Predicting forest structural parameters using the image texture derived from WorldView-2 multispectral imagery in a dryland forest,Israel

Estimation of forest structural parameters by field-based data collection methods is both expensive and time consuming. Satellite remote sensing is a low-cost alternative in modeling and mapping structural parameters in large forest areas. The current study investigates the potential of using WordView-2 multispectral satellite imagery for predicting forest structural parameters in a dryland plantation forest in Israel. The relationships between image texture features and the several structural parameters such as Number of Trees (NT), Basal Area (BA), Stem Volume (SV), Clark-Evans Index (CEI), Diameter Differentiation Index (DDI), Contagion Index (CI), Gini Coefficient (GC), and Standard Deviation of Diameters at Breast Heights (SDDBH) were examined using correlation analyses. These variables were obtained from 30 m × 30 m square-shaped plots. The Standard Deviation of Gray Levels (SDGL) as a first order texture feature and the second order texture variables based on Gray Level Co-occurrence Matrix (GLCM) were calculated for the pixels that corresponds to field plots. The results of the correlation analysis indicate that the forest structural parameters are significantly correlated with the image texture features. The highest correlation coefficients were calculated for the relationships between the SDDBH and the contrast of red band (r = 0.75, p < 0.01), the BA and the entropy of blue band (r = 0.73, p < 0.01), and the GC and the contrast of blue band (r = 0.71, p < 0.01). Each forest structural parameter was modeled as a function of texture measures derived from the satellite image using stepwise multi linear regression analyses. The determination coefficient (R²) and root mean square error (RMSE) values of the best fitting models, respectively, are 0.38 and 109.56 ha⁻¹ for the NT; 0.54 and 1.79 m² ha⁻¹ for the BA; 0.42 and 27.18 m³ ha⁻¹ for the SV; 0.23 and 0.16 for the CEI; 0.32 and 0.05 for the DDI; 0.25 and 0.06 for the CI; 0.50 and 0.05 for the GC; and 0.67 and 0.70 for the SDDBH. The leave-one-out cross-validation technique was applied for validation of the best-fitted models (R² > 0.50). In conclusion, cross-validated statistics confirmed that the structural parameters including the BA, SDDBH, and GC can be predicted and mapped with a reasonable accuracy using the texture features extracted from the spectral bands of WorldView-2 image.     

Terrestrial laser scanning to estimate plot-level forest canopy fuel properties

This paper evaluates the potential of a terrestrial laser scanner (TLS) to characterize forest canopy fuel characteristics at plot level. Several canopy properties, namely canopy height, canopy cover, canopy base height and fuel strata gap were estimated. Different approaches were tested to avoid the effect of canopy shadowing on canopy height estimation caused by deployment of the TLS below the canopy. Estimation of canopy height using a grid approach provided a coefficient of determination of R² = 0.81 and an RMSE of 2.47 m. A similar RMSE was obtained using the 99th percentile of the height distribution of the highest points, representing the 1% of the data, although the coefficient of determination was lower (R² = 0.70). Canopy cover (CC) was estimated as a function of the occupied cells of a grid superimposed upon the TLS point clouds. It was found that CC estimates were dependent on the cell size selected, with 3 cm being the optimum resolution for this study. The effect of the zenith view angle on CC estimates was also analyzed. A simple method was developed to estimate canopy base height from the vegetation vertical profiles derived from an occupied/non-occupied voxels approach. Canopy base height was estimated with an RMSE of 3.09 m and an R² = 0.86. Terrestrial laser scanning also provides a unique opportunity to estimate the fuel strata gap (FSG), which has not been previously derived from remotely sensed data. The FSG was also derived from the vegetation vertical profile with an RMSE of 1.53 m and an R² = 0.87.     

