Introduction

In Turkey, forest stands are classified by differences in species composition, crown closure and development stages. Crown closure is defined as the percentage of forest canopy projected on a horizontal plane over a unit ground area, and it becomes a very important parameter in developing forest, ecological, hydrological and climate models (Xu et al. 2003). Nevertheless, its measurement in the field is hard, time consuming and expensive (Xu et al. 2003; Pu et al. 2003). Thus, other methods of estimating forest characteristics for larger areas such as remote sensing are often used. Remote sensing data is an alternative way of taking field measurements in accurate crown closure estimation because it is considered a low-cost and large-area coverage forest information resource (Franklin et al. 2003). In other similar studies on this topic, Butera (1986) confirmed the relationship between the crown closure and the simulated Thematic Mapper (TM) spectral bands in Colorado, USA and the TM data bands predicted forest crown closure with 57–74 % accuracy. Peterson et al. (1987) analyzed the forest structure in Sequoia National Park (California, USA) using simulated TM data. Statistically significant relationships were found between the TM reflectance values and stand parameters such as crown closure and basal area. Pereira et al.(1995) used Landsat TM imagery and ground measurements to determine biomass, percent canopy cover and canopy volume in Mediterranean shrublands; the greatest results were obtained with the Normalized Difference Vegetation Index (NDVI) to estimate percent canopy cover (R2 = 0.65). Calvão and Palmeirim (2004) collected field data on Mediterranean shrublands and developed correlations between several parameters and spectral variables (single channel reflectance value and NDVI) from Landsat TM data; the highest correlation for canopy cover was obtained with TM 3 (r = −0.91) and NDVI (r = 0.91). Sivanpillai et al. (2006) evaluated the relationship between Landsat ETM + reflectance values and the stand characteristics of commercially managed loblolly pine (Pinus teada L.) in east Texas. The models for stand age and tree density respectively were obtained with an adjusted R2 (78 % and 60 %). Mohammadi et al. (2010) modeled forest stand volume and tree density using Landsat ETM + data in the Hyrcanian forests, northern Iran. The models for stand volume and tree density respectively were obtained with an adjusted R2 (43 % and 73.4 %). The aim of this study was to investigate the relationship between spectral reflectance values and VIs calculated by Landsat TM sensor to estimate crown closure and tree density in the mixed forest areas.

Materials and Methods

Study Area and Data

The study area covers the provinces of Almus, Niksar, Erbaa, Bafra, Vezirköprü, Merzifon, Kargı, Boyabat, Arac, Samatlar, Ilgaz, Dirgine, Karabük and Mengen in the Black Sea Region, in the north of Turkey. This study area ranges in latitude (North) from 40o15′28″ to 41o46′15″ and in longitude (East) from 32o28′02″ to 37o32′56″ (Fig. 1). These sampled mixed stands were naturally regenerated and uniformly stocked stands (55–97 % tree layer cover), without any evidence of historical damage such as fire or storms. Located between 750 m and 1750 m altitude, the study area is characterized geomorphologically by high mountainous land, with moderate to steep slopes ranging between 5 % and 60 % (30 % of the whole area). The mean annual temperature is between −5.8 °C and 14.6 °C, and minimum and maximum temperatures are −8.4 °C and 22.67 °C, respectively. The climatic regime is of typical Black Sea climate characterized by a mild winter, a cool summer and relatively homogeneous precipitation as high as 1000 and 1250 mm.

Fig. 1
figure 1

Map showing the locations of the mixed Scots pine-Oriental beech studied in northern of Turkey

Full size image

In this study, the data were obtained from 156 temporary sample plots with ranging stand age, site index, density and mixture percentage in mixed stands Scots pine (Pinus sylvestris L.) and Oriental beech (Fagus orientalis Lipsky). The size of circular plots ranged from 600 to 1200 m2, depending on stand density, to accomplish a minimum of 50–60 trees per plot. The diameters at breast height, total height, age, crown height, crown diameter, and spatial coordinate were measured in each tree of sample plots. In each plot altitude, aspect and slope were also measured. The spatial coordinates were measured in all trees (alive and dead). Diameter at breast height (dbh) was measured to the nearest 0.1 cm with calipers for every living tree with dbh > 8.0 cm. Tree height and crown height were measured to the nearest 0.1 m by digital hypsometer in a randomized sub-sample of 30–40 trees representing various dbh classes in the sample plots. Crown diameter was measured to the nearest 0.1 cm by tape meter in a randomized sub-sample of 30–40 trees representing various dbh classes in the sample plots.

