LiDAR and passive optical remote sensing data. However, this study took an empirical approach for estimating biomass, and the accuracies of estimation models are affected by several factors, including vegetation type, field plot size, statistical method, and LiDAR point density. These factors may result in inconsistent optimal LiDAR plot sizes and predictor variables. Therefore, to obtain the best prediction model of biomass, the optimal LiDAR plot size and predictor variables should be determined based on the specific data over the specific study area. Acknowledgments This research was funded by the National Natural Science Foundation of China (Nos. 41371350 and 41271428); the International

386

S. Luo et al. / Ecological Indicators 73 (2017) 378–387

Postdoctoral Exchange Fellowship Program 2014 by the Office of China Postdoctoral Council (20140042). The authors would like to thank three anonymous reviewers and associate editor Giovanni Zurlini for their thoughtful comments and suggestions.

References Ahmed, R., Siqueira, P., Hensley, S., 2013. A study of forest biomass estimates from lidar in the northern temperate forests of New England. Remote Sens. Environ. 130, 121–135. Anderson, J., Plourde, L., Martin, M., Braswell, B., Smith, M., Dubayah, R., Hofton, M., Blair, J., 2008. Integrating waveform lidar with hyperspectral imagery for inventory of a northern temperate forest. Remote Sens. Environ. 112, 1856–1870. Asner, G., Mascaro, J., Muller-Landau, H., Vieilledent, G., Vaudry, R., Rasamoelina, M., Hall, J., van Breugel, M., 2012. A universal airborne LiDAR approach for tropical forest carbon mapping. Oecologia 168, 1147–1160. Basuki, T.M., van Laake, P.E., Skidmore, A.K., Hussin, Y.A., 2009. Allometric equations for estimating the above-ground biomass in tropical lowland Dipterocarp forests. For. Ecol. Manage. 257, 1684–1694. Burnham, K.P., Anderson, D.R., 2002. Model Selection and Multimodel Inference: a Practical Information-Theoretic Approach, second ed. Springer, New York, USA. Cao, L., Coops, N., Hermosilla, T., Innes, J., Dai, J., She, G., 2014. Using small-footprint discrete and full-waveform airborne LiDAR metrics to estimate total biomass and biomass components in subtropical forests. Remote Sens. 6, 7110–7135. Chen, J., Gu, S., Shen, M., Tang, Y., Matsushita, B., 2009. Estimating aboveground biomass of grassland having a high canopy cover: an exploratory analysis of in situ hyperspectral data. Int. J. Remote Sens. 30, 6497–6517. Chen, Q., Lu, D., Keller, M., dos-Santos, M., Bolfe, E., Feng, Y., Wang, C., 2016. Modeling and mapping agroforestry aboveground biomass in the brazilian amazon using airborne lidar data. Remote Sens. 8, 21. Cheng, T., 2007. Research on Forest Biomass and Carbon Storage in Xiaolong Mountain, Gansu Province. Ph.D. Dissertation. College of Natural Resources and Environment, Beijing Forestry University, Beijing, China. Cho, M.A., Skidmore, A., Corsi, F., van Wieren, S.E., Sobhan, I., 2007. Estimation of green grass/herb biomass from airborne hyperspectral imagery using spectral indices and partial least squares regression. Int. J. Appl. Earth Observ. Geoinf. 9, 414–424. Clark, M.L., Roberts, D.A., Ewel, J.J., Clark, D.B., 2011. Estimation of tropical rain forest aboveground biomass with small-footprint lidar and hyperspectral sensors. Remote Sens. Environ. 115, 2931–2942. Dalponte, M., Bruzzone, L., Gianelle, D., 2012. Tree species classification in the Southern Alps based on the fusion of very high geometrical resolution multispectral/hyperspectral images and LiDAR data. Remote Sens. Environ. 123, 258–270. Dubayah, R.O., Drake, J.B., 2000. Lidar remote sensing for forestry. J. For. 98, 44–46. Duncanson, L.I., Dubayah, R.O., Cook, B.D., Rosette, J., Parker, G., 2015. The importance of spatial detail: assessing the utility of individual crown information and scaling approaches for lidar-based biomass density estimation. Remote Sens. Environ. 