A New Method for Forest Canopy Hemispherical Photography Segmentation Based on Deep Learning

Canopy leaf area index for apple tree using hemispherical photography in arid region

Abstract. Surface roughness parameters, namely the roughness length and displacement height, are an integral input used to model surface fluxes. However, most models assume these parameters to be a fixed property of plant functional type and disregard the governing structural heterogeneity and dynamics. In this study, we use large-eddy simulations to explore, in silico, the effects of canopy-structure characteristics on surface roughness parameters. We performed a virtual experiment to test the sensitivity of resolved surface roughness to four axes of canopy structure: (1) leaf area index, (2) the vertical profile of leaf density, (3) canopy height, and (4) canopy gap fraction. We found roughness parameters to be highly variable, but uncovered positive relationships between displacement height and maximum canopy height, aerodynamic canopy height and maximum canopy height and leaf area index, and eddy-penetration depth and gap fraction. We also found negative relationships between aerodynamic canopy height and gap fraction, as well as between eddy-penetration depth and maximum canopy height and leaf area index. We generalized our model results into a virtual "biometric" parameterization that relates roughness length and displacement height to canopy height, leaf area index, and gap fraction. Using a decade of wind and canopy-structure observations in a site in Michigan, we tested the effectiveness of our model-driven biometric parameterization approach in predicting the friction velocity over heterogeneous and disturbed canopies. We compared the accuracy of these predictions with the friction-velocity predictions obtained from the common simple approximation related to canopy height, the values calculated with large-eddy simulations of the explicit canopy structure as measured by airborne and ground-based lidar, two other parameterization approaches that utilize varying canopy-structure inputs, and the annual and decadal means of the surface roughness parameters at the site from meteorological observations. We found that the classical representation of constant roughness parameters (in space and time) as a fraction of canopy height performed relatively well. Nonetheless, of the approaches we tested, most of the empirical approaches that incorporate seasonal and interannual variation of roughness length and displacement height as a function of the dynamics of canopy structure produced more precise and less biased estimates for friction velocity than models with temporally invariable parameters.

The quantification of forest carbon sequestration is helpful to understand the carbon storage on the Earth. The estimation of forest carbon sequestration can be achieved by the use of leaf area index (LAI), which is derived from forest gap fraction. The hemispherical image-based technique is the most popular non-destructive means for obtaining such information. However, only the gap fraction of the top canopy is derived due to the limitation of imaging technique. The gap fraction information of understory is thus neglected. In this study, we evaluate the use of a terrestrial laser scanner (TLS) to obtain the forest canopy and understory gap fraction. The forest TLS data were manually classified as the top canopy and understory layers to facilitate the estimation of top canopy and understory gap fraction, respectively.

Canopies are major part of plant photosynthesis and have distinct architectural elements such as tree crowns, whorls, branches, shoots, etc. By measuring canopy structural parameters, the solar radiation interception, photosynthesis effects and the spatio-temporal distribution of solar radiation under the canopy can be evaluated. Among canopy structure parameters, Leaf Area Index (LAI) is the key one. Leaf area index is a crucial variable in agronomic and environmental studies, because of its importance for estimating the amount of radiation intercepted by the canopy and the crop water requirements. The LAI can be achieved by hemispheric images which are obtained below the canopy with high accuracy and effectiveness. But existing hemispheric images canopy-LAI measurement technique is based on digital SLR camera with a fisheye lens. Users need to collect hemispheric image manually. The SLR camera with fisheye lens is not suit for long-term canopy-LAI outdoor measurement too. And the high cost of SLR limits its capacity. In recent years, with the development of embedded system and image processing technology, low cost remote canopy hemispheric image acquisition technology is becoming possible. In this paper, we present a remote hemispheric canopy image acquisition system with in-field/host configuration. In-field node based on imbed platform, low cost image sensor and fisheye lens is designed to achieve hemispherical image of plant canopy at distance with low cost. Solar radiation and temperature/humidity data, which are important for evaluating image data validation, are obtained for invalid hemispherical image elimination and node maintenance too. Host computer interacts with in-field node by 3G network. The hemispherical image calibration and super resolution are used to improve image quality in host computer. Results show that the remote canopy image collection system can make low cost remote canopy image acquisition for LAI effectively. It will be a potential technology candidate for low-cost remote canopy hemispherical image collection to measure canopy LAI.

