Generating pseudo large footprint waveforms from small footprint full-waveform airborne LiDAR data for the layered retrieval of LAI in orchards

Wang Li; Zheng Niu; Jing Li; Hanyue Chen; Shuai Gao; Mingquan Wu; Dong Li

doi:10.1364/OE.24.010142

1. Introduction

Leaf area index (LAI), defined as half of the total leaf area per unit ground surface area [1], is a key parameter for the study of biogeochemical cycles in ecosystems. Vegetation LAI often drives both the within- and the below-canopy microclimate [2]. However, it is very difficult to quantify LAI because of its spatial and temporal variability [2]. Direct and indirect methods are often used for the field measurement of LAI including the destructive sampling and the use of optical instruments [3]. The most commonly used optical instruments for LAI field measurement includes the LAI2200 plant canopy analyzer (Li-COR, Inc., Lincoln, NE, USA), LI3000 (LI-COR, Lincoln, Nebraska), the Tracing Radiation and Architecture of Canopies (TRAC) (3rd Wave Engineering, Ontario, Canada), and hemispherical photography. Nevertheless, the field measurements are often time-consuming and limited to small areas. Thus, both passive and active remote sensing tools have been widely employed in the estimation of LAI based on the well correlation between the field-observed LAIs and remotely sensed spectral values. The spectral values referred to reflectance, radiance, vegetation indices, and the backscatter intensity of synthetic aperture radar (SAR) images [4–6]. Nevertheless, most of the published studies only focused on the total LAI distributed in the horizontal plane. The vertical characteristics of LAI distribution are often missed, limited by the low penetration capability of most optical remote sensing sensors. Previous studies reported that the accuracy of satellite-estimated forest LAI is often limited by the presence of understory vegetation [7]. The low penetration capability confined optical remote sensing sensors to detect the backscattered reflectance only from the top of vegetation canopies. However, reflectance returned from different height layers within vegetation canopies may give a more real and concrete reference to biologists and botanists. In addition, this will also help to make a more precise estimation of carbon budget in global climate change studies.

The presence of airborne light detection and ranging (LiDAR) or laser scanning technique is likely to make it possible to collect layered characterization of targets based on its high penetration capability. It offers the possibility of accurately detecting the understory by extracting the structure attributes in the complicated vegetation areas [8]. The laser sensor actively emits electromagnetic waves that penetrate into and interact with different components of vegetation canopies. Backscattered signals from the canopy components along the entire canopy depth are then recorded by the laser sensors. A large quantity of studies showed that small footprint (discrete return or full-waveform) airborne LiDAR data can be used to accurately estimate LAI either by parametric or non-parametric models based on the empirical relationships between the field-observed LAIs and LiDAR metrics [9–11]. Previous studies also successfully retrieved the LAIs of forests and crops by directly putting the LiDAR-derived canopy cover index or penetration index into the Beer-Lambert law [12–15]. A sufficient amount of ground returns is an important prerequisite for the LiDAR-based LAI estimation because it directly affects the precision of LiDAR metrics. Compared to the discrete return airborne LiDAR system, the full-waveform system can register the complete returned backscattered signals, providing more physical properties of the intercepted objects [16, 17]. However, many small footprint laser beams are often fully intercepted by the vegetation canopies due to its small cross-sections. This may not be the frequent problem for large footprint system such as the pioneer airborne laser sensor SLICER (Scanning LiDAR Imager of Canopies by Echo Recovery, 10 m footprint) [18], LVIS (Laser Vegetation Imaging Sensor, 25 m footprint) [19] and space-borne laser sensor ICESat (Ice, Cloud and land Elevation Satellite)/GLAS (Geoscience Laser Altimeter System, 70 m footprint) [20]. The vertical profiles of LAI were successfully estimated by the medium footprint LVIS waveform data with high correlations to the LAI from tower measurements [19]. In their study, total leaf area present within each tower sector was measured to quantify corresponding LAI using a LICOR-3100 leaf area meter.

At present, it is difficult to obtain large footprint waveform LiDAR data after the retirement of the ICESat satellite in 2009. Therefore, pseudo large footprint waveforms were proposed by aggregating or stacking the raw small footprint waveforms into a larger footprint waveform [16, 21, 22], which might be a promising alternative in cases where real medium or large footprint waveform data are unavailable. For instance, plot-level pseudo large footprint waveforms that generated from small-footprint airborne LiDAR data have been used to retrieve effective LAIs in a discontinuous canopy environment with an improved correlation to the LAIs retrieved from hemispherical photography [23]. Nevertheless, to the best of our knowledge, few studies have investigated the retrieval of the layered LAIs in orchards with different crops interplanted. Additionally, the pseudo large footprint waveforms generated from the small footprint full-waveform airborne LiDAR data have not yet been applied in the layered retrieval of LAIs in agriculture landscape of orchards. Interplanting crops under the fruit trees in orchards are very common in many agriculture landscapes. An implicit retrieval of the LAIs for different layers in orchards may provide valuable suggestions to the management and production in precision agriculture.