Spectral mixture analysis to monitor defoliation in mixed-aged Eucalyptus globulus Labill plantations in southern Australia using Landsat5-TM and EO-1Hyperion data

Defoliation is a key parameter of forest health and is associated with reduced productivity and tree mortality. Assessing the health of forests requires regular observations over large areas. Satellite remote sensing provides a cost-effective alternative to traditional ground-based assessment of forest health, but assessing defoliation can be difficult due to mixed pixels where vegetation cover is low or fragmented. In this study we apply a novel spectral unmixing technique, referred to as weighted Multiple Endmember Spectral Mixture Analysis (wMESMA), to Landsat 5-TM and EO-1 Hyperion data acquired over a Eucalyptus globulus (Labill.) plantation in southern Australia. This technique combines an iterative mixture analysis cycle allowing endmembers to vary on a per pixel basis (MESMA) and a weighting algorithm that prioritizes wavebands based on their robustness against endmember variability. Spectral mixture analysis provides an estimate of the physically interpretable canopy cover, which is not necessarily correlated with defoliation in mixed-aged plantations due to natural variation in canopy cover as stands age. There is considerable variability in the degree of defoliation as well as in stand age among sites and in this study we found that results were significantly improved by the inclusion of an age correction algorithm for both the multi-spectral (R²_{no age correction} = 0.55 vs R²_{age correction} = 0.73 for Landsat) and hyperspectral (R²_{no age correction} = 0.12 vs R²_{age correction} = 0.50 for Hyperion) image data. The improved accuracy obtained from Landsat compared to the Hyperion data illustrates the potential of applying SMA techniques for analysis of multi-spectral datasets such as MODIS and SPOT-VEGETATION.     

Tree survey and allometric models for tiger bush in northern Senegal and comparison with tree parameters derived from high resolution satellite data

A tree survey and an analysis of high resolution satellite data were performed to characterise the woody vegetation within a 10 × 10 km² area around a site located close to the town of Dahra in the semi-arid northern part of Senegal. The surveyed parameters were tree species, height, tree crown radius, and diameter at breast height (DBH), for which allometric models were determined. An object-based classification method was used to determine tree crown cover (TCC) from Quickbird data. The average TCC from the tree survey and the respective TCC from remote sensing were both about 3.0%. For areas beyond the surveyed areas TCC varied between 3.0% and 4.5%. Furthermore, an empirical correction factor for tree clumping was obtained, which considerably improved the estimated number of trees and the estimated average tree crown area and radius. An allometric model linking TCC to tree stem crosssectional area (CSA) was developed, which allows to estimate tree biomass from remote sensing. The allometric models for the three main tree species found performed well and had r²-values of about 0.7–0.8.     

Sub-footprint analysis to uncover tree height variation using ICESat/GLAS

Detailed forest height data are an indispensable prerequisite for many forestry and earth science applications. Existing research of using Geoscience Laser Altimeter System (GLAS) data mainly focuses on deriving average or maximum tree heights within a GLAS footprint, i.e. an ellipse with a diameter of 65 m. However, in most forests, it is likely that the tree heights within such ellipse are heterogeneous. Therefore, it is desired to uncover detailed tree height variation within a GLAS footprint. To the best of our knowledge, no such methods have been reported as of now. In this study, we aim to characterize tree heights’ variation within a GLAS footprint as different layers, each of which corresponds to trees with similar heights. As such, we developed a new method that embraces two steps: first, a refined Levenberg–Marquardt (LM) algorithm is proposed to decompose raw GLAS waveform into multiple Gaussian signals, within which it is hypothesized that each vegetation signal corresponds to a particular tree height layer. Second, for each layer, three parameters were first defined: Canopy Top Height (CTH), Crown Length (CL), and Cover Proportion (CP). Then we extracted the three parameters from each Gaussian signal through a defined model. In order to test our developed method, we set up a study site in Ejina, China where the dominant specie is Populus euphratica. Both simulated and field tree height data were adopted. With regard to the simulation data, results presented a very high agreement for the three predefined parameters between our results and simulation data. When our methods were applied to the field data, the respective R² become 0.78 (CTH), CL (R² = 0.76), CP (R² = 0.74). Overall, our studies revealed that large footprint GLAS waveform data have the potentials for obtaining detailed forest height variation.     

Estimates of forest structure parameters from GLAS data and multi-angle imaging spectrometer data

Quantitative estimates of forest vertical and spatial distribution using remote sensing technology play an important role in better understanding forest ecosystem function, forest carbon storage and the global carbon cycle. Although most remote sensing systems can provide horizontal distribution of canopies, information concerning the vertical distribution of canopies cannot be detected. Fortunately, laser radars have become available, such as GLAS (Geoscience laser altimeter system). Because laser radar can penetrate foliage, it is superior to other remote sensing technologies for detecting vertical forest structure and has higher accuracy. GLAS waveform data were used in this study to retrieve average tree height and biomass in a GLAS footprint area in Heilongjiang Province. However, GLAS data are not spatially continuous. To fill the gaps, MISR (multi- angle imaging spectrometer) spectral radiance was chosen to predict the regional continuous tree height by developing a multivariate linear regression model. We compared tree height estimated by the regression model and GLAS data. The results confirmed that estimates of tree height and biomass based on GLAS data are considerably more accurate than estimates based on traditional methods. The accuracy is approximately 90%. MISR can be used to estimate tree height in continuous areas with a robust regression model. The R², precision and root mean square error of the regression model were 0.8, 83% and 1 m, respectively. This study provides an important reference for mapping forest vertical parameters.     

用地基激光雷达提取单木结构参数——以白皮松为例

以白皮松(Pinus bungeana Zucc)为研究对象,针对地基激光雷达TLS扫描的3维点云数据在单株木垂直方向的分布特征,提出了一种基于体元化方法的树干覆盖度变化检测方法,获取单木枝下高;然后根据获取的枝下高引入2维凸包算法获取垂直方向分层树冠轮廓,并计算树冠体积和冠幅;同时获取的单木参数还有胸径与树高。结果表明:单木枝下高的估测精度较高,R2与RMSE分别为0.97 m和0.21 m;胸径估测结果的R2与RMSE分别为0.79 cm和1.07 cm;采用逐步线性回归方法建立单木树冠体积与其他单木参数的相关关系,模型变量包括冠幅、叶子填充树冠长度和胸径,样本数为20,模型的R2与RMSE分别是0.967 m3和2.64 m3。本文方法能较准确地估测枝下高,TLS数据具有对树冠结构3维建模的潜力。     

Retrieval of chlorophyll and nitrogen in Norway spruce (Picea abies L. Karst.) using imaging spectroscopy

The research evaluated the information content of spectral reflectance (laboratory and airborne data) for the estimation of needle chlorophyll (C_AB) and nitrogen (C_N) concentration in Norway spruce (Picea abies L. Karst.) needles. To identify reliable predictive models different types of spectral transformations were systematically compared regarding the accuracy of prediction. The results of the cross-validated analysis showed that C_AB can be well estimated from laboratory and canopy reflectance data. The best predictive model to estimate C_AB was achieved from laboratory spectra using continuum-removal transformed data (R²cv = 0.83 and a relative RMSEcv of 8.1%, n = 78) and from hyperspectral HyMap data using band-depth normalised spectra (R²cv = 0.90, relative RMSEcv = 2.8%, n = 13). Concerning the nitrogen concentration, we observed somewhat weaker relations, with however still acceptable accuracies (at canopy level: R²cv = 0.57, relative RMSEcv = 4.6%). The wavebands selected in the regression models to estimate C_AB were typically located in the red edge region and near the green reflectance peak. For C_N, additional wavebands related to a known protein absorption feature at 2350 nm were selected. The portion of selected wavebands attributable to known absorption features strongly depends on the type of spectral transformation applied. A method called “water removal” (WR) produced for canopy spectra the largest percentage of wavebands directly or indirectly related to known absorption features. The derived chlorophyll and nitrogen maps may support the detection and the monitoring of environmental stressors and are also important inputs to many bio-geochemical process models.     

