Analysis of forest structural complexity using airborne LiDAR data and aerial photography in a mixed conifer–broadleaf forest in northern Japan

Sadeepa Jayathunga•Toshiaki Owari•Satoshi Tsuyuki

Introduction

Forest structure comprises numerous components,many of which are fundamental to the functioning and diversity of ecosystems(Spies 1998).However,it is often described by vertical and horizontal distribution of foliage(Spies 1998;Staudhammer and LeMay 2001;Van Pelt and Nadkarni 2003).On the other hand,forest structural complexity is essentially a measure of the number of different attributes and the relative abundance of each and correlates strongly with many ecological processes and services,including biodiversity(McElhinny et al.2005).Natural forests typically have complex structures,including heterogeneous spatial arrangements as well as a diversity of individual structures(Franklin and Van Pelt 2004).Recently,a great deal of attention has been paid to studying forest structural complexity due to its important role in forest management and conservation.

More complex forests may have higher functional diversity and(or)they may respond differently to new or variable conditions,creating more overall stability in ecosystem functions and enabling complex forests to sustain ecosystem services in the context of changing conditions(Bradford and Kastendick 2010).It is a widely accepted fact that a new forest management paradigm based on heterogeneity,unpredictability and adaptability,rather than on uniformity,predictability,and ‘command and control’is better suited to deal with future challenges(Messier and Puettmann 2011).Therefore,forest managers are increasingly interested in forest structural complexity that provides fundamental information required for successfulmanagementplanning and decision making(Whittaker et al.2005;Egoh et al.2008).

Traditionally,forest-related studies were based on the data acquired through expensive,date-lagged and time consuming fi eld surveys(Xie et al.2008)from relatively few and often subjectively selected plots,which limited the ability to study spatial patterns of forest structural complexity in some important ways(Kane et al.2010b).It is important to incorporate new scienti fi c developments and emerging technologies into data collection(Messier and Puettmann 2011)and explore their potential to minimize resource requirements,e.g.,cost,labor,time etc.,when necessary.Remote sensing techniques have offered practical,reliable and economical means to study forest structure,especially over large areas during a comparatively short period of time and less effort(Langely et al.2001;Ma et al.2001).

However,passive optical sensors are limited in their ability to capture forest structural complexity,particularly in uneven-aged mixed species forests where multiple canopy strata often exist(Wulder 1998;Olthof and King 2000;Lovell et al.2003).Airborne laser scanning(ALS)is an active remote sensing technique,which overcomes this limitation by providing otherwise unavailable scienti fi c insights through the detailed and novel structural measurements(Lovell et al.2003;Hyyppa et al.2012).Compared to other remote sensing options,ALS measurements have been shown to predict more accurate estimates of forest structural attributes(Lefsky et al.2001).Application of ALS measurements in analyzing forest structure using individual forest structural attributes has been researched intensively for coniferous forests(Næsset 2002;Maltamo et al.2005;Reitberger et al.2008)and for some broadleaf forests in terms of both area based approach(ABA)and individual tree based methods(Næsset et al.2004).Nonetheless,analysis of forest structural complexity using multivariate statistical techniques,by integrating multiple forest structural attributes(derived from LiDAR data or other remote sensing data sources),as opposed to individual forest structural attributes,is comparatively limited in the literature(Lefsky et al.2005;Wunderle et al.2007;de la Cueva 2008;Kane et al.2010b;Pasher and King 2010),especially for mixed conifer–broadleaf forests

Materials and methods

(Maltamo et al.2005;Chasmer et al.2004;Reitberger et al.2008).Dickinson et al.(2014)is the only recent study on mixed conifer–broadleaf forests along with several other forest types.

Furthermore,previous literature indicates a good complementary relationship between image data and LIDAR data,as they contain very different information:image data provide a detailed description of the spectral signatures but no information on the height of ground covers,whereas LIDAR data give detailed information about the height and three dimensional structure of forest cover but no information on the spectral signatures(Dalponte et al.2008).In this scenario,combination of airborne LiDAR data and image data,e.g.,high-resolution aerial photography holds promise to provide more detailed information including precise identi fi cation of tree species(Wulder et al.2013).However,most of the studies do not approach the integration of structural attributes derived from airborne LIDAR data and image data from a data combination perspective but address the problems in terms of separate use of these information sources.At present,only very few studies have properly exploited the complementary information present in airborne LiDAR data and high-resolution aerial imagery and investigated the joint use of these techniques to classify forest structural complexity in natural forest areas where many tree species are present(Dalponte et al.2008;Dickinson et al.2014).In this paper,we address this issue.

The aim of this study is to examine forest structure by integrating multiple attributesderived from airborne LiDAR data and high-resolution aerial photography to understand the complexity of mixed conifer–broadleaf forests in northern Japan.In this study,our objectives were:(1)to compare fi eld-,image-and LiDAR-based measures of stand structural complexity,(2)to develop a classi fi cation scheme to understand how complex the natural forest is,and(3)to identify the relationship between forest structural complexity and site condition(e.g.,elevation).

Study site

This study was carried out in the Maeyama Forest Reserve,an unmanaged mixed conifer–broadleaf forest in the University of Tokyo(UTokyo)Hokkaido Forest(Fig.1b).The UTokyo Hokkaido Forest is in Furano City in the central part of Hokkaido Island in northern Japan(43°10′–20′N,142°18′–40′E,18–1459 m a.s.l.)and has total area of about 22,716 ha.The mean temperature was 6.4°C and precipitation was 1297 mm at the Arboretum(230 m a.s.l.)during 2001–2008(The University of Tokyo Hokkaido Forest 2015).Snow covers the ground from late November to early April with the maximum depth of about 1 m(Owari et al.2011).

