Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Modelling the vertical distribution of canopy fuel load using national forest inventory and low-density airbone laser scanning data

  • Eduardo González-Ferreiro ,

    Contributed equally to this work with: Eduardo González-Ferreiro, Stéfano Arellano-Pérez

    Current address: Department of Forest Ecosystems and Society, Oregon State University, Corvallis, Oregon, United States of America; Laboratory of Applications of Remote Sensing in Ecology (LARSE), USDA Forest Service—Pacific Northwest Research Station, Corvallis, Oregon, United States of America

    Affiliation Unidad de Gestión Forestal Sostenible, Departamento de Ingeniería Agroforestal, Universidad de Santiago de Compostela, Lugo, Spain

  • Stéfano Arellano-Pérez ,

    Contributed equally to this work with: Eduardo González-Ferreiro, Stéfano Arellano-Pérez

    Affiliation Unidad de Gestión Forestal Sostenible, Departamento de Ingeniería Agroforestal, Universidad de Santiago de Compostela, Lugo, Spain

  • Fernando Castedo-Dorado,

    Affiliation Departamento de Ingeniería y Ciencias Agrarias, Universidad de León, Campus de Ponferrada, Ponferrada, Spain

  • Andrea Hevia,

    Affiliation Sustainable Forest Management Area, Forest and Wood Technology Research Centre (CETEMAS), Pumarabule, Carbayín, Siero—Asturias, Spain

  • José Antonio Vega,

    Affiliation Centro de Investigación Forestal de Lourizán, Pontevedra, Spain

  • Daniel Vega-Nieva,

    Affiliation Instituto de Silvicultura e Industria de la madera, Universidad Juárez del Estado de Durango, Ciudad Universitaria, Durango, Durango, México

  • Juan Gabriel Álvarez-González ,

    juangabriel.alvarez@usc.es

    ‡ These authors also contributed equally to this work.

    Affiliation Unidad de Gestión Forestal Sostenible, Departamento de Ingeniería Agroforestal, Universidad de Santiago de Compostela, Lugo, Spain

  • Ana Daría Ruiz-González

    ‡ These authors also contributed equally to this work.

    Affiliation Unidad de Gestión Forestal Sostenible, Departamento de Ingeniería Agroforestal, Universidad de Santiago de Compostela, Lugo, Spain

Modelling the vertical distribution of canopy fuel load using national forest inventory and low-density airbone laser scanning data

  • Eduardo González-Ferreiro, 
  • Stéfano Arellano-Pérez, 
  • Fernando Castedo-Dorado, 
  • Andrea Hevia, 
  • José Antonio Vega, 
  • Daniel Vega-Nieva, 
  • Juan Gabriel Álvarez-González, 
  • Ana Daría Ruiz-González
PLOS
x

Abstract

The fuel complex variables canopy bulk density and canopy base height are often used to predict crown fire initiation and spread. Direct measurement of these variables is impractical, and they are usually estimated indirectly by modelling. Recent advances in predicting crown fire behaviour require accurate estimates of the complete vertical distribution of canopy fuels. The objectives of the present study were to model the vertical profile of available canopy fuel in pine stands by using data from the Spanish national forest inventory plus low-density airborne laser scanning (ALS) metrics. In a first step, the vertical distribution of the canopy fuel load was modelled using the Weibull probability density function. In a second step, two different systems of models were fitted to estimate the canopy variables defining the vertical distributions; the first system related these variables to stand variables obtained in a field inventory, and the second system related the canopy variables to airborne laser scanning metrics. The models of each system were fitted simultaneously to compensate the effects of the inherent cross-model correlation between the canopy variables. Heteroscedasticity was also analyzed, but no correction in the fitting process was necessary. The estimated canopy fuel load profiles from field variables explained 84% and 86% of the variation in canopy fuel load for maritime pine and radiata pine respectively; whereas the estimated canopy fuel load profiles from ALS metrics explained 52% and 49% of the variation for the same species. The proposed models can be used to assess the effectiveness of different forest management alternatives for reducing crown fire hazard.