168, 102–112. Ediriweera, S., Pathirana, S., Danaher, T., Nichols, D., 2014. Estimating above-ground biomass by fusion of LiDAR and multispectral data in subtropical woody plant communities in topographically complex terrain in North-Eastern Australia. J. For. Res. 25, 761–771. Englhart, S., Keuck, V., Siegert, F., 2011. Aboveground biomass retrieval in tropical forests—the potential of combined X- and L-band SAR data use. Remote Sens. Environ. 115, 1260–1271. Estornell, J., Ruiz, L.A., Velázquez-Martí, B., Fernández-Sarría, A., 2011. Estimation of shrub biomass by airborne LiDAR data in small forest stands. For. Ecol. Manage. 262, 1697–1703. Foody, G.M., Boyd, D.S., Cutler, M.E.J., 2003. Predictive relations of tropical forest biomass from Landsat TM data and their transferability between regions. Remote Sens. Environ. 85, 463–474. Frazer, G.W., Magnussen, S., Wulder, M.A., Niemann, K.O., 2011. Simulated impact of sample plot size and co-registration error on the accuracy and uncertainty of LiDAR-derived estimates of forest stand biomass. Remote Sens. Environ. 115, 636–649. Gao, S., Niu, Z., Huang, N., Hou, X., 2013. Estimating the Leaf Area Index, height and biomass of maize using HJ-1 and RADARSAT-2. Int. J. Appl. Earth Observ. Geoinf. 24, 1–8. ˜ D., Chuvieco, E., Danson, F.M., 2010. Estimating biomass carbon García, M., Riano, stocks for a Mediterranean forest in central Spain using LiDAR height and intensity data. Remote Sens. Environ. 114, 816–830. Geerling, G.W., Labrador-Garcia, M., Clevers, J.G.P.W., Ragas, A.M.J., Smits, A.J.M., 2007. Classification of floodplain vegetation by data fusion of spectral (CASI) and LiDAR data. Int. J. Remote Sens. 28, 4263–4284. Ghamisi, P., Benediktsson, J.A., Phinn, S., 2015. Land-cover classification using both hyperspectral and LiDAR data. Int. J. Image Data Fusion 6, 189–215. Glenn, N.F., Spaete, L.P., Sankey, T.T., Derryberry, D.R., Hardegree, S.P., Mitchell, J.J., 2011. Errors in LiDAR-derived shrub height and crown area on sloped terrain. J. Arid Environ. 75, 377–382. Gobakken, T., Næsset, E., Nelson, R., Bollandsås, O.M., Gregoire, T.G., Ståhl, G., Holm, S., Ørka, H.O., Astrup, R., 2012. Estimating biomass in Hedmark County,

Norway using national forest inventory field plots and airborne laser scanning. Remote Sens. Environ. 123, 443–456. Guyot, G., Baret, F., Major, D.J., 1988. High spectral resolution: determination of spectral shifts between the red and the near infrared. Int. Arch. Photogramm. Remote Sens. 11, 750–760. Hansen, E., Gobakken, T., Bollandsås, O., Zahabu, E., Næsset, E., 2015. Modeling aboveground biomass in dense tropical submontane rainforest using airborne laser scanner data. Remote Sens. 7, 788–807. Hayashi, R., Kershaw, J.A., Weiskittel, A., 2015. Evaluation of alternative methods for using LiDAR to predict aboveground biomass in mixed species and structurally complex forests in northeastern North America. Math. Comput. For. Nat. Resour. Sci. 7, 49–65. He, Q., Chen, E., An, R., Li, Y., 2013. Above-ground biomass and biomass components estimation using LiDAR data in a coniferous forest. Forests 4, 984–1002. Hernández-Stefanoni, J., Dupuy, J., Johnson, K., Birdsey, R., Tun-Dzul, F., Peduzzi, A., Caamal-Sosa, J., Sánchez-Santos, G., López-Merlín, D., 2014. Improving species diversity and biomass estimates of tropical dry forests using airborne LiDAR. Remote Sens. 6, 4741–4763. Hopkinson, C., Chasmer, L., Lim, K., Treitz, P., Creed, I., 2006. Towards a universal lidar canopy height indicator. Can. J. Remote Sens. 32, 139–152. Houghton, R.A., 2005. Aboveground forest biomass and the global carbon balance. Glob. Chang. Biol. 11, 945–958. Hudak, A.T., Strand, E.K., Vierling, L.A., Byrne, J.C., Eitel, J.U.H., Martinuzzi, S., Falkowski, M.J., 2012. Quantifying aboveground forest carbon pools and fluxes from repeat LiDAR surveys. Remote Sens. Environ. 123, 25–40. Huete, A., Justice, C., Liu, H., 1994. Development of vegetation and soil indices for MODIS-EOS. Remote Sens. Environ. 