3.1. Detection of Negative Changes in Forest cover (Deforestation)
The classification of Landsat 5 and 8 images resulted in two maps showing the forest/non-forest areas (Fig. 4). As the classification of satellite imagery was carried out in two well-defined classes (forest and non-forest) and training samples were selected carefully, it was expected that the accuracy of the classification would be acceptable. The results also indicated high classification accuracy for the two images (Table 4).
Table 4
Results of accuracy assessment of classification of satellite images
Landsat 8-2022 | Landsat 5-1991 |
Kappa Coefficient | Overall Accuracy | Kappa Coefficient | Overall Accuracy |
0/94 | 95/61% | 0/91 | 93/14% |
The map of negative changes in forest cover (Fig. 5), indicating areas with deforestation, was generated by intersecting the two classification maps. Before dealing with the details of negative changes in forest cover, it is important to note that the studied catchments are located in the mountainous region and are relatively far from the anthropogenic interventions of urban centers situated in the plains in the eastern part of the region. As a result, these catchments can reflect the relationship between physical factors and deforestation. In these areas, human interventions are mainly in the form of agriculture and livestock breeding. Although a comparison of the forest/non-forest classification maps reveals fairly the expansion of deforestation from the plains towards the mountains, the deforestation map shows more clearly this fact. This map shows that deforestation is not only occurring along the main valleys towards the headwaters but that forest destruction is also traceable in different altitude zones and landforms, indicating a hazardous ecogeomorphic event. The rate of forest loss in the catchments is shown in Table 5. Considering that the catchments are listed in this table from north to south, it is determined that the percentage of deforestation in the northern catchments is relatively higher than in the southern ones, even though these catchments are smaller. Out of the total area of catchments, which is equal to 1993 km2, 174 km2 (8.7%) has been exposed to deforestation during 32 years. The Havigh catchment has the highest percentage of deforestation (7.7%) among the catchments. The spread of deforestation in the main valley of this catchment towards the headwaters is visible, indicating the degradation of the riparian zone. The lowest percentage of deforestation (1.8%) belongs to the Dinachal catchment.
Table 5
The rate of deforestation in Talesh catchments over 32 years (1991–2022)
Catchment | Area (km2) | Deforestation (km2) | Deforestation (%) |
Lavandvil | 37.44 | 1/4 | 3/7 |
Cheloond | 61.60 | 3/7 | 6 |
Lamir | 51.59 | 3/2 | 6/2 |
Choobar | 61.89 | 4/5 | 7/2 |
Havigh | 126.78 | 9/8 | 7/7 |
Shirabad | 85.76 | 2/6 | 3 |
Lisar | 174.86 | 5/3 | 3/1 |
Korghanrood | 528.08 | 27/2 | 5/1 |
Navrood | 264.83 | 8/7 | 3/3 |
Khalehsara | 49.22 | 1/2 | 2/4 |
Dinachal | 202.27 | 3/7 | 1/8 |
Shafarood | 348.59 | 18/6 | 5/3 |
Although riparian forests and shrubs in small and rugged catchments are more abundant than in large and less rugged catchments (Engelhardt et al. 2011), it should also be noted that small catchments are more sensitive to land degradation and soil erosion compared to large catchments and have a faster hydrological response. Important studies by Milliman et al. (1999) in the East Indies and Vanmaercke et al. (2011) in the European catchments refer to this high response of small catchments and their sensitivity to floods and landslides. Such catchments characterized by the narrow valley and bedrock dominance, are typically under the control of floods and may exhibit significant channel variations during high flows. As a result, these catchments are highly sensitive to both natural and human disturbances and require preventive land-use management practices such as grazing, roads, and recreation. Compared to small catchments, large catchments have an abundance of grasslands and meadows (Engelhardt et al. 2011) and have a slower hydrogeomorphic response due to the extensive river network and the large distance between the upstream and downstream. However, it is clear that the protective role of forests in protecting water and soil resources is greater than grasslands, and the lower forest cover of large catchments such as Korghanrood and Shafarood has made the risk of floods and erosion in these catchments serious. The issue of the treeline is important in this case. The tree line in the region is located at an altitude of 2000 to 2200 meters. As the area of the catchments increases by moving from north to south of the region, the catchment divides recede to the west and cross the tree line. From another perspective, as the area of the catchments increases, the percentage of bare land or rock outcrops in the upper reaches of the catchments increases. Therefore, it is clear that maintaining environmental balance in all catchments is essential and depends on the conservation and restoration of forests.
3.2. Relationship between geomorphologic variables and deforestation (Logistic Regression model)
The spatial logistic regression test resulted in a spatial predictive model of the probability of deforestation based on geomorphological variables (Eq. 5), which indicates the direction and intensity of the relationships between the dependent and independent variables. This can be understood by considering the b coefficients of the predictive regression model:
Logit (Deforestation) = -1.1803 - − 2.097567* alt − 0.292042*tpi − 2.819719*s − 0.000031*north − 0.000017*east + 0.000312*ls + 0.118545*sl − 1.921002*tri + 0.744707*plc − 0.170077*prc − 0.272819*ci − 0.510281*ca + 0.000751*str-dens − 0.000124*str-dist − 1.262366*twi (5)
Considering that the five variables including northness, eastness, erosion factor of slope length, river density, and distance to the river have had a negligible effect on deforestation, these variables can be removed from the above equation, so that the following final equation can be presented:
Logit (Deforestation) = -1.1803–2.097567* alt − 0.292042*tpi − 2.819719*s + 0.118545*sl − 1.921002*tri + 0.744707*plc − 0.170077*prc − 0.272819*ci − 0.510281*ca − 1.262366*twi (6)
The first geomorphological variable considered in explaining the spatiotemporal changes in vegetation cover is altitude. As elevation increases to a certain extent, precipitation usually increases, and if the maximum elevation of the region does not reach the snow line, favorable topoclimatic circumstances are generally provided for the growth and development of forests. As expected, a strong relationship is observed between altitude (Alt) and the probability of deforestation in the Talesh catchments. The negative relationship between the two variables of altitude and topographic position index (tpi) and the occurrence of deforestation indicates that deforestation has occurred at lower altitudes and in valleys (Fig. 6).
Undoubtedly, the lack of easy access and unfavorable topoclimatic conditions for human activities in highlands have contributed to less deforestation in these areas. In contrast, the accessibility of forests located in lowlands has induced the deforestation process in these areas. The main valleys provide a permanent water reserve for livestock breeding, and in addition, these landforms accelerate the adjacency of roads to the dense forest (Vanacker et al. 2003). Therefore, it is clear that anthropogenic interventions and natural resource destruction have been more concentrated in river valleys. This inverse relationship between altitude and forest loss was widely mentioned by researchers (e.g. Arekhi et al. 2013; Kumar et al. 2014; Bonilla-Bedoya et al. 2018; Pujiono et al. 2019; Bera et al. 2022; Saha et al. 2022). Furthermore, this spatial-statistical relationship indicates that the riparian zone and coastal buffers of the catchments are under threat of degradation (Fig. 7).
“This transitional zone (ecotone) plays a prominent role in the equilibrium of the entire catchment and acts as a regulator of flowing and transportation of water, sediment, and nutrients between the river and the adjacent uplands” (Patten et al. 1998). So this feature is particularly sensitive to flooding in mountainous catchments. In addition, assuming the significance of the flood forest as a complete part of the food chain and the high scarcity of the remaining flood forest, the conservation of this forest should be a main conservation strategy (Lohani et al. 2020).
Changes in elevation across a land surface are represented by the slope, and landforms are formed by the interconnection of these slopes. Slope is an important variable that controls erosion, runoff, and soil development. As the slope increases, the rate of surface and subsurface drainage and soil erosion increases, and the occurrence of mass movements such as landslides and falls is directly related to the increase in slope. Therefore, the stabilization and sustainability of sediment, water, and nutrients may be limited on steep slopes, and subsequently, the establishment and stabilization of plant communities may become difficult. Despite this fact, from the perspective of human activities, the inaccessibility of steep slopes and their unsuitable conditions for the establishment and continuation of human activities, especially agriculture, can be a cause of the preservation of forest cover against deforestation. The inverse relationship between deforestation and slope in the above equation refers to this issue and the refuge of forests in steep areas. Gonzalez-Gonzalez et al. (2021) in their analysis of the spatiotemporal changes of deforestation in Colombia stated that low slope acts as a preventive factor to deforestation, whereas high slope acts as an attractor factor to deforestation. Most researchers, such as Fox et al. 2012; Arekhi et al. 2013; Kumar et al. 2014; Bonilla-Bedoya et al. 2018; Pujiono et al. 2019; Plata-Rocha et al. 2021; Saha et al. 2022, have found this inverse relationship between deforestation and slope.