Therefore, this study is aimed to further explore the capability of small footprint full-waveform airborne LiDAR to retrieve the layered LAIs in agricultural orchards with understory crops inter-planted. Pseudo large footprint waveforms were generated based on the small footprint waveforms. Plot-level vertical profiles of canopy height and LAI at different vegetation layers (overstorey, understory and total) were retrieved from the large footprint waveforms, which were validated by field measurements.

2. Materials

2.1 Study area

The study site is located in the Huailai area, 84 km north of Beijing, China. It is a flat area in the Huailai-Yanqing Basin, along the south of the Guanting Reservoir (Fig. 1). The Huailai area is dominated by maize with a simple landform type, and a mean elevation of 30 m above sea level. Besides maize, agriculture orchards (apricot, crabalppl, and pear) are also common vegetation covers in the study site. Usually, short crops including beam, maize as well as disperse herb are interplanted under the fruit trees. Precipitation is distributed unevenly in the four seasons with an average annual precipitation of 396 mm, and the greatest precipitation occurs in summer.

Fig. 1 The study site of Huailai area with field plots overlain on the Google Earth image.

Download Full Size | PDF

2.2 Small footprint full-waveform airborne LiDAR data

Repetitive airborne flights with 50% side-overlap were conducted at a nominal height of 2800 m above ground to acquire small footprint full-waveform LiDAR data for the study site on July 27, 2014 (Table 1). The ALS70 airborne laser scanning system (Leica Geosystems Ltd., Aarau, Switzerland) was used and configured to emit laser pulses in the near-infrared wavelength (1064 nm) with a scan angle of ± 12°, a beam divergence of 0.15 mrad, and an average footprint size of 42 cm. A dual-frequency differential Global Positioning System (GPS) with an inertial measurement unit was used to record geographical coordinates (easting, northing and elevation) of each returned pulse with horizontal and vertical accuracy of 0.1 m and 0.3 m, respectively. The transmitted and returned waveform signals were recorded by the ALS70 system during the flight. The small footprint full-waveform data was stored in a LAS format file. Besides the three-dimensional position of first returned laser sample and the indexing information to associate the waveform data, a waveform vector consisting of 128 returned signal values is also stored in each waveform record. Discrete point clouds with geographical coordinates (UTM, Zone 50N/WGS-84), intensity, and the number of returns were produced by the flight vendor using proprietary processing procedures. The average density of point clouds is 3.6 points/m². Detailed processing procedure for the vendor-provided point clouds is referred to our previous studies, which included noise removal, classification of ground and non-ground points, and the interpolations of digital elevation model (DEM), digital surface model (DSM), and normalized DSM with a grid size of 1 m [11]. The height accuracy of DEM was assessed using GPS points with a height error of 0.05 ± 0.12 m [21], which is accurate enough to capture the normalized height in this study. The raw small footprint waveform data were used to generate pseudo large footprint waveforms for the layered retrieval of canopy heights and LAIs.

Table 1. Specifications of airborne LiDAR flights in the Huailai area.

View Table | View all tables in this article

2.3 Field measurements

Field measurements were conducted on August 5, 2015. In each plot, at least 10 representative trees and 10 crop plants were selected, respectively. The canopy height of each overstorey tree was measured using a laser hypsometer, and the plant height of the understory crop was also measured from the ground to the canopy top using a tape. Then, the average canopy height was calculated as the canopy height for overstorey (H_over) and understory (H_under), respectively. Totally, 20 plots with a relative high difference in land cover and canopy cover (Table 2) were measured. The spatial distribution of the field plots is shown in Fig. 1. Each field plot has an area of around 10 m × 10 m. The geographic location (latitude and longitude) of each plot was positioned using a GPS.

Table 2. Main attributes of the field plots

View Table | View all tables in this article

The canopy LAIs for the overstorey and understory vegetation cover were measured using LAI2200 plant canopy analyzer (Li-COR, Inc., Lincoln, NE, USA) under diffuse radiation conditions with a 45° view cap. Only one single optical sensor was used during the measurements. The optical sensor was first placed in a clearing with a view of the sky. The data of solar radiation was automatically logged at specified intervals. After that, the optical sensor was used to make below-canopy readings around the fruit tree trunk and above the understory canopy in each cardinal direction. For each measurement, one reading of solar radiation and nine readings of under radiation were obtained along two different diagonal directions, which were further used to determine the overstorey LAI (LAI_over) based on the gap fraction theory. Similar steps were used to measure the total LAI (LAI_total). Since the optical sensor is small and the understory vegetation is short, the optical sensor was placed in the short understory canopies without altering the canopy structure. For each layer, LAI was measured for at least three times with the average value as the final measurement for the plot. The LAI for the understory (LAI_under) can be determined as the difference between LAI_total and LAI_over. The basic statistics of the field-measured LAI and canopy height are shown in Table 3.

Table 3. Basic statistics of field-measured leaf area index (LAI) and canopy height for the different layers in the orchard plots (N = 20)

View Table | View all tables in this article

3. Methodology

The main methodological workflow for this study is shown in Fig. 2, consisting of alignment and aggregation of raw small footprint full-waveform data, height normalization, signal normalization, Gaussian fitting, layered retrieval of canopy height and LAI as well as validation by field measurements. More details of each step are described in the following subsections.