49, 224–234. Jensen, J., Humes, K., Vierling, L., Hudak, A., 2008. Discrete return lidar-based prediction of leaf area index in two conifer forests. Remote Sens. Environ. 112, 3947–3957. Jochem, A., Hollaus, M., Rutzinger, M., Hofle, B., 2011. Estimation of aboveground biomass in alpine forests: a semi-empirical approach considering canopy transparency derived from airborne LiDAR data. Sensors (Basel) 11, 278–295. Jordan, C., 1969. Derivation of leaf-area index from quality of light on the forest floor. Ecology 50, 663–666. Kajimoto, T., Matsuura, Y., Sofronov, M.A., Volokitina, A.V., Mori, S., Osawa, A., Abaimov, A.P., 1999. Above- and belowground biomass and net primary productivity of a Larix gmelinii stand near Tura central Siberia. Tree Physiol. 19, 815–822. Kaufman, Y.J., Tanre, D., 1992. Atmospherically resistant vegetation index (ARVI) for EOS-MODIS. IEEE Trans. Geosci. Remote Sens. 30, 261–270. Kelloway, E.K., 2014. Using Mplus for Structural Equation Modeling: A Researcher’s Guide. SAGE Publications. Koch, B., 2010. Status and future of laser scanning: synthetic aperture radar and hyperspectral remote sensing data for forest biomass assessment. ISPRS J. Photogramm. Remote Sens. 65, 581–590. Koetz, B., Morsdorf, F., van der Linden, S., Curt, T., Allgöwer, B., 2008. Multi-source land cover classification for forest fire management based on imaging spectrometry and LiDAR data. For. Ecol. Manage. 256, 263–271. Kronseder, K., Ballhorn, U., Böhm, V., Siegert, F., 2012. Above ground biomass estimation across forest types at different degradation levels in Central Kalimantan using LiDAR data. Int. J. Appl. Earth Observ. Geoinf. 18, 37–48. Kulawardhana, R.W., Popescu, S.C., Feagin, R.A., 2014. Fusion of lidar and multispectral data to quantify salt marsh carbon stocks. Remote Sens. Environ. 154, 345–357. Latifi, H., Fassnacht, F., Koch, B., 2012. Forest structure modeling with combined airborne hyperspectral and LiDAR data. Remote Sens. Environ. 121, 10–25. Lefsky, M.A., Cohen, W.B., Parker, G.G., Harding, D.J., 2002. Lidar remote sensing for ecosystem studies: lidar, an emerging remote sensing technology that directly measures the three-dimensional distribution of plant canopies, can accurately estimate vegetation structural attributes and should be of particular interest to forest, landscape, and global ecologists. Bioscience 52, 19–30. Li, X., Cheng, G., Liu, S., Xiao, Q., Ma, M., Jin, R., Che, T., Liu, Q., Wang, W., Qi, Y., Wen, J., Li, H., Zhu, G., Guo, J., Ran, Y., Wang, S., Zhu, Z., Zhou, J., Hu, X., Xu, Z., 2013. Heihe watershed allied telemetry experimental research (HiWATER): scientific objectives and experimental design. Am. Meteorol. Soc. 94, 1145–1160. Li, W., Niu, Z., Huang, N., Wang, C., Gao, S., Wu, C., 2015. Airborne LiDAR technique for estimating biomass components of maize A case study in Zhangye City, Northwest China. Ecol. Indic. 57, 486–496. Lim, K.S., Treitz, P.M., 2004. Estimation of above ground forest biomass from airborne discrete return laser scanner data using canopy-based quantile estimators. Scand. J. For. Res. 19, 558–570. Lin, Y., West, G., 2016. Reflecting conifer phenology using mobile terrestrial LiDAR A case study of Pinus sylvestris growing under the Mediterranean climate in Perth, Australia. Ecol. Indic. 70, 1–9. Listopad, C.M.C.S., Masters, R.E., Drake, J., Weishampel, J., Branquinho, C., 2015. Structural diversity indices based on airborne LiDAR as ecological indicators for managing highly dynamic landscapes. Ecol. Indic. 57, 268–279. Lu, D., Chen, Q., Wang, G., Moran, E., Batistella, M., Zhang, M., Vaglio Laurin, G., Saah, D., 2012. Aboveground forest biomass estimation with landsat and LiDAR data and uncertainty analysis of the estimates. Int. J. For. Res. 2012, 1–16. Lucas, R.M., Lee, A.C., Bunting, P.J., 2008. Retrieving forest biomass through integration of CASI and LiDAR data. Int. J. Remote Sens. 29, 1553–1577. Luedeling, E., Gassner, A., 2012. Partial least squares regression for analyzing walnut phenology in california. Agric. For. Meteorol. 158–159, 43–52.