The tree density per hectare was obtained by multiplying the tree number per plot by the hectare conversion coefficient. Crown area and stand canopy were calculated. Also, two different Landsat TM satellite images were used in this study. The Landsat TM was acquired on 24th October 2005 and 21st July 2006.

Digital Processing of Landsat TM

Data processing, interpreting and analysis were performed using Erdas Imagine 9.1TM image processing software (Erdas 2002). The Landsat TM satellite image was ortho-rectified and its accuracy was checked using Global Position System (GPS) data collected from the study area. A subset of Landsat TM was rectified using 1/25.000 scale topographical maps with UTM projection (ED 50 DATUM, Zone 36) using first order nearest neighbor rules. In total 40 ground points were used to register the TM image subset with a rectification error less than 1 pixel image. Solar zenith angle and atmosphere influence the spectral value of a satellite image. Thus, radiometric correction must be done to convert the digital number to the reflectance value. In the process of radiometric correction, the digital number of Landsat TM must be converted to the radiance value, and then to reflectance. The information for the radiometric correction (solar zenith angle, acquisition date, and so on) can be obtained from Landsat TM ancillary data.

Determining Crown Canopy Ratio

To determine the crown canopy ratio of each sample plot, firstly the center coordinates of each sample plot was entered into an Excel file and that file was added to a GIS (Geographic Information System) environment and converted to a point layer file using Arc GIS 10.0 ™ functions. Then, a buffer layer including all plots was created according to sample plot size ranges from 600 to 1200 m2. After that, all tree x, y coordinates were also entered into an Excel file and buffers were created using crown diameters for each tree using the same method explained above (Fig. 2).

Fig. 2
figure 2

General framework of sample plot crown closure ratio calculation

Full size image

Finally, the two layers, the sample plot area layer and the tree area layer, were intersected using GIS analysis tools to obtain a sample plot crown closure (Fig. 3). Subsequently, crown canopy and crown canopy ratio values were calculated using GIS area calculation procedures using “statistics” and “field calculator” commands.

Fig. 3
figure 3

Calculation of a sample plot crown closure

Full size image

Calculating Vegetation Indices

In this study, Landsat TM satellite images were used to produce six VIs such as SR, DVI, SAVI, NDVI, TVI and NLI. The VIs formulas and references for these indices are shown in Table 1.

Table 1 Definition of vegetation indices used in the study area
Full size table

SR, Simple Ratio; DVI, Difference Vegetation Index; SAVI, Soil Adjusted Vegetation Index; NDVI, Normalized Difference Vegetation Index; TVI, Transformed Vegetation Index; NLI, Nonlinear Vegetation Index; VIS, Visible wavelengths (TM1, TM2, TM3); NIR, near infrared wavelengths (TM4), MID, middle infrared wavelengths (TM 5, TM 7). L is equal to 0.5 in this study.

Statistical Analysis

The digital number of pixels within a 3 × 3 pixel window was extracted from all Landsat TM bands and VIs. To examine and model the relationships between remote sensing data based on the spectral reflectance values, TM 1–5 and 7, and six VIs, and stand parameters, e.g., tree density and crown closure, the correlation analysis and multiple linear regression analysis were used in the studied species and stands. These multiple regression models were developed to estimate the models for crown closure and tree density using remote sensing data, reflectance values and VIs, their combination as an independent variable and which dependent variables were stand attributes, e.g. tree density and crown closure. Also, the regression models were developed to estimate the stand parameters as a function of the suite of remote sensing data variables gathered in the studied forest site. Multiple linear regression technique was used to model the variation in some stand parameters based on a stepwise variable selection method. In a stepwise variable selection method, the process starts out just as in forwards selection, but at each step a variable that is already in the model is first evaluated for removal, and if any are eligible for removal, the one whose removal would least lower the R2 is removed. The multiple stepwise regressions and partial correlation analysis were performed using SPSS version 15.0 (SPSS Institute Inc. 2007). The stepwise regression technique was used to select the best site variables that are significant (p < 0.05) with the determination of the highest value of the coefficient adjusted by the number of parameters (R 2adj ), also called adjusted by the coefficient of determination. In this study, the following linear relationship was assumed:

$$ SP={\beta}_0+{\beta}_1\cdot {X}_1+{\beta}_2\cdot {X}_2+\dots .+{\beta}_n\cdot {X}_n+\varepsilon $$
(1)

where SP is the forest stand parameter, e.g. tree density and crown closure, X1….Xn are variable vectors corresponding to remote sensing data, e.g. the reflectance values, TM 1–5 and TM 7, and 6 VIs variables, β 1 ….β n , represent model coefficients and ε is the additive error term (Corona et al. 1998; Fontes et al. 2003).

To compare the predictive power of the spectral reflectance values, e.g. TM 1–5 and 7, and VIs, separate regression analysis was performed using relevant remote sensing data. Hence, four regression models were developed (two forest stand parameters and two remote sensing data selection models, e.g. band reflectance values and VIs). For example; there was a model predicting tree density using the spectral band reflectance values, another model using VIs and predicting tree density by using the band reflectance values and another model using VIs. In each sub-group, the related forest stand parameters were predicted using the spectral reflectance values, e.g. TM 1–5 and 7, and VIs.

The estimate of each parameter for the variables of these regression models should be statistically significant at a 95 % probability level. The null hypothesis H 0  = β 0   = .β 1  = … β n  = 0 was tested and parameters that were not significantly different from zero were rejected (Fontes et al. 2003).

The regression models were evaluated based on the accuracy statistics. The accuracy statistics covered the absolute and relative biases and the root mean square error (RMSE and RMSE %). These statistics were calculated for the models as follows:

$$ bias=\frac{{\displaystyle \sum \left({y}_i-{\widehat{y}}_i\right)}}{n} $$
(2)
$$ bias\%=100\frac{{\displaystyle \sum \left({y}_i-{\widehat{y}}_i\right)/n}}{{\displaystyle \sum {\widehat{y}}_i/n}} $$
(3)
$$ RMSE=\sqrt{\frac{{\displaystyle \sum {\left({y}_i-{\widehat{y}}_i\right)}^2}}{n-1}} $$
(4)
$$ RMSE\%=100\frac{\sqrt{{\displaystyle \sum {\left({y}_i-{\widehat{y}}_i\right)}^2/\left(n-1\right)}}}{{\displaystyle \sum {\widehat{y}}_i/n}} $$
(5)

Where n is the number of observations, and y i and \( {\widehat{y}}_i \) are the observed and predicted values of stand parameters e.g. tree density and crown closure from the developed models.

Descriptive statistics including mean, minimum, maximum, standard deviation and the coefficient of variation of the plot characteristics, reflectance values and VIs are listed below (Table 2).

Table 2 Summary statistics for sample data in the study area
Full size table

Results and Discussion

Relationships Between Crown Closure, Tree Density, Reflectance Values and Vegetation Indices

Table 3 shows the Pearson correlations between the crown closure, tree density, reflectance values (TM 1–5 and TM 7) and VIs (SR, DVI, SAVI, NDVI, TVI and NLI) representing the ecological variation in the study area. Not all VIs and reflectance values were significantly related to the crown closure and tree density. TM 1, TM 2, TM 5 and SR were significantly correlated with stand crown closure at a 95 % probability level. In addition, all spectral bands and VIs (except for DVI and NLI) were significantly correlated with stand tree density at a 95 % probability level. The independent variables TM 1 and TM 2 and TM 5 were positively correlated with the dependent variable stand crown closure, whereas the other variable SR was negatively correlated. Moreover, the predictor variables spectral reflectance values TM 1–5 and TM 7 were positively correlated with the dependent variable stand tree density, whereas the other variables vegetation indices NDVI, SR, TVI and SAVI were negatively correlated. The stand crown closure showed a moderate correlation with TM 1 (r = 0.34) and SR (r = −0.31). Furthermore, stand tree density showed a moderate correlation with TM 2 (r = 0.47), TM 2 (r = 0.49), TM 2 (r = 0.47), TM 3 (r = 0.42), TM 4 (r = 0.40), NDVI (r = 0.36), TVI (r = 0.36) and SAVI 2 (r = 0.36).

Table 3 Pearson correlations between crown closure, tree density, reflectance values and vegetation indices
Full size table

Estimating Tree Density and Crown Closure Using Landsat TM

A linear combination of VIs (DVI and SAVI) was a better estimation of tree density (R2 = 0.702) than other VIs and band reflectance values. An inverse relationship was observed between DVI and SAVI, whereas a direct relationship was observed between TM 1 and tree density. Linear combinations of VIs (SR and DVI) were better predictors of crown closure than other VIs and band reflectance values. Inverse relationships were observed between SR and DVI, and TM 1 with TM 3 and TM 4 (Tables 4 and 5).

Table 4 Parameters of the ‘best fit’ regression models of stand attributes based on the spectral reflectance values, TM 1–5 and TM 7
Full size table
Table 5 Parameters of the ‘best fit’ regression models of stand attributes based on vegetation indices
Full size table

A linear combination of SAVI and DVI explained more variance in tree density than other combinations of bands and VIs. There was a significant relationship at the 95 % probability level, normality of the residuals, 70.2 % adjusted R2, and RMSE of 83.208n ha−1. Tree density values compared favorably to RMSE and R2 values obtained by Sivanpillai et al. (2006) (RMSE = 312.5 n ha−1; R2 = 60.4 %) and Freitas et al. (2005) (R2 = 66.6 %). The R2 value obtained from this study was lower than the one that was obtained through direct estimation to predict tree density (R2 = 0.73, Mohammadi et al. 2010). In addition, the regression model with ETM4 and ETM5 as independent variables could predict tree density (R2 = 73.4 %; RMSE = 170.13 ha−1) better compared with other combinations of ETM + bands and vegetation indices. In this study, the regression model with only TM 1 as an independent variable could estimate tree density (R2 = 61.3 %; RMSE = 102.08 n ha−1) and the other TM bands did not show relationships.

A linear combination of SR and DVI explained more variance in crown closure than other combinations of original bands and VIs. There was a significant relationship at the 95 % probability level, normality of the residuals, 67.4 % adjusted R2, and RMSE of 0.02103. Falkowski et al. (2005) compared the regression models for each response variable (crown closure). Red and green band reflectance values had a strong relationship to crown closure with scores of (R2 = 0.71) and (R2 = 0.75), respectively. The relationship between NIR band reflectance values and crown closure was poor (R2 = 0.08). In the same study, the model including both the NDVI and GRVI vegetation indices as predictors of crown closure obtained the highest relationship (R2 = 0.77). Xu et al. (2003) found that the regression model with TM 4, TM3 and TM2 as independent variables could better predict stand crown closure (R2 = 0.802). Berberoglu et al. (2009) found that Envisat MERIS data (as 13 independent waveband variables) can be used to predicted crown closure (R2 = 0.433) with considerable spatial detail. Franklin et al. (2003) found that for jack pine, the relationship was positive and its value was moderate to strong (significant R ranged from 0.40 to 0.81). For white spruce, the relationship was negative and weak to moderate (significant R ranged from −0.20 to −0.57). In this study, the regression model with TM 1, TM 3 and together with the TM 4 as independent variables could estimate tree density (R2 = 61.0 %; RMSE = 0.02604 %) and the other TM bands did not show relationships. Stand crown closure and stand tree density estimations by both Landsat TM reflectance values and VIs using the models are plotted against these observed forest variable values (Figs. 4 and 5).

Fig. 4
figure 4

Comparison between observed and predicted stand crown closure by regression models based on the spectral reflectance values, TM 1–5 and TM 7 (a) and VIs (b)

Full size image
Fig. 5
figure 5

Comparison between observed and predicted stand tree density by regression models based on the spectral reflectance values, TM 1–5 and TM 7 (a) and VIs (b)

Full size image

Conclusions

In this study, the relationship between reflectance values and some VIs obtained from Landsat TM recorded with tree density and crown closure were analyzed through multiple linear regression analysis in mixed forest stands. Statistically significant relationships were found between corresponding reflectance values and VIs recorded by the TM satellite image with tree density and crown closure. A linear combination of SAVI and DVI explained more variance in tree density than other combinations of bands and VIs. The model for tree density resulted in adjusted R2 = 70.2 %; RMSE = 83.208 n ha−1. Similarly, a linear combination of SR and DVI explained more variance in crown closure than other combinations of bands and VIs. The model for crown closure resulted in adjusted R2 = 67.4 %; RMSE = 0.02103 %. Based on the results of this study, the Landsat TM satellite image data are useful to estimate tree density and crown closure in mixed forest stands.