Fig.1 Maps to show locations of study area and inventory plots.a The University of Tokyo Hokkaido Forest in Japan,b Mayeama Forest Reserve in the University of Tokyo Hokkaido Forest and c inventory plots in Mayeama forest reserve

The UTokyo Hokkaido Forest is a pan-mixed conifer–broadleaf forest(Tatewaki 1958).Pan-mixed conifer–broadleafforestsare regarded asthetransitionzone between cool temperate broadleaf forests and sub-arctic coniferous forests(Matsuda et al.2002).Abies sachalinensis(Sakhalin fi r),one of the dominant tree species in the pan-mixed forest type,can grow at a wide range of altitudes(200 to about 1200 m a.s.l.;The University of Tokyo Hokkaido Forest 2015).Other common tree species include,among others,Picea jezoensis,P.glehnii,Fraxinus mandshurica,Kalopanax pictus,Quercus crispula,Betula maximowicziana,Taxus cuspidataandTilia japonica(Horie et al.2013).The forest fl oor is often occupied by dwarf bamboo(Sasa senanensisandS.kurilensis)(Owari et al.2011).Several vegetationtypescanbeobservedinthisforestareaincluding naturalforest,secondaryforestrecoveringfrom fi redamage,forest plantations and alpine zone vegetation.

Mayeama forest reserve

The speci fi c study site,the Mayeama Forest Reserve,has an area of 1239.38 ha and its elevation ranges from 450 to 1459 m a.s.l.This area was purposely selected for the present study due to the availability of area-wide remote sensing data including airborne LiDAR data and highresolution aerial photography and to the presence of different forest structure types.Main soil types found in Mayeama are black soil,light-colored black soil(both considered as paleosols with a tendency for weak podsolization)and brown forest soil.

All these soil types are closely correlated with vegetation(Nakata et al.1994).Major tree species includeA.sachalinensis,P.jezoensis,Betula ermanii,T.japonicaandP.glehnii(Owari et al.2011).Nakata et al.(1994)identi fi ed three altitudinal vegetation types in Mayeama on the basis of vegetation similarity and species composition:(1)Pinus pumilashrubs,which are distributed at altitudes exceeding 1250 m a.s.l.;(2)sub-alpine mixed forest(consisting mainly ofP.jezoensis,A.sachalinensisandB.ermanii),distributed between 700 m and 1250 m a.s.l.;(3)cool temperate mixed forest(consisting mainly ofA.sachalinensis,P.jezoensis,B.ermaniiand many cool temperate deciduous tree species),below 700 m a.s.l.

Field measurements

We used existing fi eld inventory data collected in April 2011 from a total of 131 circular sample plots(each having the radius of 17.84 m and an area of～0.1 ha)that were established in Mayeama and the surrounding area using a systematic sampling scheme with a spatial interval of 400 m.These data include the location of each plot center,DBH and species name of all the living trees with DBH≥5 cm,number of juveniles(DBH＜5 cm,height≤1.3 m),and height of three largest trees in each plot.Of the 131 plots,76 plots(30 at elevations lower than 800 m a.s.l.,40 between 800–1200 m a.s.l.and 6 above 1200 m.)were located completely inside Mayeama forest reserve(Fig.1c).

Only data from canopy layer trees were included when calculating fi eld metrics because only the fi rst returns of the LiDAR data were used to derive LiDAR metrics.The common practice of canopy tree discrimination is to use a single DBH threshold.However,a single DBH threshold may not discriminate canopy trees accurately in heterogeneous uneven-aged forests,especially in unmanaged mixed conifer–broadleaf forests,which are subjected to partial disturbances over time.Therefore,we used plot-speci fi c DBH thresholds calculated based on the percentage cumulative DBH2of each plot to discriminate canopy trees from the inventory data.

Five fi eld metrics that are commonly used to characterize the forest structure were selected for this study(Spies 1998;Zenner 2000;Dewalt et al.2003)and are summarized in Table 1.Mean DBH and stem density are the most basic and common attributes that explain size and distribution of trees,respectively(Avery and Burkhart 2002).Mean dominant height was selected as the fi eld measurement to represent biomass.Basal area(BA),i.e.,the area(usually in square meters)occupied by tree stems(Avery and Burkhart 2002),is often used to estimate tree growth rate(Pastur et al.2008).Proportion of broadleaf canopy trees represents the vegetation composition in the canopy layer(Dickinson et al.2014),which often in fl uences the amount of light penetrating the canopy and the composition and distribution of understory vegetation(Whittney and Foster 1988).

LiDAR data

LiDAR data were collected in September 2012 using a Leica ALS70 sensor mounted on a fi xed wing aircraft(Pasco Corp.,Japan).Nominal fl ying height was 1000 m above the ground.Data were acquired with a pulse rate of 267,200 Hz,a scan angle of±10°,and course overlap of 50–60%.Average point density was 8.4 points/m2.

Previous studies have identi fi ed and used many different LiDAR metrics for various purposes including direct and indirect estimation of forest attributes.However,vertical and horizontal complexity of forest structure can be characterized using three categories of LiDAR metrics.Firstly,LiDAR metrics to represent biomass such as height measurements,secondly,LiDAR metrics to measure vertical complexity such as variability of height,and fi nally,LiDAR metrics to measure canopy gaps such as canopy density(Lefsky et al.2005).We selected seven metrics that describe vertical and horizontal variations of forest structure(Table 1).