Fig. 2 The main methodological workflow for this study.

Download Full Size | PDF

3.1 Generation of pseudo large footprint waveforms

Pseudo large footprint waveforms were generated by aggregating the raw small footprint waveforms [23]. Since the small footprint waveforms do not always have ground returns, aggregating the small footprint waveforms can assure a large enough amount of ground returns that needed for the estimates of gap fraction and LAI in Beer-Lambert law. The aggregation approach was also called waveform stacking approach in other similar studies [22]. In this study, the aggregation approach was applied with a gird size of 10 m that corresponded to the plot size. The aggregation grid size (10 m) was believed to be large enough to capture the local maximum height of common individual trees meanwhile small enough to clarify the contributions from canopy and ground. Before aggregation, the absolute elevations that stored in the 128 samples of the raw small footprint waveform were normalized using corresponding absolute elevation from the DEM that produced by the discrete point clouds (Section 2.2). According to the normalized height and the location of the first pulse, the raw small footprint waveforms with 128 signal samples were aligned in each plot. Then, pseudo large footprint waveforms were generated by adding up the returned signal values from different small waveforms in each vertical height bin (h_bin = 0.15 m). The height bin size (0.15 m) was determined according to the vertical ranging resolution of the laser pulse. To remove the effects from the difference on emitted pulses and atmospheric conditions [24], the pseudo large footprint waveform in each plot was normalized as follows:

V_{norm, i} = \frac{V_{i}}{V_{T}}

V_{T} = \sum_{i = 1}^{N} V_{i}

where

V_{i}

is the sum of the returned signal values contributed from the raw small footprint waveforms at the i_th vertical bin,

V_{norm, i}

is the normalized value of

V_{i}

, N is the number of vertical bins in the large footprint waveform,

V_{T}

is the sum of the returned signal values in all vertical bins. Different from the real large footprint waveform of satellite LiDAR data in [24], the value of N was not constant in this study, which was determined by the maximum height of normalized waveform in each plot. After the normalization of signal values, the area of the pseudo large footprint waveform in all plots equaled one (Fig. 3).

Fig. 3 The Gaussian-fitted waveform and corresponding residuals based on the pseudo large footprint waveform.

Download Full Size | PDF

3.2 Discrimination of different vegetation layers from LiDAR

The normalized pseudo large footprint waveforms were used to identify the vertical positions for different layers including the ground, understory and overstorey part. Since each plot possessed a two-layered vegetation distribution, two peaks can be obtained in the vegetation part of the large footprint pseudo waveform. Therefore, evident separation points should be found at the start and end edges of the overstorey and understory component, because the distance between the canopy base of overstorey and the canopy top of the understory are usually large enough to be captured by a number of vertical bins. Therefore, a Gaussian fitting was conducted to automatically capture the canopy base of different cover layers. In this study, the pseudo large footprint waveform was treated as a sum of several single Gaussian functions [25]:

f (t_{^{i}}) = h_{i} \exp [- \frac{{(t_{i} - α_{i})}^{2}}{w_{i}^{2}}]

f (t) = \sum_{i = 1}^{n} f (t_{^{i}})

where

f (t_{^{i}})

is a single Gaussian component,

h_{i}

is amplitude,

α_{i}

is central position,

w_{i}

is pulse width, n is the number of peaks. In fact, the Gaussian fitting is a procedure to determine the i_th Gaussian components with 3i unknown parameters by solving a nonlinear least-square problem [26, 27]. According to the number of peaks in the normalized waveforms, the number of Gaussian model terms was set as four. The model fitting results was evaluated using two goodness-of-fit statistics including the adjust coefficient of determination (adj-R²) and the root mean square error (RMSE). After the Gaussian fitting, the canopy edges for different layers were identified from the fitted waveforms, which were further used to determine the canopy height for overstorey and understory.

3.3 Retrieving LAIs for different layers

The gap theory that quantifies the relationship between LAI and the gap probability $P (θ)$ was used to retrieve the vertical distribution of canopy LAI in this study [19].

P (θ) = e^{- G (θ) * L A I / \cos (θ)}

G (θ)

is the projection coefficient with a view zenith angle of

θ

. In this study, all the bin values in the pseudo large footprint waveform were cumulatively added up, the sum of which was further used to divide the bin values, forming the canopy cover profile

f_{c o v e r} (z)

and gap probability profile

P (z)

as a function of height [19].

P (z) = 1 - f_{c o v e r} (z) = 1 - \frac{R_{v} (z)}{R_{v} (0)} \frac{1}{1 + \frac{ρ_{v}}{ρ_{g}} \frac{R_{g}}{R_{v} (0)}}

R_{v} (z)

,

R_{v} (0)

and

R_{g}

are the integrated signal values from the canopy to height z, from canopy top to canopy bottom, and from the ground, respectively.

