1. Introduction
Leaf area index (LAI) is an important parameter characterizing the structure of vegetation canopy. It is defined as half of the total leaf area on the ground at the unit level [
1]. It is closely related to vegetation photosynthesis [
2] and the carbon cycle [
3] and is a key parameter in quantitatively describing the growth and development characteristics of forest leaves [
4]. LAI provides an effective method for quantitative growth analysis of plant communities and has been widely applied to crop yield estimation [
5], the breeding of improved varieties [
6] and forest productivity comparation at the landscape level [
7]. The measurement of LAI has been divided into the direct method and indirect method [
8]. The direct method usually includes litter collection and destructive sampling to obtain LAI through manual measurement. However, it is difficult to collect litter data, and the processing process is complicated. The destructive method often causes irreversible damage to vegetation [
9]. Meanwhile, the indirect method, which involves tools such as optical instruments and remote sensing [
10], is used to calculate LAI by analyzing the light absorption mode and reflection characteristics of the vegetation canopy [
11]. Optical instruments can obtain stand-scale LAI with high precision and repeatability but can only obtain spatially discrete sample data, thus hindering large-scale continuous collection of vegetation information and posing high technical requirements [
12]. For example, the LAI-2200C plant canopy analyzer is commonly used to measure the ground LAI, and it specifies that the measurement should take place on a cloudy day or in the early morning [
13]. Before the measurement, the measurement points of the sample site should be calculated and laid out first. Then, all sky data should be measured in the open space outside the sample site and recorded as value A; the measurement points should be measured successively and recorded as value B. Finally, the ground LAI of the sample site should be calculated by matching values A and B [
14]. Remote sensing technology allows for macroscopic and efficient ground object observation and can overcome the defects of traditional methods in parameter extraction [
15]. However, optical passive remote sensing is susceptible to light saturation and LAI underestimation, so accurate ground data are needed for accuracy verification [
16].
Light Detection and Ranging (LiDAR) is an active remote sensing technology that uses a laser as a light source [
17]. The laser pulses emitted by LiDAR can penetrate vegetation canopy and obtain the three-dimensional structure information of forests accurately, making up for the light saturation defect of passive remote sensing. LiDAR has gradually become the main means by which to obtain the effective leaf area index (LAIe) [
18,
19,
20]. LAIe is the macroscopic LAI value obtained through optical instrument measurement or remote sensing under the assumption of random leaf distribution [
21]. Both LAIe and LAI reflect the life vitality, living environment and canopy structure of vegetation. At the macro scale, an LAIe obtained using optical instruments or remote sensing methods has more scientific basis and validity than an LAI acquired through complex calculation, and LAIe can replace LAI as an effective parameter for monitoring vegetation information dynamics. LAIe has the same function as LAI—characterizing the number of leaves—and the direct measurement result of LAI-2200C is LAIe. However, due to the limitations of the measurement angle of the sensor and the occlusion effect of LiDAR from different platforms [
22,
23], some point cloud data are missing, leading to uncertainty in LAI estimation. For example, You et al. [
24] carried out a study on the inversion accuracy of LAI by employing Airborne Laser Scanning (ALS) in coniferous forests and found that the laser could not penetrate the dense canopy, resulting in the overestimation of LAI. Moreover, in the estimation of LAI in a north tropical forest, Xie et al. [
25] mentioned that when Terrestrial Laser Scanning (TLS) was severely obscured, the estimation accuracy was lower than the LAI-2200C value. In order to overcome such a limitation, Guo et al. [
26] estimated the LAI of a
Pinus koraiensis forest by combining TLS and ALS data. It was confirmed that the combination of multi-source data could reduce the canopy shielding effect and improve LAI inversion accuracy. Many researchers have found that compared with data combination, data fusion can provide rich ground feature information and reflect the spatial structure characteristics of vegetation directly, while retaining the repeatability of information observation [
27,
28,
29]. In 2022, Piao et al. [
30] carried out LAIe inversion research by means of data fusion and found that the method had higher prediction accuracy and could avoid redundant errors caused by combining a large number of data. Therefore, the three-dimensional structure information of forests can be described more comprehensively using fusion data, but the factors that affect the accuracy of point cloud information still exist, such as point cloud density and point cloud thinning. In recent years, there have been studies showing the relationship between point cloud density, point cloud thinning and point cloud data [
31,
32], while the impact of point diameter on point cloud data has rarely been reported. Point clouds are composed of a massive collection of points with target surface characteristics. The point diameter can drive the internal structure of target point clouds to change to different degrees, affecting the accuracy of obtained forest structure parameters. Therefore, it is of great significance to explore the extraction of forest structure parameters from point cloud data based on the optimal point diameter to improve LAIe inversion accuracy.