Digital terrain model(DTM;resolution of 0.5 m)was generated from the LiDAR data and calibrated for the height of forest fl oor vegetation(e.g.,dwarf bamboo).All canopy structural metrics were based on heights above this DTM.In this study,only the fi rst return LiDAR data and calibrated DTM were used when calculating the seven LiDAR metrics for each plot.LiDAR data were processed using TNTmips Pro 2015(MicroImages,USA)and when the software did not provide in-built functions to derive certain metrics(e.g.,percentile canopy heights and canopy density),custom TNTmips SML(spatial manipulation language)scripts were used to calculate such parameters.Surface area ratio(Table 1)was calculated using triangulated irregular network(TIN)and surface properties function in TNTmips Pro 2015.

Table 1 Description of fi eld-and LiDAR-derived metrics used in this study(Lefsky et al.2002;Parker et al.2004;Kane et al.2010a,b)

Orthographic aerial photography

Digital color aerial photographs with 0.5 m pixel resolution(acquired in June 2011 taken with a DMC camera mounted on a platform fl ying about 3170 m above the ground)were provided by the Japan Forestry Agency and Asia Air Survey Co.,Ltd.Aerial photography had forward overlap of 59–65%and lateral overlap of 32%.

Because the proportion of broadleaf cover can be considered as a fundamental variable characterizing mixed forests(Hame et al.1997),we quanti fi ed the proportion of broadleaf cover using high-resolution aerial photography following the same procedure as Dickinson et al.(2014).The area occupied by broadleaf species in aerial photographs was identi fi ed using supervised classi fi cation of three visible light bands with maximum likelihood method.The proportion of broadleaf cover was determined for each plot using the number of 50 cm pixels classi fi ed as broadleaf vegetation within each plot.

In this study,we used remote sensing data acquired in September 2012(airborne LiDAR)and June 2011(aerial photography)and fi eld data collected in April 2011.Time difference between the data acquisitions had no impact on the results because any signi fi cant events that disturbed the forest structure in the study site were not observed between April 2011 and September 2012.

Statistical analyses

The R statistical package(release 3.1.2)(R Core Team 2014)was used for the statistical analysis.It is important to perform a spatial autocorrelation test before the statistical analysis because some statistical procedures applied in this study are based on the assumption of independence of observations,samples or units of study.We tested the spatial autocorrelation in the data set by calculating Moran’sIvalue(Dale and Fortin 2002).

First,correlations between individual fi eld,LiDAR and image metrics were compared using Pearson coef fi cients to assess whether the LiDAR measurements capture the actual fi eld structural measurements in mixed conifer–broadleaf forests.Then,a principal component analysis(PCA)was performed for multivariate sets of metrics.PCA was appropriate for these data,as selected variables had approximately linear relationships with each other(McCune and Grace 2002).In the PCA ordinations,the length of each arrow indicates the variance of metrics,whereas the angles between arrows indicate the strength of the correlation between the metrics(smaller angles＜90°indicate higher positive correlations,90°anglesare uncorrelated,＞90°angles are negatively correlated,and 180°angles are perfectly negatively correlated).

LiDAR metrics for classi fi cation were selected following Lefsky et al.(2005):a height measurement,a canopy cover measurement and a measurement of height variability.The three selected LiDAR metrics,i.e.,95th percentile of canopy height,canopy density and surface area ratio,and proportion of broadleaf cover derived from aerial photography are considered to be attributes to describe dominant characteristics and the overall complexity of a multifaceted forest structure(Lefsky et al.2005;Kane et al.2010a;Dickinson et al.2014).Therefore,these four metrics were utilized to develop the forest structure classi fi cation scheme by using thek-means clustering algorithm to divide plots into forest structure classes.This simple unsupervised classi fi cation method was preferred as it results in forest structure classes that equally represent the metrics of forest structural complexity and has relatively less data processing requirements.A fundamental and frequent problem in cluster analysis is to determine the best estimate of the number of clusters,which is usually decided before clustering data(Yan 2005).The best estimate of the number of clusters for thek-means algorithm was identi fi ed by repeating the analysis for 1–25 clusters and creating a scree plot of the within-cluster sum of squares against the number of clusters(Everitt and Hothorn 2011).

Previous studies found relationships between elevation and tree attributes such as tree height,DBH and density,which play import roles in forest structural complexity(Vazquez and Givnish 1998;Koch et al.2004;Asner et al.2014).Therefore,altitude might have similar in fl uence on the degree of forest structural complexity.To identify the presence of such relationships,we fi rst used scatter plots and then simple linear regressions a basis to understand and explore this multivariate relationships.

Results

Correlation between fi eld metrics and metrics derived from remote sensing data

Most fi eld and LiDAR metrics showed signi fi cant correlations with each other(Table 2).The LiDAR metric 95th percentile height of canopy height was strongly and positively correlated with the fi eld metrics mean DBH,dominant height and BA.A strong correlation was found between 95th percentile height and dominant height,both of which are canopy height measurements.Except broadleaf tree density,all the other fi eld metrics were positively correlated with canopy density,which explains the continuity of the canopy(strongest correlation was found between canopy density and BA).Surface area ratio,which is the LiDAR metric sensitive to both vertical and horizontal variation of the canopy,showed signi fi cant positive correlations with dominant height,mean DBH and BA.Canopy tree density showed signi fi cant correlations with only three LiDAR metrics;SD of canopy height,CV of canopy height and canopy density.Broadleaf tree density showed signi fi cant correlations with all the LiDAR metrics,except for CV of canopy height and canopy density.In contrast,all the other fi eld metrics showed signi fi cant correlations with all the LiDAR metrics.Proportion of broadleaf cover was found to have a strong positive correlation with broadleaf tree density.

Ordination of all the fi eld,LiDAR and image metrics is shown Fig.2.Only the fi rst two axes are presented and interpreted in the ordination because a substantial portion(76.7%)of the variance was explained in those two axes.The fi rst principal component(PC1)was primarily correlated with measures of size(DBH or height)and size variation,whereas the second principal component(PC2)was primarily correlated with measures of tree density.Ordination explained similar relationships as observed in the results of the Pearson correlation test.BA,mean DBH,surface area ratio and all height measurements,e.g.,maximum canopy height,95th percentile height,mean canopy height and dominant height,showed higher positive correlations with each other,whereas the CV of canopy height was negatively correlated with all those metrics.A higher positive correlation was observed between broadleaf tree density and the proportion of broadleaf cover.Also,a higher positive correlation was observed between SD of canopy height and surface area ratio.

Table 2 Pearson correlations of fi eld metrics with LiDAR and image metrics

Fig.2 Ordination resulting from PCA of all fi eld,LiDAR and image metrics.The bottom and left axes show actual observations multiplied by loadings;the top and right axes show weights assigned to each variable after centering and scaling(BA basal area)

Classi fi cation of forest structure

Based on the results of the cluster analysis,six forest structure classes with different levels of complexity were identi fi ed.The number of plots per class ranged from 8 to 17 plots(mean=12.7).Complex canopies have taller canopy height,greater surface area ratio and higher canopy density(Kane et al.2010a).Higher forest structure class numbers were assigned to the classes with characteristics of comparatively more complex canopies.Table 3 summarizes the means and SD values of the LiDAR,image and fi eld based metrics for the identi fi ed six canopy structure classes.

Class 1 exhibited the least complex forest structure found in this study site because of the shortest canopy height(mean of 95th percentile canopy height=9.9 m),lower canopy density(mean=50.6),and the lowest surface area ratio(mean=2.73).Further,class 1 showed the highest proportion of broadleaf cover indicating that it includes broadleaf dominant forest structures.When the fi eld metrics of class 1 were considered,they also showed similar variations in mean values for dominant height,BA and mean DBH.

Class 2 had canopies that were neither very short nor very tall canopy.Also,class 2 showed lower canopy density(mean=48.6)and comparatively lower surface area ratio(mean=3.23).Proportion of broadleaf cover in class 2 was 0.43,indicating almost an even composition of broadleaf and conifer species.

Classes 3–6 had 95th percentile heights that overlapped with each other.However,class 6 had the highest mean canopy.Moreover,classes 3–6 had very dense,heterogeneous canopies.Class 6 had the greatest surface area ratio(mean=4.98).Classes 3 and 5 are broadleaf-dominant classes,whereasclasses4and6areconifer-dominantclasses.Similarly,dominant height,BA,mean DBH,and tree density had comparatively greater values for classes 3–6.Results of the relationships between LiDAR and image metrics are summarized in Fig.3.It is clear that our classi fi cation grouped similar forest structures because the identi fi ed structure classes formed distinct clusters in the scatter plots(Fig.3)and the PCA ordination(Fig.4).In the ordination ofselected LiDAR and image metrics(Fig.4),the fi rst two principal components explained 88.2% of the variation.Although,the ordination did not show a distinct pattern for the distribution of complexity classes,more complex classes,e.g.,class 3–6 are distributed close to each other.

Table 3 LiDAR,image and fi eld-derived metrics for the six canopy classes de fi ned by classi fi cation

Relationship between forest structural complexity and elevation

The LiDAR-derived 95th percentile of canopy height and canopy density was negatively correlated with plot elevation at the 1%signi fi cance level(Table 4).These relationships were also evident in the regression analysis(Fig.5).In contrast,the surface area ratio was not related to plot elevation.No clear relationship was identi fi ed in the scatter plot of proportion of broadleaf cover versus plot elevation(Fig.5).A weak correlation was observed between plot elevation and proportion of broadleaf cover(correlation coef fi cient=0.20 at 5%signi fi cance level).Few clear relationships between plot elevation and class were apparent(Fig.6).Class 1,which had the lowest forest structural complexity,was found at elevations above 1000 m.Class 2,which showed intermediate complexity,was distributed above 800 m.More complex classes(Class 3–6)were found below 1200 m altitudes.

Discussion

Comparison of fi eld,image and LiDAR-based measures of stand structural complexity and selection of metrics for classi fi cation

In this study,we observed strong correlations between BA and canopy density,between mean DBH and surface area ratio and between dominant height and 95th percentile of canopy height.In previous studies also,authors found similar signi fi cant and strong correlations between the fi eld and LiDAR metrics(Lefsky et al.1999;Kodani and Awaya 2005,2006;Kane et al.2010a).Therefore,strong correlations found between fi eld metrics and metrics derived using remote sensing data and ordination indicate that these two sets of measurements captured similar patterns of forest structure.

Fig.3 Relationships between LiDAR and image metrics to each other.Each dot represents one fi eld plot and plots are coded by canopy structure class

However,it is important to note that the strength of the correlations between fi eld and LiDAR-derived metrics differ among forest types.For example,Kodani and Awaya(2006)found signi fi cant correlations between stem numbers and LiDAR-derived mean tree height(in a broadleaf forest in Shikoku Island,Japan),but the correlation we found between canopy tree density and mean canopy height was not signi fi cant at the 5%signi fi cance level.Consistent with our fi ndings,Lefsky et al.(1999)found strong correlations between BA and LiDAR derived canopy heights(in deciduous forests in the United States).We found signi fi cant negative correlations between canopy tree density and two LiDAR metrics(surface area ratio and SD of height)and a weak positive correlation between canopy tree density and canopy density.In contrast,Kane et al.(2010a)did not fi nd signi fi cant correlations between LiDAR-derived height and canopy tree density(in conifer forests in the United States).These differences in correlations between LiDAR and fi eld metrics in different forest types may be due to the differences in the canopy structure,composition,and shapes of canopy trees,i.e.,broadleaf treesare sphericaland coniferoustreesconical,in coniferous,broadleaf,and mixed conifer–broadleaf forests.Therefore,the LiDAR metrics that best explain the canopy structure may be different among these forest types.