Estimating the parameters of forest inventory using machine learning and the reduction of remote sensing features

Locally computed statistics of image texture and a case-based reasoning (CBR) system were evaluated for mapping of forest attributes. Cluster analysis was preferred to regression models, as a pre-selection method of features. The best stand-based accuracy using satellite sensor images was 74.64 m⁻³ ha⁻¹ (36%) RMSE for stand volume, 1.98 m⁻³ ha⁻¹ a⁻¹ (49%) for annual increase in stand volume, where κ = 0.23 for stand growth classes and κ = 0.41 for dominant tree species in stands. The top pixel-based accuracy using orthophotos was 76.54 m⁻³ ha⁻¹ (41%) RMSE for stand volume, 1.87 m⁻³ ha⁻¹ a⁻¹ (44%) for annual increase in stand volume, where κ = 0.24 for stand growth classes and κ = 0.38 for dominant tree species in stands. Mean saturation in 30 m radius was the most useful feature when orthophotos were used, and standard deviation of Landsat ETM 6.2 values in 80 m radius was the best when satellite sensor images were used. The most valuable feature components (radii, channels and local statistics) for orthophotos were: 30 m kernel radius, lightness and the mean of pixel values; for satellite sensor images: 80 m kernel radius, near-infrared channel (ETM 4) and the mean of pixel values. Locally computed statistics.     

Exploring hyperspectral bands and estimation indices for leaf nitrogen accumulation in wheat

Hyperspectral sensing can provide an effective means for fast and non-destructive estimation of leaf nitrogen (N) status in crop plants. The objectives of this study were to design a new method to extract hyperspectral spectrum information, to explore sensitive spectral bands, suitable bandwidth and best vegetation indices based on precise analysis of ground-based hyperspectral information, and to develop regression models for estimating leaf N accumulation per unit soil area (LNA, g N m⁻²) in winter wheat (Triticum aestivum L.). Three field experiments were conducted with different N rates and cultivar types in three consecutive growing seasons, and time-course measurements were taken on canopy hyperspectral reflectance and LNA under the various treatments. Then, normalized difference spectral indices (NDSI) and ratio spectral indices (RSI) based on the original spectrum and the first derivative spectrum were constructed within the range of 350–2500 nm, and their relationships with LNA were quantified. The results showed that both LNA and canopy hyperspectral reflectance in wheat changed with varied N rates, with consistent patterns across different cultivars and seasons. The sensitive spectral bands for LNA existed mainly within visible and near infrared regions. The best spectral indices for estimating LNA in wheat were found to be NDSI (R₈₆₀, R₇₂₀), RSI (R₉₉₀, R₇₂₀), NDSI (FD₇₃₆, FD₅₂₆) and RSI (FD₇₂₅, FD₅₁₆), and the regression models based on the above four spectral indices were formulated as Y = 26.34x^1.887, Y = 5.095x − 6.040, Y = 0.609 e^3.008x and Y = 0.388x^1.260, respectively, with R² greater than 0.81. Furthermore, expanding the bandwidth of NDSI (R₈₆₀, R₇₂₀) and RSI (R₉₉₀, R₇₂₀) from 1 nm to 100 nm at 1 nm interval produced the LNA monitoring models with similar performance within about 33 nm and 23 nm bandwidth, respectively, over which the statistical parameters of the models became less stable. From testing of the derived equations, the model for LNA estimation on NDSI (R₈₆₀, R₇₂₀), RSI (R₉₉₀, R₇₂₀), NDSI (FD₇₃₆, FD₅₂₆) and RSI (FD₇₂₅, FD₅₁₆) gave R² over 0.79 with more satisfactory performance than previously reported models and physical models in wheat. It can be concluded that the present hyperspectral parameters of NDSI (R₈₆₀, R₇₂₀), RSI (R₉₉₀, R₇₂₀), NDSI (FD₇₃₆, FD₅₂₆) and RSI (FD₇₂₅, FD₅₁₆) can be reliably used for estimating LNA in winter wheat.     

Estimation of gross primary production in wheat from in situ measurements

Gross primary production (GPP) is a parameter of significant importance for carbon cycle and climate change research. Remote sensing combined with other climate and meteorological data offers a convenient tool for large scale GPP estimation. This paper presents a study of GPP estimation using three methods with in situ measurements of canopy reflectance, LAI, and the photosynthetically active radiation (PAR). First, because LAI is considered as an indicator of the factor of absorbed PAR (f_APAR), it provides reasonable estimates of GPP for all types of wheat with coefficient of determination R² of 0.7353. The second method uses four kinds of vegetation indices (VIs) to estimate GPP because these indices are suggested to be reliable candidates in the estimation of light use efficiency (LUE). Good determination coefficients were acquired in estimating GPP with R² ranging from the lowest of 0.7604 for NDVI to the highest of 0.8505 for EVI. A new method was proposed for the estimation of GPP following the Monteith logic, which considering GPP as a product of VI × VI × PAR. Results indicated that this method can provide the best estimates of GPP as determination coefficient R² increased largely compared to the other two methods. EVI × EVI × PAR was demonstrated to be the most suitable for the estimation of GPP with the highest R² of 0.9207, which was about 10% larger as compared to GPP estimated from the single EVI. These results will be helpful for the development of new models of GPP estimation with all remote sensing inputs.     

