Soil erosion estimation for the Göta Älv river using remote sensing, GIS and the Revised Universal Soil Loss Equation (RUSLE) model

(1)

Master’s thesis

Physical Geography and Quaternary Geology, 60 Credits

Department of Physical Geography

Soil erosion estimation for the Göta Älv river using remote sensing, GIS and the Revised Universal Soil Loss

Equation (RUSLE) model

Konstantinos Sourlamtas

NKA 237

2019

(2)

Cover picture copyright: SGI, Mikael Svensson / Scandinavian photo agency

(3)

Preface

This Master’s thesis is Konstantinos Sourlamtas’s degree project in Physical Geography and Quaternary Geology at the Department of Physical Geography, Stockholm University. The Master’s thesis comprises 60 credits (two terms of full-time studies).

Supervisor has been Norris Lam at the Department of Physical Geography, Stockholm University. Examiner has been Ian Brown at the Department of Physical Geography, Stockholm University.

The author is responsible for the contents of this thesis.

Stockholm, 12 July 2019

Björn Gunnarson Vice Director of studies

(4)

IIn the memory of my Grandfather who passed away during my thesis.

(1929-2018)

(5)

1

Abstract

According to previous studies, the study area of Göta Älv river has high risk of landslides along the river banks due to the water flow. Soil erosion can affect the increase of the landslides in an area with unstable soils caused by the increase rainfall. The Swedish climate is getting more vulnerable thus there is a potential increased risk in erosion and landslides due to unpredictable rainfall intensity. This study aims to calculate soil erosion for the Göta Älv river using the Revised Universal Soil Loss Equation (RUSLE) where a comparison of data from remote sensing and meteorological and geological agencies were completed. Two research questions will be addressed, first if the different calculation of the soil erodibility (K) factor affects RUSLE result, and second how much soil erosion occurs and will potentially occur in the future. Factors including rainfall erosivity (R), soil erodibility (K), slope length and steepness (LS), land cover management (C) and conservation practices (P) were analyzed and used as inputs for the RUSLE model. Moreover, three scenarios were applied for the calculation of K factor in order to show how each one can affect the soil erosion result. The scenarios includes the K-scenario 1, 2 and 3, where the values were derived from a world soil database, a table with literature values and estimated field measurements, respectively. Also, three scenarios for R factor were applied for the periods 2000-2018, 2021-2050 and 2069-2098 (R-scenarios 1, 2 and 3) in order to show how future changes to rainfall patterns could affect soil erosion in the Göta Älv river and if it increases the risk of the landslides. The results suggest that the soil erosion varied between 0 – 0.5 t/ha for all the time periods with mean annual soil loss between 20 – 22 t/ha/yr and maximum soil loss between 2158- 5443 t/ha. The difference between the three K factor scenarios is almost 4%, which is pretty low thus, no influence on the soil erosion results. In conclusion, the different calculations of the K factor affected more the estimated maximum soil loss instead of the mean annual soil loss. The different calculations of R factor showed that more than 90%

of the total area was not affected by the soil erosion when the soil loss will not be increased considerably in the future due to the rainfall increase.

(6)

2

Acknowledgments

I would like to sincerely thank my supervisor, Norris Lam, for his valuable advices, help, and guidance throughout my master thesis. His assistance was helpful to me to fulfill and complete my study on time. I appreciate the fact that he was always there when I had a question or trouble with my thesis.

Moreover, I would like to thank the Geological Survey of Sweden for their help and guidance in the methods of the soil erodibility (K) factor and for providing the data of soil and Geotechnical Institute and especially David Bendz who gave me the first idea of my thesis.

Last but not least, I would like to thank my family and Andy for their psychological support all this time.

(7)

3

1. Introduction

Soil erosion is a global climate and environmental problem and can be potentially dangerous for the natural environment since it can ruin agriculture and human settlements and can increase the risk of the landslides. Soil erosion is a process due to (1) the extraction of soil particles from the soil surface, (2) the transport of the soil particles due to moving water or wind, (3) deposition of the soil when there is no more energy to move the particles (Renard et al., 1997; Morgan, 2005).

Three different mechanisms cause soil erosion. These are water, wind, and tillage. Water is the one that this study focuses on and plays a significant role in the erosion process.

The erosion caused by water consists of splash and sheet erosion (Figure 1). Splash erosion happens during rainfall that dislocates the soil particles from the land surface while splash erosion occurs on the surface of the area. As rain falls, the soil begins to become saturated, it cannot retain any more water, and additional rainfall will start to flow across its surface as runoff. This water layer removes soil particles along declining surfaces and various amounts of erosion will happen. During these rain events, rills are created, which are minor waterways and if the water continues to fall in the area, then those waterways become permanent and are called gullies. Continued erosion in these areas is called sheet erosion, and the amount of erosion depends on the severity of the rainfall, slope of the area, and soil (Amangabara et al., 2016; Morgan, 2005).

Figure 1: Soil erosion process, (Amangabara et al., 2016)

One of the main controlling factor is rainfall intensity where this is the amount of rain that falls within a given time. Higher rainfall intensity will lead to higher soil erosion.

However, land cover plays a vital role in the prevention of erosion with the presence of vegetation such as grasslands or forests. They safeguard the soil and make it more stable to splash erosion which will therefore reduce the potential for soil erosion. Soil erosion can only happen when there are exposed soils and often occurs on exposed agricultural lands and along river banks. Soil erosion can affect infrastructure like roads, river banks and urban land and can cause changes in land cover, for example from agriculture to barren lands where there is no more soil (Morgan, 2005).

(10)

6

In urban areas soil erosion is a significant problem due to the loss of vegetation and urban surfaces that are resistant to water gather flows of water and transports it to the exposed soil resulting in water pollution through deposition of sediment. One example comes from Morgan, (2005) for Maryland, USA where the urban area reached the agricultural land and the land started to converted in scrub and the soil erosion increased up to 7000 t/ha annual rates when the area was bared during the construstions.

Since soil erosion occurs at large spatial scale areas and, soil erosion is often difficult to measure in the field (Morgan, 2005). One alternative is to use models that characterize the different physical properties of the affected areas. For example, one well applied model is the Revised Universal Soil Loss Equation (RUSLE) (Alexakis et al., 2013;

Alkharabsheh et al., 2013; Demirci et al., 2012; Khare et al., 2016; Kouli et al., 2009;

Zhou et al., 2014) that can be used to calculate soil loss due to rainfall runoff for slopes in an area. It shows how factors such as climate, soil, topography and land use/cover management affect soil erosion caused by rainfall intensity and ground runoff (Alexakis et al., 2013; Renard et al., 1997) and can be used to estimate soil loss due to rainfall across a evenly spaced raster (e.g. DEM). RUSLE requires physical properties, such as land use/cover (LULC), topography, climate elements (precipitation), and soil characteristics (Alexakis et al., 2013). These can be field measurements or estimated from remote sensing products (Demirci et al., 2012; Kouli et al., 2009; Rwanga et al., 2017).