Fig.4 Ordination resulting from PCA of selected LiDAR and image metrics.Each number indicates the forest structure class

Table 4 Pearson correlations between elevation and fi eld metrics

Multifaceted forest structure is commonly characterized using three categories of LiDAR metrics:LiDAR metrics to represent biomass,LiDAR metrics to measure vertical structure and LiDAR metrics to measure canopy gaps(Lefsky et al.2005).Canopy height,a dominant characteristic of forest structure(Spies 1998),has been used commonly due to its direct and increasingly well-understood relationship to various forest structural attributes including aboveground biomass.Although,many height metrics,e.g.,maximum height,mean height and percentiles of canopy height(e.g.,95th,90th,75th),can be derived using LiDAR data,tree crown development pattern in the study site should be considered when selecting the most suitable height measurement(Kane et al.2010a).In this study,95th percentile height was preferred because foliage distribution in trees is generally concentrated in the top canopy layers and foliage is more evenly distributed throughout the canopy space due to greater light penetration.However,studied LiDAR canopy height metrics(95th percentile of canopy height,maximum canopy height and mean canopy height)were highly correlated suggesting that the selection of a particular height measurement might not adversely affect the result.In this study,surface area ratio was preferred as the metric to represent vertical structure and height variation over previously used metrics of height variations,e.g.,SD of canopy height(Lefsky et al.2005)and CV of canopy height(Zimble et al.2003)because it provided a three-dimensional measure of canopy structural heterogeneity and was correlated with several measures of forest structural complexity.However,the high positive correlation found between surface area ratio and SD of canopy height suggests that SD of canopy height could be used as an alternative metric of canopy height variation when calculating surface area ratio with remote sensing data is problematic.Although not able with our fi eld data,determining whether the surface area ratio can be used to detect structural complexity that is not detectable using the SD of canopy height would be interesting.LiDAR canopy density,which explains canopy gaps(Vepakomma et al.2010)and hence canopy continuity,played an important role in forest structural complexity classi fi cation because canopy gaps are formed periodically in the mixed conifer–broadleaf forests in central Hokkaido due to frequent snow and wind damage.

Fig.5 Relationships between LiDAR and image metrics and plot elevation

Fig.6 Relationship between forest structure class and plot elevation

Moreover,vegetation composition plays an important role in litter decomposition rate(Prescott et al.2000),availability and effect of ectomycorrhizal fungal communities(Ishida et al.2007)and understory light environment(Messier et al.1998).However,none of the LiDAR metrics used in this study provided information on the composition of broadleaf and coniferous species in mixed conifer–broadleaf forests.Image-derived metric included in our classi fi cation explained the vegetation composition in mixed conifer–broadleaf forests;speci fi cally whether a particular forest structure is conifer dominant or broadleaf.Therefore,combination of the three categories of LiDAR metrics with a metric derived from high resolution aerial photography provided us a good understanding of the structural complexity in mixed conifer–broadleaf forests in northern Japan.

Classi fi cation of forest structural complexity using airborne LiDAR and aerial photography

Six forest structure classes identi fi ed to be present in this study area indicate that our study site has different forest structure types which show different characteristics to each other.Most of the plots were classi fi ed into comparatively more complex classes(classes 3–6).Therefore,it is evident that this study site accommodates complex forest structure types having dense,heterogeneous canopy.Dickinson et al.(2014)found that mixed conifer–broadleaf forests with no active management activities tend to have different forest structural types,which have comparatively higher complexity.Therefore,our results can be considered as a good interpretation of the complexity of unmanaged mixed conifer–broadleaf forests.

However,the disadvantage of classifying the forest structure into distinct classes using classi fi cation techniques is that such techniques attempt to sort forest types into discrete classes or clusters.Due to the multifaceted nature of the forest structure some of the facets form gradients without clear breaks,making it dif fi cult to de fi ne distinct forest structure types.For example,when discrete classes are de fi ned;there may be some forest structure types that have characteristics that do not belong to any de fi ned class but fall in between two de fi ned classes.For that reason,many authors suggest using continuous fi eld classi fi cation methods for land cover analysis and mapping by producing continuous values for ecosystem characteristics rather than using discrete classes(Zimble et al.2003;Koy et al.2005).However,continuous fi eld classi fi cations are not very effective in communicating information to forest managers.Therefore,it can be argued that the classi fi cation of forest structure types into distinct classes provides better interpretation than continuous classi fi cations in terms of their applicability in forest management.However,the classi fi cation method and metrics used in classi fi cation and the distinction of classes should be based on the requirements of the user and the end product.

Explanations for variation observed

Trees that grow in homogeneous site conditions such as light,water and nutrients and experience few partial disturbances may develop a very even canopy(Van Pelt and Franklin 2000)whereas young forests exposed to many partial disturbances can have more complex structures than older forests that had not had such partial disturbances(Zenner 2004).Therefore,a portion of variation can be explained based on the impact of physical and biological disturbances because these disturbances might also have triggered developmentofdifferentforeststructures.Mayeama is an unmanaged natural forest reserve;hence,regeneration in the forest structural development process is natural,which is often affected by the distribution and growth of dwarf bamboo,which in turn is related to the reduction in overstory trees over time(Noguchi and Yoshida 2005;Owari 2013).In addition,pathogenic fungi(Racodium therryanumorHerpotrichia juniper)can cause a serious snow blight disease(Igarashi and Cheng 1988).Moreover,wind throw damage and crown damage in young forests caused by periodic wind and snow might have resulted in greater surface area ratio and lower canopy density.All these disturbances might have given rise to the emergence of many different forest structure types and also greater complexity in the canopy structure.

In this study,the plots found to have low-complex canopies may be sites that simply have failed to experience partial disturbances frequently or experienced disturbances in small scale.The plots that had greater surface area ratio and greater complexity may have undergone disturbances more frequently,and thus developed more heterogeneous canopy structures.Therefore,the complex forest structures observed in this study may be maintained not only by the differences in habitat but also by a balance between less-frequent large disturbances and more-frequent smaller ones.

Relationships between forest complexity classes and elevation

The negative correlation we found between tree height and elevation is consistent with the fi ndings of Koch et al.(2004)and Asner et al.(2014)that tree height decreases along an altitudinal gradient.In this study,we found a negative correlation between plot elevations and canopy density,indicating that the number of gaps increases with elevation.This fi nding is consistent with Asner et al.(2014)who also found that elevation is positively correlated with LiDAR-derived canopy gap density,understory vegetation cover and the vertical partitioning of vegetation in canopies.We found few clear relationships between forest structural complexity and plot elevation in agreement with Kane et al.(2010a)who found that the canopy structural complexity does not develop in a fashion that can be linearly related to elevation.Moreover,Vazquez and Givnish(1998)found that species composition and structure change signi fi cantly along altitudinal gradients in tropical forests,but they did not fi nd any clear linear relationship either.However,it was evident that the forest structure types found at higher elevations(above 1000 m)are less complex compared with the forest structure types found in lower elevations.Therefore,a more detailed study would provide a better understanding of the relationship between forest structural complexity developments along an elevation gradient.

Results of this study suggest new directions for future research such as analyzing the spatial dynamics of forest structural complexity in mixed conifer–broadleaf forests because such spatial dynamics can provide the fundamental information required for successful management planning and decision making.Also,studies that focus on understanding the impact of partial disturbances in forest structural complexity development will inform the development of a new forest management paradigm that is based on heterogeneity,unpredictability and adaptability for dealing with future challenges.Most of the applications of airborne LiDAR data in mixed conifer–broadleaf forests had been limited to scienti fi c research purposes due to its high cost and lack of expertise.Therefore,the potential of new lowcost methods such as the use of unmanned aerial vehicles(UAVs)in analyzing structural complexity of mixed conifer–broadleaf forests should be researched to enable frequent studies needed for understanding how complex mixed conifer–broadleaf forests respond to new or variable conditions and how well they can sustain ecosystem services in the context of changing conditions.

AcknowledgementsThis study used Juro Kawachi Memorial data set of forest spatial information.We thank Mr.Yuji Nakagawa and other technical staff members of the UTokyo Hokkaido Forest for their kind support and providing existing fi eld inventory and remote sensing data for this research.We also thank the reviewers for their valuable comments.

Conclusions

The strong correlations found between fi eld and remote sensing metrics indicate that these two sets of measurements captured similar patterns in forest structure in mixed conifer–broadleaf forests.Therefore,it is simple and easy to calculate metrics derived from airborne LiDAR measurements and high resolution aerial photography to effectively determine the structural complexity in mixed conifer–broadleaf forests in northern Japan.Also,combined application of airborne LiDAR and orthographic aerial photography provided a good interpretation of the vegetation cover.The degree of forest structural complexity showed few clear relationships with plot elevations,suggesting the need for a detailed analysis to determine the relevance of ground elevation to the development of forest structural complexity in mixed conifer–broadleaf forests.On the basis of the results of this study,we can conclude that the structural complexity of mixed conifer–broadleaf forests can be characterized effectively by integrating multiple forest structural attributes derived using airborne LiDAR data and high-resolution aerial photography.

Asner GP,Anderson CB,Martin RE,Knapp DE,Tupayachi R,Sinca F,Malhi Y(2014)Landscape-scale changes in forest structure and functional traits along an Andes-to-Amazon elevation gradient.Biogeosciences 11:843–856

工作人员倾听、记录群众意见后，先在锻炼对象中开展认领，超出能力权限的通过临时党支部交办到所驻社区城市基层党建联建委员会，在驻区单位党员、住区党员中开展认领；对需要市委、市政府层面解决的，通过城市基层党建街道共建委员会上交到城市基层党建工作领导小组，上报到市委市政府，变成全市共同行动，让党及时听到了群众的“声音”，为市委市政府决策提供了参考。

Avery TE,Burkhart HE(2002)Forest measurements.McGraw Hill,Boston

Bradford JB,Kastendick DN(2010)Age-related patterns of forest complexity and carbon storage in pine and aspen-birch ecosystems of northern Minnesota,USA.Can J For Res 40:401–409

Chasmer L,Hopkinson C,Treitza P(2004)Assessing the threedimensional frequency distribution of airborne and ground-based LiDAR data for red pine and mixed deciduous forest plots.Int Arch Photogramm Remote Sens Spat Inf Sci XXXVI-8/W2:66–70

Dale MRT,Fortin MJ(2002)Spatial autocorrelation and statistical tests in ecology.EcoSc 9(2):167

Dalponte M,Bruzzone L,Gianelle D(2008)Fusion of hyperspectral and LIDAR remote sensing data for classi fi cation of complex forest areas.IEEE Trans Geosci Remote Sens 46(5):1416–1427

De la Cueva AV(2008)Structural attributes of three forest types in central Spain and Landsat ETM plus information evaluated with redundancy analysis.Int J Remote Sens 29(19):5657–5676

Dewalt SJ,Maliakal SK,Denslow JS(2003)Changes in vegetation structure and composition along a tropical forest chronosequence:implications for wildlife.For Ecol Manag 182:139–151

Dickinson Y,Zenner EK,Miller D(2014)Examining the effect of diverse management strategies on landscape scale patterns of forest structure in Pennsylvania using novel remote sensing techniques.Can J For Res 44(4):301–312

