COMPARISON OF DIGITAL ELEVATION MODEL METHODS TO ASSESS ERRORS IN MICROTOPOGRAPHY AND HYDROLOGIC PATTERNS IN SOUTHEAST-SPAIN

(1)

Bachelor Thesis

Earth Sciences

COMPARISON OF DIGITAL ELEVATION MODEL METHODS TO ASSESS

ERRORS IN MICROTOPOGRAPHY AND HYDROLOGIC PATTERNS IN

SOUTHEAST-SPAIN

Jeroen Kooijman

10718397 - University of Amsterdam Future Planet Studies

Abstract

Digital Elevation Models (DEMs) have become an important tool in deriving topographic information, as they allow extraction of basic geomorphologic and topographic features. With the emergence of Unmanned Aerial Vehicles (UAV), it has become easier to take photographs needed to create a DEM. However, errors in DEMs derived from aerial photographs are still present. Vegetation can be the cause of errors, creating false elevations in the DEM. These errors affect parameters used in topographic and hydrologic analysis (Wechsler, 2007). To improve the amount of errors in a DEM, field measurements can be made using differential GPS. These measurements are highly accurate due to the addition of a base station correcting for GPS measurements made by the user. With the correction of the DEM the effect of due errors in elevation on microtopography and hydrology analysis is evaluated. This theory was applied to the research area in the Alquería catchment in Southeast-Spain. High resolution UAV aerial photographs were used to create a DEM. dGPS measurements under the vegetation were used to create an improved DEM of the research area. Interpretation of microtopography and hydrological analysis were performed on both DEMs to assess the effect of errors in the DEM. From the results in this research, it can be concluded that improving the DEM made from aerial photography is beneficial to a more accurate interpretation of microtopography and hydrological patterns and is an effective method in removing the error caused by vegetation in UAV derived DEMs.

Keywords: Digital Elevation Model, Differential GPS, Unmanned Aerial Vehicle, Microtopography, Hydrology

(2)

IBED

Table of Contents

Introduction 3

Fieldwork Area 5

Knowledge Gap 8

Research Aim & Research Questions 8

Methods 9 Results 13 Discussion 21 Upscaling 23 Recommendations 23 Conclusion 24 Acknowledgements 25 References 26 Appendices 29 Appendix A: Maps 29

Appendix B: Plot Cross Sections 41

Appendix C: Terrace Cross Sections 43

(3)

IBED

Introduction

The use of remote sensing data is an important tool in environmental research, since the supply of data has grown significantly (Lam et al., 1998). An important tool in deriving topographic information is a Digital Elevation Model (DEM). Since DEMs allow for the extraction of basic geomorphologic and topographic features of an area, they have become the foundation of environmental and topographic queries (Mokarram & Hojati, 2015). DEMs can be derived from photogrammetric data capture (Moore et al., 1991, in Wilson & Gallant, 2000), which rely on stereoscopic interpretation of satellite imagery or aerial photographs (Carter 1988, Weibel and Heller 1991, in Wilson & Gallant, 2000). Traditionally, these photographs are taken by aircrafts at an elevation of approximately 6000 metres. With the emergence of Unmanned Aerial Vehicles (UAVs) the collection of these aerial photographs has become cheaper and more accessible to individual users. Furthermore, due to the low altitude at which these UAVs fly, the resolution of photographs has improved significantly (DiBiasi & Dutton, 2017).

This research extents on the MSc thesis of Jan Timmerman (2017). A DEM was provided at the start of the research and is based on aerial UAV photos of the Alquería catchment in Southeast Spain. These photos were taken at an altitude of 100 metres. This allowed for a cell size of 2.3 by 2.3 cm (Timmerman, 2017). This cell size of the DEM permits a very precise analysis of the research area. However, this DEM contains elevation errors due to vegetation present in the research area. The effect of vegetation and shadows on DEM errors was studied by Fabris & Pesci (2005). Their analysis of four areas with different vegetation covers showed that areas with higher amounts of vegetation caused a higher standard deviation in accuracy of the DEM. This is a recurring problem with Digital Elevation Models, as a large amount of errors in a DEM can be attributed to errors in data collection (Chaplot et al., 2006). Furthermore, inaccuracies in DEMs increase when the research area has high differences in slopes (Toz & Erdogan, 2008). Hydrologic analysis frequently uses topographic attributes derived directly from DEMs. This means that the errors in DEMs directly affect derived parameters (Wechsler, 2007). To improve this DEM detailed elevation, measurements must be made with a Differential Global Positioning System (dGPS). dGPS improves accuracy of positioning compared to GPS by adding a base/reference station that has a dedicated radio link with the GPS receiver/rover (Matosevic, Salcic & Berber, 2006). A visual representation of the principles behind dGPS is shown in figure 1.

Removal of the error in the DEM could serve as a helpful tool in interpreting microtopography in the area. Furthermore, the improved DEM helps to predict exact water and sediment flow lines in the landscapes. Accuracy of a DEM is essential to derive the correct topography and physical processes in a natural ecosystem (Chaplot et al., 2006). Additionally, hydrological processes rely highly on the topography of an area. Surface form is an important factor in deriving surface water flow. Moreover, parameters such as flow direction are determined from slope and aspect of an area (Wechsler, 2007). Therefore, it is needed to assess the accuracy of the corrected DEM created with measurements. This assessment is needed to determine if water and sediment flows can be accurately derived from the microtopography in the area from this DEM. Moreover, a correct estimation of the elevation under the vegetation in the area gives a better insight into water and sediment flow in the research area. This is because vegetation has a shielding effect on rainfall in semi-arid regions. Since rainfall is intercepted by vegetation which influences the drop falling height and therefore the splash erosion. Furthermore, vegetation increases infiltration and reduces direct evaporation (Puigdefábregas, 2005). The research area in the Alquería catchment is

(4)

IBED

characterised by high intensity rainfall events (Cammeraat, 2004). The shielding effect of vegetation could therefore be of significant influence on the water and sediment flow in the research area. The fact that there is more infiltration into the soil under vegetation could help predict exact water flows in the research area. Previous research by Cammeraat (2004), which measured water flow, concluded that water does not flow downslope in areas covered with vegetation. Additionally, plant clumps such as Stipa tenacissima are characterized by sediment deposition in mounds below or in micro fans above the plant (Puigdefábregas, 2005). This process was also confirmed in the research area by Cammeraat & Imeson (1999), stating that the soil surface around the Stipa tenacissima was slightly higher compared to the bare surface around it. Additionally, Cammeraat & Imeson (1999) state that the microtopography resembles micro terracing on the slopes. This is due to the accumulation of coarse and fine materials in front of the Stipa tenacissima tussocks. A measurement of the correct elevation under the vegetation errors can therefore give a better insight into sediment flow in a semi-arid area like the research area in the Alquería catchment.

(5)

IBED

Fieldwork Area

The research area is located east of the city of Lorca in the Southeast of Spain in the Guadalentín catchment (Cammeraat, 2002). The research area has a total area of 800 by 1500 metres. A combination of tussock grasses (Stipa tenacissima) and few trees (Pinus Halepensis) characterize the vegetation cover of the area. Pedimenation surfaces attached to limestone ridges dominate the geomorphology of the research area (Cammeraat, 2004). The research area has a semi-arid climate and is in the driest area of Europe, with an annual precipitation of approximately 277 mm (Cammeraat & Imeson, 1999). The research area contains three plots that are a former research area, containing Gerlach troughs at the bottom of the slopes (Timmerman, 2017). For this research, two of the three plots have been chosen. The chosen plots are the two plots visible in figure 2 as the red bordered areas. Both plots have a south facing slope. Plot 2 is the southern plot and can be seen in figure 3 and has an area of 320 m2

(Cammeraat, 2004). The vegetation in this plot is dominated by grass tussocks and contains two trees (Timmerman, 2017). Plot 3 is the northern plot and can be seen in figure 4 and has an area of 400 m2_{(Cammeraat, 2004). The vegetation in this plot is dominated by grass}

tussocks and contains one tree (Timmerman, 2017).

(6)

IBED

Figure 3: Aerial photograph of plot 2 in the research area surrounded by red lines and Stipa tussocks location and numbers.

(7)

IBED

Figure 4: Aerial photograph of plot 3 in the research area surrounded by red lines and Stipa tussocks location and numbers.

(8)

IBED

Knowledge gap

Errors in DEMs have been an interesting field of research for over a long period. The addition of feature specific data to photogrammetrically derived elevation models has proven to decrease errors in these models (Li, 1994). However, even though methods to assess and remove errors have been created (Hengl et al., 2004; Li, 1991; Lopez, 2002), errors do persist (Wechsler, 2004). With the high resolution UAV photos from 2013 of the Alquería catchment and the derived DEM from this area, the foundations of a DEM error assessment are well established. Furthermore, the chosen approach in this research differs from previously mentioned research due to the amount of measurements to correct the DEM using the dGPS equipment. Research from Li (1991) and Li (1994) used a set of control points to reduce error in the DEM and research from Hengl et al. (2004) and Lopez (2002) used modelling to reduce errors. Moreover, this approach varies from previous research in DEM error determination in the fact that a new DEM is created to compare with the DEM containing errors, instead of improving upon the original DEM with error elimination. Another difference in this research is the high resolution of the DEM that is used, this higher resolution allows for a more precise analysis. Therefore, this research could provide new insights into the determination and elimination of photogrammetrically derived DEM errors. This could also serve topographic and hydrologic analysis that make use of DEMs for their research.

Research aim & questions

The aim of the research is to see if the Digital Elevation Model corrected by collected field data differs significantly from the UAV collected Digital Elevation Model data. In addition, the research aims to predict water and sediments flows, and interpret microtopography using Digital Elevation Models. The following research question will be answered in this research project: ‘How does improvement of Digital Elevation Model accuracy help interpret microtopography in the Alquería catchment?’. This research question is divided into multiple sub-questions. The first sub-question is: ‘How can a Digital Elevation Model be corrected using dGPS in the field?’. The second sub-question is: ‘How can a Digital Elevation Model be used to interpret microtopography in the Alquería catchment?’. The third sub-question is: ‘How do hydrological patterns in the Alquería catchment look based on the area’s microtopography?’. The first sub-question requires fieldwork in the Alquería catchment, as new measurements are needed to correct the Digital Elevation Model. The second and third sub question use the new measurements to draw conclusion on the microtopography and hydrologic patterns in the research area.

(9)

IBED

Methods

The following methods were chosen to answer the research question. At the start of the research, a week of fieldwork was required. This week was used to collect data needed for the correction of the elevation data in the Digital Elevation Model. Before starting in the field, multiple maps were created based on the provided DEM using GIS software. These maps were based on the collected orthophotos from the UAV and include: a map of the research area, maps of the smaller plots and a larger map of the area surrounding the research area. These maps were used to navigate the fieldwork area. The measurements were made using a Topcon HiPer Pro dGPS. This dGPS equipment used real time differential GPS, which means that a reference station is set up for the experiment.

Furthermore, using the Bluetooth connection between the reference station and the GPS receiver, the GPS data is corrected (Trimble Navigation Limited, 2004). This method differs from post processing differential GPS, as it corrects the GPS receiver in real time, leading to a higher accuracy in the field (Trimble Navigation Limited, 2004). The measurements were made at the base of each plant in the fieldwork area to get the precise x, y and z coordinate. This x, y and z coordinate was saved and transferred to a spreadsheet. To get an accurate estimate of the correct height under the vegetation around 30 measurements were made per Stipa tussock. This number differs slightly depending on the size of the of Stipa tussocks. Measurements were made in plot 2 and plot 3 of the research area. The first contained approximately 165 Stipa tussocks. The second plot contained approximately 141 Stipa tussocks. To create a systematic approach to measuring the vegetation the fieldwork area was divided into vertical strips. These strips were measured from the bottom to the top. Each of the grass tussocks was assigned a number, which can be seen in figure 3 & 4.

After all elevations under the vegetation had been measured the measurements were converted from point measurements into a DEM by using Kriging interpolation. Kriging interpolation was chosen through a statistical comparison between IDW and Kriging in Matlab (see Appendix D). The measurements were separated into a training dataset containing 70 percent of the measurements and a test dataset containing 30 percent of the measurements. Using the training dataset, the values on the test dataset were interpolated using both IDW and Kriging. The values from the interpolation were then compared to the actual values in the test dataset. Using ANOVA and the Wilcoxon rank-sum test, the variance between the interpolation values and the measurements were then calculated for the two interpolation methods. This showed the least error using Kriging, hence the choice for Kriging for the interpolation to create the DEM. After the creation of the DEM from the dGPS measurements it was compared to the provided DEM created from the aerial photographs. However, before a comparison could be made between the two DEMs, data correction needed to be performed for a correct analysis. This is done by checking if the bare areas of the DEMs have the same elevation. These areas were not influenced by errors due to vegetation and can therefore serve as control points between the two DEMs. Fifty measurements on each of the DEMs on the same bare areas were compared. These measurements were then subtracted from each other and then averaged. This gave the average difference in elevation between the two DEMs. Using the raster calculator this value could then be added or subtracted from the dGPS created DEM to ensure the comparison could be made starting at the same elevation.

Furthermore, this DEM was then used to predict several microtopographical features. These predictions were made on both the provided DEM and the corrected DEM. Comparison of both models allowed for a quantification of the error made by the DEM made with the UAV photos from 2013. To quantify the differences between the DEM created from the aerial

(10)

IBED

photographs and the DEM created from the dGPS measurements, the raster calculator was used to create difference maps between the maps. A first method of interpreting microtopography is the slope and aspect analysis. This contributed to the assessment of the change in slope around vegetation patches in the two plots. Like for the quantification of the DEMs the slope maps were subtracted from each other using the raster calculator to create a difference map. Creating a difference map of the aspect of the plots was not possible as it would change the aspects of the areas, if subtraction from each other was used. Consequently, the comparison of the aspect maps was mostly done by visual comparison of the maps. To analyse the differences in elevation and slope between the DEMs, cross sections were made. This was firstly done to the entire plot, to analyse the disappearance of the peaks caused by the vegetation in the aerial photograph DEM. Furthermore, a selection of smaller cross sections was made to analyse the presence of terraces in the dGPS DEM and the difference in elevation compared to the aerial photograph map.

The rest of the analysis was mostly focused on water and sediment flows in the area. An important first calculation that needed to be made to help find water and sediment flow was flow direction. Flow direction is calculated by measuring the flow direction for each cell in a raster dataset. The direction the cell flows to has a value corresponding to one of the powers of two up till 128 (Jenson & Domingue, 1988). This in turn created a raster with values showing the flow direction. Flow direction could be calculated in ArcGIS software using the flow direction tool. This helped find water and sediment flow in the research area. A further hydrological analysis could be carried out, calculating flow accumulation, watershed and stream network. Flow accumulation is the sum of number of cells that flow to it. Using flow direction and flow accumulation the different watersheds can be calculated. Flow accumulation can also be used to distinguish a stream network (Jenson & