Wildland fires are major environmental concerns in many ecosystems. In forest ecosystems, fires usually start in surface fuels, and then further spread and develop into large fires depending on the environmental conditions ( Kucuk et al. , 2012 ). In the Mediterranean Basin, fire plays a major role in maquis and pine forests. Surface fuels of Mediterranean pine forests are highly flammable due to their high dead fine fuel loading, aerated litter and chemical volatility ( Fernandes et al. , 2009 ; Kucuk & Aktepe, 2017 ), leading to potentially very high rates of fire spread in pine stands ( Fernandes & Rigolot, 2007 ). Fire spread models have been developed and calculations for surface and crown fuels have been carried out globally ( e.g. Alexander & Lanoville, 2004 ; Morvan, 2014 ), and several studies have provided data and models for surface and crown fire behavior in the Mediterranean Basin ( e.g. , Dimitrakopoulos, 2002 ; Bilgili et al. , 2006 ; Kucuk et al. , 2007 ; Fernandes, 2009 ; Kucuk et al. , 2015 ).

The prediction of surface fire behavior is crucial for fuel and fire management planning. However, it has been one of the most challenging issues in fire behavior modeling as the fuel conditions and stand structure characteristics can be highly variable in various stand types ( e.g. Sullivan, 2009 ; Fernandes, 2014 ), making the successful prediction of fire behavior difficult in these fuel types. The variability in fuel characteristics is usually a result of stand growth and development as well as fuel management. Silvicultural interventions, such as thinning, radically alter the structure and distribution of fuels within a forest stand ( Bilgili & Methven, 1994 ; Bilgili, 2003 ), thereby resulting in changes in microclimatic conditions. Changes in fuel characteristics and weather conditions greatly influence fire behavior characteristics ( Graham et al. , 1999 ; Bilgili, 2003 ).

Anatolian black pine ( Pinus nigra J.F. Arnold subsp. nigra var. caramanica (Loudon) Rehder) (aka black pine) is the second most widely distributed conifer species in Turkey, covering a land area of 4.7 million ha ( GDF, 2014 ). Pure stands of black pine are generally found in fire susceptible environments ( Küçük et al. , 2008a ; Kucuk et al. , 2017 ). Silvicultural interventions ( i.e. , thinnings) are common practices in black pine stands. However, there has been no study on the prediction of rate of fire spread in thinned black pine forests in Turkey. The main objective of this study was to model the rate of surface fire spread in a thinned, even-aged Anatolian black pine stand. Results of this study can be very useful for fire and fuel management.

Summary of surface fuel, weather and fire behavior parameters is given in Table 1. The correlation analysis indicated that wind speed and some fuel properties were correlated with ROS (Table 2). The regression models that best explained the relationships between ROS and fuel and weather variables are given in Table 3.

ROS was closely related to wind speed, length of ignition line, moisture content of fine fuels and fine fuel load (needle and 1h fuels). Wind speed alone explained 85% of the observed variation in the rate of fire spread (Table 3; R ² = 0.854; p < 0.01, log ROS model 1). The results obtained are in agreement with many studies conducted under field conditions ( e.g. , Van Wagner, 1993 ; Fernandes et al. , 2009 ). The addition of length of ignition line as the second independent variable improved slightly the percent variability explained ( R ² = 0.866; p < 0.01, log ROS model 2). Although the MAPE decreased from 42.99% to 31.08%, a tendency towards under prediction of ROS at high values was confirmed by the MBE . For all measured ROS values, the MBE changed from -0.093% to -15.699% (Table 3). The effect of LIL on fire rate of spread have also been reported by other researches ( e.g. , Wotton et al. , 1999 ; Dupuy et al. , 2011 ; Morandini et al. , 2001 ; Fernandes, 2014 ). However, the potential use of LIL in a model or in practice can be limited from a practical point of view except for prescribed burns for which the length of ignition lines is determined by a practitioner; and given the sizes of the burning plots as compared with the fire lines achievable in larger experiments and in large wildfires, the contribution of this study should be evaluated with respect to the comparative analysis of ROS only, and not the actual prediction of ROS . A correction factor such as that proposed by Fernandes (2014) may be used to better justify and compare fire spread differences under the same environmental conditions but different ignition line lengths.

The dependence of ROS on fuel properties was also evident from the results. Addition of fuel moisture and fine fuel load ( needle FL and 1h FL ) significantly increased the percent variability explained ( log ROS model 3 and 4, Table 3). While the inclusion of other fuel and weather parameters to the model did not significantly change the standard error of the estimates, MBE decreased significantly, and, except for log ROS model 3, MAE and MAPE values were comparable (Table 3). Differences in fire spread can be ascribed to the considerable heterogeneity in fuel load and fuel moisture content (Table 1). The relationships with fuel moisture content and fuel load in addition to wind ( log ROS model 4, Fig. 2; ROS model 5 and 6, Table 3) indicate that variability and heterogeneity within-stand weather and fuel characteristics greatly influence fire spread.

The results of the study were based on a total of 33 experimental fires in surface fuels under canopy. Regression models were fitted for the estimation of the rate of surface fire spread. Analyses indicated that midflame wind speed was the most significant factor affecting rate of fire spread. Length of ignition line, fine fuel load and fuel moisture characteristics were also influential on fire spread. Although the experimental burns covered a limited range of weather and fuel conditions, the impact of weather and fuel variability within the stand were evident on observed rates of spread. This preliminary results could help managers to plan fire hazard, suppression operations and pres­cribed burnings in these forests. Nevertheless, further experimentation is required for developing a robust rate of fire spread prediction model for surface fires in thinned Anatolian black pine stands.