Egoh BN,Reyers B,Rouget M,Richardson DM,Le Maitre DC,Van Jaarsveld A(2008)Mapping ecosystem services for planning and management.Ecosyst Environ 127:135–140

Everitt B,Hothorn T(2011)Cluster analysis.In:Everitt B,Hotohorn T(eds)An introduction to applied multivariate analysis with R.Springer,New York,pp 163–200

Franklin JF,Van Pelt R (2004)Spatial aspects of structural complexity in old-growth forests.J For 102(3):22–28

Hame T,Salli A,Andersson K,Lohi A(1997)A new methodology for the estimation of biomass of conifer dominated boreal forest using NOAA AVHRR data.Int J Remote Sens 18:3211–3243

Harper KA,Bergeron Y,Gauthier S,Drapeau P(2002)Post- fi re development of canopy structure and composition in black spruce forests of Abitibi,Quebec:a landscape scale study.Silva Fenn 36(1):249–263

Horie K,Miyamoto Y,Kimura N,Oikawa N(2013)List of vascular plant of the University of Tokyo Hokkaido Forest.Univ For(Tokyo Univ)54:59–106

Hyyppa J,Yu XW,Hyyppa H,Vastaranta M,Holopainen M,Kukko A,Kaartinen H,Jaakkola A,Vaaja M,Koskinen J,Alho P(2012)Advances in forest inventory using airborne laser scanning.Remote Sens 4:1190–1207

Igarashi T,Cheng D(1988)Fungal damage caused byRacodium therryanumto regeneration of Japanese larch by natural seeding.Bull Coll Exp Hokkaido Univ 45(1):213–219

Ishida TA,Nara K,Hogetsu T(2007)Host effects on ectomycorrhizal fungal communities:insight from eight host species in mixed conifer–broadleaf forests.New Phytol 174(2):430–440

Kane VR,Bakker JD,McGaughey RJ,Lutz JA,Gersonde RF,Franklin JF (2010a)Examining conifer canopy structural complexity across forest ages and elevations with LiDAR data.Can J For Res 40(4):774–787

Kane VR,McGaughey RJ,Bakker JD,Gersonde RF,Lutz JA,Franklin JF(2010b)Comparisons between fi eld-and LiDAR-based measures of stand structural complexity.Can J For Res 40:761–773

Koch GW,Sillett SC,Jennings GM,Davis SD(2004)The limits to tree height.Nature 428:851–854

Kodani E,Awaya Y(2005)Estimating forest stand parameters ofCryptomeria japonicaandChamaecyparis obtusausing LiDAR.In:116th conference proceedings of Japan Forestry Society,pp 62–63

Kodani E,Awaya Y(2006)Estimating height and volume in broadleaved forest stand using LiDAR.In:Proceedings of Silvilaser 2006,pp 173–176

Koy K,McShea WJ,Leimgruber P,Haack BN,Aung A(2005)Percentage canopy cover—using Landsat imagery to delineate habitat for Myanmar’s endangered Eld’s deer(Cervus eldi).Anim Conserv 8:289–296

Langely SK,Cheshire HM,Humes KS(2001)A comparison of single date and multitemporal satellite image classi fi cations in a semiarid grassland.J Arid Environ 49:401–411

Larson AJ,Lutz JA,Gersonde RF,Franklin JF,Hietpas FF(2008)Potential site productivity in fl uences the rate of forest structural development.Ecol Appl 18(4):899–910

Lefsky MA,Harding D,Cohen WB,Parker G,Shugart HH(1999)Surface LiDAR remote sensing of basal area and biomass in deciduous forests of eastern Maryland,USA.Remote Sens Environ 67:83–98

Lefsky MA,Cohen WB,Spies TA(2001)An evaluation of alternate remote sensing products for forest inventory,monitoring,and mapping of Douglas- fi r forests in western Oregon.Can J For Res 31:78–87

Lefsky MA,Cohen WB,Parker GG,Harding DJ(2002)LiDAR remote sensing for ecosystem studies.Bioscience 52(1):19–30

Lefsky MA,Hudak AT,Cohen WB,Acker SA(2005)Patterns of covariance between forest stand and canopy structure.Remote Sens Environ 95(4):517–531

Lovell JL,Jupp DL,Culvenor DS,Coops NC(2003)Using airborne and ground-based ranging lidar to measure canopy structure in Australian forests.Can J Remote Sens 29(5):607–622

Ma ZK,Hart MM,Redmond RL(2001)Mapping vegetation across large geographic areas:integration of remote sensing and GIS to classify multisource data.Photogramm Eng Remote Sens 67(3):295–307

Maltamo M,Mustonen K,Hyyppa J,Pitka nen J,Yu X(2005)Identifying and quantifying structural characteristics of heterogeneous boreal forests using laser scanner data.For Ecol Manag 216(1–3):41–50

Matsuda K,Shibuya M,Koike T(2002)Maintenance and rehabilitation of the mixed conifer–broadleaf forests in Hokkaido,Northern Japan.Eur J For Res 5(2):119–130

McCune B,Grace JB(2002)Analysis of ecological communities.MjM Software Design,Gleneden Beach

McElhinny C,Gibbons P,Brack C,Bauhus J(2005)Forest and woodland stand structural complexity:its de fi nition and measurement.For Ecol Manag 218(1–3):1–24

Messier C,Puettmann KJ(2011)Forests as complex adaptive systems:implications for forest management and modelling.Ital J For Mt Environ 66(3):249–258

Messier C,Parent S,Bergeron Y(1998)Effects of overstory and understory vegetation on the understory light environment in mixed boreal forests.J Veg Sci 9(4):511–520