Introduction

Accurate knowledge of fuel characteristics has been shown to be critical in forest fire management, as fuel constitutes a primary component of fire risk [1,2]. Moreover, accurate fuel mapping is required for using fire behaviour models and simulation systems, which are essential for establishing fuel treatment priorities and evaluating the effectiveness of fuel management actions [3].

One of the current main objectives of fuel management in conifer forests is the mitigation of crown fire hazard. Crown fires are usually intense and spread rapidly, which makes them difficult and dangerous to suppress. They can cause severe economic damage and ecological disruption [4].

Assessing the risk of crown fire initiation and spread is a key element in gauging fire behaviour. Both crown fire initiation and spread are widely recognised to be determined by structural characteristics of the canopy fuel complex such as available canopy fuel load (CFL), canopy bulk density (CBD) and canopy base height (CBH) [4]. CFL is defined as the mass of available canopy fuel per unit ground area; CBD indicates the amount of available fuel per volume unit in the aerial layer, and CBH is the lowest height above ground level at which there is sufficient available canopy fuel to propagate fire vertically through the canopy [2,5]. Fire behavior simulation systems that assess crown fire potential thus require accurate estimates of these variables. For modelling purposes, needles and fine twigs are commonly considered available fuel because these tree biomass components are usually consumed within the flaming front of a crown fire [6].

Two main approaches can be used to estimate CBD and CBH [2,7,8]. The simplest approach is based on the assumption that available canopy fuel is homogeneously distributed throughout the aerial layer of the stand. CFL is thus estimated by dividing the total available biomass in the canopy by the stand surface area; CBD is estimated by dividing CFL by the canopy depth, and CBH is estimated as the average crown base height of all trees in the stand. This approach is consistent with the criteria for crown fire initiation and spread proposed by Van Wagner [9] and is integrated in most empirical crown fire models. However, important advances are expected to be made in modelling crown fire behaviour in the future. Therefore, estimation of the vertical distribution of canopy fuels should focus on a more realistic and complex approach based on evidence that the fuel distribution in each tree crown affects the general fuel distribution in the stand [7]. Such a realistic approach assumes that the crown of each tree in the stand has a particular shape that defines the available fuel distribution along the crown. The distribution of the available fuel in the stand is predicted by taking into account the shape of each tree and the height at which the crown starts and ends. CBD and CBH are subsequently estimated from the predicted vertical distribution according to certain criteria. CBD is usually estimated as the maximum bulk density in a vertical profile that shows the distribution of this variable from the ground to the top of the tallest tree in the stand, and CBH as the height at which a certain bulk density value is reached in that profile [2]. Different bulk density values were arbitrarily proposed for estimating CBH: 0.037 kg m-3 [5], 0.012 kg m-3 [10] and 0.067 kg m-3 [11]. This option enables us to take into account information provided by Airborne Laser Scanning (ALS) data, which can be used to predict the forest structure in three dimensions.

In Spain, the only information currently available regarding canopy fuel complex characteristics at national and regional scales is that provided by the National Forest Inventory (NFI). The NFI is based on a systematic sampling design (1x1 km sample grid) with permanent plots and previous forest stratification by photo-interpretation. However, this type of inventory does not yield full spatial coverage and often takes several years to complete [12]. For example, in the region of Galicia (north-west Spain), the field work involved in the fourth NFI (NFI-4) lasted about 11 months, and data processing took approximately 3 years [13]. Moreover, the NFI is costly and the information provided could become quickly outdated, especially for fast growing species.