This paper develops an algorithm to estimate vegetation canopy gap fraction (GF), taking advantage of the full Terrestrial Laser Scanner (TLS) resolution. After calculating the TLS angular resolution, the algorithm identifies the missing laser hits (gaps) within an angular grid in the azimuthal and zenithal directions. The algorithm was first tested on angular data simulations with random (R), cluster (C) and random and cluster together (RC) gap pattern distributions. Noise introduced in the simulations as a percentage of the resolution accounted for the effect of TLS angular uncertainty. The algorithm performs accurately if angular noise is <6% of the angular resolution. To assess the impact of the change in projection, this study compared these GF estimates from angular grid simulations to their transformation into simulated hemispherical images (SHI). SHI with C patterns perform accurately, but R and RC patterns underestimate GF, especially for GF values below 0.6. The SHI performance to estimate GF was always far below the algorithm developed here with the angular grid simulations. When applied to actual TLS data acquired over individual Quercus ilex L. trees, the algorithm rendered a GF between 0.26 and 0.40. TLS had an angular noise <6%. Converting the angular grid into simulated HI (TLS-SHI) provided a better agreement with actual HI acquired in the same location as the TLS data, since they are in the same projection. The TLS-SHI underestimated GF by an average of 4% compared to HI. HI and TLS-SHI presented 14% and 17% lower values than the GF calculated from the angular grids, respectively. Nevertheless, the results from the simulations indicate that the algorithm based on the angular grid should be closer to the actual GF of the tree canopy.

We used field measurements and Monte Carlo simulations of canopy gap-size distribution and gap fraction to examine how beam radiation interacts with clumped boreal forest canopies of aspen (Populus tremuloides Michx.), black spruce (Picea mariana (Mill.) B.S.P.) and jack pine (Pinus banksiana Lamb.). We demonstrate that the Beer-Lambert law can be modified to accommodate transmission of radiation through a clumped forest canopy as a function of path length or sun zenith angle. Multiband Vegetation Imager (MVI) measurements and Monte Carlo simulations showed that values of the zenith element clumping index (Omega(e)(0)) are typically between 0.4 and 0.5 in jack pine and black spruce and 0.65 in aspen. Estimates of LAI obtained from MVI measurements of the canopy gap fraction and adjusted for canopy clumping and branch architecture yielded LAI values of 3.0 in jack pine, 3.3 in aspen, and about 6.0 in black spruce. These LAI estimates were within 10-25% of direct measurements made at the same sites. Data obtained with the MVI, along with numerical simulations, demonstrated that assumptions of random foliage distributions in boreal forests are invalid and could yield erroneous values of LAI measured by indirect techniques and false characterizations of atmosphere-biosphere interactions. Monte Carlo simulations were used to develop a general equation for beam radiation penetration as a function of zenith angle in clumped canopies. The essential measurements included stem spacing, crown diameter, crown depth, and within-crown gap fraction.

The clumping index (CI) is an important factor for deriving forest leaf area index (LAI) at both pixel and point scale. Due to the non-uniform distribution of leaves within trees and random layout of trees under natural conditions, the CI varies with view zenith angle (VZA) as well as the gap fraction. The variation tendency of gap fraction and CI with VZA dominates the potential uncertainty of deriving LAI from directional observations. In this research, the continuous change of gap fraction and CI with VZA was simulated under virtual tree scenarios. Meanwhile, the directional CI was calculated using an algebraic model developed based on previous researches on gap fraction. Both results were validated using ground measurements. The CI discrepancy distributes within ±0.06 coming from the random number distribution restriction and the influence from tree trunk. The analysis using the algebraic model shows that the forest CI reaches an asymptotic level after 55° for forest canopy dense enough. This is an interesting and important phenomenon, which reveal a stable feature of CI to be further used to extract forest information.

Evaluation of Canopy Biophysical Variable Retrieval Performances from the Accumulation of Large Swath Satellite Data

Biotic and abiotic factors how to influence soil respiration in different young monoculture plantations are not clearly understood. Soil respiration and its controlling factors were studied in six monoculture plantations in the coastal area of Shanghai, China. Soil respiration was significant difference among six stands. Variations of soil respiration in six plots were not directly related to changes in soil water content, but significant relationship was observed between soil respiration and soil temperature. The variation of soil respiration was firmly correlated to the variation of leaf area index (LAI) or gap fraction (GF), soil respiration enhanced with the increase of GF (or decreasing LAI). The microclimate within forest and soil temperature also had positively correlation with soil respiration, but which mainly were affected by GF or LAI. There was no significant relationship between soil respiration and either root biomass or soil nutrients.

The structure of a forest canopy is the key determinant of light transmission, use and understory availability. Airborne light detection and ranging (LiDAR) has been used successfully to measure multiple canopy structural properties, thereby greatly reducing the fieldwork required to map spatial variation in structure. However, lidar metrics to date do not reflect the full extent of the three-dimensional information available from the data. To this end, we developed a new metric, the polar grid fraction (GRID), based on gridding lidar returns in polar coordinates, in order to more closely match measurements provided by field instruments on leaf area index (LAI), gap fraction (GF) and percentage photosynthetically active radiation transmittance (tPAR). The metric summarizes the arrangement of lidar point returns for a single ground location rather than to an area surrounding the location.Compared with more traditional proportion-based and height percentile-based estimators, the GRID estimator increased validation R2 by 14.5% for GF and 6.0% for tPAR over the next best estimator. LAI was still best estimated with the more traditional statistic based on the proportion of ground returns in 14 m × 14 m moving kernels. By applying the models to a 2 × 2 m grid across the lidar coverage area, extreme values occurred in the estimations of all three response variables when using proportion-based and height percentile-based estimators. However, no extreme values were estimated by models using the GRID estimator, indicating that models based on GRID may be less influenced by spurious data. These results suggest that the GRID estimator is a strong candidate for any project requiring estimates of canopy metrics for large areas.