The effect of leaf-on and leaf-off forest canopy conditions on LiDAR derived estimations of forest structural diversity

Forest structural diversity metrics describing diversity in tree size and crown shape within forest stands can be used as indicators of biodiversity. These diversity metrics can be generated using airborne laser scanning (LiDAR) data to provide a rapid and cost effective alternative to ground-based inspection. Measures of tree height derived from LiDAR can be significantly affected by the canopy conditions at the time of data collection, in particular whether the canopy is under leaf-on or leaf-off conditions, but there have been no studies of the effects on structural diversity metrics. The aim of this research is to assess whether leaf-on/leaf-off changes in canopy conditions during LiDAR data collection affect the accuracy of calculated forest structural diversity metrics. We undertook a quantitative analysis of LiDAR ground detection and return height, and return height diversity from two airborne laser scanning surveys collected under leaf-on and leaf-off conditions to assess initial dataset differences. LiDAR data were then regressed against field-derived tree size diversity measurements using diversity metrics from each LiDAR dataset in isolation and, where appropriate, a mixture of the two. Models utilising leaf-off LiDAR diversity variables described DBH diversity, crown length diversity and crown width diversity more successfully than leaf-on (leaf-on models resulted in R² values of 0.66, 0.38 and 0.16, respectively, and leaf-off models 0.67, 0.37 and 0.23, respectively). When LiDAR datasets were combined into one model to describe tree height diversity and DBH diversity the models described 75% and 69% of the variance (R² of 0.75 for tree height diversity and 0.69 for DBH diversity). The results suggest that tree height diversity models derived from airborne LiDAR, collected (and where appropriate combined) under any seasonal conditions, can be used to differentiate between simple single and diverse multiple storey forest structure with confidence.     

Tree species classification using UAS-based digital aerial photogrammetry point clouds and multispectral imageries in subtropical natural forests

Tree species composition of forest stand is an important indicator of forest inventory attributes for assessing ecosystem health, understanding successional processes, and digitally displaying forest biodiversity. In this study, we acquired high spatial resolution multispectral and RGB imagery over a subtropical natural forest in southwest China using a fixed-wing UAV system. Digital aerial photogrammetric (DAP) technique was used to generate multi-spectral and RGB derived point clouds, upon which individual tree crown (ITC) delineation algorithms and a machine learning classifier were used to identify dominant tree species. To do so, the structure-from-motion method was used to generate RGB imagery-based DAP point clouds. Then, three ITC delineation algorithms (i.e., point cloud segmentation (PCS), image-based multiresolution segmentation (IMRS), and advanced multiresolution segmentation (AMRS)) were used and assessed for ITC detection. Finally, tree-level metrics (i.e., multispectral, texture and point cloud metrics) were used as metrics in the random forest classifier used to classify eight dominant tree species. Results indicated that the accuracy of the AMRS ITC segmentation was highest (F₁-score = 82.5 %), followed by the segmentation using PCS (F₁-score = 79.6 %), the IMRS exhibited the lowest accuracy (F₁-score = 78.6 %); forest types classification (coniferous and deciduous) had a higher accuracy than the classification of all eight tree species, and the combination of spectral, texture and structural metrics had the highest classification accuracy (overall accuracy = 80.20 %). In the classification of both eight tree species and two forest types, the classification accuracies were lowest when only using spectral metrics, indicated that the texture metrics and point cloud structural metrics had a positive impact on the classification (the overall accuracy and kappa accuracy increased by 1.49–4.46 % and 2.86–6.84 %, respectively).     