ALS has proven to be a useful source of auxiliary data for describing the canopy fuel stratum because it can provide information that can be used to predict the three-dimensional structure of vegetation and other forest features at different scales [1416]. Research aimed at characterizing forest resources by use of commercial ALS sensors operating at moderate to low spatially dense sampling (up to 0.5 pulses m-2) has focused on the stand-level approach, which often establishes empirical relationships between metrics estimated from ALS data and field-measured stand variables. Specific stand-level studies that describe the canopy fuel stratum are scarce and often restricted to a few tree species and small sites [1723]. Some ALS studies have also attempted to predict diameter probability density function parameters; for example, Gobakken and Næsset [24,25] used the Weibull distribution to predict diameter distributions from laser scanner data; Breidenbach et al. [26] applied a generalized linear model (GLM) to estimate the shape and scale parameters of the Weibull distribution by using ALS metrics as predictors; and Thomas et al. [27] applied ALS-based Weibull parameter prediction in different types of forests including coniferous, hardwoods and mixed woods.

Spain now has a wide coverage of coarse resolution (0.5 first returns m-2), small-footprint ALS data, thanks to the Plan Nacional de Ortofotografía Aérea 2008–2015 (PNOA 2008–2015), which has carried out 5 annual flight surveys covering an area of 406,550 km2 (slightly more than 80% of the surface of Spain). The main aim is to produce fine resolution Digital Elevation Models (DEM) and to control the quality of PNOA conventional photogrammetric products. Although PNOA ALS data were not originally conceived for the purpose of characterizing forest vegetation, it is expected that wall-to-wall, georeferenced point clouds could provide accurate estimates of the available canopy fuel profile.

In this study, we aimed to model the vertical distribution of the available canopy fuel load in Pinus pinaster Ait. and Pinus radiata D. Don stands in Galicia (north-west Spain), by combining extensive (NFI) field and countrywide ALS data sets. Although the study focuses on two species, the method could be applied to other species when new data acquisitions become available. Two different set of models were used to predict the vertical distributions of available canopy fuel loads and were fitted using different sets of predictors: the first based on stand variables calculated from field measurements in the NFI plots and the second based on ALS metrics.

Material and methods

Study area

The region of Galicia (north-west Spain) is characterized by rugged orography and an oceanic climate with mild temperatures, low thermic oscillation between winter and summer and frequent rainfall. Galicia is one of the most important regions in Spain in terms of forestry production. Around 30% of the 1.4 million hectares of tree-covered land comprises pure and even-aged pine stands, mainly maritime pine (Pinus pinaster) (271,281 ha) and radiata pine (Pinus radiata) (96,177 ha), and more than 68 million cubic metres of standing timber [13] provide 51% and 27% of respectively the conifer and total annual harvest volume in Spain [28]. Fig 1 shows the location of Galicia in Spain and Europe, and the spatial distribution of pure P. pinaster and P. radiata NFI plots.

thumbnail
Fig 1. Location of the study area and spatial distribution of the NFI-4 plots where P. pinaster and P. radiata were dominant.

https://doi.org/10.1371/journal.pone.0176114.g001

Pine stands are among the most flammable types of vegetation, particularly when they are not thinned or pruned and carry large surface fuel loads [29,30]. This, together with the low moisture contents reached in the litter in dry summer periods [31], has led to wildfires (especially crown fires) becoming of ecological, economic and social concern in Galicia.

Field data

The field data used for this study were obtained from the NFI-4 carried out in Galicia [13] (data available at http://www.magrama.gob.es/es/biodiversidad/servicios/banco-datos-naturaleza/; last accessed on 7-8-2016). The NFI project maintains a network of sample plots throughout Spain, which are aimed at providing continuously updated information on the status of nationwide forest resources, including timber volumes and species composition [32]. Sample plots are established at the intersections of a 1-km x 1-km UTM grid and consist of four circular concentric subplots of radii 5, 10, 15 and 25 m (Fig 2). Diameter at breast height (d) and total tree height (h) are measured in trees selected on the basis of their diameter and distance to the plot centre (d ≥ 42.5 cm for the 25-m radius; d ≥ 22.5 cm for the 15-m radius; d ≥ 12.5 cm for the 10-m radius and d ≥ 7.5 cm for the 5-m radius). The number of trees of diameter 2.5–7.5 cm (saplings) is also recorded in the 5-m radius subplot. Diameter at breast height is measured to the nearest 0.1 cm with graduated trees calipers in two perpendicular directions. Total tree height is measured to the nearest 0.1 m with a hypsometer. The following stand variables were calculated from the tree variable measurements, by using tree expansion factors: number of stems per hectare (N), quadratic mean diameter (dg), stand basal area (G), mean stand height () and dominant height (H, defined as the mean height of the 100 thickest trees per hectare). These variables were used as predictors for statistical analyses. The tree expansion factor adjusts to a per hectare basis for the ith tree; it expresses the number of trees per hectare that each selected tree represents in the inventory, according to the subplot radius.