One study by the European Soil Data Centre (ESDAC) by Panagos et al., (2015) used remotely sensed data for the land cover management and the estimation of the vegetation density, the CORINE land cover dataset, the European Soil database, Digital Elevation Model and support practices observations and then used these data to calculate RUSLE for all of Europe at a 100 m resolution. For that study, where soil loss was calculated for the entire country of Sweden was calculated at 100 m resolution and is the only study that has been made for the country with RUSLE therefore a need to estimate soil erosion that considers climate change is needed.

Soil erosion estimation requires many factors such as land cover/use, climate and environmental change, topography, land cover management and support practices of the area. Applying soil erosion modeling can give a better understanding of this process, can be quicker with less cost (Chen et al., 2011) and also can be a useful tool for testing effects of future climate change scenarios. The RUSLE model is often applied within a GIS environment and modeling using GIS for spatial processing and in combination with remote sensing can be a robust decision-making tool that helps in the use of spatial information and influence with erosion modeling to deal with the problem of soil erosion (Alexakis et al., 2013; Kouli et al., 2009).

(11)

7

Remote sensing is cost effective since it can provide historical information without physical contact of the earth and can be used in the calculation of the parameters of the RUSLE model. The products of remote sensing can be used in the estimation of land cover/use (LULC) and also for estimated land cover management by using the Normalized Difference Vegetation Index (NDVI) (Chen et al., 2011; Demirci et., 2012;

Kouli et al., 2009; Prasannakumar et al., 2011). NDVI values range from -1 to +1 and is a normalized ratio between electromagnetic energy in the infrared and red wavelengths.

NDVI values can be related to landcover management through an empirical equation and then relating this to LULC by overlaying the calculated C factor with the LULC (Alexakis et al., 2013; de Carvalho et al., 2014; Demirci et al., 2012;Jianping et al., 2015;

Van der Knijff et al., 2000). This process can be quicker compared to field surveys that are time-consuming and costly. Data from long running satellite missions such as Landsat can be very useful in the application of RUSLE (Alexakis et al., 2013; Kouli et al., 2009;

Rwanga et al., 2017). Factor such as the support practices can be derived from remote sensing but with satellite images with higher geometric and spectral characteristics can give better information on the support practices in the area if they exist (Kouli et al., 2009).

In semi-urban areas where agriculture and urban areas are mixed, soil erosion can be a significant problem due to the exposed agricultural lands and urban surfaces that do not slow water flows and transports it to the exposed soil causing soil erosion (Renard et al., 1997; Morgan, 2005). Since majority of the world’s population lives near rivers, these increase of soil erosion can cause increased damage to infrastructure and human life (Morgan, 2005; Statens Geotekniska Institute, 2018). Climate and environmental change can cause and increase soil erosion since rainfall intensity is predicted to be increased with climate change. Soil erosion can also affect the increase of the landslides in an area with unstable soils caused by the increase rainfall (Khare et al., 2016; Statens Geotekniska Institute, 2018; Xu et al., 2008). According to Andersson-Sköld et al., (2014) the Swedish climate is getting more vulnerable thus there is increase risk in erosion and landslides due to unpredictable rainfall intensity.

In general, the impact of soil erosion is more significant on steep lands because the effects of gravity are higher and therefore increase the mobility of the soil during rainfall events. Riverbanks can easily be eroded especially during high rainfall (David & Nadia, 1998) and can cause damage to infrastructure such as bridges or urban areas located along the rivers. Therefore, predicting soil erosion would be beneficial for fighting against climate change (Andersson-Sköld et al., 2014) but since there is almost no applications of the RUSLE model in Sweden to estimate soil loss, it can be difficult to plan for soil erosion due to climate change. It is important to study the effects of climate change on soil erosion in river valley such as the Göta Älv as it is an important river that drains the Vänern lake and then flows out to the Kattegatt sea. This river valley is prone

(12)

8

to landslides with a number of them occurring each year as a result of soil erosion (Andersson-Sköld et al., 2014) and will likely be affected by climate change.

2. Aim and Scope

Therefore, the main aim of this thesis is to calculate soil erosion for the Göta Älv river using RUSLE where a comparison of data from remote sensing and meteorological and geological agencies were completed. The RUSLE model needs a number of parameters or factors in order to calculate the soil erosion. One of these factors is the soil erodibility (K) factor that represents the potential of the soil for erosion. In this study, three different methods were selected for the calculation of the K factor in order to show the differences and limitations for each method. Specifically, the K-scenarios 1, 2 and 3 estimated values from a world soil database, from table with literature values and from field measurements, respectively.

Moreover, three scenarios of rainfall were used to calculate the R factor in the RUSLE and these scenarios were derived from the Swedish Meteorological and Hydrological Institute (SMHI). Rainfall data for the first scenario represents measured rainfall from 2000-2018 while the second and third scenario are modelled rainfall by the SMHI, for the periods 2021-2050 and 2069-2098, respectively. Specifically, the R-scenarios 1, 2 and 3 are for the periods 2000-2018, 2021-2050 and 2069-2098, respectively. All the RUSLE factors were the same for all three scenarios except for the scenario 2 and 3 where the R factor was calculated for the 2021-2050 and 2069-2098 time periods. The aim of this was to see how future changes to rainfall patterns could affect soil erosion in the Göta Älv river.

Overall, the research questions of this thesis are:

1) How much soil erosion occurs and will occur in the Göta Älv river area during the three time scenarios (i.e. for the years 2000-2018, 2021-2050 and 2069- 2098)?

2) Does the different calculations methods of the K factor affects the results of the RUSLE model for soil loss estimation?

(13)

9

3. Related study of soil erosion with RUSLE in Sweden

The paper presented by Panagos et al., (2015) estimated the soil erosion using the RUSLE model for Sweden and was used to visually compare my results of the soil erosion. The results were not available as a raster but only as a map in the published article, as shown in Figure 2. The K factor was calculated based on the European Soil Database, which is a combination of field measurements, where the sample points and visualized based on the interpolation method of Cubist regression resulting in a raster of 500 m resolution. The R factor was estimated based on rainfall data from weather stations from SMHI and interpolated with 500 m resolution. The C factor was calculated based on the Normalized Difference Vegetation Index and LULC map with 100 m resolution. The LS factor was based on a SRTM DEM with 25 m resolution, and the P factor was based on protection measures from the Good Agricultural Environmental Condition (GAEC) database and LUCAS Earth Observation database with 1 km resolution. The final soil erosion model has 100 m resolution for 2010.

Figure 2: Map of soil loss rates in EU (2010) based on RUSLE, (Panagos et al., 2015)

4. Study Area

This study was carried out in the Göta Älv river, Sweden (58^o 3΄ N, 12^o 3΄ E) and is located north of Gothenburg (Figure 3). It is the largest river in Sweden (Göransson et al., 2013) that flows north to south, has 93 km length and originates from the lake of

(14)

10

Vänern and flow to the Kattegatt sea at the south (Statens Geotekniska Institute, 2018).

One major problem that affects the area is landslides, and is caused by soil erosion along the river banks. The sediment deposits along the river are made up of glacial and post- glacial deposits with thin layers of mud and sand (Göransson et al., 2013). In the Göta Älv river, quick clay is visible, so if a landslide happens, it can affect an extensive area since quick clay can destabilize the soil stability causing larger landslides (Statens Geotekniska Institute, 2018). The river flow is controlled by a dam which causes increase in erosion when the reservoir is filled and thus increases landslides along the river beach (Andersson-Sköld et al., 2014).

The land cover along the river valley is heterogeneous with agriculture, exposed soil, dense and sparse forests, and urban lands. The main urban land cover is the city of Gothenburg, the second biggest in Sweden (Göransson et al., 2013). Closest to the river, exists urban areas and most of the agriculture and grassland while forests and bare lands are located further from the river bank. The topography of the study area is made up of flat regions where the elevation is between 0 - 40m along the river bank at the north and south parts while high elevation 190m is located in the center. The average yearly rainfall for the area is 883 mm during the period 2000-2018.

The selection of the study areas was done following the study by Göransson et al., (2013) from the Swedish Geotechnical Institute (SGI). They used as study area not the catchment of the Göta Älv but only its section from the lake Vänern to Kattegatt sea and I did the same in order to estimate the soil erosion in the local scale of the Göta Älv.

Figure 3: Study area of Göta Älv river area.

(15)

11

5. Data and Methods

In this section, a description of the data (Table 1), their sources, type and how they have been used is shown. Also, the methodology behind every equation was presented and how each factor in RUSLE model was calculated.

5.1 Datasets

The Table 1 shows the data that were used in order to calculate the soil erosion with RUSLE. Each data was collected from different Swedish agencies as raw data and developed as inputs for each factor for the RUSLE equation.

Table 1: Description, sources, and usage of every data that was used .

Type Description Source

Containing rainfall data for

2000-2018 for every month.

CSV

Annual precipitation data. The unit is mm from 10 rain gauges distributed around the study area. The sum over a month based on continuous measurements for 2000 - 2018. Mean Annual precipitation

was calculated from this data. It was used for the estimation of the R factor for R-scenario 1.

SMHI (www.smhi.se/klimatdat

a/meteorologi/)

Contain the modeled Mean

Annual Precipitation (MAP) for 2021- 2050 and 2069-

2098.

Vector

Climate index database for Sweden. Vector data for the Representative Concentration Pathway (RCP) 8.5 climate model that shows the mean annual precipitation. For the time periods 2021-2050

and 2069-2098.

SMHI (www.smhi.se/klimatdat

a/meteorologi/)

DEM(Digital Elevation Data)

Raster dataset

2m resolution produced in 2012 and was used for the estimation of topography factor.

Lantmäteriet, www.zeus.slu.se

Landsat 8 Satellite images

Two images from Landsat 8 for June 2018. They used for the calculation of the LULC map for C and P factor and NDVI for C factor.

Landsat, USGS www.earthexplorer.usg

s.gov/

Raster dataset, 1:1000000 scale

Soil data containing information about sand, silt, clay, organic carbon

% from the Food and Agriculture Organization (FAO). It was used for the estimation of the K factor for K-scenario 1.

FAO, www.fao.org

Soil type map (1:25000)

Vector

Used for the estimation of K factor for K- scenario 2. The soil map was connected with 188 sample points with the specific soil type of the map, and K factor was assigned based on a table from literature.

The data were interpolated into a continuous raster. They were used for the estimation of the K factor for K-scenario 2.

SGU, www.sgu.se

Soil samples CSV

188 soil data contain information only for the type of the soil. They were used for the estimation of the K factor for K-scenario 2.

20 soil data contain information about particle size, sample depth, and chemical and organic information. They were used for the

estimation of the K factor for K-scenario 3.

SGU, www.sgu.se

Landslides, Points

Used for the identification of the landslides n the study area and the connection of the maximum soil loss pixels with the steep areas and

the landslides.

SGI, www.swedgeo.se

(16)

12

5.2 Methodology

In this section of the thesis, each RUSLE parameter that was determined from GIS and Remote sensing data will be quantified and described. Each factor and the RUSLE model were implemented in ArcGIS desktop 10.5. The C and P factor calculated from the 2018 Landsat 8 images was used as input for the calculation of soil erosion for all three R- scenarios (i.e. 2000-2018, 2021-2050 and 2069-2098).

5.3 Revised Universal Soil Loss Equation (RUSLE)

The RUSLE model can be used to estimate the spatial distribution of soil erosion. It is an easily applicable model that uses few data and less time and alongside with the GIS is the most applicable empirical soil erosion model worldwide (Demirci et al., 2012). ArcGIS software was used to implement the RUSLE factors, and the equation by Renard et al., (1994):

ۯ ൌ ܀ כ ۹ כ ۺ܁ כ ۱ כ ۾

Equation (1)

Where A is the mean annual soil loss per unit area (ton/ha/year), R is the rainfall/runoff erosivity factor in mm, K is the soil erodibility factor in (ton/ha/ hr/ha/MJ/mm), LS is the slope length steepness factor without dimensions, C is the land cover management factor without dimensions; and P is the support practice factor without dimensions.

5.4 Rainfall erosivity factor (R)

The rainfall erosivity factor (R) describes the rain effect as well as the amount and rate of rainfall. The R factor is defined by the maximum 30 minutes intensity I₃₀ multiplied by the rain energy and the input rainfall data should be for a period of 20 years (Renard et al., 1997). According to the Equation(2) suggested by Lee et al. (2006), the erosivity (R) for the annual precipitation (mm/year) can be calculated from:

ࡾ ൌ ૜ૡǤ ૞ ൅ ૙Ǥ ૜૞ כ ࡹ࡭ࡼEquation(2)

Where the annual rainfall erosivity (R) is calculated in mm and the Mean Annual Precipitation (MAP) was estimated from the meteorological data by the SMHI for each year. Ten rainfall stations (Figure 4) were used to estimate the precipitation data within the study area. Then their Mean Annual Precipitation was calculated for the time period of 2000-2018 (Table 2). This period was chosen since rainfall data were not available for all the stations before the year of 2000. Also, it has to be clarified that the MAP of the future periods, 2021-2050, and 2069-2098 were already modeled by SMHI using the climate model RCP8.5 and these were given as monthly averages for each entire period.

These were then used for the calculation of the future R factors, for R-scenarios 2 and 3.

It has to be clarified that the MAP was calculated for each station, for all the months of

(17)

13

the R-scenario 1 since the MAP for the future time periods had already been modelled by SMHI. Specifically, the R-scenarios 1, 2 and 3 are the dates 2000-2018, 2021-2050 and 2069-2098 respectively. The interpolation was done before the calculation of Equation(2) and the method that was selected for the visualization of the MAP for each station was the Inverse Distance Weighted (IDW), since this method takes into account that if the closer the point is to the center of the cell the more is affected and based on the other studies (Demirci et al., 2012; Dubber et al., 2008; Krishna Bahadur, 2009;

Prasannakumar et al., 2012; Van der Knijff et al., 2000). Lastly, the raster calculator was used for the calculations.

Table 2: Mean Annual Precipitation in mm for each period of study

Station Latitude Longitude MAP(mm)

2000-2018 2021-2050 2069-2098

Goteborg A 57715 11992 933.48 1016.12 1140.74

Alingsas D 57889 12552 906.24 958.53 1076.56

Uddevalla D 57609 12067 1029.02 1017.64 1129.66

Gendalen 58370 11936 1049.52 853.06 937.26

Trokorna 58158 12647 856.32 801.55 882.24

Vanersborg 58268 12655 794.30 875.48 973.60

Kroppefjall-Granan A 58355 12361 847.44 936.81 1032.89

Kallered D 58606 12199 1052.73 1120.12 1253.19

Lidkoping 58660 13150 677.18 693.53 768.51

Traneberg 58483 13147 683.94 703.63 770.92

Figure 4: Spatial distributions of meteorological stations and soil samples in the study area

(18)

14

5.5 Soil erodibility factor (K)

Soil erodibility factor (K) is the sensitiveness of soil to erosion. It is the annual soil loss per unit of R for a standard condition of the soil that has tilled-up and downslope and without conservation practice and on the slope of 9% and 22 m length (Morgan, 2005).

Generally, clay is resistant to disengagement, sandy soils also have low values due to high filtration, and the soil is difficult to move. Silt loam soils have moderate to high values since this soil type can easily be transported. Silt soils have the highest K values since this soil type can be easily eroded with high runoff rates (Ganasri et al., 2015).

The most accurate way to estimate the K factor is by field measurements and for more than one year. However, direct measurements on the field for the K factor are not efficient since it is an expensive, time-consuming process and spatially variable (Wischmeier et al., 1978). There are many equations for the calculation of the K factor (Wischmeier et al., 1978;Renard et al. 1997;David 1998;El-Swaify et al., 1976).

However, all of them need detailed information of the soil data such as particle size, organic matter, soil structure code permeability profile, soil pH or fraction of the soil type (Benavidez et al., 2018). Williams and Renard (1983) proposed an equation which needs only soil organic carbon and soil particle as parameters, and has been shown to be very helpful for areas with no data about soil characteristics. The equation by Williams and Renard (1983) is:

K ൌ ࢌࢉ࢙ࢇ࢔ࢊ כ ࢌࢉ࢒ െ ࡿ࢏ כ ࢌ࢕࢘ࢍࢉ כ ࢌࢎ࢏࢙ࢇ࢔ࢊ ۳ܙܝ܉ܜܑܗܖ(3) Where,

ࢌࢉ࢙ࢇ࢔ࢊ= 0.2 + 0.3 * exp (-0.256 * Sa * (1 – Si/100)), where Sa is sand (%) and Si is Silt (%). It is the factor that gives low soil erodibility factors for soils with high rough sand substances and high values for soils with little sand,

ࢌࢉ࢒ െ ࡿ࢏ = (Si / Cl + Si )^0.3, where Si is silt(%) and Cl is clay(%). It is the factor that gives low soil erodibility factors for soils with high clay to slit ratios,

ࢌ࢕࢘ࢍࢉ= (1 – 0.0256* orgC / orgC + exp(3.72 – 2.95* orgC)), where orgC is the organic carbon(%). It is the factor that reduces soil erodibility for soils with high organic carbon content,

ࢌࢎ࢏࢙ࢇ࢔ࢊ = (1 – 0.7*SN / SN + exp( -5.51+ 22.9SN)), where SN = 1 – (Sand/ 100. It is the factor that reduces soil erodibility for soils with super high sand contents

(19)

15

5.5.1 K-Scenario 1 – Minimum availability of data

This scenario represents cases where there is no available data for the local study area and therefore relies on using the world soil database (Figure 5) from the Food and Agricultural Organization (FAO) which includes soil properties for Sweden and it is freely available (FAO et al., 2012). This scenario was based on similar studies (Breiby, 2006; Xu et al., 2008; Yang et al., 2003; Zhou et al., 2014; Kouli et al., 2009).

For the K- scenario 1, ۳ܙܝ܉ܜܑܗܖ(4) was applied to calculate the K factor using the data from the FAO database for the Göta Älv river. All the calculations were made in excel and then connected with the raster dataset from FAO, and the K factor visualized in ArcGIS.

Figure 5: Soil database from FAO

(20)

16 5.5.2 K-Scenario 2- Table based approach

The motivation behind this scenario is when there is some data, such as soil type map and soil samples but without further information like the organic content, available for the study area. This scenario was based on similar studies (Alexakis et al., 2013;

Alkharabsheh et al., 2013; Benavidez et al., 2018; Chen et al., 2011; Emeribeole &

Iheaturu, 2015; Khare et al., 2016).

The K factors used in the K-scenario 2 was determined from table 2 by Robert P. Stone &

Don Hilborn, (2012). In order to assign the K factor values, first identification of the each soil type had to be done by using the soil map and the soil samples from the SGU. Then the soil types were available for all the 188 soil samples collected from SGU and the K factor was assigned to every one based on the Table 3.

Table 3: Soil erodibility, K factor for each soil type in the study area, (Stone et al., 2012). Highlighted are the identified soil types in the study area.

Soil type K factor

Clay 0.22

Clay loam 0.30

Coarse sandy loam 0.07

Fine Sand 0.08

Fine sandy loam 0.18

Heavy clay 0.17

Loam 0.30

Loamy fine sand 0.11

Loamy Sand 0.04

Loamy very fine sand 0.39

Sand 0.02

Sandy clay loam 0.20

Sandy Loam 0.13

Silt Loam 0.38

Silty clay 0.26

Silty clay loam 0.32 Very fine sand 0.43 Very fine sandy loam 0.35

Ten of the soil types from Table 3 were found in the study area and all the assignments were done in ArcGIS. Also, in order to create a continuous raster dataset as input for the RUSLE (Equation 1), interpolation was applied and IDW as a method for the 188 points (Figure 4). IDW selected as a method to create a continuous surface based on other studies (Dubber et al., 2008; Krishna Bahadur, 2009; Prasannakumar et al., 2011; Van der Knijff et al., 2000). IDW provides more weight to the close points than the distanced ones.

(21)

17

5.5.3 K-Scenario 3 - Maximum availability of data

The motivation behind this scenario is when there is much information available for soil data in the study area from field measurements by a geological agency. This scenario was based on similar studies (de Carvalho et al., 2014; Dubber et al., 2008; Kalambukattu &

Kumar, 2017; Koutalakis et al., 2013; Yang et al., 2003; Demirci et al., 2012; Ganasri et al., 2015)

In this scenario, the calculation of the K factor was based on the availability of the soil content for 20 point samples collected in the field and laboratories (Table 4). These were acquired by SGU and contained information about sand, silt, clay in percentage and organic material and organic carbon for each one. These data are often not available for a study area as it is labor intensive to collect. The K factor was calculated based on the Equation 1, since all the required information to run the equation were available. After that, the interpolation method that used was the IDW and the 20 points were used (Figure 4) to create a continuous raster in order to use it as input for the RUSLE equation.

Table 4: The calculated K factor for the 20 soil samples where sand, silt, clay organic content and carbon % were already available from SGU. The parameters fcsand, fcl-si, forgc and fhisand were calculated with Equation (3).

Soil type Sand

(%) Silt (%)

Clay (%)

Oranic Content

(%)

Organic carbon

(%) fcsand fcl-si forgc fhisand K factor

clay loam 2 60 38 1.04 1.80 0.44 0.86 0.99 0.99 0.38

clay loam 7 61 32 0.52 0.90 0.34 0.88 0.99 0.99 0.30

fine clay 1 36 63 0.81 1.40 0.45 0.73 0.99 0.99 0.33

fine clay 1 29 70 0.98 1.70 0.45 0.69 0.99 0.99 0.30

fine clay 3 34 60 0.69 1.20 0.38 0.73 0.99 0.99 0.27

clay loam 0 59 41 1.16 2.00 0.50 0.85 0.98 0.99 0.42

clay loam 11 64 25 1.51 2.60 0.30 0.90 0.98 0.99 0.27

clay loam 1 41 58 1.22 2.10 0.45 0.76 0.98 0.99 0.34

clay loam 7 70 23 1.33 2.30 0.37 0.91 0.98 0.99 0.33

clay loam 2 39 59 1.45 2.50 0.41 0.75 0.98 0.99 0.31

clay loam 1 31 68 2.15 3.70 0.45 0.70 0.97 0.99 0.31

clay loam 0 67 33 1.45 2.50 0.50 0.88 0.98 0.99 0.43

fine clay 0 48 52 0.52 0.90 0.50 0.80 0.99 0.99 0.40

fine clay 9 45 46 0.69 1.20 0.28 0.80 0.99 0.99 0.22

fine clay 4 52 44 0.11 0.20 0.38 0.83 0.99 0.99 0.31

clay 10 74 16 1.80 3.10 0.35 0.94 0.97 0.99 0.32

clay loam 16 43 41 1.33 2.30 0.22 0.81 0.98 0.99 0.18

clay loam 2 40 58 3,.66 6.30 0.42 0.76 0.97 0.99 0.31

fine clay 2 55 43 0.69 1.20 0.43 0.84 0.99 0.99 0.36

clay loam 6 50 44 2.61 4.50 0.33 0.82 0.97 0.99 0.27

(22)

18

5.6 Slope Length and Steepness Factors (LS)

The slope length and slope steepness factor (LS) depends on the slope percentage and length of the slope and is determined as the ratio of soil loss under a certain situation at an area with slope steepness of 9% and slope length of 22.13 m. The steeper and longer the slope, the higher the soil loss. Thus the soil loss increases as the slope length increases and the LS values that are close to 0 shows no steep and longer slopes in the area so no high risk in soil erosion. The effect of gravity increases where larger slopes can mean higher erosion (Demirci et al., 2012; Morgan, 2005). The fundamental factor in LS estimation is the topography of the area since it is the dominant factor in the sediment runoff (Alexakis et al., 2013). LS factor was estimated from a Digital Elevation Model (DEM) with 2 m resolution. Moore and Burch, (1986) developed the following equation for the estimation of the LS factor:

ࡸࡿ ൌ ሺࡲ࢒࢕࢝࡭ࢉࢉ࢛࢓࢛࢒ࢇ࢚࢏࢕࢔ כ ࡯ࢋ࢒࢒࢙࢏ࢠࢋȀ૛૛Ǥ ૚૜ሻ^0.4 כ ሺ࢙࢏࢔ࢨȀ૙Ǥ ૙ૡૢ૟ሻ^1.3Equation(5) Where:

Flow Accumulation was calculated with the use of the DEM and is a raster that shows the cells that flow into each downslope cell. Cell size is the resolution of the elevation dataset while the sinΘ is the slope of the raster cell in radians.

5.7 Cover Management Factor (C)

The cover management factor C shows the impact of soil operations, plants, crop coherence, yielding, and soil cover and below surface biomass on soil erosion (Emeribeole et al., 2015). Generally, it shows the cropping and management operations in provincial administration (Kouli et al., 2009). The vegetation cover of the study area protects the soil from erosion caused by rainfall intensity and thus soil erosion can be reduced with the appropriate land cover management of the vegetation (Demirci et al., 2012; Alexakis et al., 2013; Kouli et al., 2009).

The C factor is related to the LULC according to Jianping et al., (2015) and Van der Knijff et al., (2000) and in order to estimate the C factor, NDVI should be first estimated.

Using the empirical Equation 5 of Jianping et al., (2015) the C factor was derived where a correlation pixel by pixel was made. Moreover, the LULC map and NDVI were both estimated from Landsat 8 satellite images taken on June 2018 at a spatial resolution of 30 m.

࡯ ൌ ૙Ǥ ૛૛ૠ࢞ࢋ࢞࢖ሺെૠǤ ૜૜ૠ࢞ࡺࡰࢂࡵሻEquation (5) Where, the NDVI with the equation by (Tucker, 1979):

ࡺࡰࢂࡵ ൌ ࡺࡵࡾ െ ࡾࢋࢊȀࡺࡵࡾ ൅ ࡾࢋࢊ Equation (6)

(23)

19

Where NIR refers to the spectral reflectance acquired at the near-infrared band and Red in the red band.

The LULC map for 2018 was created with a maximum likelihood supervised classification of seven LULC types (agriculture, grassland, dense and sparse forest, bare land, urban and water). The motivation of choosing these 7 classes comes from the literature, where Emeribeole et al., (2015) and Lee et al., (2006) proposed these landcover types for the estimation of the C and the P factor. Finally, C factor values were assigned to each of the LULC classes by overlaying analysis between the LULC and C factor according other studies (Breiby, 2006; Emeribeole et al., 2015; Ganasri et al., 2015).