机载激光雷达及高光谱的森林乔木物种多样性遥感监测

利用机载LiDAR和高光谱数据并结合37个地面调查样本数据,基于结构差异与光谱变异理论,通过相关分析法分别筛选了3个最优林冠结构参数和6个最优光谱指数,在单木尺度上利用自适应C均值模糊聚类算法,在神农架国家自然保护区开展森林乔木物种多样性监测,实现了森林乔木物种多样性的区域成图。研究结果表明,(1)基于结合形态学冠层控制的分水岭算法可以获得较高精度的单木分割结果(R~2=0.88,RMSE=13.17,P0.001);(2)基于LiDAR数据提取的9个结构参数中,95%百分位高度、冠层盖度和植被穿透率为最优结构参数,与Shannon-Wiener指数的相关性达到R~2=0.39—0.42(P0.01);(3)基于机载高光谱数据筛选的16个常用的植被指数中,CRI、OSAVI、Narrow band NDVI、SR、Vogelmann index1、PRI与Shannon-Wiener指数的相关性最高(R~2=0.37—0.45,P0.01);(4)在研究区,利用以30 m×30 m为窗口的自适应模糊C均值聚类算法可预测的最大森林乔木物种数为20,物种丰富度的预测精度为R~2=0.69,RMSE=3.11,Shannon-Wiener指数的预测精度为R~2=0.70,RMSE=0.32。该研究在亚热带森林开展乔木物种多样性监测,是在区域尺度上进行物种多样性成图的重要实践,可有效补充森林生物多样性本底数据的调查手段,有助于实现生物多样性的长期动态监测及科学分析森林物种多样性的现状和变化趋势。     

Multi-level filtering segmentation to measure individual tree parameters based on Lidar data: Application to a mountainous forest with heterogeneous stands

This paper presents a method for individual tree crown extraction and characterisation from a canopy surface model (CSM). The method is based on a conventional algorithm used for localising LM on a smoothed version of the CSM and subsequently for modelling the tree crowns around each maximum at the plot level. The novelty of the approach lies in the introduction of controls on both the degree of CSM filtering and the shape of elliptic crowns, in addition to a multi-filtering level crown fusion approach to balance omission and commission errors. The algorithm derives the total tree height and the mean crown diameter from the elliptic tree crowns generated. The method was tested and validated on a mountainous forested area mainly covered by mature and even-aged black pine (Pinus nigra ssp. nigra [Arn.]) stands. Mean stem detection per plot, using this method, was 73.97%. Algorithm performance was affected slightly by both stand density and heterogeneity (i.e. tree diameter classes’ distribution). The total tree height and the mean crown diameter were estimated with root mean squared error values of 1.83 m and 1.48 m respectively. Tree heights were slightly underestimated in flat areas and overestimated on slopes. The average crown diameter was underestimated by 17.46% on average.     

Estimation of vegetation LAI from hyperspectral reflectance data: Effects of soil type and plant architecture

The retrieval of canopy biophysical variables is known to be affected by confounding factors such as plant type and background reflectance. The effects of soil type and plant architecture on the retrieval of vegetation leaf area index (LAI) from hyperspectral data were assessed in this study. In situ measurements of LAI were related to reflectances in the red and near-infrared and also to five widely used spectral vegetation indices (VIs). The study confirmed that the spectral contrast between leaves and soil background determines the strength of the LAI–reflectance relationship. It was shown that within a given vegetation species, the optimum spectral regions for LAI estimation were similar across the investigated VIs, indicating that the various VIs are basically summarizing the same spectral information for a given vegetation species. Cross-validated results revealed that, narrow-band PVI was less influenced by soil background effects (0.15 ≤ RMSE_cv ≤ 0.56). The results suggest that, when using remote sensing VIs for LAI estimation, not only is the choice of VI of importance but also prior knowledge of plant architecture and soil background. Hence, some kind of landscape stratification is required before using hyperspectral imagery for large-scale mapping of vegetation biophysical variables.     

