Neşe Gülci

Abstract Diff erent forest fires causing different degrees of effects occur in fire-sensitive forests due to various reasons such as climate change.Useful as well as harmful aspects of forest fires are a multi-disciplinary research topic.Geographical information systems (GIS) and remote sensing(RS) methods off er a number of benefits for researchers and operators in the field of forest fire research.The present study analyses timber pricing based on forest contractor demands of post-salvage logging processes.The effect of timber obtained from compartment units on producers’ pricing policy was modelled.Sapadere forest f rie area (2500 ha)located in Antalya in Turkey was selected as the main study area.Topography parameters (aspect,slope and slope position),stand types (diameter class and crown closure),and burn severity were analyzed together using GIS and R software packages.A multi-linear regression model (R2=0.752)demonstrated that factors that had the most impact on pricing were slope position,aspect,stand age,crown closure and burn severity.This model can be used to estimate salvage logging prices in Calabrian pine (Pinus brutia Ten.) stands with similar parameters.Forest administrators and contractors may readily address the unit price of timber by estimating approximate costs in a given forest area for which they are going to bid.This will help reduce operational planning times of harvesting procedures in burned stands.

Keywords Timber extraction·Forest fires·Stumpage sale·Cost estimation·Sustainability·Crown fire

Introduction

It is often difficult to identify disturbances in forest ecosystems caused by climate change such as wildf ires,drought,harmful pests,epidemics,species changes,hurricanes,storms and landslides due to their complex relationships(Dale et al.2001).Studies on firef ighting and prevention have increased globally due to ecological and economic losses caused by forest fires (Thomas et al.2017).Forest fires are an indisputable factor of Mediterranean forest ecosystems.In the Turkish Mediterranean coast,crown fires are common in the economically important Calabrian pine forests at low altitudes (Keeley et al.2011).Therefore,forestry researches on habitats and firef ighting technologies have attracted considerable attention.

In order to prevent forest fires,researchers from different disciplines have aimed to reduce the negative effects of forest fires (loss of lives and ecosystem services),and to provide optimal conditions for forest development (Moritz et al.2014;Lee et al.2015).Studies focusing on fuel reduction in forest areas within a framework of precautions against forest fires have been a prioritized management policy for many years (Agee and Skinner 2005;Stephens and Ruth 2005).

Recent studies within the scope of fire management have been carried out using mechanical devices at different scales.Comparisons and evaluations related to forest production and fire management are conducted within the scope of sustainable ecosystem planning.For example,production-oriented forestry operations affecting different stand parameters at different levels due to spatial differences are active intervention methods used during pre-and post-f ire timelines to ensure forest sustainability (Allen et al.2019).

To reduce the negative ecological and economic effects of wildf ires,harvesting is carried out for salvage logging soon after the fire by forest administrations or by contractors in Turkey (Bilici et al.2017).This is because the burnt trees become more vulnerable to pests such as insects and fungi (Sessions et al.2004;Akay et al.2007).Large forest fires,especially in the Mediterranean basin,require largescale evaluation and studies according to harvesting schedules.These studies,which require different management approaches based on various regional differences and stand parameters,usually focus on logging as a common point.

A difficult step is the determination of salvage operation costs on forest lands due to working under pressure and planning and organizing successfully to meet tight deadlines.Both forest managers and contractors lose time during research on unit prices of timber.Thus,an estimation model can be helpful for both sides in order to estimate the unit price of burned stands.

Forestry production costs display regional differences around the world due to social,environmental,political and economic factors.Production costs (loss and prof it) are modelled by taking temporal and spatial differences into account (Ackerman et al.2014).The costs of planning and merchantable logging in different regions of the world are calculated using a variety of analytical methods (Eker 2004;Prestemon and Holmes 2008).Allometric formulas may be used to calculate timber logging costs as far as timber losses are concerned in burnt forest areas by forest administrations(Akay et al.2006;GDF 2015).In addition,in order to reveal post-f ire timber losses,economic losses are analyzed by modelling the relationships between GIS-assisted harvesting(economic value) and fire behaviour (potential distribution of fires) (Rodríguez y Silva et al.2 012).For example,economic losses occurring in forested areas affected by wildf ires may be calculated using satellite images (Rodríguez y Silva et al.2013).

In recent years,researchers and forest operators have reported useful scientific findings based on data obtained from analysis,modelling and visualization studies using GIS and RS.Several successful personal or institutional GIS and RS software can be now found in the market and most trade mark and open-source software are actively used in forestry.Optical or non-optical catalogue data such as old and new satellite images are processed and evaluated using GIS and RS to monitor and analyze changes occurring in vast forest ecosystems (Buma 2012;Aini et al.2019).In today’s world,a large-scale cloud information system which eliminates data processing and analysis limitations such as data storage,processor capacity,and cost is available (Nemani et al.2011).For instance,Google Earth Engine (GEE) off ers a number of advantages in a variety of research topics such as drought,forest fire,water management,and climate and vegetation observation (Gorelick et al.2017).

Economic losses observed in salvage logging following a fire can be easily modelled and explained in GIS environment or RS techniques using the data obtained from allometric formulas.When it comes to salvage logging in particular,numerous studies in the literature have analyzed operational production costs by determining the relationships between unit timber price and sale price or volume loss.In addition,social,ecological and ecosystem impact assessments are also performed using various terms such as fire severity,f ire intensity and burn severity (Keeley 2009;Kavgaci et al.2016).However,there are few studies in the literature which approach production costs and timber unit prices from a scientific perspective except for those focusing on the relationship between operational planning and costs (Thompson and Anderson 2015;Pacheco and Claro 2018;She et al.2019).Therefore,the present study focuses on analyzing the relationships among salvage logging costs,unit price(set by contractors for post-f ire timbers) and various factors in a post-f ire forested area.It aims to off er an enlightening data production model for operators who need an alternative salvage logging operation and production system.

Materials and methods

Study area

The Mediterranean Coast of Turkey was selected as the study area because it is socio-economically and ecologically affected by forest fires.It is located within the boundaries of the Alanya Directorate of Forestry and has a typical Mediterranean forest cover consisting of Calabrian pine (Pinus brutiaTen.),cedar (Cedrus libaniA.Rich.) and oak spp.It has a mild Mediterranean climate and bears a high tourism economy with abundant tourist activities throughout the year.The main study area was the Sapadere forest fire area located within the boundaries of Demirtaş and Mahmutlar Forest Enterprises affiliated with the Alanya Directorate of Forestry.The Sapadere forest fire burnt for almost 24 h 30 June 2017,affecting an area of nearly 2500 ha (Fig.1).

Fig.1 Location of the study area (a),elevation (b),Sentinel 2 (Band 4,3 and 2) pre-f ire (c),and post-f ire satellite images (d)

It was one of the largest forest fires in the history of Turkey,recorded as a difficult one which suddenly spread over a large area and became challenging for firef ighting operations.The fire was eventually extinguished thanks to helicopters,water trucks,f ire stations and firemen from numerous forest enterprise units in the area and by teams organized by the local administration.Salvage logging operations were carried out using unit price and stumpage sale methods (GDF 2015) and timber removed using conventional ground-based extraction methods (man power,cable skidding,chainsaws,and farm tractors).

Data and equipment

In order to determine stand types in the pre-and post-f ire period (species,stand development stages,degree of crown closure,and compartment boundaries),forest maps from 2010 and 2017 (GDF 2017) and 1:25000 scale topographic maps were used.Copernicus Sentinel-2 satellite images have a resolution of 10-m in multispectral bands to calculate burn severity (Drusch et al.2012).Google Earth Engine (GEE),which enables access to a data catalogue at multiple petabytes,such as Landsat and Sentinel satellite data,was used for remote sensing data processing (Gorelick et al.2017).Satellite images were obtained using Internet-accessible application programming interface (API) and an associated web-based interactive development environment (IDE) in GEE which relies on cloud information system.Satellite images captured in the main study area during the pre-f ire(26 June 2017) and post-f ire (7 July 2017) periods were used.These images were captured on certain dates when no logging operations were performed in the burnt forest area and the cloudiness ratio was the lowest level.

ArcGIS 10 software was used to visualize digital maps(ESRI 2011).R statistics program and packages were used for statistical data analysis (Revelle and Revelle 2015;Wei et al.2017;R Core Team 2018).Both unit price and stumpage sale methods were used for production.The present study considers Calabrian pines in 30 different compartments where stumpage sale is preferred.Costing tables from salvage logging operations after clear cutting and offi cial sale leaf lets of the related forest enterprise marketing unit were also used.Unit prices (VAT excluded) were determined through a public auction attended by forest contractors who bid for salvage logging operations.The final unit prices in the stumpage sale (Turkish Lira (TRL)m3) were converted to Euro (€) (According to the European Central Bank Currency web portal 1 €=1.17 US Dollar=4.13 TRL;31-07-2017).In addition,a manual GPS device and video camera were used to obtain spatial data during the field study.

Estimation of burn severity

Remote sensing devices are actively used to determine areas affected by global,regional and local forest fires and to estimate burn severity as well as to prevent,analyze,and monitor forest fires (Chuvieco 2009).In the present study,the delta Normalized Burn Ratio (dNBR) technique was used to create a map for estimating burn severity.dNBR is calculated by subtracting post-f ire NBR from pre-f ire index images (Eq.(1),Miller and Thode 2007;Eq.(2),Petropoulos et al.2014).In Eq.(1),NIR is near-infrared band and SWIR represents short wave infrared spectral range.