5.8 Support Practice Factor (P)

The Support practice factor (P) refers to the ratio of soil loss as a response to upslope and downslope cultivation that can have positive effects to soil loss by decreasing and minimizing the quantity and proportion of water runoff to reduce the erosion (Emeribeole et al., 2015). The values of the factor are between 0, and 1 and values close to 0 shows good support practice in addition to values close to 1, which show no impact on the results. Those values are determined based on some certain practices like the contouring, strip-cropping and terracing of the agriculture (Ganasri et al., 2015).

As for the Göta Älv river, according to the SGI report (Swedish Geotechnical Institute, 2012) during 1960 – 1970 support practices were applied where shore erosion was visible these are still existing today (Figure 6). Specifically, from the town of Trollhättan to the town of Älvängen erosion, protection is visible on both sides of the river except along some parts. However, south of the town of Älvängen, the support practices are not visible except on the eastern banks between the urban areas of Bohus and Lärje and along the city of Göteborg. The support practice is mainly revetments of blasted rocks but none of the other support practices such as contouring, terracing, and stripping are found in the study area (Swedish Geotechnical Institute, 2012).

Figure 6: Erosion protection measures south of Trollhättan, (Swedish Geotechnical Institute, 2012)

(24)

20

For this thesis, the P factor was assigned to LULC map created in the previous section and the P factor values were based on values from Yang et al., (2003) and Zhou et al., (2014) as shown in Table 5.

Table 5: P factor values of each LULC type of the study area, (Yang et al., 2003; Zhou et al., 2014)

LULC P factor

Urban area 1.0

Bare land 1.0

Sparse Forest 1.0

Dense Forest 1.0

Agriculture 0.5

Grassland 0.8

Water 1.0

5.9 Accuracy assessment of the LULC classification

For this study, supervised classification was carried out on two Landsat 8 images with ENVI, where the user creates spectral signatures of known categories, like agriculture, grassland, bare land, dense and sparse forest, urban land and waterbodies and then the software specifies every pixel in the satellite image to the LULC type to which spectral signature is most similar. Each Area of Interest (AOI) was selected from the Landsat satellite images and Google Earth.

Overall, three steps that were applied: (1) the on-screen digitization of the AOIs for each LULC class. This visual selection was based on easily identified areas in both satellite images and Google Earth. A total of 362 AOIs were digitized into polygons where 33 AOIs for agriculture and bare land, 98 for the dense forest, 46 for grasslands, 22 for the sparse forest, 62 for urban areas and 51 for waterbodies. (2) Spectral signatures were exported, after the digitization of the AOIs. (3) The process of the supervised classification was applied, where the AOIs were used for the representation of each class.

After the classification process, accuracy assessment was performed to evaluate the supervised classification. This was done by an error matrix where total number of 140 point was generated on the classified image while a total number of 96 validation points across the study area were made from Google Earth that was used as a reference source to classify the selected points.

Table 6 shows the calculated error matrix for the LULC 2018 classification. The columns corresponds to the ground truth, in which classes the pixels in the validation fit and the rows depict which classes the pixels have been specified. The diagonal shows the pixels that have been classified correctly. The Kappa value is a statistic value that compares the observed accuracy with the expected one. In general values = 1 shows perfection among the categories, where values > 0.75 shows very good connection and values between 0.40

(25)

21

– 0.75 a good connection. However, kappa value <= 0.40 shows poor connection among the classes (Khare et al., 2016; Rwanga & Ndambuki, 2017).

Table 6: Error matrix for LULC classification.

LULC Water Grassland Agriculture

Sparse forest

Dense forest

Bare

land Urban Total U_Accuracy Kappa

Water 20 0 0 0 0 0 0 20 1 0

Grassland 0 9 12 0 1 0 1 23 0 0

Agriculture 0 1 26 0 0 0 1 28 1 0

Sparse

forest 0 0 0 3 0 2 0 5 1 0

Dense

forest 0 0 0 6 20 0 0 26 1 0

Barren land 0 0 3 1 1 6 3 14 0 0

Urban 0 0 0 0 0 2 19 21 1 0

Total 20 10 41 10 22 10 24 137 0 0

P_Accuracy 1 1 1 0 1 1 1 0 1 0

Kappa 0 0 0 0 0 0 0 0 0 0.703

6.

Results

After calculating the above factors, R, K, LS, C, and P, Equation 1 was applied to estimate soil erosion for six scenarios. Specifically, the soil erosion was calculated using three different K factors (i.e. K-scenario 1 uses soil information from FAO soil database, K-scenario 2 uses table with K values derived from literature and K-scenario 3 uses extensive soil information from soil samples from SGU) and for the different time periods (i.e. R-scenario 1 is for years 2000-2018, R-scenario 2 is for years 2021-2050 and R-scenario 3 is for the years 2069-2098). Several maps and tables containing the calculated factors and the total soil loss calculated for every scenario are presented.

(26)

22

6.1 Rainfall erosivity factor (R)

The R factor was calculated for R-scenario 1, R-scenario 2 and R-scenario 3, as shown in the Figure 7.

Figure 7: Calculated R factor for R-scenario 1 (2000-2018), R-scenario 2 (2021-2050) and R-scenario 3 (2069-2098).

It is noticeable for all the three R-scenarios that the R-values are higher at the south around Göteborg and Kåhög and lower at the north and the center of the study area around Trollhättan and Lödöse, respectively. This difference occurred due to the difference in the elevation where the R factor is higher at higher elevation areas. Also, rainfall values are estimated to get higher in the futures, thus the intensity of the rain in the study area can be assumed to also get higher.

Specifically, for the R-scenario 1 the minimum, maximum and mean R value is 333.91, 393.73 and 357.40 mm, respectively. For the R-scenario 2 the minimum, maximum and mean R value is 339.09, 424.88 and 370.76 mm with increase from the R-scenario 1 of 1.55, 7.91 and 3.73%, respectively. Lastly, for R-scenario 3 the minimum, maximum and mean R value is 371.28, 470.93 and 408.92 mm with increase from R-scenario 2 of 9.49, 10.83 and 10.29%, respectively. The increase from R-scenario 1 to R-scenario 3 is 11.19, 19.60 and 14.41% in minimum, maximum and mean R value, respectively.

(27)

23

Figure 8: Rainfall plot of the gauges stations in the study area showing the changes between the time periods.