S. Luo et al. / Ecological Indicators 73 (2017) 378–387 Luo, S., Wang, C., Xi, X., Pan, F., 2014. Estimating FPAR of maize canopy using airborne discrete-return LiDAR data. Opt. Express 22, 5106–5117. Luo, S., Wang, C., Pan, F., Xi, X., Li, G., Nie, S., Xia, S., 2015. Estimation of wetland vegetation height and leaf area index using airborne laser scanning data. Ecol. Indic. 48, 550–559. Luo, S., Chen, J.M., Wang, C., Xi, X., Zeng, H., Peng, D., Li, D., 2016a. Effects of LiDAR point density, sampling size and height threshold on estimation accuracy of crop biophysical parameters. Opt. Express 24, 11578–11593. Luo, S., Wang, C., Xi, X., Zeng, H., Li, D., Xia, S., Wang, P., 2016b. Fusion of airborne discrete-return LiDAR and hyperspectral data for land cover classification. Remote Sens. 8, 3. Mauya, E., Hansen, E., Gobakken, T., Bollandsas, O., Malimbwi, R., Naesset, E., 2015. Effects of field plot size on prediction accuracy of aboveground biomass in airborne laser scanning-assisted inventories in tropical rain forests of Tanzania. Carbon Balance Manag. 10, 10. Mitchard, E.T.A., Saatchi, S.S., Woodhouse, I.H., Nangendo, G., Ribeiro, N.S., Williams, M., Ryan, C.M., Lewis, S.L., Feldpausch, T.R., Meir, P., 2009. Using satellite radar backscatter to predict above-ground woody biomass: a consistent relationship across four different African landscapes. Geophys. Res. Lett. 36. Morsdorf, F., Kötz, B., Meier, E., Itten, K.I., Allgöwer, B., 2006. Estimation of LAI and fractional cover from small footprint airborne laser scanning data based on gap fraction. Remote Sens. Environ. 104, 50–61. Mutanga, O., Skidmore, A.K., 2004. Narrow band vegetation indices overcome the saturation problem in biomass estimation. Int. J. Remote Sens. 25, 3999–4014. Næsset, E., Gobakken, T., 2008. Estimation of above- and below-ground biomass across regions of the boreal forest zone using airborne laser. Remote Sens. Environ. 112, 3079–3090. Næsset, E., Bollandsås, O.M., Gobakken, T., 2005. Comparing regression methods in estimation of biophysical properties of forest stands from two different inventories using laser scanner data. Remote Sens. Environ. 94, 541–553. Næsset, E., Gobakken, T., Bollandsås, O.M., Gregoire, T.G., Nelson, R., Ståhl, G., 2013. Comparison of precision of biomass estimates in regional field sample surveys and airborne LiDAR-assisted surveys in Hedmark County, Norway. Remote Sens. Environ. 130, 108–120. Nelson, R.F., Hyde, P., Johnson, P., Emessiene, B., Imhoff, M.L., Campbell, R., Edwards, W., 2007. Investigating RaDAR–LiDAR synergy in a North Carolina pine forest. Remote Sens. Environ. 110, 98–108. Nie, S., Wang, C., Dong, P., Xi, X., 2016. Estimating leaf area index of maize using airborne full-waveform lidar data. Remote Sens. Lett. 7, 111–120. Olsoy, P.J., Mitchell, J.J., Levia, D.F., Clark, P.E., Glenn, N.F., 2016. Estimation of big sagebrush leaf area index with terrestrial laser scanning. Ecol. Indic. 61 (Part 2), 815–821. Parresol, B.R., 2001. Additivity of nonlinear biomass equations. Can. J. For. Res. 31, 865–878. Popescu, S.C., Zhao, K., Neuenschwander, A., Lin, C., 2011. Satellite lidar vs. small footprint airborne lidar: comparing the accuracy of aboveground biomass estimates and forest structure metrics at footprint level. Remote Sens. Environ. 115, 2786–2797. Qi, J., Chehbouni, A., Huete, A.R., Kerr, Y.H., Sorooshian, S., 1994. A modified soil adjusted vegetation index. Remote Sens. Environ. 48, 119–126. Qin, Y., Li, S., Vu, T.-T., Niu, Z., Ban, Y., 2015. Synergistic application of geometric and radiometric features of LiDAR data for urban land cover mapping. Opt. Express 23, 13761–13775. ˜ D., Valladares, F., Condés, S., Chuvieco, E., 2004. Estimation of leaf area index Riano, and covered ground from airborne laser scanner (Lidar) in two contrasting forests. Agric. For. Meteorol. 124, 269–275. Richardson, J.J., Moskal, L.M., Kim, S.-H., 2009. Modeling approaches to estimate effective leaf area index from aerial discrete-return LIDAR. Agric. For. Meteorol. 149, 1152–1160. Rondeaux, G., Steven, M., Baret, F., 1996. Optimization of soil-adjusted vegetation indices. Remote Sens. Environ. 55, 95–107.