Næsset E(2002)Predicting forest stand characteristics with airborne scanning laser using a practical two-stage procedure and fi eld data.Remote Sens Environ 80:88–99

Næsset E,Gobakken T,Holmgren J,Hyyppa H,Hyyppa J,Maltamo M,Nilssonb M,Olssonb H,Perssonf A˚,So derman U(2004)Laser scanning of forest resources:the Nordic experience.Scand J For Res 19(6):482–499

Nakata M,Tanaka H,Yagi H (1994)Altitudinal changes in vegetation and soils on Mt.Dairoku,central Hokkaido,Japan.Jpn J Ecol 44:33–47

Noguchi M,Yoshida T(2005)Factors in fl uencing the distribution of two co-occurring dwarf bamboo species(Sasa kurilensisandS.senanensis)in a conifer-broadleaved mixed stand in northern Hokkaido.Ecol Res 20:25–30

Olthof I,King DJ(2000)Development of a forest health index using multispectral airborne digital camera imager.Can J Remote Sens 26:166–176

Owari T(2013)Relationship between the abundance of Abies sachalinensis juveniles and site condition in selection forests of central Hokkaido,Japan.FORMATH 12:1–20

Owari T,Matsui M,Inukai H,Kaji M(2011)Stand structure and Geographic conditions of natural selection forests in central Hokkaido,Nothern Japan.J For Plan 16:207–214

Parker GG,Harmon ME,Lefsky MA,Chen J,Van Pelt R,Weis SB,Thomas SC,Winner WE,Shaw DC,Franklin JF(2004)Threedimensional structure of an oldgrowthPseudotsugatsuga canopy and its implications for radiation balance,microclimate,and gas exchange.Ecosystems 7:440–453

Pasher J,King DJ(2010)Multivariate forest structure modelling and mapping using high resolution airborne imagery and topographic information.Remote Sens Environ 114:1718–1732

Pastur GJ,Cellini JM,Lencinas MV,Peri PL(2008)Stand growth model using volume increment/basal area ratios.J For Sci 54(3):102–108

Prescott CE,Zabek LM,Staley LM,Kabzems R(2000)Decomposition of broadleaf and needle litter in forests of British Columbia:in fl uences of litter type,forest type,and litter mixtures.Can J For Res 30(11):1742–1750

R Development Core Team(2014)R:a language and environment for statistical computing.www.R-project.org

Reitberger J,Krzystek P,Stilla U(2008)Analysis of full waveform LiDAR data for the classi fi cation of deciduous and coniferous trees.Int J Remote Sens 29(5):1407–1431

Spies TA(1998)Forest structure:a key to the ecosystem.Northwest Sci 72:34–39

Staudhammer CL,LeMay VM(2001)Introduction and evaluation of possible indices of stand structural diversity.Can J For Res 31(7):1105–1115

Tatewaki M(1958)Forest ecology of the islands of the North Paci fi c Ocean.J Fac Agric Hokkaido Univ 50(4):371–486

The University of Tokyo Hokkaido Forest(2015)http://www.uf.a.utokyo.ac.jp/ fi les/gaiyo_hokkaido.pdf.Accessed 03 June 2016

Van Pelt R,Franklin JF(2000)In fl uence of canopy canopy structure on the understorey environment in tall,old-growth,conifer forests.Can J Remote Sens 30(8):1231–1245

Van Pelt R,Nadkarni NM(2003)Development of canopy structure inPseudotsuga menziesiiforests in Southern Washington cascades.For Sci 50(3):326–341

Vazquez JA,Givnish TJ(1998)Altitudinal gradients in tropical forest composition,structure and diversity in the Sierra de Manan.J Ecol 86:999–1020

Vepakomma U,Kneeshaw D,St-Onge B(2010)Interactions of multiple disturbances in shaping boreal forest dynamics:a spatially explicit analysis using multi-temporal LiDAR data and high resolution imagery.J Ecol 98(3):526–539

Whittaker RJ,Araújo MB,Jepson P,Ladle RJ,Watson JE,Willis KJ(2005)Conservation biogeography:assessment and prospect.Divers Distrib 11:3–23

Whittney GG,Foster DR(1988)Overstorey composition and age as determinants of the understorey fl ora of woods of central New England.J Ecol 76:867–876

Wulder M (1998)Optical remote sensing techniques for the assessment of forest inventory and biophysical parameters.Prog Phys Geogr 22(4):449–476

Wulder WA,Coops NC,Hudak AT,Morsdorf F,Nelson R,Newnham G,Vastaranta M(2013)Status and prospects for LiDAR remote sensing of forested ecosystems.Can J Remote Sens 39(Supplement 1):S1–S5

Wunderle AL,Franklin SE,Guo XG(2007)Regenerating boreal forest structure estimation using SPOT-5 pan-sharpened imagery.Int J Remote Sens 28(19):4351–4364

Xie YC,Sha ZY,Yu M(2008)Remote sensing imagery in vegetation mapping:a review.J.Plant Ecol 1(1):9–23

Yan MJ(2005)Methods of determining the number of clusters in a data set and a new clustering criterion.Doctoral Thesis,Blacksburg:Virginia.https://theses.lib.vt.edu/theses/available/etd-12062005-153906/unrestricted/Proposal-Face.pdf.Accessed 03 June 2016

Zenner EK(2000)Do residual trees increase structural complexity in Paci fi c Northwest coniferous forests?Ecol Appl 10(3):800–810

Zenner EK(2004)Does old-growth condition imply high live-tree structural complexity?For Ecol Manag 195(1–2):243–258

Zimble DA,Evans DL,Carlson GC,Parker RC,Grado SC,Gerard PD(2003)Characterizing vertical forest structure using smallfootprint airborne LiDAR.Remote Sens Environ 87:171–182