Looking at the Figure 8, we can notice that the south station of Kallerad D has the highest amount of annual rainfall for all the time periods and increasing substantially from 2000-2018 to 2069-2098 with over 1200 mm of annual rainfall. The station of Goterborg A, Alingas D, Vanersborg, Kroppefjall-Granan A, Lidkoping and Traneberg follow the same trend, which shows increase of rainfall during the two periods of the future, 2021-2050 and 2069-2098. On the other hand, the stations of Uddevalla D, Gendalen, Trokorna show a different trend, where the period of 2000-2018 shows higher rainfall than the future periods.

0 200 400 600 800 1000 1200 1400

Mean annual Precipitation (mm)

2000-2018 2021-2050 2069-2098

(28)

24

Figure 9: Interpolation of MAP from each gauge station and every time period.

Moreover, looking at the interpolation (Figure 9) of M.A.P from the rainfall stations, it is noticeable that the stations of Uddevalla D, Goterborg A and Kallerad D have the higher rainfall values for all the time periods. The lowest values are from the stations of Traneberg and Lidkoping.

(29)

25

6.2 Soil erodibility factor (K)

The K factor was calculated for K-scenario 1, K-scenario 2 and K-scenario 3, as shown in the Figure 10.

Figure 10: Calculated K factor for the K-scenario 1, K-scenario 2 and K-scenario 3.

The results show that the calculated K factor is different for every scenario. For the K- scenario 1, the result shows only two different K factor in the study area, that have been estimated from two soil types, vertical cambisols, and orthic podzols, with K values 0.1456 to 0.1824, respectively. For the K-scenario 2, the calculated K factor shows a range of K factor values between 0.08 – 0.38 where these have been estimated by 188 soil types. Higher values are in soils that have a small particle size such as clay loam and silt with values 0.3 and 0.38, respectively and lower values for soils with larger particle size like sand and fine clay with values 0.08 and 0.22, respectively. For the K-scenario 3, the results are closer to the second scenario since both have estimated the K factor based on soil samples. However, the differences on the values are higher since the range is between 0.1843 and 0.4351 where higher values are seen in the soil types of fine clay and lower in clay loam.

(30)

26

6.3 Slope length and slope steepness factor (LS)

The values of the LS factor were estimated from the DEM, which is assumed that it represents the past, present and future conditions of the Göta Älv River and was also used for the R-scenario 2 and R-scenario 3. The LS factor ranges from 0 to 91.475, and the high values, above 25, are located in the larger stream network and steep slopes along the river bank. On the other hand, low values below 5 are the majority and are found in flat areas with no streams near them.

6.4 Cover management factor (C)

The cover management factor C for the study area of the Göta Älv river was calculated for the year 2018 (Table 7) and was used as an input for the calculation of RUSLE for all the scenarios, R-scenario 1, R-scenario 2 and R-scenario 3.

Table 7: C factor and NDVI values for each LULC type

LULC C factor 2018 NDVI 2018

Agriculture 0.0254 0.6297

Bare land 0.0277 0.5318

Sparse forest 0.0100 0.7928

Dense forest 0.0077 0.8249

Grassland 0.0467 0.7883

Urban 0.3409 0.3452

Water 1 -0.3924

According to the Table 7, high values are found in water and urban regions since these areas have little or no vegetation and low values of NDVI. Low values are found on agriculture, forest, and grass, since these areas contain dense or medium quantity of vegetation and high values of NDVI. It is clear that low values dominate the area since the most of the study area is covered by forests with dense vegetation.

6.5 Support practice - P factor

The support practice factor of the Göta Älv river for 2018 was calculated based on the LULC type of the study area and the P values taken from previous literature (Table 5).

The P factor dataset of 2018 was used for all the scenarios. Specifically, the agricultural class was given the value 0.5, grassland 0.8, and all the other classes 1 according to the Table 5.

(31)

27

6.6 Soil Erosion Estimation using the K-scenarios

The calculation of the soil erodibility factor was done using three scenarios. The 3 different K-scenarios showed higher differences in the estimated maximum soil loss rather than in the mean annual soil loss which it varies between 20-22 t/ha/yr for all the methods. These differences ranged up to 1164 t/ha more for K-scenario 1 compared to K- scenario 2, up to 3000 t/ha more for K-scenario 1 compared to K-scenario 3 and up to 1802 t/ha more for K-scenario 2 than K-scenario 3. Moreover, according to the results of the soil erosion for different time periods with the different K-scenarios (Table 8), the majority of the soil loss occurred between 0 – 0.5 t/ha.

Based on the results of soil erosion using the K-scenario 1, for the period 2000-2018, this is a summation of the eighteen years with the values range from 0 to 2158.32 t/ha with a mean of 20.59 t/ha/yr and standard deviation of 16.36 t/ha/yr. For the future periods, which are a summation of the 29 years, for the period 2021-2050, the values range from 0 to 2244.28 t/ha with a mean of 20.74 t/ha/yr and standard deviation of 16.68 t/ha/yr. For the period 2069-2098, the values range from 0 to 2477.21 t/ha with a mean of 20.88 t/ha/yr and a standard deviation of 17.27 t/ha/yr.

Based on the results of soil erosion using the K-scenario 2, for the period 2000-2018 the values range from 0 to 3175 t/ha with a mean of 20.98 t/ha/yr and a standard deviation of 17.80 t/ha/yr. For the period 2021-2050, the values range from 0 to 3299.53 t/ha with a mean of 21.10 t/ha/yr and a standard deviation of 18.10 t/ha/yr. For the period 2069- 2098, the values range from 0 to 3641.58 t/ha with a mean of 21.32 t/ha/yr and a standard deviation of 18.80 t/ha/yr.

Based on the results of soil erosion using the K-scenario 3, for the period 2000 - 2018, the values range from 0 to 4713 t/ha with a mean of 21.94 t/ha/yr and a standard deviation of 20.94 t/ha/yr. For the period 2021 - 2050, the values range from 0 to 4927.47 t/ha with a mean of 22.08 t/ha/yr and a standard deviation of 21.33 t/ha/yr. For the period 2069 - 2098, the values range from 0 to 5443 t/ha with a mean of 22.40 t/ha/yr and a standard deviation of 22.26 t/ha/yr.

(32)

28

Table 8: Calculated soil erosion of each K-scenario for each time period, where K-sc1, K-sc2 and K-sc3 are K-scenario 1, K-scenario 2 and K-scenario 3, respectively.

The high values are found mainly on the riverbank and in high slope steepness areas (Figure 11). The erosion rate class of 0 – 0.5 t/ha is the prevailing one in the study area for all the periods of study The calculated difference between the three K-scenarios in the soil erosion class 0 – 0.5 t/ha is only about 4% (Table 8).