thumbnail
Fig 2. A typical sample plot used in the fourth National Forest Inventory (NFI-4).

Grey circles represent trees selected on the basis of tree diameter and the distance to the plot centre.

https://doi.org/10.1371/journal.pone.0176114.g002

A total of 8515 sample plots were measured in the NFI-4 in Galicia, which corresponds to approximately one sample location per 166 ha of afforested land.

All plots in which P. pinaster or P. radiata were dominant (more than 90% of trees and more than 90% of total stand basal area) were selected. In addition, plots in which dead trees accounted for more than 10% of basal area were rejected.

ALS data

ALS data were acquired under the direction of the Spanish Ministerio de Fomento (Dirección General del Instituto Geográfico Nacional and Centro Nacional de Información Geográfica). The data were obtained for the PNOA project, in autumn 2009 in eastern Galicia (provinces of Lugo and Ourense) and in autumn 2010 in western Galicia (provinces of A Coruña and Pontevedra). The data were acquired with a RIEGL LMS-Q680 sensor, installed in a fixed wing aerial platform, operated at 1064 nm, a pulse repetition rate of 70 kHz, a scan frequency of 46 Hz, a maximum scan angle of ±30°, a maximum beam divergence of 0.5 mrad, an average flying height of 1300 m above sea level, and a minimum swath overlap of 15%. Square ALS blocks of 2 km side and covering the whole region of Galicia were obtained from the CNIG computer server (data available at http://centrodedescargas.cnig.es/CentroDescargas/buscadorCatalogo.do?codFamilia=LIDAR; last accessed on 5-27-2016). A maximum of 4 returns per pulse were registered, with a theoretical laser pulse density required for the PNOA project of 0.5 first returns m-2.

Extraction of ALS metrics.

ALS metrics are descriptive structural statistics calculated from the normalized height of the ALS data cloud (NHD) and processing of raw ALS data is necessary to obtain these metrics. We clipped the ALS cloud within the limit of buffer areas of radius 30 m (25 m of NFI plot plus a buffer of 5 m) and then each cloud was saved as an independent ALS file. We buffered all the selected NFI-4 plots by 5 m to avoid errors (especially near of the plot boundaries) in the subsequent height normalization processes. After that, the outliers were removed considering a window size of 100 m and a maximum and minimum ellipsoidal height bound of ± 5.0*Std. dev. A filtering algorithm (adapted from [33]), based on linear prediction [34], was used to extract ground returns from the ALS point cloud. All returns were used in this process, the cell size was set to 2 m (based on available ALS data density) and the other parameters (a:1.0; b:4.0; g:-2.5; w:2.5; iterations:5) were obtained from Barreiro-Fernández et al. [35], as these performed well with a variety of land cover, forest types and slopes. Then, a 2 m cell size DEM grid was generated by estimating the elevation of each grid cell from the average elevation of all points within the cell; if the cell does not contain any points, it is filled by interpolation from the neighbouring cells. In the following step, the normalized ALS point cloud was obtained by subtraction of the ellipsoidal height of the DEM from the ellipsoidal height of each ALS return; returns below a normalized height of 0 m were excluded. The normalized ALS point cloud within the limits of each field plot (which were previously stored as polygons in shapefiles) were clipped, and an independent file for each plot of 25 m radius was created. Finally, height and canopy cover metrics of the clipped and normalized point clouds were estimated using all returns. The minimum height threshold (MHT), which is commonly specified as the lower boundary for calculating height metrics (central tendency, dispersion, shape and percentile statistics), was established at 1.5 m. The height break threshold (HBT), which is the limit for separating the point cloud data into two sets to separate canopy returns from the under canopy returns, in order to estimate canopy cover metrics, was established as 4 m (based on field observation). A new filter was applied to remove any remaining outliers; only returns with normalized heights between 0–60 m were included in the analysis, after considering the highest tree apexes observed in the field. In total, 38 metrics widely used to estimate variables related to canopy cover and height distribution [23,36] were extracted from ALS pulses and used as predictors in the statistical analysis. The ALS metrics and the corresponding descriptions for height distribution and canopy cover are summarised in Table 1. Note that all variables were computed from all ALS returns in the database, i.e. all returns per laser pulse.

thumbnail
Table 1. Potential explanatory variables related to height distribution and canopy cover.

https://doi.org/10.1371/journal.pone.0176114.t001

Quality control and plot selection

The plot positioning procedures used in the NFI-4 were applied by using standard GPS equipment; the plot coordinates therefore had an expected average accuracy of approximately 3–5 m. Gobakken and Næsset [37] observed that the use of larger plot sizes (300–400 m2) compensates for errors in positioning sample plots for estimating biophysical properties derived from ALS data. The size of the sample plot used in the NFI is around 1964 m2; for positioning errors of 5 and 10 m (much larger than the assumed error), the overlapping areas between a plot in true position and a plot located in an altered position were 84.3 and 74.7%, respectively. Moreover, the precision of the ALS estimates tends to be less sensitive to positioning errors in even-aged plots (as in the present study) and dense forests [37]; therefore, only plots with a crown projection area greater than 90% were selected. The final number of sample plots was 646 (512 P. pinaster and 134 P. radiata plots). In Galicia, the NFI-4 measurements were made between 4 and 11 months before ALS measurements. Some of the plots may have been thinned or cropped during this interval, and the ALS metrics would therefore not reflect the stand structure of the plots on the measurement date. In order to prevent errors in the subsequent modelling process, we tried to remove all plots that had undergone silvicultural treatments during the interval between measurements. Visiting all NFI inventory plots is very expensive and time-consuming and was deemed unfeasible. Visualization of ALS data was also ruled out as a means of plot selection because it does not guarantee detection of all disturbed plots during this interval. Finally, the following two (rejection) criteria were used in sample plots selection: 1) all sample plots with a percentage of first returns above 2 m less than 80% were rejected, and 2) plots for which the field-measured dominant height (H) and ALS-derived 99th height percentile (h99) differed by more than 3 times the root mean square error (RMSE) of a pre-existing model for the species and study area were rejected. This model was fitted from a network of 25 permanent plots installed in pine stands through Galicia and related the field-measured H to the ALS metric h99 (for more details, see [23]).

(1)

Taking the second criterion into account, plots with absolute differences between field-measured H and ALS-derived h99 larger than 5.48 m were disregarded for further analysis. Under these decision-making criteria, the number of plots selected was reduced from 646 to 554 (436 for P. pinaster and 118 for P. radiata). The mean, maximum, minimum values and standard deviation for the main tree and stand variables for both species are shown in Table 2.

thumbnail
Table 2. Descriptive statistics for the main tree and stand variables corresponding to the sample plots used in this study.

Std. Dev., standard deviation; d, tree diameter; h, total tree height; cl, crown length (defined as the distance from the crown base to the tree top); N, stand density; dg, quadratic mean diameter; G, stand basal area and H, dominant height (estimated as the mean height of the 100 thickest trees per ha).

https://doi.org/10.1371/journal.pone.0176114.t002

Vertical distribution of available canopy fuel estimated from field data