Methods for estimating LAIe using LiDAR are divided into the physical model method and the empirical model method [
33]. The physical model method is based on the spectrophotometric method and calculates the light attenuation through the vegetation canopy and the estimation of LAI [
34]. It has the advantages of clear physical meaning, wide practicability and strong universality. Compared with the physical model, the empirical model can be used to establish a regression model by extracting the statistical parameters of point cloud data and measuring LAI [
35]. With high flexibility and few influencing factors, this method has significant advantages over remote sensing methods in LAI estimation [
36]. In recent years, in order to meet the accuracy requirements of LAI estimation via remote sensing technology, a large number of scholars built LAI inversion models based on vegetation index, laser penetration index and echo intensity [
20,
37,
38]. For example, Luo et al. [
39] estimated LAI in the forest area of Dayakou City based on the laser penetration index in 2013 and obtained a high-precision LAI inversion model. However, in previous research, the echo intensity needed to be corrected, and in mountainous areas with a high slope, the echo intensity was greatly affected by distance and incidence angle, resulting in a decrease in the estimation accuracy of LAI. In 2022, Stobbelaar et al. [
40] inverted the LAI of mixed broadleaf–conifer forest based on seven vegetation indices and compared it with the LAI value estimated using surface reflectance, proving that the prediction accuracy of modeling with additional variables was higher. However, due to the heterogeneity of leaf characteristics and growth rhythm among different tree species [
41], it was difficult to avoid the influence of the vegetation’s canopy size, diameter at breast height (DBH) and tree height parameters, resulting in large fluctuations in the estimated results of LAIe. Therefore, the inversion of LAIe based on multiple forest structure parameters of different forest types can reveal the relationship between LAIe and canopy structure more comprehensively, thus improving the LAIe inversion accuracy.
This study took the LAIe of three different conifer forests in Changchun City as the research object. Based on the fusion point cloud data derived from TLS and ALS, we explored the influence of different point diameters on forest structure parameter extraction and correlation. Ten forest structure parameters, including DBH and stand density, were extracted at the optimal diameter, and the relationship between different forest structure parameters and measured LAIe was analyzed. Using the polynomial regression method, LAIe inversion models driven by point cloud data were constructed for different forest types, and the variation rules of LAIe in different forest types were also investigated. Specifically, the three objectives of this study were as follows: (1) to explore the influence of point cloud diameter on the estimation of forest structure parameters and clarify the best parameters and formulas for LAIe estimation of different forest types; (2) to develop a high-precision LAIe estimation method driven by LiDAR point cloud data and explore the spatial distribution of and differences in LAIe in different forest types; and (3) to simplify LAIe estimation at the sample plot scale and provide a reference for fine LAIe estimation using remote sensing inversion on a large scale. This study attempted to make use of the accurate restoration characteristics of LiDAR point cloud data to evaluate forest structure and proposed an LAIe inversion method completely based on point cloud data, so as to break through the limitations of traditional methods regarding light environment changes and leaf spatial distribution and thus accelerate data acquisition efficiency, enhance estimation accuracy and improve calculation speed. Also, it provided a new way to estimate the LAIe of urban forests and validate large-scale LAI products.