Enhancing Connectivity Index to Assess the Effects of Land Use Changes in a Mediterranean Catchment

Study Area

The study area is included in the Arba river drainage catchment. The Barués catchment (23 km²) is an ephemeral stream catchment located in the central part of the Ebro Basin in northeast Spain (Figure 1). From a geological point of view, it is situated on the distal part of the pre-Pyrenean range with characteristically S–SW low bedding between 5 and 8°. The rock outcrops include two concordant Oligomiocene lithostratigraphic units of the Uncastillo Formation composed of sandstones, claystones and siltstones (Sole et al., 1972; Teixel et al., 1992; Pardo & Arenas, 1996). The geomorphological setting is clearly conditioned by the low bedding of the strata. This sets up the path of the streams following the strata dip.

The climate is continental Mediterranean, characterised by cold winters and hot and dry summers. Rainfall events mainly occur in spring (April and May) and autumn (September and October) and a summer drought between the two humid periods. The mean annual temperature is 13·4 °C, and the mean annual rainfall is about 500 mm. Most abundant soils were classified during field surveys as Calcisols and Cambisols (FAO, 2014). The soils developed on Quaternary deposits are mainly formed by alluvial deposits and have basic pH, low soil organic carbon contents (between 0·13 and 5·65%) and the secondary accumulation of carbonates.

Multitemporal Analysis

The assessment of sediment connectivity was carried out by applying a topography-based index in two different scenarios by using two land use maps for 1957 and 2010. The first map was created by orthorectification of the 1957 American army aerial photographs by using a supervised classification in erdas after photographic enhancement. The actual map was digitised over 2010 National Plan of Aerial Orthophotography and fieldwork maps.

The streams with higher connectivity ratios were digitised over 2010 orthophotography, and the IC was created with the 2010 digital elevation model (DEM) and compared with the 1957 aerial photography drainage net to visualise whether a remarkable displacement or modification in the slopes occurred between both periods (Figure 2).

Stream displacement and topographic changes between the periods can only be measured for what it is visible in the multitemporal aerial photography due to the absence of the 1957 DEM. Figure 2 shows the near absence of displacement or modification of the secondary slope streams in a highly degraded area of the catchment. Most visible differences between both streams in Figure 2 are produced for the aerial stereoscopic photography deformation. These low grades of displacement in the secondary streams imply that most changes in the topography are below the detection limit of our DEM.

However, the main channel has variations in its morphology, being higher in the medium and lower parts where sections are deeper and surrounded by crop fields, and nearly insignificant in the upper parts of the catchment. Due to the impossibility of obtaining a 1957 DEM and after checking the absence of substantial modifications to the topography between 1957 and 2010, it was decided to use a 2010 DEM with a low grade of uncertainty.

Digital Elevation Model Refinement

Nowadays, the most popular IC from Borselli et al. (2008), the modified version by Cavalli et al. (2013) and geomorphological studies related to slopes or erosion (Kawabata & Bandibas, 2009; Gutiérrez & Lizaga, 2016; Masselink et al., 2016) are based on DEMs. Thus, the accuracy of model results is fully dependent on DEM quality and resolution: It is needed to apply and develop a topography-based index such as IC correctly. Therefore, developing a model based on lidar data was necessary to refine the existing lidar points to create a high-resolution DEM (1 × 1 m). Accordingly, we used Spanish National Geographic Institute (IGN) raw lidar data points following Montealegre et al. (2013) methodology. First, we deleted the noise of the data, deleting lidar points classified as noise by using ArcGIS. Second, we proceeded to filter our lidar data in order to generate a bare ground 1 × 1 m raster DEM by using mcc–lidar, a command-line tool for processing discrete-return lidar points in forested environments based on the multiscale curvature classification algorithm developed by Evans & Hudak (2007). The root mean square error (RMSE) was calculated for both DEMs by using a cross-validation method (Table 1).

Total Aerial Biomass

lidar data represent a tool that supports the calculation of forest inventories and allows the vegetation to be measured in three dimensions (Ruiz, 2012; Reyes & Delia, 2014). TAB was calculated for the reforestation forests (Rfs) by using Equation 1, developed by Domingo et al. (2016), and the software fusion. This equation was created and optimised in a nearby reforested area with Pinus halepensis Mill that had similar forest dasymetric characteristics as the pine reforestation in the study catchment. Equation 1 is a good complement to understand the change produced by pine reforestation.

(1)

Where: e refers to the Euler's number, A2m is the percentage of the first lidar laser return above 2 m ground height, produced by the reflectance of the treetop canopy, and P40 is the 40th percentile of lidar data. A 25 × 25 m per pixel size was selected to create a raster image comparable with the study plots accomplished by Domingo et al. (2016) to obtain such an equation.

Connectivity Index and Adaptation to Steep Slope Agriculture

In headwater fluvial catchments, the more relevant aspect of coupling is the connection between hillslope and channel, that is, hillslope–channel coupling (Harvey, 2002). Sediment connectivity was estimated by using a geomorphometric approach to simulate how connectivity changes due to different land covers. Hence, it is very important to determine how the system connectivity responds to human-induced cover changes (Harvey, 2002). For this reason, we applied the IC (Equation 1) proposed by Borselli et al. (2008) by using the C-factor from Revised Universal Soil Loss Equation as a weighting factor (W) in the index.

(2)

Where: D_up and D_dn are the upslope and downslope components, defined by:

(3)

Where: W is the average weighting factor of the upslope contributing area, S is the average slope gradient of the upslope contributing area (m/m), and A is the upslope contributing area (m²).

(4)

Where: d_i is the length of the flow path along each i cell according to the steepest downslope direction (m) and W_i and S_i are the weighting factor and the slope gradient of the i cell respectively.

Borselli et al. (2008) proposed S =  sin α + 0 · 005 including 0·005 as the minimum slope value to avoid infinite values in Equation 4. Thus, we preserved the original values of our 1 × 1 m DEM to obtain more realistic results after checking the absence of 0 slope values. Besides calculating the contributing area, we used the procedure of multiple flow D-infinity approach (Tarboton, 1997) instead of the single flow direction algorithm (O'Callaghan & Mark, 1984) used in the hydrology ArcGIS toolbox as proposed in Cavalli et al. (2013). Using D-infinity allows us to calculate the flow accumulation of converging and diverging flow directions to create a more realistic topographic index.

In addition, the IC was improved by using a terrain roughness index (RI) as a W. Cavalli & Marchi (2008) and Cavalli et al. (2013) developed and later implemented in the IC. RI was calculated as the standard deviation of the residual topography by using the mean of the 25 neighbourhood cells values as a 5 × 5 moving window. However, for our index, we created a new roughness index (SdRI; Equation 5) calculated as the standard deviation of the slope as recommended by Grohmann et al. (2011). SdRI was also applied by using a 3 × 3 moving window to accentuate steep slope terrace characteristics of the low strata bedding preserved and modified by agricultural practices in our study catchment.

(5)

Where: x_i is the pixel value and x_m is the average of the 3 × 3 moving window.

Figure 4 shows the good adjustment of the SdRI to the steep slope terraces compared with the RI, which also shows a good fit but appears to be better for larger terraces than those present in our study area.

Furthermore, in Equations 2 and 3, we also used our two land use maps (C-factor) as weighting factor to show the differences between both land uses (1957 and 2010) and how the IC changed over time. The C-factor values assigned to our land uses are 0·0011, 0·0010, 0·06, 0·2 and 0·26 for reforestation forest, Mediterranean forest, abandoned land, cultivated land and trails respectively extracted from Panagos et al. (2015). Combining both approximations, a weighting factor was applied as the link between the C-Factor and SdRI index, trying to develop a better adjustment to reality. The model was tested in the entire catchment to facilitate visualisation at a more detailed scale in an area where the four land uses occur. We selected the same area as the red square of Figure 1 to show how connectivity changed over 50 years due to land use and cover variations.

To introduce the probable variations of forest density in our model, the TAB layer created in the previous section was implemented inside the C-factor layer by using it as the inverse of TAB to produce lower C-Factor in the areas with higher densities and higher C-Factor in the areas with lower densities. Even with a 1-m resolution DEM, it was impossible to implement the same methodology for the other land covers: It was only used on reforested areas by creating a mask layer.

(6)

Where: W is the weighting factor, C is the C-factor of the different land covers, TAB is the variance of the total aerial biomass (only used over pine reforestation), and SdRI is the roughness index. These data are normalised by scaling them between 0 and 1. This normalisation is needed because the W is a dimensionless factor and should range from 0 to 1 to be weighted equally to slope in the index. For this reason, the W factor was normalised by using unity-based normalisation (Equation 7):

(7)

Where: x_i is the pixel value and min/max x_i are the minimum and maximum values respectively, of the moving window.

Digital Elevation Model

Increasing DEM quality allowed a significant improvement in the accuracy of the connectivity model. Without a good enhancement and a high-resolution digital terrain model, the possibilities of terrain error or the lack of reality adjustment increase exponentially (Li & Wong, 2010; Vaze et al., 2010), therefore increasing the error of the IC. Thus, enhanced lidar data over regular IGN filtered lidar data allowed us to create a more accurate 1 × 1 m high-resolution DEM instead of the 5 × 5 m resolution IGN DEM. Figure 3 compares the hillshade created by using IGN and mcc-filtered data, both compared after increasing resolution to 1 × 1 m. Table 1 shows the RMSE reduction in mcc DEM and also how RMSE increases with higher vegetation canopy density and height.

Even increasing the resolution of the IGN DEM with their filtered data, there was a high adjustment error regarding the vegetation, but this error was greater in scrubland and riparian vegetation than in forests. In the Figure 3 lidar profiles, it is clearly visible how the mcc filter DEM (Figure 3B) successfully removed the scrubland points situated in the upper part and in the middle part of the profile, unlike the IGN DEM (Figure 3A), which is clearly visible in the hillshade image. For this reason, the DEM optimised with mcc-filtered data was selected (Figure 4).

Land Use Distribution Maps

Numerous studies in a variety of environments have demonstrated the significant effects of the vegetation cover increase on the reduction of runoff connectivity and water erosion (Elwell & Stocking, 1976; Zuazo & Pleguezuelo, 2008; Mohammad & Adam, 2010; Sandercock & Hooke, 2011; Fox et al., 2012). The importance of land cover can be summarised in two main effects: the direct physical protection of the soil surface by the canopy and leaf cover preventing rainfall impact and soil detachment particles and the indirect improvement of the soil resistance and quality (García-Ruiz et al., 1995; Boix-Fayos et al., 1998; Dunjó et al., 2004; Navas et al., 2008).

The high variation in total runoff and consequently in transported sediment reflects the major importance of total land cover and land use type on runoff generation and soil loss; these have significant implications on soil erosion (Kosmas et al., 1997). The results obtained for the 1957 and 2010 land cover maps reveal the major variation that took place during the last five decades in the study area.

Our results showed that Mediterranean forest and grassland have been the main land uses in the catchment over the years, together accounting for more than 50% of the total catchment surface area. A decrease in cultivated land was observed between 1957 and present, when the area dedicated to agriculture decreased from 13·4 to 3·8 km² (corresponding to a decrease of ~71%; Table 2). On the other hand, forested areas increased from 9·2 km² in 1957 to 15·8 km² at present. Vegetation and land use are important factors on the catchment hydrology, as indicated by Bryan & Campbell (1986); both are key controls on the intensity and frequency of runoff and surface sheet erosion.

Figure 5A shows a classic example of steep slope agriculture with almost 70% of the slopes and stream terraces cultivated. Conversely, Figure 5B shows an important modification caused by natural revegetation and reforestation over steep slopes. It has been shown in a nearby large mountain catchment that badlands and severely eroded areas have higher connectivity values than agricultural, forest and scrubland land uses/covers (Palazón & Navas, 2014).

Steep slope terraces are common in Mediterranean mountain agroecosystems for rudimentary agriculture. The near absence of tectonics in the Barués catchment determined the low-dip strata, resulting in an easily farmable terrain. Steep slope agriculture not only increases erosion and runoff ratios on the hillslopes but also produces slope instability, fostering the probability of mass movements due to the absence of vegetation cover that protects the soil from erosion and prevents its displacement. This can be seen in Figure 6, which shows a typical rotational stream-bank failure movement. These types of landslide are usually developed during storm events on poorly cohesive materials favoured by the absence of vegetation cover, high slope and dryland crops situated on the top of the hills of the catchment. These small slide movements are developed on stream walls induced by erosion at the toe of the stream bank walls.

Quantification of riparian vegetation over time was difficult due to image resolution, but a minimum 10% increment was detected. The presence of riparian forest and the reinforcement of bank soils by herbaceous riparian vegetation significantly reduce the likelihood of erosion by mass failure, in agreement with the observations of Hubble et al. (2010). Moreover, riparian vegetation increases the apparent cohesion through root reinforcement of bank soils reducing bank migration rates, increasing bank strength and reducing bank failure frequency (Micheli & Kirchner, 2002).

Total Aerial Biomass

The Barués catchment TAB was divided into three major groups with distinctive ranges of TAB values (Figure 7). These three groups coincided with the three different reforestation forests periods (MOP-CHE, 1976; Ortigosa et al., 1990) that conditioned the greater or lesser development of trees and forest. The three groups reported distinctive ranges of TAB values; thus, higher values were found on north faces and lower ones appeared on south-facing slopes (Figure 7). Nevertheless, younger reforestation was too minor as to test this assumption with TAB values. Table 3 shows major differences among the three different pine forests (Rf 1, Rf 2 and Rf 3). The oldest (Rf 1) nearly doubled the mean value of the TAB, while Rf 2 doubled the youngest one (Rf 3). The TAB mean values of the north-sloping faces (WNW–ENE, 292·5–67·5°) in contrast to the mean values of the south-sloping faces (WSW–ESE, 112·5–247·5°) calculated for each pine reforestation showed substantial differences. The solar angular ranges, being higher on the south-facing slopes, increasing temperature and reducing the accumulation of moisture, regulate this effect. Hence, the less developed soil biological activity and the lower organic matter content probably affect tree growth (Cerdà, 1998).

The TAB equation is a good complement in the model to understand the change produced by pine reforestation, not only related to coupling and connectivity but also to the increment of the vegetation volume. This increment probably reduced greenhouse environmental effects, as Fang et al. (2001) observed in China, suggesting that carbon sequestration through forest management practices could help offset carbon dioxide emissions.

Connectivity Index

With the modification in land cover, mostly in reforested areas, the decrease of runoff might interfere, reducing sediment displacement and probably concentrating highest connectivity and erosion areas in streams that remain coupled, as can be seen in the yellow–green streams within the reforested areas in Figure 8.

The IC with different land covers as a W gives an approximation of the effect of human activity over the study area. In Figure 8, the decrease in water and sediment fluxes can easily be recognised due to new covers such as natural revegetation, abandoned fields and the great variation induced by reforestation (Table 4).

The decrease of IC values could be related to the increase in trees and vegetation cover, which minimises the kinetic energy of the raindrops preventing soil detachment (Llorens, 1997; De Luna et al., 2000). Hence, vegetation cover probably improves soil quality by favouring infiltration and preventing runoff. On the contrary, high erosion rates and low slope resistance could easily produce topless and landslides as the most frequent mass movements (Figure 6). These feed large volumes of sediment into the stream system, as topless and bank failures are found in most stream banks along the principal streams in the catchment.

Implementation of SdRI gives greater relevance to the steep slope terrain adapting IC model to Mediterranean mountain agroecosystems. This is clearly visible on the top hills and also produces little decentralisation of the hydrological network caused by the alternating flat-steep slopes produced by the combination of ancient agriculture terraces and the strata (Figure 8). In addition, the TAB layer provides a real variation inside the pine reforestation, showing the difference between biomass and also proving that vegetated soils situated on the north-facing slopes had greater biomass. Higher proportions of biomass surely result in more developed soils and probably lower sediment yield, runoff and erosion, while bare soils on the south-facing slopes might have higher runoff rates.

The implementation of different land uses (C-factor) in the model shows the relevance of reforestation areas to the development of hydrological connectivity. Reforestation seems to be homogenous but has variations in above-ground biomass density inside reforested areas due to different factors such as topography, solar radiation and organic matter among others. Figure 4C shows the variation produced for the inclusion of the C-factor and TAB layer in the W factor over the roughness index, including the variation of the vegetation cover.

Figure 9 shows the zoomed biomass variation introduced into the IC with the TAB layer. The connectivity decline is clearly visible on the north faces in relation to the other orientations. This is probably due to the distribution of soil moisture not only in the reforested areas but also in other areas because of the patchy vegetation structure (Marchamalo et al., 2015).

Comparing both IC output maps, land cover maps (Figures 5 and 8) and IC values (Table 4), the importance of land cover for preventing runoff and the benefits of landscape restoration can be seen. Strong changes in reforestation cover, land abandonment, natural revegetation and especially the reversal to Mediterranean forest probably have major effects on the loss of coupling and, as a consequence, on the decrease of discharge, as Navas et al. (2011) observed in the Yesa reservoir and probably also the streamflow reduction, as observed by many authors across the Iberian Peninsula (Morán-Tejeda et al., 2010; Lorenzo-Lacruz et al., 2012). Other authors have also reported runoff reduction as a consequence of the increase of infiltration, interception and evapotranspiration rates. Furthermore, Table 4 confirms that the highest connectivity changes are produced between the actual pine forest and the same area in 1957; this also occurs, though to a lesser extent, over the natural revegetated forest.