The table prepared by the United States Geological Survey-Fire Effects Monitoring and Inventory Protocol(USGS-FIREMON) was used to determine categories for the level of burn severity (USDA 2006).With the online platform of the United Nations Platform for Space-based Information for Disaster Management and Emergency Response (UN-SPIDER),NIRSentinel-2andSWIRSentinel-2images at multiband spectral range were processed using GEE,and thus a burn severity estimation map was created(UNOOSA 2018) (Eq.(1);Fig.2).

Fig.2 Burn severity estimation map created using GEE (a) and defined levels

Topographic data

Shuttle Radar Topography Mission (SRTM) 1 Arc-Second Global (SRTM1N36E032V3,～30 m) was used to analyze land structure in the study area (h ttp://earth explo rer.usgs.gov) (Rodríguez et al.2006).SRTM V3 data were used to obtain slope percent,aspect and position of the fire area.ARCGIS 3D Analyst add-on created slope and aspect maps(ESRI 2011).A toolbox add-on operating in ArcGIS environment and calculating slope position index (SPI) was also used (Dilts 2015) (Fig.3).

Fig.3 Slope % (a),aspect (b) and slope position (c) maps and defined classes

Slope categories recommended by the International Union of Forest Research Organizations (IUFRO) were considered for the analysis of slope categories.The classification proposed by Weiss (2001) and Jenness (2006)was used for slope position.Given the land structure of the burnt forest area,5 × 5 pixel small-neighborhood size relationships were analyzed to create a SPI map.Some decision variables were given numerical values for the database analysis (1=Flat (−1),2=North (0—22.5) and (337.5—360)to 6=South (157.5—202.5)) (Table 1).

Data preparation in GIS environment

Slope percent,aspect and position maps created in Raster format and burn severity estimation maps were converted to vector data in GIS environment where decimals were converted to whole numbers.Burn severity estimation,slope percent,aspect and slope position maps converted in vector format and forest maps were intersected.Feature data containing burn severity,slope percent,aspect,position,stand types and compartment boundaries were brought together.Unit prices per m3(mean unit prices for Calabrian pine timber) set by the forest contractor for Calabrian pine salvage logging were also added to the feature tables.All necessary variables for the estimation model were prepared in GIS environment.

Statistical data analysis

Correlations among the data were statistically analyzed.First,it was determined whether variances among burn severity (BS) estimation value,slope percent,aspect,stand development stages (DBH),crown closure (CC) and slope position (SP) data obtained were distributed homogenously.Correlations were revealed using Pearson analysis.Simple and multiple linear models were created among burn severity,the dependent variable,and independent decision variables.At this stage,the correlation among data with different CC,DBH,slope percent,SP,aspect and BS values was analyzed using simple regression models.

The correlation of unit price (€/m3) for Calabrian pine salvage logging in the stumpage sale compartments,which is the dependent variable,and the mean crown closure,DBH,slope percent,position,aspect and burn severity values was also analyzed using simple and multiple linear regression models.In addition,the root mean square error (RMSE)value of the estimation model for salvage logging prices was also calculated.Descriptive statistical data related to the variables for Calabrian pine stands in thirty different compartments are given in Table 2.

Table 1 Criteria for the defined classes of variables in GIS database

Results and discussion

Correlation among burn severity and variables

The correlation of unit price (€/m3) for Calabrian pine salvage logging and other variables was statistically analyzed.The highest and lowest burn severity values were estimated as 999 and −146,respectively.Some statistical data for the independent variables are given in Table 2.The correlation among burn severity and DBH,aspect,slope percent and SP(p<0.001) revealed a low negative and a slightly positive correlation with burn severity,DBH and aspect (Fig.4).

Fig.4 Correlation among variables according to Pearson correlation (***=p <0.001)

Burn severity tended to increase in inverse proportion to diameters and in direct proportion to crown closure.Additionally,it was also determined that burn severity increased from valley bottoms to hilltops and in southern areas in terms of SP (Alexander et al.2006).BS also increased in direct proportion to the slope percent.It can be therefore concluded that age and diameters of Calabrian pines influenced burn severity more compared to topographic factors(Turner et al.1999).

Linear regression analysis indicated that a model which could account for the correlation between independent variables and burn severity could not be created.These findings confirm those of Wu et al.(2013),that no models could be created using single variable models.Therefore,the linear correlation among BS and all variables was analyzed using a multiple regression model.The resultsof multiple linear regression model (AdjustedR2=0.081,p=0.00) are also not sufficient for burn severity estimation based on all variables (Fig.5).