Applications of AnnAGNPS model for soil loss estimation and nutrient loading for Malaysian conditions

The study evaluated the performance and suitability of AnnAGNPS model in assessing runoff, sediment loading and nutrient loading under Malaysian conditions. The watershed of River Kuala Tasik in Malaysia, a combination of two sub-watersheds, was selected as the area of study. The data for the year 2004 was used to calibrate the model and the data for the year 2005 was used for validation purposes. Several input parameters were computed using methods suggested by other researchers and studies carried out in Malaysia. The study shows that runoff was predicted well with an overall R² value of 0.90 and E value of 0.70. Sediment loading was able to produce a moderate result of R² = 0.66 and E = 0.49, nitrogen loading predictions were slightly better with R² = 0.68 and E = 0.53, and phosphorus loading performance was slightly poor with an R² = 0.63 and E = 0.33. The erosion map developed was in agreement with the erosion risk map produced by the Department of Agriculture, Malaysia. Rubber estates and urban areas were found to be the main contributors to soil erosion. The simulation results showed that AnnAGNPS has the potential to be used as a valuable tool for planning and management of watersheds under Malaysian conditions.     

Comparison of airborne laser scanning methods for estimating forest structure indicators based on Lorenz curves

The purpose of this study was to compare a number of state-of-the-art methods in airborne laser scanning (ALS) remote sensing with regards to their capacity to describe tree size inequality and other indicators related to forest structure. The indicators chosen were based on the analysis of the Lorenz curve: Gini coefficient (GC), Lorenz asymmetry (LA), the proportions of basal area (BALM) and stem density (NSLM) stocked above the mean quadratic diameter. Each method belonged to one of these estimation strategies: (A) estimating indicators directly; (B) estimating the whole Lorenz curve; or (C) estimating a complete tree list. Across these strategies, the most popular statistical methods for area-based approach (ABA) were used: regression, random forest (RF), and nearest neighbour imputation. The latter included distance metrics based on either RF (NN–RF) or most similar neighbour (MSN). In the case of tree list estimation, methods based on individual tree detection (ITD) and semi-ITD, both combined with MSN imputation, were also studied. The most accurate method was direct estimation by best subset regression, which obtained the lowest cross-validated coefficients of variation of their root mean squared error CV(RMSE) for most indicators: GC (16.80%), LA (8.76%), BALM (8.80%) and NSLM (14.60%). Similar figures [CV(RMSE) 16.09%, 10.49%, 10.93% and 14.07%, respectively] were obtained by MSN imputation of tree lists by ABA, a method that also showed a number of additional advantages, such as better distributing the residual variance along the predictive range. In light of our results, ITD approaches may be clearly inferior to ABA with regards to describing the structural properties related to tree size inequality in forested areas.     

Combining UAV-based plant height from crop surface models,visible, and near infrared vegetation indices for biomass monitoring in barley

In this study we combined selected vegetation indices (VIs) and plant height information to estimate biomass in a summer barley experiment. The VIs were calculated from ground-based hyperspectral data and unmanned aerial vehicle (UAV)-based red green blue (RGB) imaging. In addition, the plant height information was obtained from UAV-based multi-temporal crop surface models (CSMs). The test site is a summer barley experiment comprising 18 cultivars and two nitrogen treatments located in Western Germany. We calculated five VIs from hyperspectral data. The normalised ratio index (NRI)-based index GnyLi (Gnyp et al., 2014) showed the highest correlation (R² = 0.83) with dry biomass. In addition, we calculated three visible band VIs: the green red vegetation index (GRVI), the modified GRVI (MGRVI) and the red green blue VI (RGBVI), where the MGRVI and the RGBVI are newly developed VI. We found that the visible band VIs have potential for biomass prediction prior to heading stage. A robust estimate for biomass was obtained from the plant height models (R² = 0.80–0.82). In a cross validation test, we compared plant height, selected VIs and their combination with plant height information. Combining VIs and plant height information by using multiple linear regression or multiple non-linear regression models performed better than the VIs alone. The visible band GRVI and the newly developed RGBVI are promising but need further investigation. However, the relationship between plant height and biomass produced the most robust results. In summary, the results indicate that plant height is competitive with VIs for biomass estimation in summer barley. Moreover, visible band VIs might be a useful addition to biomass estimation. The main limitation is that the visible band VIs work for early growing stages only.