AnInternationalScientificResearchJournal

ORIGINALRESEARCH

JournalofResearchinEcology

Authors: SaeidShabani1, AkbarNajafi2, BarisMajnonian3, JalilAlavi4and AliSattarian5.

Institution:

1.DepartmentofForest Engineering,FacultyofNatural Resources,TarbiatModares University,Iran.

2.FirstAffiliation:Departmentof ForestEngineering,Facultyof NaturalResources,TarbiatModares University,Iran.Secondary affiliation:Departmentof Harvesting,Logisticsand Amelioration,facultyofForestry, TechnicalUniversityinZvolen, Slovakia

3.DepartmentofForestry,Faculty ofNaturalResources,Universityof Tehran,Iran,

4.DepartmentofForest Engineering,FacultyofNatural Resources,TarbiatModares University,Iran,

5.FacultyofNaturalResources, UniversityofGonbad-e-Kavous,

Correspondingauthor:

AkbarNajafi, EmailId:

WebAddress:

http://eologyresearch.info/ documents/EC0031.pdf

ABSTRACT:

Stand ecology can be aﬀected during forest harvesëng. The aim of the presentstudywastodeterminepotenëalareaextentofstanddamagesontheforests basedonstandmonitoringdatacollectedfrom351circleplotswiththesurfacearea of 314 m2 before the beginning of logging operaëons. Slope percent, land form, roughness,forest type, standdensity, extracted volumeand extracëon system were collectedaspredictorvariables. Residual treeand regeneraëondamageas response variables were measured during postharvest assessments too. In this context a geospaëallyexplicitpredicëvemodelofstanddamagewasdevelopedusingLRandFR asaninterfaceandcollecëonoffuncëonsinR.Geospaëally,standdamagesinterms of residual tree and regeneraëon map were depicted by GIS package, separately. According to success and predicëon rates, LR had the best performance model comparedwithFRinresidualtreeandregeneraëondamagemodelling.Basedonthe LRresultsslope,foresttype,standdensity,harvestedvolumeandextracëonsystem were the most eﬀecëve variables controlling stand damages. These models by local calibraëonaredecisionsupporttoolstopredictstanddamagesduringground-based harvestoperaëonsonotherforestlands.

Keywords: AIC,ForestEcosystems,Logging,PredictorVariables

Ar!cleCita!on: SaeidShabani,AkbarNajaﬁ,BarisMajnonian,JalilAlavi,andAliSaWarian. EcologicalModellingasaToolforDamagesReducingtoForestEcosystemsduring Skidding. JournalofResearchinEcology(2015)3(1):021-030

Dates: Received:04May2015 Accepted:25May2015 Published:06June2015

ThisarticleisgovernedbytheCreativeCommonsAttributionLicense(http://creativecommons.org/ licenses/by/2.0), which gives permission for unrestricted use, non-commercial, distribution and reproductioninallmedium,providedtheoriginalworkisproperlycited.

JournalofResearch

021-030|JRE|2015|Vol3|No1

inEcology

AnInternational

INTRODUCTION

In all of the hyrcanian forests in the north of Iran, selection silviculture is a common forest management system and it is progressively becoming a more common forest management practice globally (Marvie Mohadjer, 2004). In single-tree selection logging, 1-2 percent of standing volume is removed from the stand and residual trees are retained across a full range of size classes approximating a target tree diameter distribution, intended to ensure recruitment of trees into successively larger size classes.

Uncontrolled selection logging causes vast soil disturbances (Najafi et al., 2009), reduced floral and faunal diversity (MacDonald et al., 2014), extensive canopy cover removal (Rockwell, 2007), residual tree damages, regeneration loss and long-term changes in tree species composition and forest types (Saga and Selas, 2012).

The guiding idea of sustainability orders that a commercial forest should be managed to decrease as much undesirable damage as possible to the residual stand and all over forest ecosystem. This is an essential purpose for both ecological and economic reasons (Muller, 1998). Reduced Impacts on Logging (RIL) is one of the harvests methods that minimize residual stand damages via directional felling, pre-harvest inventory, precise planning of forest roads and skid trails, finding an optimal log landing location, and mapping.

Mapping of stands susceptibility to damages during skidding is performed by modelling; whereas it is assumed which conditions that lead to stand damages in the past are likely to them in the future as well (Vorpahl et al., 2012). Modelling predict damages by a set of predictor variables in any location in a landscape over a period of time without the necessity of mathematically describing the underlying processes in a physical way. Few studies have reported the prediction of stand damages in forest operations (Sowa and Kulak, 2008; Reeves et al., 2012). Logistic Regression (LR) with a logit link function, is one of the most frequently used techniques in damages susceptibility modelling. LR consists of an additive combination of single parametric terms, each representing a linear function of a single predictor. Due to their simplicity LR allow for a broad variety of statistical analyses. Nonlinearities as well as predictor interactions can be regarded by explicit definition of additional predictor variables (Vorpahl et al., 2012).

Besides LR, Frequency Ratio (FR) is such a models that have not been used for prediction of forest stand damages despite of extensive application in other natural resource sciences. FR as a bivariate statistical method can be applied as a suitable geospatial assessment tool to determine the hypothetical relationship between response and predictor variables, consist of multi-classified maps (Pourghasemi et al., 2012).

The objectives of this study was to apply LR and FR ecological modelling techniques to model the relationships between stand damages (consisting residual tree and regeneration) and predictor variables and evaluate the performance of above models using Area Under the Curve (AUC).

MATERIALS AND METHODS Study Area

The study area is located in the western part of Mazandaran province, Iran (Figure 1). The application forest lies between the latitudes 36° 30′ to 36° 35′ N, and the longitudes 51° 27′ to 51° 33′ E, that covers an area of 108 ha. Altitude in the study area varies between 1400 and 2000 m above sea level. The slope percent of the area range from 18 to as much as 120. Based on Iranian Meteorological Department, the temperature of region area ranges between 0 °C and 32 °C. The mean annual rainfall is around 800 mm, most of which falls during the month of September and December. The study area is covered by two soil textures clay and clay – loamy according to laboratory results. The dominant species are

Journal of Research in Ecology (2015) 3(1): 021-029

beech (Fagus orientalis Lipsky), with other species consisting maple (Acer velutinum Boiss), horn (Carpinusbetulus L.), alder (Alnus subcordata C. A. Mey), and oak (Quercus castaneifolia C. A. Mey).

The trees were felled using manual chain saw and logs transported from stump to forest road side by mule and Zetor (LTT-100A) fuel woods and timbers,

Figure 1. Study design for single plot

respectively. Extraction was done between mid-June to July 2013.

Sampling Methods

From mid-June to July in 2013 we used random-systematically methods to determine response and predictor variables for forest. We established 50 * 50 network to set up 351 sampling plots of 314 m2 (radius 10 m) in network crossing. Prior to skidding were recorded; predictor variables consisting of slope percent (<30, 30-60, >60), land form (ridge, hillside, canyon), roughness (even, uneven, rough) (Owende et al., 2002) forest type (pure beech, beech-maple, mixed: alder-oakhornbeam), stand density per ha (<150, 150-200, >200), extraction system (Mule or Zetor), and were carried out extracted volume m3/ha (<10, 10-13, >13) during skidding.

Post-harvest stand damages data (response variables) had been assigned 0 for absent of damages (non-occurrence) or 1 for present of damages (occurrence) for each plot. Taking into account studies concerning ecological damages to stands by other studies (Iskandar et al., 2006) the following classification was assumed: Residual tree damage: bark-wood damage > 25 cm2; Regeneration Damage: breaking and regeneration destruction (DBH < 7.5 cm2)

Geospatial-Ecological Modelling Logistic Regression Model

In general, to predict ecological damage, it is necessary to assume that damages occurrence is determined by damage-related variables, and that future damages will occur under the same situations as previous damages. In the present study, was fitted a third order polynomial was fitted for each predictor and additionally performed a backwards stepwise variable selection based on the Akaike Information Criterion (AIC) and displayed prediction values in GIS. The purpose of using AIC is to find an optimal trade-off between an unbiased approximation of the underlying model and the loss of accuracy caused by estimating a number of parameters, and the number of data points used in its calibration (Dawson et al., 2006). The standard logistic model equation can be expressed in its simplest form as (Eq. 1): (Eq. 1):

α+ β1X1 +β2X2 + ... +β iXi

Exp

Υ=

α+β1X1 +β2X2 + ... +βiXi

1 + Exp Where:

Y: varies from 0 to 1,

a: intercept of the model, βi (i=1, 2, 3, …, n): slope coefficient of the model

i: number of independent variables Xi (i= 1, 2, 3, …, n) is the independent variable.

Frequency Ratio Model

To quantitatively construct the damage susceptibility map, the FR model should be in a GIS environment. The FR is the ratio of the area where damages occurred in the total study area, and is also the ratio of the damage occurrence to non-occurrence for the given attribute. At first, the FR was calculated for each range or type of variable; the FR were then summed to calculate the Damage Susceptibility Index (DSI) (Eq. 2) (Pourghasemi et al., 2012):

(Eq. 2):

Where DSI is damage susceptibility index and n is the number of variables. The FR method is very easy to apply, and results reported by many authors (Yilmaz, 2009) are readily intelligible. Models Comparison

After training, both models will be grouped of one type that were trained on the same functional part of stand damages and calculated quality measures from a tenfold internal cross-validation to determine success and prediction rates.

RESULTS

(Eq. 3):

According to success and prediction rates, LR had the best performance model compared to FR in residual tree damage modelling (Figure 2, 3).

The results of spatial relationship between residual tree damage and predictor variables using frequency ratio model is shown in Table 1. In the mentioned table, slope<30% and mule extraction system indicated low probability of residual tree damage occurrence (FR: 0). Also zetor extraction system had the highest residual tree damage (FR: 2.434).

Significant predictor variables on residual tree damage based on LR model is shown in table 2. Additionally, the resultant beta coefficients and test statistics for each independent variable in the logistic regression equation are given in Eq. 3. Based on the analysis of logistic regression coefficients, the results showed that slope, stand density and extracted volume have positive effect on residual tree damage occurrence. It also showed that stand density has the highest beta coefficient (0.1), followed by slope (0.04) and extracted volume (0.01). In contrast, mule extraction system has negative effect on residual tree damage (Beta coefficient: -22.34).

-5.11-22.34extractionsystem(mule)+ 0.01extractedvolume+ 0.1standdensity + 0.04slope

e

Υ=

-5.11-22.34extractionsystem(mule)+ 0.01extractedvolume+ 0.1standdensity + 0.04slope

1+ e

Figure 2. Success rate curves for the residual trees Figure 3. Prediction rate curves for the residual susceptibility maps produced in this study trees susceptibility maps produced in this study

Journal of Research in Ecology (2015) 3(1): 021-029.

Table 1. Spatial relationship between residual tree damage and predictor variables by FR model

Predictor Variables FR

No. of pixels Percentage No. of Percentage in domain of domain Damage Of Damage

Slope (%) | <30 | 200 | 1.871 | 0 | 0.000 | 0.000 |
---|---|---|---|---|---|---|

30-60 | 6575 | 61.518 | 52 | 46.847 | 0.762 | |

>60 | 3913 | 36.611 | 59 | 53.153 | 1.452 | |

Land Form | canyon | 3733 | 34.927 | 36 | 32.432 | 0.929 |

hillside | 2170 | 20.303 | 20 | 18.018 | 0.887 | |

ridge | 4785 | 44.770 | 55 | 49.550 | 1.107 | |

Roughness | even | 4504 | 42.141 | 49 | 44.144 | 1.048 |

uneven | 2971 | 27.798 | 30 | 27.027 | 0.972 | |

rough | 3213 | 30.062 | 32 | 28.829 | 0.959 | |

Forest Type | pure beech | 2646 | 24.757 | 23 | 20.721 | 0.837 |

beech-maple | 5430 | 50.805 | 71 | 63.964 | 1.259 | |

mixed | 2612 | 24.439 | 17 | 15.315 | 0.627 | |

Stand Density (ha) | <150 | 4138 | 38.716 | 23 | 20.721 | 0.535 |

150-200 | 3631 | 33.973 | 43 | 38.739 | 1.140 | |

>200 | 2919 | 27.311 | 45 | 40.541 | 1.484 | |

Extracted Volume (m3/ha) | <10 | 3190 | 29.847 | 10 | 9.009 | 0.302 |

10-13 | 1831 | 17.131 | 29 | 26.126 | 1.525 | |

>13 | 5667 | 53.022 | 72 | 64.865 | 1.223 | |

Extraction System | mule | 6297 | 58.917 | 0 | 0.000 | 0.000 |

zetor | 4391 | 41.083 | 111 | 100.000 | 2.434 |

Table 2. Significant predictor variables on residual tree damage according to LR model

Variables

DF AIC pr (>Chi)

Slope

1 224.17 0.00581

Volume Extraction

1 429.39 0.00000

System

Finally based on the best model (LR) was mapped area susceptibility was mapped to residual tree damage. The map reclassified it in two categories of susceptibility: 0-0.25 (low) (41.05%) and 0.50-0.75 (high) (58.95%) (Figure 4), because there wasn’t any pixel in two classes’ 0.25-0.50 (medium) and 0.75-1 (very high).

In the regeneration damage modelling, LR model had the best success and prediction rate compared to FR model (Figure 5, 6).

Figure 4. Residual tree susceptibility map produced

It was found that regeneration damage in stand

by LR model

density<150 has a lower frequency ratio (0), but in stand

Figure 5. Success rate curves for the regeneration

Figure 6. Prediction rate curves for the regeneration susceptibility maps produced in this study susceptibility maps produced in this study

density>200 there was a higher probability of a regeneration damage occurrence (FR: 2.34) (Table 3).

The resultant beta (β) coefficients and test statistics for each independent variable in the logistic regression equation are shown in table 4 and Eq. 4. Based on the analysis of logistic regression coefficients, the results showed that slope, forest type, stand density and extracted volume have positive effect on regeneration damage occurrence. It was also observed that mixed forest had type has the highest beta coefficient (3.21), and then, beech-maple forest type, extracted volume, stand density and slope have most effect by coefficients of 2.18, 0.4, and 0.1, respectively. In contrast, the extraction system has negative effect on regeneration damage of study area, having beta coefficient of (β)

1.96.

Finally based on the best model (LR) area susceptibility was mapped to regeneration damage. The map reclassified it in four category of susceptibility: low (60.09%), moderate (10.09%), high (8.70%) and very high (21.12%) (Figure 7).

DISCUSSION

According to the results, presence of extensive

(Eq.

ecological damages to stands is due to activities such as skidding operations and site susceptibility degrees. Slope, forest type, stand density, harvested volume, and extraction system had significant effect on modelling of stand damages based on the LR model.

Skidding perform unstable in steep slopes, therefore timbers move around that leads to residual tree and regeneration damages. Sist et al., (2003) and Naghdi et al., (2009) reported that skidding on steep train increase stand damages compared to gentle train.

Site susceptibility to damages is different as for forest types. Sowa and kulak (2008) showed variation among forest types could ascend stand damages occurrence probability from 0.5 to .97 odds ratio.

Stand density is one of the most important effective variables on stand damage during forest skidding. Timber transport is difficult with increasing in tree and regeneration density per ha, because should be kept trees and regeneration groups should be kept against harvesting operators. This condition increases the collision chance of timbers and residual stands. Sist et al., (2003) and Iskandar et al., (2006) showed that stand

-17.85 + 2.18 forest typ e (beech -maple) + 3.21 forest typ e (mixed) -1.96 extraction system (mule) + 0.40 extracted volume + 0.1 stand density + 0.08 slope

e

Υ=

-17.85 + 2.18 forest typ e (beech -maple) + 3.21 forest typ e (mixed) -1.96 extraction system (mule) + 0.40 extracted volume + 0.1 stand density + 0.08 slope

1 + e

Journal of Research in Ecology (2015) 2(1): 021-029

Table 3. Spatial relationship between regeneration damage and predictor variables by FR model

Predictor Variables | No. of pixels | Percentage | No. of | Percentage | FR | |
---|---|---|---|---|---|---|

in domain | of domain | Damage | Of Damage | |||

Slope (%) | <30 | 200 | 1.871 | 2 | 1.639 | 0.876 |

30-60 | 6575 | 61.518 | 74 | 60.656 | 0.986 | |

>60 | 3913 | 36.611 | 46 | 37.705 | 1.030 | |

Land Form | canyon | 3733 | 34.927 | 45 | 36.885 | 1.056 |

hillside | 2170 | 20.303 | 16 | 13.115 | 0.646 | |

ridge | 4785 | 44.770 | 61 | 50.000 | 1.117 | |

Roughness | even | 4504 | 42.141 | 42 | 34.426 | 0.817 |

uneven | 2971 | 27.798 | 43 | 35.246 | 1.268 | |

a | rough | 3213 | -30.062 | 37 | 30.328 | 1.009 |

c Forest Type | pure beech | 2646 | 24.757 | 14 | 11.475 | 0.464 |

( | beech-maple | 5430 | 50.805 | 63 | 51.639 | 1.016 |

mixed | 2612 | 24.439 | 45 | 36.885 | 1.509 | |

h Stand Density (ha) | <150 | 4138 | 38.716 | 0 | 0.000 | 0.000 |

150-200 | 3631 | 33.973 | 44 | 36.066 | 1.062 | |

>200 | 2919 | 27.311 | 78 | 63.934 | 2.341 | |

Extracted Volume | <10 | 3190 | 29.847 | 20 | 16.393 | 0.549 |

(m3/ha) | 10-13 | 1831 | 17.131 | 29 | 23.770 | 1.388 |

>13 | 5667 | 53.022 | 73 | 59.836 | 1.129 | |

Extraction System | mule | 6297 | 58.917 | 48 | 39.344 | 0.668 |

zetor | 4391 | 41.083 | 74 | 60.656 | 1.476 |

density has the most effect on stand damages in short and long term, spatially about selective logging method.

In the present study it is shown in relation to variables influenced by harvesting intensity such as stand damages, number of harvested trees per hectare etc.; plays a key role. According to results there have a linear relation between harvested volume and both stand damages, however Panfil and Gullison (1998) reported that the relationship between variables is unlikely such as percent ground area damaged and harvested volume are

Table 4. Significant predictor variables on regeneration damage according to LR model

Variables

DF | AIC | pr (>Chi) | |
---|---|---|---|

Slope | 1 | 203.86 | 0.00000 |

Forest Type | 2 | 162 | 0.00147 |

Stand Densi | |||

1 | 354.58 | 0.00000 | |

ty | |||

Extracted | |||

1 | 213.90 | 0.00000 | |

Volume | |||

Extraction | |||

1 | 160.72 | 0.00179 | |

System |

linear (Jackson et al., 2002). Additionally they mentioned this relationship affected by other variables of spatial forest types.

This greater damages is primarily due to the extremely large area disturbed by skidder relative to mule logging especially in residual tree damages (Jackson et al., 2002). The much larger area damaged by skidder was the resulting of greater size of transported timbers by it. Also, skidders need large spaces for relocation compared to animals therefore increased neighbor stand damages. Skidder impact on residual trees was more than regeneration compared to mule logging. This can be observed in the residual trees map (Figure 4), the area was divided into two categories low and high susceptibility. Because all of the residual tree damages created by skidder and its role was more than those of others variables. Thus residual tree susceptibility map largely constructed is similar to the extraction system map.

The critical strategy in ecological prediction modelling is the validation of predicted results so that the results can provide a meaningful interpretation with respect to future stand damages. However, the success rate curve can help to determine the resulting of stand damages as well. The susceptibility maps have been classified in the areas of existing damages, but the prediction rate just can just explain how well the model and predictor variables predict the damages.

AUC analysis was used to validate the best model (LR) for stand damages mapping. This curve is a standard methodology to evaluate the model precision (Pourghasemi et al., 2012). The AUC curve can display imagination of the trade-off between the false-negative and false-positive rates for every possible cutoff value. The AUC curve shows the false-positive rate on the X axis and the true positive rate on the Y axis. The AUC plots the accuracy of a prediction model system by describing the system's ability to expect the correct occurrence or non-occurrence of pre-defined “events”.

According to Pourghasemi et al., (2012) qualitative relationship between AUC and prediction accuracy can be classified into the following categories: 0.9–1 (excellent); 0.8–0.9 (very good); 0.7–0.8 (good); 0.6–0.7 (average); and 0.5–0.6 (poor). According to prediction rate of the best model (LR), it is showed that the model used during this study had a reasonably good precision in predicting the sand damages occurrence mapping in the mapping of study area.

The results of current study also showed that the FR model can be used as an easy tool in the evaluation of stand damages susceptibility when a enough number of data are available.

CONCLUSION

Damage to forest ecosystems consisting residual trees and regeneration may have important negative implications for future harvests in the forests of Iran. The present study suggests that via modelling and prediction of susceptibility forest sites can decrease the number and extent of logging-induced damages on residual trees and regeneration. These approaches are the part of controlled selective logging and reduced impact logging that lead to reduced stand damages. These results are the most appropriate tools for managers to manage forests more accurately.

REFERENCES Dawson CW. Abrahart RJ, and Hydrotest LM. 2006. A web-based toolbox of evaluation metrics for the standardised assessment of hydrological forecast. Environ. Model. Softw. 22(7): 1034–1052.

Iskandar H. Snook L, Toma K, Kenneth T, MacDicken G, and Kanninen M. 2006. A comparison of damage due to logging under different forms of resource access in East Kalimantan, Indonesia. For. Ecol. Manage. 237: 83–93.

Journal of Research in Ecology (2015) 3(1): 021-029

Jackson SM. Fredericksen TS, and Malcolm JR. 2002. Area disturbed and residual stand damage following logging in a Bolivian tropical forest. For. Ecol. Manage. 166: 271–283.

MacDonald RL. Chen HYH, Palik BP, and Prepas EE. 2014. Influence of harvesting on understory vegetationalong a boreal riparian-upland gradient. For. Ecol. Manage. 312: 138–147.

Marvie Mohadjer MR. 2004. Silviculture of the oriental Beech (Fagus orienalis Lipsky); experiences made in Caspian forests, North of Iran. Proceedings from the 7th International Beech Symposium, IUFRO Research Group 1.10.00, 10-20 May, Tehran, Iran. 15

17.

Muller U. 1998. Effects of refinements on the development of a neotropical rainforest after controlled logging in Suriname. [MSc thesis]. Wageningen Agricultural University, Wageningen, Netherlands.

Naghdi R. Lotfalian M, Baheri I, and Moradmand Jalali A. 2009. Damages of skidder and animal logging to forest soils and natural regeneration. Croat. J. For. Eng. 30 (2): 141–149.

Najafi A. Solgi A, and Sadeghi SH. 2009. Soil disturbance following four wheel rubber skidder logging on the steep trail in the north mountainous forest of Iran, Soil & Tillage Research. 103(1): 165–169.

Owende PMO. Lyons J, Haarlaa R, Peltola A, Spinelli R, Molano J, and Ward SM. 2002. Operations protocol for eco-efficient wood harvesting on sensitive sites. Ireland, 74 pp. http://www.ucd.ie/ foresteng/html/ecowood/op.pdf.

Panfil SN. Gullison RE. 1998. Short-term impacts of experimental timber harvest intensity on forest structure and composition in the Chimanes forest. Bolivia. For.

Ecol. Manage.102(9): 235–243.

Pourghasemi HR. Moradi HR, Mohammady M, Bednarik M, and Pradhan B. 2012. A Comparative assessment between index of entropy, logistic regression, and frequency ratio models for landslide susceptibility mapping in Iran. Natural Disasters. 30 pp.

Reeves DA. Reeves MC, Abbott AM, Page-Dumroese DS, and Coleman MD. 2012. A detrimental soil disturbance prediction model for ground-based timber harvesting. Can. J. For. Res. 42(5): 821–830.

Rockwell C. Kainer K, Marcondes N, and Baraloto

C. 2007. Ecological limitations of reduced-impact logging at the smallholder scale. For. Ecol. Manage.

238: 365–374.

Saga Ø Selas, V. 2012. Nest reuse by Goshawks after timber harvesting: Importance of distance to logging, remaining mature forest area and tree species composition. For. Ecol. Manage. 270: 66–70.

Sist P. Fimbel R, Nasi R, Sheil D, and Chevallier MH. 2003. Towards sustainable management of mixed dipterocarp forests of South East Asia: moving beyond minimum diameter cutting limits. Environmental Conservation. 30(4): 364–374.

Sowa J. Kulak D. 2008. Probability of occurrence of soil disturbance during timber harvesting. Croat. J. For. Eng 29 (1): 29–39.

Vorpahl P. Elsenbeer H, Mrker M, and Schrder S. 2012. How can statistical models help to determine driving factors of landslides? Ecol. Model. 239: 27–39.

Yilmaz I. 2009. Landslide susceptibility using frequency ratio, logistic regression, artificial neural networks and their comparison: A case study from Kat landslide (Tokat-Turkey). Computer and Geosciences. 35(6): 1125–1138.

Journal of Research in Ecology (2015) 3(1): 021-030