Fig.5 BS estimation model created using multiple linear regression

Correlation results and model estimation

Thirty different stumpage sale compartments were analyzed and the correlation of unit prices for Calabrian pines bid by the contractors and the mean burn severity (MBS),mean slope percent (Mslope),mean aspect (Maspect),mean position (MSP),mean DBH (MDBH),and mean crown closure(MCC) within the compartment boundaries was analyzed using linear regression models (Table 3).

Table 3 Descriptive statistical data for the dependent variable(cost) and independent variables

There was a positive correlation between unit price for salvage logging and MDBH (p<0.001),aspect (p<0.001)and MSP.On the other hand,a negative correlation was found between MBS (p<0.01) and MCC (p<0.05) and price.A slight positive correlation was found in terms of Mslope.The relationships among each variable andpand r values are shown in Fig.6.Their correlation with the cost variable is Mean diameter at breast height >Mean aspect >Mean burn severity >Mean crown closure >Mean slope position >Mean slope.

Fig.6 The relationships among each variable according to Pearson correlation (* p <0.05,** p <0.01,*** p <0.001)

When the linear relationship between timber unit price and each variable is analyzed,MCC (R=0.378) and MBS(R=0.481) is inversely proportional to price.On the other hand,the unit price increased in direct proportion to MSP(R=0.360) towards the hilltops,MDbH (R=0.644),and Maspect (R=0.620) in the southern and southwestern areas.Although it is difficult to explain the relationship between unit price and Mslope (R=0.154) using a linear model,an increase in slope resulted in a slight increase in unit price(Fig.7).However,it is quite demanding to estimate an acceptable unit price using each single variable,and therefore,all variables were integrated into the model.

Fig.7 Relationships between burn severity and stand and topography factors of forest fires in 2017

A multiple linear regression was created by generalizing compartment size in order to increase the price estimation accuracy.The model was designed in such a way that it could accurately calculate contractor costs rather than revealing the net costs in salvage logging.Thus,the effect of independent variables used in linear regression analysis on salvage logging price (€/m3) (R2=0.75,RMSE=3.17) was determined (Fig.8).

Fig.8 Linear model for the estimation of salvage logging unit prices

Model estimates acceptable price of salvage logging were ±3.17 €/m3for private companies or those with a legal entity.The most decisive factors in price setting may be listed as Mean slope position >Mean aspect >Mean DBH >Mean crown closure >Mean slope >Mean burn severity.However,this finding can only be generalized for forest fire areas with a size and stand structure similar to the Mediterranean coastline (Figs.1 and 3).This model makes it possible to estimate unit prices in a short period of time and thus salvage logging may be rapidly performed to prevent further value losses in burnt forest areas (Sessions et al.2004).Thus,forest enterprises may easily find a balance between prof it and loss in terms of stumpage sales on the basis of legal procedures (GDF 2015).

Salvage logging unit price will affect changes in supply and demand balance for firewood (Ackerman et al.2014).Interdisciplinary studies and spatial measurements are needed in order to create more realistic,detailed and accurate models and therefore new studies must be carried out to determine unit prices and standard times for salvage logging.A new estimation model with a higher accuracy rate may be created when weather and fuel material conditions are integrated into the correlation among burn severity and preferred variables (Oliveras et al.2009).In addition,factors related to geographical differences must also be taken into account.

Conclusion

In the present study,spatial data such as burn severity,stand attributes and topography were readily modelled and reported in a GIS environment.Owing to the cloud information system,the need for a high capacity computer and expensive software was eliminated.The results demonstrate that the unit price of salvaged Calabrian pine was inversely proportional to burn severity and crown closure.Slope,aspect and DBH classes as independent variables had a positive effect on unit price.Slope position,aspect and DBH had greater effects on price setting.

The unit price estimation model (€/m3) created for Calabrian pine salvage logging in stumpage sales compartments similar to the main study area can be used effectively by all forest enterprises with certain standards set.Other spatial data such as distance to main roads,warehouses or batching areas can be integrated into the model in this study which will improve the accuracy of unit price estimation.In addition,the determination of unit cost and standard times in forestry operations is of utmost importance.

AcknowledgementsThe author would like to thank the Turkish Academic Network and Information Center (TÜBİTAK-ULAKBİM),the Google Earth Engine (GEE),the U.S.Geological Survey (USGS) and the European Space Agency (ESA) for providing open-access data,would like to thank Dr.Sercan Gülci for his comments and help in field work.Besides,the author would thank to editors of the journal as well as the anonymous reviewers who have helped improve and clarify this manuscript.

Author contributionsThe author designed and prepared all stages of this manuscript.

Compliance with ethical standards

Conf lict of interestThe author declares that there are no conf licts of interest regarding the publication of this manuscript.