The gap probability is sensitive to the canopy/ground reflectivity ratio $ρ_{v} / ρ_{g}$ and the wavelength of detection laser [28]. To clarify the difference between average canopy and ground reflectance of near-infrared wavelength (1064 nm) at 0° phase angle, the ground waveform amplitude was scaled by a factor of 2 by assuming that the reflectance of the ground was half that of the canopy [29, 30]. In fact, the reflectance ratio of ground and vegetation can also be adjusted according to the conditions of different forest stands [23, 28]. It has been demonstrated that the spatial heterogeneity in the ground and vegetation reflectance was closely dependent on the footprint size, and aggregating the small footprint waveforms to lager footprint sizes could decrease their impacts on canopy gap [30]. Hereafter, the apparent foliage profile $F_{app} (z)$ was derived from $P (z)$ , which was further used to derive the cumulative LAI profile $L A I_{cum}$ [19]:

F_{app} (z) = \frac{d \log P (z)}{d z}

L A I_{cum} (z) = C * \int_{z_{0}}^{z} \frac{F_{app} (z)}{G} d z

The clumping index C was set to an empirical value of 1.58 [28]. And the projection coefficient G was set as 0.5, assuming a random foliage distribution within the canopy [19]. Finally, the LAIs for different vegetation layers (LAI_over, LAI_under, and LAI_total) were extracted from

L A I_{cum}

as the LAI difference between the canopy top and base, respectively.

3.4 Validation

The LiDAR-retrieved LAIs for overstorey, understory and total vegetation cover were compared to the field-observed LAIs by simple linear regression models. Three statistical estimators were used to evaluate the model accuracy, namely, coefficient of determination (R²), root mean square error (RMSE), and relative root mean square error (rRMSE). R² refers to the fraction of variance explained by the model.

R M S E = \sqrt{\frac{1}{n} \sum_{i = 1}^{n} {(p_{i} - \overset{\land}{p_{i}})}^{2}}

r R M S E = \frac{R M S E}{\bar{p}}

where

p_{i}

is the field-observed value and

\overset{\land}{p_{i}}

is the LiDAR-retrieved value.

4. Results

4.1 Pseudo large footprint waveforms

The Gaussian-fitted waveform and corresponding residuals based on the normalized pseudo large footprint waveform in one field plot (overstorey: pear trees, understory: maize) are shown in Fig. 3. Results showed that the Gaussian-fitted model obtained an adjust-R² of 0.99 and RMSE of 0.004 [Fig. 3(b)]. Three evident peaks can be found in the normalized waveform, which corresponded to the overstorey, understory and the ground part along the vertical height bins. An evident large proportion of ground energy was obtained in the large footprint waveform, showing a different construction from that of raw small footprint waveforms. The pseudo waveforms for all the field plots were fitted following the same way as above. The canopy heights for the two vegetation layers were extracted from the fitted waveforms, which were further compared to the field-observed ones. Results showed that satisfactory relationships were found between the LiDAR-retrieved and field-observed canopy heights with R² of 0.82 for the overstorey and 0.76 for the understory, respectively [Figs. 5(a) and 5(b)]. The regression model for H_over showed a better performance than that of H_under as significantly lower errors were obtained (rRMSE: 9.52% < 31.12%). The retrieved canopy height varied from 2.76 to 6.11 m for H_over, and 0.41 to 2.88 m for H_under, respectively (Table 4). Compared with field measurements, the mean of canopy heights were slightly overestimated by 0.09 m for H_over and 0.1 m for H_under.

Table 4. Basic statistics of the retrieved canopy height and LAI for the field plots (N = 20)

View Table | View all tables in this article

4.2 Retrieval of layered LAI in orchards

The retrieved LAIs for different vertical height bins in two contrasting LAI plots (plot #08 and plot #12) with the highest and lowest retrieved LAI_total are shown in Fig. 4. The retrieved canopy height for these two plots obtained a H_over of 4.32 m and 2.76 m, H_under of 1.66 m and 0.51 m, respectively. The two LAI profiles in Fig. 4 showed significantly different shapes that resulted from the different vegetation types with an estimated total LAI of 4.48 and 1.26, respectively. The high LAI plot consisted of pear trees with maize as understory while the low LAI plot consisted of apricot trees and beams. The maize canopy in plot #08 contributed more to the total LAI than the pear trees due to its denser canopy cover. In contrast, the overstorey apricot contributed the most to the total LAI in plot #12. These indicated that the understory layer did play an important role in the leaf area accumulation in the orchards.

Fig. 4 The distribution of LAIs along the vertical direction in (a) a high LAI plot and (b) a low LAI plot retrieved from the pseudo large waveforms.

Download Full Size | PDF

Results showed that the retrieved LAIs for different layers in all the field plots were highly comparable to the field-observed ones [Figs. 5(c)-5(e)]. In general, the LAIs of overstorey achieved higher estimation accuracy than the understory LAIs with much lower RMSE (0.28 m < 0.40 m) and rRMSE (14.03% < 35.48%). The estimation of total LAI obtained the highest robustness with the lowest rRMSE of 12.16% by explaining 69% variance of the field-observed LAIs. Compared to the field measurements, the mean values of LAI_over and LAI_total were slightly overestimated by 0.05 and 0.14, respectively [Table 3 and Table 4]. In contrast, the mean of LAI_under was underestimated by 0.24, showing a relatively poorer estimation performance. This indicated that estimating the layered LAIs from the pseudo large footprint waveforms in the multi-layer agriculture orchards are feasible and the estimation errors are acceptable.