In addition to topographic factors such as altitude and slope, some variables indicate the land roughness. Among these, two factors can be mentioned: slope length (SL) and terrain ruggedness index (TRI). Slope length (SL) is a geomorphological variable that is measured along the maximum slope, and its high values are observed in large and main valleys. The slope length variable reflects the altitudinal changes along the longitudinal extent of the catchment. The positive relationship between slope length and deforestation indicates that deforestation has occurred in the downstream areas and main valleys (Fig. 8). In contrast, forests in the upstream and first-order tributaries have been less affected by deforestation. Although the volume of floods and sediments increases in the downstream part of the catchment, and this factor may be effective in the destruction of forests due to the removal of plants or the burial of neonate shrubs and seeds under sediments, this fact is less observed in the Talesh catchments. Goebel et al. (2012) believe that the occurrence of floods leads to the generation and conservation of habitats for various plant species due to increasing erosion and sediment transport. So, it can be said that the same factor of easy access and the special attractions of the downstream coastal areas have led to the tendency and concentration of human activities (agriculture, tourism, etc.) in these areas which in turn, have caused forest destruction. Meanwhile, the expansion of the river channel and the formation of a floodplain downstream have led to the extended use of water resources and suitable soil in this section of the catchment, resulting in deforestation there. In addition to the SL factor, which is to some extent a reflection of hydrogeomorphic processes across the catchments, the terrain ruggedness index (tri) is a reflection of the morphodynamic conditions of the slopes and morphogenetic strength. This index is also a rate of topographic heterogeneity (Riley et al. 1999) so the high values of this index indicate low homogeneity of landforms in a specific area. The negative relationship between TRI and deforestation indicates that uneven and topographically heterogeneous areas haven't been exposed to deforestation. Although the adaptation and balance of vegetation cover is greater in homogeneous environments, the ecological diversity of heterogeneous environments is high. A review of the literature also indicates that researchers have acknowledged and emphasized the effects of the heterogeneity of geomorphological conditions on the abundance and diversity of vegetation cover (e.g. Hoersch et al. 2002; Stallins 2006; Reinhardt et al. 2010). In addition, high values of the TRI indicate vigorous altitude gradients that provide an unfavorable environment for human activities and limit the uniform spread of human works.
Landscape curvature variables are useful measures for interpreting the significant water and sediment transport processes in a landscape. Plan curvature (PlC) represents the degree of divergence or convergence perpendicular to the flow direction, while profile curvature (PRC) represents convexity or concavity along the flow direction. The impact of these variables on spatial and temporal changes in vegetation cover (forest) occurs due to control of transport and deposition processes, as well as linear and point accumulation of materials such as water and sediment. The availability of the necessary moisture for soil-forming biophysical and biochemical processes in concave and depressed areas makes them fertile and productive, and concave areas have characteristics that make them more conducive to plant growth and net primary production (Gessler et al. 2000). The direct relationship between the PLC and the forest cover loss in the Talesh catchments indicates this matter. This relationship shows that deforestation has occurred on the convex surfaces of the slopes. This fact can be attributed to the dispersion and movement of sediments and water and the limitation of soil formation and development on convex surfaces, which constrains the stability and regeneration of forests. From an anthropogenic perspective, it appears that the existence of limited, relatively flat surfaces at the top of anticlines makes access to these areas easy, especially for livestock Breeders, leading to deforestation. Furthermore, the avoidance of humans from the shade and humidity of depressions and hollows can also be a reason for the lower probability of deforestation in these areas. The reverse relationship between the second curvature variable (PRC) could be in the completion of the former relations (PLC vs. Deforestation) indicating loss of forest cover on convex surfaces in the direction of the slope. In other words, areas with accelerated flow have been exposed to deforestation. In contrast, depressions and hollows located along the slope with decelerated flow have been less affected by deforestation.
In addition to land curvature variables, which are both indicators of landform and process, there is another variable called the convergence index (CI) that is not sensitive to absolute elevation changes and emphasizes more on the hydrogeomorphic process. This index indicates the convergence/divergence of flow, and its high values are observed on ridges and its low values are observed in river valleys. The resulting inverse relationship between this variable and the dependent variable indicates that the probability of deforestation is higher in the areas with flow convergence and valleys than in other parts of the catchment. Lohani et al. (2020) also found that deforestation is more prevalent in floodplains than upland areas. From an anthropogenic perspective, this fact refers to the attractiveness and easier access to floodplains, which was explained in relation to elevation variables. From a natural perspective, the occurrence of geomorphological disturbances such as landslides and floods can also be involved in this matter. Geomorphological processes such as floods and landslides are important factors in ecological disturbances (Rice et al. 2012). This is particularly important in large catchments such as Korghanrood and Shafarood, which have extensive floodplains (Fig. 9). Typically, the distributions of riparian tree species were limited to riverine corridors where floodplains are preserved from wearing floods (Shaw and Cooper 2008). However, the expansion of human activities such as agriculture and gardening, construction of roads, dams, villas, etc. is evident in the Talesh region from the plains (east) towards the mountains and across the catchments (west), resulting in deforestation.
Although the morphological and roughness variables described refer to the hydrologic conditions of the catchments, some variables are more indicative of the hydrological characteristics across the catchments. The variables of contributing area (CA) and topographic wetness index (TWI) are among them. CA indicates the potential of flow accumulation at a specific location. This location can be any point in the catchment that gathers upstream water flow. The inverse relationship between this variable and deforestation indicates that areas with high flow accumulation are less affected by deforestation. Conversely, areas with unfavorable hydrological conditions and scarce water resources have experienced more deforestation. This fact can be more clearly and completely traced in the relationship between the last geomorphological variable, the topographic wetness index (TWI), and deforestation. The involvement of slope and specific contributing area parameters in the calculation of the TWI makes this variable a composite variable that reflects both roughness and hydrological circumstances. Testing the relationships between TWI and the probability of deforestation in the study watersheds shows that there is an inverse relationship between them. This means that deforestation is more prevalent in areas with good drainage and low moisture than in areas with high surface and subsurface moisture. In other words, steep areas that are prone to rapid runoff and lack of moisture accumulation are more likely to experience deforestation. Areas that are representative of wetlands, where flow accumulation and sedimentation occur, naturally have more nutrients due to biogeochemical processes for the development and stabilization of forests. This allows for faster forest regeneration in these areas.
If we want to summarize the relationship between geomorphological variables and deforestation in the Talesh catchments, we can state that the effect of geomorphic form and process on deforestation is well-represented in the relationship between geomorphometry variables and the probability of deforestation. At the same time, the anthropogenic effects on changing environmental conditions through deforestation are implicit in these interactions. The results correspond to this fact: "Hydrology, geomorphology, and ecology of fluvial systems cannot be fully understood in isolation from one another" (Cadol and Wine 2017). This interweaved relationship and interdependency can be better studied in some geomorphic environments. The results indicate the dominance of deforestation in main and large valleys and the alteration of forest cover in the floodplain and riparian section. The available evidence also shows that the deforestation of these areas in recent years has caused significant damage to human settlements and facilities due to increased flooding. Although the clearing of riparian trees is sometimes associated with coastal erosion, the human effects on this event dominate the purely natural effects. In this regard, Bera et al. (2022) stated that “In high altitude zones, deforestation mainly occurs due to physical factors such as weathering, mass wasting, aeolian process, landslide, etc. whereas, in low altitude zones, deforestation mainly happens due to anthropogenic factors". This is because these areas have favorable geomorphic settings for the expansion of human activities. Humans impact on riverside ecosystems through the exploitation and management of land and water and the introduction or elimination of plant and animal species (Patten 1998). So, discovering the relationships between geomorphological characteristics and riparian vegetation can enhance our ability to spatio-temporal prediction and aware restoration actions (Engelhardt et al. 2011).