On the correct estimation of effective leaf area index: Does it reveal information on clumping effects?

Non-destructive estimation of canopy gap fractions and shrub canopy volume of dominant shrub species in the Mojave desert

Vegetation crown cover is a popular structural descriptor in both the remote sensing and ecological fields. However, several different definitions of vegetation crown cover are commonly measured and used, and relating one against the other is often difficult. In this study we derive simple expressions relating mature forest crown cover (CC), foliage projective cover (FPC) and leaf area index (LAI) using canopy gap fraction and canopy clumping theory. The relationship between FPC and CC is shown to be related to the ratio of the within canopy to between canopy clumping. A comprehensive field data set collected in forests and woodlands across Queensland is used to demonstrate the validity of the models. The results have widespread application in the modelling and analysis of forest structure. Introduction Cover of natural systems is complex to describe and map. In remote sensing, several measurements of vegetation crown cover are commonly used. These are crown cover (CC), foliage projective cover (FPC) and leaf area index (LAI). CC is defined as the percentage of ground area covered by the vertical projection of crowns, which in this case are assumed opaque. FPC is the percentage of ground area covered by the vertical projection of foliage (Walker and Hopkins 1990). LAI is defined here as half the total developed area of leaves per unit ground horizontal surface area (Chen et al. 1997). CC is a primary stand density variable in many forest inventories since it is readily understood by a variety of practitioners. It is relatively easily estimated in the field (Walker and Hopkins 1990) and from aerial photography (Fensham et al. 2003). FPC is more complex to measure in the field and is typically measured using a transect based method (Johansson 1985). FPC is more closely related to the photosynthetic and evaporative potential of a plant community than CC due to the low density foliage typical of many Australian woody species (Specht 1981). Herein, FPC refers to the overstorey component only, which includes woody life forms above 2 m (Specht 1983). The LAI is the main variable used to model many processes, such as canopy photosynthesis and evapotranspiration. It determines the size of the plant–atmosphere interface and thus plays a key role in the exchange of energy and mass between the canopy and the atmosphere (Chen et al. 1997; Simioni et al. 2003). CC treats the crown as an opaque object, however the FPC of individual crowns for most Australian woody plants is between 40% and 70% depending on crown architecture (Walker and Hopkins 1990). For the same leaf area, variation in foliage clumping and leaf orientation angle determines the FPC of individual crowns (Campbell 1990; Henry et al. 2002; Kucharik et al. 1999). Indeed, high variation in leaf orientation angle and foliage clumping across Australian woody species has been reported in the literature and shown to function to reduce levels of intercepted light at high sun elevations (Falster and Westoby 2003; King 1997). Forests are defined as vegetation greater than 2 m in height with a minimum crown cover of 20 %. A CC of 20% is suggested as being equivalent to a FPC approximating of 10 to 12% (Australia National Forest Inventory et al. 2003.). Despite the use of both FPC and CC as indicators of vegetation condition and structure in Australia there is a paucity of work quantifying the relationship across a range of native plant communities in Australia. This research links established models of canopy light interception, and used field data collected from plant communities across Queensland, Australia to fit the parameters in the models, resulting in an improved understanding of the usage, linkage and applicability of these fundamental cover metrics

How to quantify tree leaf area index in an open savanna ecosystem: A multi-instrument and multi-model approach

The forest canopy vertical structure, understory spatial distribution as well as bare soil ground under a forest canopy strongly affect the accuracy of retrieving leaf area index (LAI) using optical remotely sensed data. Combining the reflectance from two different view directions and geometric optical (GO) radiative model, we investigated the effects of background information on retrieving forest canopy LAI using moderate resolution data for temperate forest in Northeastern China. Firstly, the forest background reflectance (BR) was computed using the bidirectional reflectance distribution function (BRDF) product from MODIS data. Then the forest canopy effective leaf area index (LAIe) was retrieved using the four scale geometric optical radiative model after removing the effects of forest background. It was found that the LAIe estimates without considering and with considering forest background contribution can explain 46 % and 36 % variations of field measurement-based LAIe (p<; 0.01, N=28), respectively. The forest canopy LAIe estimation without removing background influence can result in the LAIe overestimation ranging from 20 % to 46 %, thus it is necessary to separate the contribution of forest background to total forest canopy reflectance.