Fig. 5 The observed and retrieved (a) H_over (m); (b) H_under (m); (c) LAI_over; (d) LAI_under; and (e) LAI_total. The red dashed line represents the 1:1 line. **p<0.01.

Download Full Size | PDF

5. Discussion

In this study, pseudo large footprint waveforms were generated from small footprint full-waveform airborne LiDAR data, which were further used to retrieve the vertical LAI distribution in different orchard plots. The aggregation or stacking approach was used to generate the large footprint waveforms, which highly helped to capture large enough amount of ground returns at plot level for the estimation of LAIs. The satisfactory relationships between the retrieved and field-observed LAIs for different vegetation layers proved that the pseudo large footprint waveforms can be promising alternative in the quantitative evaluation of vegetation LAI. In this study, LAI values were directly retrieved from the LiDAR-derived cumulative LAI profiles, making airborne LiDAR act as a direct tool to measure LAI as common optical instruments do (e.g., hemisphere photography and LAI2200). This is especially important for the layered detection of LAIs because common optical instruments like hemisphere photography are often difficult to obtain the LAIs for canopies together with short understory layers. Although the LAI2200 was operational to separately measure the canopy LAIs for different vegetation layers in the field experiment, it is still infeasible to measure a wall-to-wall LAI distribution in the three dimensional space. Furthermore, all field measurements may have their own biases [30]. For instance, the LAI2200 cannot fully cover the field plot and the weather conditions and solar radiation varied among different plots during the field measurements.

In spite of these, several cautions should be considered when the pseudo large footprint waveform data are used to retrieve layered LAIs. In this study, the aggregation grid size was set to 10 m that was identical to the plot size, which was believed to be large enough to capture the local maximum height of common individual trees meanwhile small enough to clarify the contributions from canopy and ground. No further exploration on different grid sizes were conducted in this study, due to the fact that the field plot size and the plot spatial heterogeneity in our study was relatively small compared with those in natural forest applications. According to [10], the grid size mainly depends on the pulse density, sensitivity of the detection algorithm to weak returns, and site heterogeneity. Therefore, the grid size should be tested when the vegetation types are changed (e.g., natural forest) in future applications due to different canopy structure complexity and heterogeneity. When the grid size was smaller than the plot size, the plot LAI should be calculated as a weighted average of the estimated LAIs from all the grids.

The canopy heights H_over and H_under were slightly overestimated based on our data. This is quite different from previous studies where canopy heights were tended to be underestimated due to the miss detection of canopy tops [13, 31, 32]. This missing detection is common during LiDAR measurements, particularly in vegetated areas (e.g., forests and crop lands). However, the Gaussian fitting applied to the normalized large footprint waveform in this study probably counteracted this kind of underestimation of canopy height. On the other hand, the difference between the LiDAR-retrieved and field-observed canopy heights were also resulted from the field measurement by the laser hypsometer and tape. In addition, the highest canopy tops might be missed because only several representative overstorey trees and understory crop plants were measured.

Usually, an insufficient amount of ground returns can lead to a high canopy cover, making an overestimated LAI from the Beer-Lambert law [13, 33]. The large footprint waveform data provided more information from the ground compared to discrete return LiDAR data, which helped to reduce the overestimation of low LAI and the underestimation of high LAI. It is because of the fact that some tiny returned information that recorded by the full-waveform laser scanner can be easily missed by the discrete return laser system. The miss detection of tiny returned information is usually caused by the relatively high thresholds used in the pulse detection methods for the discrete return laser system. This is especially true for the cases in this study, because large range of LAI can be found in the group of vegetation layers (overstorey, understory and total). However, in this study, only slight overestimation of LAI_under and underestimation of LAI_over and LAI_total occurred. This indicated that the generated pseudo large footprint waveform helped to reduce the overestimation problem especially in short vegetation types.

The biases of LAIs in this study were likely caused by the approximation of clumping index and projection coefficient, and might have been affected to an unknown degree by the one-year time lag between the field measurements and LiDAR campaign. Additionally, the assumption of the reflectance of the ground being half that of the canopy might also contributed to the estimation errors of LAI. In spite of these, the estimation errors obtained in this study for the layered LAIs were acceptable compared with previous studies [11, 34]. It is worth noting that the canopy structure complexity and spatial pattern of the overstorey and understory in this study was relatively simple and regular. This facilitated the airborne laser more easily to penetrate the overstorey to detect the understory canopies. It will be of higher significance when our approaches can be applied to the other natural forest stands like tropical forests where dominated by a more complicated vegetation structural system with higher spatial variations. It is because of that the layered LAI distribution of natural forests can provide a much broader inference than the agriculture landscape of orchards including the assessment of forest health, wildlife habitat, species diversity and nutrient cycling [35].

In this study, both the generation of large footprint waveforms and the layered retrieval of LAIs were based on the single-wavelength backscattered laser signals from the small footprint airborne LiDAR data. Compared with multi- and hyper-spectral remote sensing data, the single-wavelength laser signals can provide limited spectral information but important three-dimensional structure attributes. Nevertheless, a wider range of spectral detection could greatly increase the possibility of capturing the weak variations of vegetation components [36]. Previous study reported that a combined use of multi–spectral and LiDAR data can be a promising alternative to improve the retrieval accuracy of LAI and other biophysical parameters [11]. However, this kind of combination may difficult to obtain the layered distribution of vegetation characteristics. Thus, the recently developed hyper-spectral LiDAR technique might be a more promising alternative for the study of vegetation biogeochemical characteristics in the three dimensional space [36, 37]. Since the development of hyperspectral LiDAR technique has just started, there is still a long way to go before it is applied in wide applications. This study gave a possible solution for the layered retrieval of LAIs based on the single-wavelength LiDAR data when multi- and hyper-spectral data are unavailable.

6. Conclusion

This study tried to generate a kind of pseudo large footprint waveform data from the small footprint full-waveform airborne LiDAR data. The layered distribution of canopy heights and LAIs were successfully retrieved based on the large footprint waveform data in an agricultural landscape of orchards with typical multi-layer vegetation covers. Statistically significant simple linear regression models were fitted between the LiDAR-retrieved and field-observed values for the canopy height and LAI in different layers. Satisfactory results were obtained with a RMSE of 0.36 m for H_over, 0.29 m for H_under, 0.28 for LAI_over, 0.40 for LAI_under, and 0.38 for LAI_total, respectively. The simple linear regression model successfully explained 82%, 76%, 75%, 64%, and 69% of the variance in the field-observed H_over, H_under, LAI_over, LAI_under, and LAI_total, respectively. To conclude, estimating the layered LAIs in the multi-layer agriculture orchards from the pseudo large footprint waveforms is feasible and the estimation errors are acceptable, which will provide some new ideas and methods for the quantitative remote sensing with vegetation.

Acknowledgments

This work was supported by China’s Special Funds for Major State Basic Research Project of China under grant 2013CB733405; the National Natural Science Foundation of China under grant 41201345 and 41401399; the National High Technology Research and Development Program of China (863 Program) (2014AA06A511); the High Resolution Scientific and Technological Major Project of China (05-Y30B02-9001-13/15-9), and the Special Foundation for Young Scientists from the director of the Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences.

References and links

1. J. M. Chen and T. A. Black, “Defining leaf area index for non-flat leaves,” Plant Cell Environ. 15(4), 421–429 (1992). [CrossRef]

2. N. J. J. Bréda, “Ground-based measurements of leaf area index: a review of methods, instruments and current controversies,” J. Exp. Bot. 54(392), 2403–2417 (2003). [CrossRef] [PubMed]

3. I. Jonckheere, S. Fleck, K. Nackaerts, B. Muys, P. Coppin, M. Weiss, and F. Baret, “Review of methods for in situ leaf area index determination,” Agric. For. Meteorol. 121(1-2), 19–35 (2004). [CrossRef]

4. S. Gao, Z. Niu, N. Huang, and X. Hou, “Estimating the Leaf Area Index, height and biomass of maize using HJ-1 and RADARSAT-2,” Int. J. Appl. Earth. Obs. 24, 1–8 (2013). [CrossRef]

5. A. Kross, H. McNairn, D. Lapen, M. Sunohara, and C. Champagne, “Assessment of RapidEye vegetation indices for estimation of leaf area index and biomass in corn and soybean crops,” Int. J. Appl. Earth. Obs. 34, 235–248 (2015). [CrossRef]

6. C. Y. Wu, Z. Niu, J. D. Wang, S. A. Gao, and W. J. Huang, “Predicting leaf area index in wheat using angular vegetation indices derived from in situ canopy measurements,” Can. J. Rem. Sens. 36(4), 301–312 (2010). [CrossRef]

7. Y. Qi, F. Li, Z. Liu, and G. Jin, “Impact of understory on overstorey leaf area index estimation from optical remote sensing in five forest types in northeastern China,” Agric. For. Meteorol. 198–199, 72–80 (2014). [CrossRef]

8. K. K. Singh, A. J. Davis, and R. K. Meentemeyer, “Detecting understory plant invasion in urban forests using LiDAR,” Int. J. Appl. Earth. Obs. 38, 267–279 (2015). [CrossRef]

9. K. Zhao and S. Popescu, “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(8), 1628–1645 (2009). [CrossRef]

10. J. J. Richardson, L. M. Moskal, and S.-H. Kim, “Modeling approaches to estimate effective leaf area index from aerial discrete-return LIDAR,” Agric. For. Meteorol. 149(6-7), 1152–1160 (2009). [CrossRef]

11. W. Li, Z. Niu, C. Wang, W. Huang, H. Chen, S. Gao, D. Li, and S. Muhammad, “Combined Use of Airborne LiDAR and Satellite GF-1 Data to Estimate Leaf Area Index, Height, and Aboveground Biomass of Maize During Peak Growing Season,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 8(9), 4489–4501 (2015). [CrossRef]

12. S. Luo, C. Wang, X. Xi, and F. Pan, “Estimating FPAR of maize canopy using airborne discrete-return LiDAR data,” Opt. Express 22(5), 5106–5117 (2014). [CrossRef] [PubMed]

13. W. Li, Z. Niu, N. Huang, C. Wang, S. Gao, and C. Wu, “Airborne LiDAR technique for estimating biomass components of maize: A case study in Zhangye City, Northwest China,” Ecol. Indic. 57, 486–496 (2015). [CrossRef]

14. L. Korhonen, I. Korpela, J. Heiskanen, and M. Maltamo, “Airborne discrete-return LIDAR data in the estimation of vertical canopy cover, angular canopy closure and leaf area index,” Remote Sens. Environ. 115(4), 1065–1080 (2011). [CrossRef]

15. S. Luo, C. Wang, F. Pan, X. Xi, G. Li, S. Nie, and S. Xia, “Estimation of wetland vegetation height and leaf area index using airborne laser scanning data,” Ecol. Indic. 48, 550–559 (2015). [CrossRef]

16. T. Hermosilla, N. C. Coops, L. A. Ruiz, and L. M. Moskal, “Deriving pseudo-vertical waveforms from small-footprint full-waveform LiDAR data,” Remote Sens. Lett. 5(4), 332–341 (2014). [CrossRef]

17. S. Gao, Z. Niu, G. Sun, D. Zhao, K. Jia, and Y. Qin, “Height Extraction of Maize Using Airborne Full-Waveform LIDAR Data and a Deconvolution Algorithm,” IEEE Geosci. Remote. 12(9), 1978–1982 (2015). [CrossRef]

18. M. A. Lefsky, W. B. Cohen, S. A. Acker, G. G. Parker, T. A. Spies, and D. Harding, “Lidar remote sensing of the canopy structure and biophysical properties of Douglas-fir western hemlock forests,” Remote Sens. Environ. 70(3), 339–361 (1999). [CrossRef]

19. H. Tang, R. Dubayah, A. Swatantran, M. Hofton, S. Sheldon, D. B. Clark, and B. Blair, “Retrieval of vertical LAI profiles over tropical rain forests using waveform lidar at La Selva, Costa Rica,” Remote Sens. Environ. 124, 242–250 (2012). [CrossRef]

20. S. Luo, C. Wang, G. Li, and X. Xi, “Retrieving leaf area index using ICESat/GLAS full-waveform data,” Remote Sens. Lett. 4(8), 745–753 (2013). [CrossRef]

21. S. Nie, C. Wang, P. Dong, and X. Xi, “Estimating leaf area index of maize using airborne full-waveform lidar data,” Remote Sens. Lett. 7(2), 111–120 (2016). [CrossRef]

22. L. A. Magruder, A. L. Neuenschwander, and S. P. Marmillion, “Lidar waveform stacking techniques for faint ground return extraction,” J. Appl. Remote Sens. 4(1), 043501 (2010). [CrossRef]

23. K. D. Fieber, I. J. Davenport, M. A. Tanase, J. M. Ferryman, R. J. Gurney, V. M. Becerra, J. P. Walker, and J. M. Hacker, “Validation of Canopy Height Profile methodology for small-footprint full-waveform airborne LiDAR data in a discontinuous canopy environment,” ISPRS J. Photogramm. Remote Sens. 104, 144–157 (2015). [CrossRef]

24. V. H. Duong, R. Lindenbergh, N. Pfeifer, and G. Vosselman, “Single and two epoch analysis of ICESat full waveform data over forested areas,” Int. J. Remote Sens. 29(5), 1453–1473 (2008). [CrossRef]

25. Y. Qin, S. Li, T. T. Vu, Z. Niu, and Y. Ban, “Synergistic application of geometric and radiometric features of LiDAR data for urban land cover mapping,” Opt. Express 23(11), 13761–13775 (2015). [CrossRef] [PubMed]

26. Y. Qin, W. Yao, T. Vu, S. Li, Z. Niu, and Y. Ban, “Characterizing Radiometric Attributes of Point Cloud Using a Normalized Reflective Factor Derived From Small Footprint LiDAR Waveform,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 8(2), 740–749 (2015). [CrossRef]

27. Y. Qin, T. T. Vu, and Y. F. Ban, “Toward an Optimal Algorithm for LiDAR Waveform Decomposition,” IEEE Geosci Remote Sens. 9(3), 482–486 (2012). [CrossRef]

28. W. Ni-Meister, D. L. B. Jupp, and R. Dubayah, “Modeling lidar waveforms in heterogeneous and discrete canopies,” IEEE Geosci Remote. 39(9), 1943–1958 (2001). [CrossRef]

29. D. J. Harding, M. A. Lefsky, G. G. Parker, and J. B. Blair, “Laser altimeter canopy height profiles - Methods and validation for closed-canopy, broadleaf forests,” Remote Sens. Environ. 76(3), 283–297 (2001). [CrossRef]

30. J. Armston, M. Disney, P. Lewis, P. Scarth, S. Phinn, R. Lucas, P. Bunting, and N. Goodwin, “Direct retrieval of canopy gap probability using airborne waveform lidar,” Remote Sens. Environ. 134, 24–38 (2013). [CrossRef]

31. C. Hopkinson, L. E. Chasmer, G. Sass, I. F. Creed, M. Sitar, W. Kalbfleisch, and P. Treitz, “Vegetation class dependent errors in lidar ground elevation and canopy height estimates in a boreal wetland environment,” Can. J. Rem. Sens. 31(2), 191–206 (2005). [CrossRef]

32. W. Li, Z. Niu, S. Gao, N. Huang, and H. Y. Chen, “Correlating the Horizontal and Vertical Distribution of LiDAR Point Clouds with Components of Biomass in a Picea crassifolia Forest,” Forests 5(8), 1910–1930 (2014). [CrossRef]

33. S. Solberg, “Mapping gap fraction, LAI and defoliation using various ALS penetration variables,” Int. J. Remote Sens. 31(5), 1227–1244 (2010). [CrossRef]

34. A. Peduzzi, R. H. Wynne, T. R. Fox, R. F. Nelson, and V. A. Thomas, “Estimating leaf area index in intensively managed pine plantations using airborne laser scanner data,” For. Ecol. Manage. 270, 54–65 (2012). [CrossRef]

35. S. Martinuzzi, L. A. Vierling, W. A. Gould, M. J. Falkowski, J. S. Evans, A. T. Hudak, and K. T. Vierling, “Mapping snags and understory shrubs for a LiDAR-based assessment of wildlife habitat suitability,” Remote Sens. Environ. 113(12), 2533–2546 (2009). [CrossRef]

36. W. Li, Z. Niu, G. Sun, S. Gao, and M. Wu, “Deriving backscatter reflective factors from 32-channel full-waveform LiDAR data for the estimation of leaf biochemical contents,” Opt. Express 24(5), 4771–4785 (2016). [CrossRef]

37. W. Li, G. Sun, Z. Niu, S. Gao, and H. Qiao, “Estimation of leaf biochemical content using a novel hyperspectral full-waveform LiDAR system,” Remote Sens. Lett. 5(8), 693–702 (2014). [CrossRef]

Parameter	Specifications	Parameter	Specifications
Laser Scanning System	Leica ALS70	Average Footprint Size	42 cm
Laser Wavelength	1064 nm	Flight Line Overlap	50%
Beam Divergence	0.15 mrad	Average Point Density	3.6 pts/m²
Scan Angle	± 12°	Flight Area	145 km²
Nominal Flight Height	2800 m	Flight Date	2014.07.27

Vegetation Layer		Number of Plots	Plot Size
Overstorey	Understorey	Number of Plots	Plot Size
Apricot	Beam	5	10m
Apricot	Maize	5	10m
Crabalppl	Beam	8	10m
Pear	Maize	2	10m

Statistics	Canopy height (m)		Leaf area index
Statistics	H_over	H_under	LAI_over	LAI_under	LAI_total
Mean	3.82 (0.68)	0.85 (0.15)	2.00 (0.19)	0.82 (0.65)	3.15 (0.18)
Max	5.72 (1.76)	2.37 (0.54)	2.76 (0.78)	2.57 (1.62)	4.63 (0.51)
Min	2.36 (0.26)	0.38 (0.03)	1.11 (0.01)	0.29 (0.12)	2.55 (0.01)
SD	0.79 (0.31)	0.57 (0.13)	0.43 (0.19)	0.55 (0.35)	0.54 (0.14)

Statistics	H_over (m)	H_under (m)	LAI_over	LAI_under	LAI_total
Mean	3.91	0.95	1.95	1.06	3.01
Max	6.11	2.48	2.99	2.31	4.48
Min	2.76	0.41	1.07	0.14	1.91
SD	0.85	0.56	0.55	0.64	0.65

Parameter	Specifications	Parameter	Specifications
Laser Scanning System	Leica ALS70	Average Footprint Size	42 cm
Laser Wavelength	1064 nm	Flight Line Overlap	50%
Beam Divergence	0.15 mrad	Average Point Density	3.6 pts/m²
Scan Angle	± 12°	Flight Area	145 km²
Nominal Flight Height	2800 m	Flight Date	2014.07.27

Generating pseudo large footprint waveforms from small footprint full-waveform airborne LiDAR data for the layered retrieval of LAI in orchards

Abstract

1. Introduction

2. Materials

2.1 Study area

2.2 Small footprint full-waveform airborne LiDAR data

2.3 Field measurements

3. Methodology

3.1 Generation of pseudo large footprint waveforms

3.2 Discrimination of different vegetation layers from LiDAR

3.3 Retrieving LAIs for different layers

3.4 Validation

4. Results

4.1 Pseudo large footprint waveforms

4.2 Retrieval of layered LAI in orchards

5. Discussion

6. Conclusion

Acknowledgments

References and links

Cited By

Figures (5)

Tables (4)

Equations (10)

Optics Express