From the view of ecogeomorphic sensitivity, the importance of riparian and coastal buffers in regulating and moderating water flow and sediment yield in small mountainous catchments, such as the northern Talesh catchments that have experienced a higher percentage of deforestation compared to southern catchments, is greater. Floods in forested mountain landscapes are distinctly different from lowland floods. Floods in mountainous areas are different from floods in plains and relatively flat areas. Maximum floods in such places are short-term (hours to days) and have sharp peaks, inducing an intense flow of water foot-slopes and steep channels. Meanwhile, the rugged topography strengthens the mass movement downslope, where a steep channel provides the necessary conditions for the sudden transfer of coarse sediments and wood debris within the drainage networks. So, flood disturbance in mountainous rugged units is characterized by physical damage to rivers and riparian ecosystems (Swanson et al. 1998). In addition to the stream flow regime characteristics of such catchments, the characteristics of the riparian ecosystem itself contribute to its ecogeomorphic sensitivity. Due to the incision of valleys and the lack of lateral channel expansion, this zone is narrower in the small northern Talesh catchments. Consequently, their forest strip is also narrower than in larger catchments. Therefore, the preservation and survival of the riparian zone in these catchments is a priority in terms of flood occurrence.
3.3. Accuracy assessment of the logistic regression model
After examining the quality and quantity of the effects of independent variables on the dependent variable, it is time to evaluate the validity and performance of the logistic regression model. This evaluation is performed based on the Pseudo R2 and ROC statistics. The PR2 rate of the resulting regression model in the study area is 0.12, which indicates an acceptable fit of the regression line. Pir Bavaghar (2015) discriminated that if the values of this statistic are between 0.2 and 0.4, a good fit is obtained for the model. However, due to the high dispersion of experimental data and their pixel nature, values lower than 0.2 are expected. Arekhi et al. (2013) also obtained a PR2 < 0.2. Kumar et al. (2014) only achieved a value of 0.29 in one of the four predictive models. However, the ROC statistic is more important than the PR2 statistic and refers to the agreement between the actual deforestation map and the map obtained from the deforestation prediction model. This statistics which "signifies the predictive capability of the models for future probability of deforestation” (Saha et al. 2020), is affected by the quality and quantity of the input data. The ROC value in the Talesh catchments was 0.75, which indicates that the predictive regression model is good. This statistic was also 0.76 in the model of Arekhi et al. (2013). However, Siles 2009; Kumar et al. 2014; Pir Bavagar 2015; Pojiono et al. 2019; and Saha et al. 2020; achieved higher rates. The ROC values obtained in the studies of these researchers were 0.85, 0.87, 0.81, 0.86, and 0.87, respectively. These differences in the evaluation values of predictive models can be due to various reasons that may occur in all spatial-statistical analyses. The sampling method, both in preparing forest cover maps and in using deforestation samples, is one of the factors that affect the final results of the model. In this study, an attempt was made to correctly apply the principles of sampling in both stages with a good distribution. However, the concentration or dispersion of deforestation samples and their general spatial arrangement aren't under the authority of the researcher. The existence of forest loss areas in a scattered and point-like manner may make it somewhat difficult to explain the spatial distribution of deforestation. The reference for generating maps of independent variables and the spatial resolution of digital elevation models are other important factor in modeling changes in forest cover. The existence of redundant and inaccurate data in some places on the one hand, and the inadequacy in converting raw elevation data, is a common topic in terrain analysis. Resampling of the DEM, improvement of spatial resolution by moving windows, and applying a majority filter can lead to better fitting of regression models by reducing environmental heterogeneity and integrating erroneous pixels with accurate ones, which requires further studies. Finally, the selection of appropriate independent variables for modeling the occurrence of deforestation is effective in the predictive power of the models. In this research, although some variables such as aspect and distance to river, which were present in the initial model, had a negligible effect on deforestation and were removed from the final model, it should be noted that one of the advantages of using geomorphometry variables in modeling land cover changes is the stability of such variables, which reduces the uncertainty in modeling land cover change and increases their accuracy.