387

Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1973. Monitoring vegetation systems in the Great Plains with ERTS. Third Earth Resources Technology Satellite-1 Symposium, 309–317. Ruiz, L., Hermosilla, T., Mauro, F., Godino, M., 2014. Analysis of the influence of plot size and LiDAR density on forest structure attribute estimates. Forests 5, 936–951. Sarrazin, M.J.D., van Aardt, J.A.N., Asner, G.P., McGlinchy, J., Messinger, D.W., Wu, J., 2011. Fusing small-footprint waveform LiDAR and hyperspectral data for canopy-level species classification and herbaceous biomass modeling in savanna ecosystems. Can. J. Remote Sens. 37, 653–665. Strunk, J., Temesgen, H., Andersen, H.-E., Flewelling, J.P., Madsen, L., 2012. Effects of lidar pulse density and sample size on a model-assisted approach to estimate forest inventory variables. Can. J. Remote Sens. 38, 644–654. Sun, G., Ranson, K.J., Guo, Z., Zhang, Z., Montesano, P., Kimes, D., 2011. Forest biomass mapping from lidar and radar synergies. Remote Sens. Environ. 115, 2906–2916. Swatantran, A., Dubayah, R., Roberts, D., Hofton, M., Blair, J.B., 2011. Mapping biomass and stress in the Sierra Nevada using lidar and hyperspectral data fusion. Remote Sens. Environ. 115, 2917–2930. Thomas, V., Finch, D.A., McCaughey, J.H., Noland, T., Rich, L., Treitz, P., 2006. Spatial modelling of the fraction of photosynthetically active radiation absorbed by a boreal mixedwood forest using a lidar–hyperspectral approach. Agric. For. Meteorol. 140, 287–307. Tsui, O.W., Coops, N.C., Wulder, M.A., Marshall, P.L., McCardle, A., 2012. Using multi-frequency radar and discrete-return LiDAR measurements to estimate above-ground biomass and biomass components in a coastal temperate forest. ISPRS J. Photogramm. Remote Sens. 69, 121–133. Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 8, 127–150. Vaglio Laurin, G., Chen, Q., Lindsell, J.A., Coomes, D.A., Frate, F.D., Guerriero, L., Pirotti, F., Valentini, R., 2014. Above ground biomass estimation in an African tropical forest with lidar and hyperspectral data. ISPRS J. Photogramm. Remote Sens. 89, 49–58. Wang, C., Menenti, M., Stoll, M.P., Feola, A., Belluco, E., Marani, M., 2009. Separation of ground and low vegetation signatures in LiDAR measurements of salt-marsh environments. IEEE Trans. Geosci. Remote Sens. 47, 2014–2023. Weng, Q., 2014. Scale Issues in Remote Sensing. Wiley. Wold, S., 1995. PLS for multivariate linear modelin. In: van der Waterbeemd, H. (Ed.), Chemometric Methods in Molecular Design: Methods and Principles in Medicinal Chemistry. Chemie, Weinheim, Germany, pp. 195–218. Woodcock, C.E., Strahler, A.H., 1987. The factor of scale in remote sensing. Remote Sens. Environ. 21, 311–332. Wu, H., Li, Z.L., 2009. Scale issues in remote sensing: a review on analysis, processing and modeling. Sensors (Basel) 9, 1768–1793. Xiao, Q., Wen, J., 2014. HiWATER: Airborne LiDAR Raw Data in the Middle Reaches of the Heihe River Basin. Heihe Plan Science Data Center. Zhao, K., Popescu, S., 2009. Lidar-based mapping of leaf area index and its use for validating GLOBCARBON satellite LAI product in a temperate forest of the southern USA. Remote Sens. Environ. 113, 1628–1645. Zhao, K., Popescu, S., Nelson, R., 2009. Lidar remote sensing of forest biomass: a scale-invariant estimation approach using airborne lasers. Remote Sens. Environ. 113, 182–196. Zolkos, S.G., Goetz, S.J., Dubayah, R., 2013. A meta-analysis of terrestrial aboveground biomass estimation using lidar remote sensing. Remote Sens. Environ. 128, 289–298. d’Oliveira, M.V.N., Reutebuch, S.E., McGaughey, R.J., Andersen, H.-E., 2012. Estimating forest biomass and identifying low-intensity logging areas using airborne scanning lidar in Antimary State Forest Acre State, Western Brazilian Amazon. Remote Sens. Environ. 124, 479–491.