Landslides connected with the high values pixels in the area (Figure 12), thus maximum soil loss in the area has occurred near the position of these high value pixels. At the Figure 12, some of those pixels are visible on the riverbank and at the edges of the river where the slope is steeper (i.e. 45%, 65% and 46% slope values, respectively) and also landslides have been occurred in the past. For the first image, the landslide occurred in year 2000 and the soil is sand. For the second image the landslide occurred in 2000 and the soil is clay. For the third image the landslide occurred in 2000 and the soil is clay.

K-Sc1 K-Sc2 K-Sc3 Dif(K-sc1-3) K-Sc1 K-Sc2 K-Sc3 Dif(K-sc1-3) K-Sc1 K-Sc2 K-Sc3 Dif(K-sc1-3) 0 - 0,5 94.63 92.86 90.87 3.76 94.42 92.53 90.63 3.79 93.92 91.96 90.13 3.79

0,5 - 1 2.76 3.38 3.82 1.06 2.81 3.50 3.79 0.98 2.96 3.64 3.77 0.89

1 - 2 1.50 2.11 2.77 1.27 1.59 2.21 2.87 1.28 1.77 2.40 3.06 1.29

2 - 5 0.75 1.12 1.73 0.99 0.80 1.20 1.84 1.04 0.91 1.37 2.06 1.16

5 - 10 0.21 0.30 0.46 0.25 0.22 0.32 0.49 0.27 0.25 0.36 0.56 0.31

10 - 20 0.09 0.13 0.20 0.11 0.10 0.14 0.22 0.12 0.11 0.16 0.24 0.13

20 - 50 0.05 0.07 0.10 0.06 0.05 0.07 0.11 0.06 0.06 0.08 0.12 0.07

>50 0.02 0.03 0.04 0.03 0.02 0.03 0.05 0.03 0.02 0.03 0.05 0.03

Soil Loss (t/ha/y)

Total area(%)

2000-2018 2021-2050 2069-2098

(33)

29

Figure 11: Example of high soil loss values along the riverbank and streams.

Figure 11 shows the high values where the calculated mean yearly soil classes are between 2 – 5, 5 – 10, 10 – 20, 20 – 50 and >50 t/ha. For all the three scenarios, those high values occurred along the riverbank, in the streams and steep slopes. This example represents all scenarios.

Figure 12: Landslides and pixels with extreme high values in the study area.

(34)

30

7. Discussion

7.1 How the K-scenarios affect the RUSLE result and their limitations

The calculation of the soil erodibility factor was done using three scenarios but the difference between the three K factor scenarios is almost 4%, which is pretty low. This shows that the very different K factor methods do not have much influence on the soil erosion results, maybe due to the soil types in the area. The K-scenario (Figure 10) show that most of the values ranged around 0.15-0.2 so that is likely the reason of the small difference between the K-scenarios in the calculation of the soil erosion. The soil erodibility (K) factor does not influence the soil erosion result considerably and also Jain

& Singh, (2003) found that K factor is not the most important factor for the RUSLE.

One possible reason for the major differences is the way the K factor was calculated in the different K-scenarios. In order to visualize the K factor values from the soil points, the IDW interpolation method was used based on other studies (Dubber et al., 2008;

Krishna Bahadur, 2009; Prasannakumar et al., 2011; Van der Knijff et al., 2000) which applied the same method. The IDW clustered the K values around each point indicating that the same soil exists in this area. According to Campell, (1979) the soils are spatially variable which means that the soil is different in an area and not the same, something that IDW does not assume, even though the K values are similar for all the scenarios.

For the K-scenario 2, 188 soil samples were connected with literature values of soil erosion type (Table 3) with a specified K factor. The K factor of the K-scenario 2 showed a completely different trend than the K-scenario 1 since the identification of those samples have been collected by SGU for the study area and are more reliable. Higher K values found in soils that have a small particle size such as clay loam and silt and lower K values for soils with larger particle size like sand and fine clay. Also, a limitation of this scenario is the relation between soil type map and soil samples since the scale of the soil map was 1:25000 and the relation cannot be accurate enough.

The K-scenario 3 was based on SGU soil data measurements for each soil type. It is clear that the smaller number of soils for the third scenario was a limiting factor in the calculation of the K values. Also, the soil samples have not big variety than the second scenario and only clay loam, fine clay and clay exists in contrast to the 188 soil samples with bigger variety in soil types.

(35)

31

7.2 Rainfall erosivity (R) factor influence

The rainfall erosivity factor showed values ranging between 393.73 to 333.91, 424.88 to 339.091 and 470.93 to 371.28 mm for R-scenario 1, R-scenario 2 and R-scenario 3, respectively. The R-scenario 2 and R-scenario 3, 2021-2050 and 2069-2098, respectively are modelled scenarios and looking at the plot of Figure 8, the 2 future scenarios are very similar except 1 dataset (Kallered D) which is consistently higher than the other.

Comparing with the actual data from R-scenario 1(2000-2018), the future scenarios do not show variability in the results. Also, the future rainfall does not change drastically, thus no extreme change in the soil loss occurred. Looking at the results of interpolation and plot for rainfall, Figure 8 and Figure 9, the calculated R factor for the R-scenarios have affected by the rainfall in the area, measured from the gauges. The higher R factor values were found in the south of the study area, where the stations of Kallered D and Goterborg A showed higher rainfall values for all three time periods. Also, for the period 2000-2018 the R factor affected by the station of Uddevalla D and Kroppefjall-Granan A where the R values are higher at the northwest in contrast to the future periods that were not affected by those stations. The lower values for all three R-scenarios found on the central and northeast side of the study area where the stations of Trokorna, Gendalen, Lidkoping and Traneberg gave the lowest values. These differences of R-scenarios showed that the R factor has not affected considerably the soil erosion for 2000-2018 and will not for the future. Other studies by Dike et al., (2018) and Lu et al., (2004) with mean annual precipitation of 2200 and 2016mm, respectively and elevation between 100 – 400 m showed that rainfall will affect significantly the soil erosion in the future when higher rainfall occurs.

7.3 Cover management (C) factor influence

The cover management (C) factor affects the soil erosion since it was based on the NDVI estimation of the area for June 2018. It would have been different for another month like in December since the vegetation cover is lower than in June, and that affects the NDVI result; thus, the result of the C factor. Also, since the C factor reduces the soil loss in an area, due to vegetation cover, the soil loss would be higher in December. It can be different also for another day depending on the sun angle according to the study by Epiphanio & Huete, (1995). Steep areas can cause problems to the NDVI and thus to C factor due to the shadow effect which depends on the position of the sun.

7.4 LULC map and classification influence

The LULC map was the same for all the time periods since it was not feasible to estimate them for the future, which is an issue since the land cover/use changes through the years.

That is a limitation for the study. Also, the error matrix has affected the results of the classification where the kappa value was 70% and that indicates a good connection between the classes but not perfect or very good (Khare et al., 2016). That shows a confusion of sparse forest and grassland with other land cover classes like the dense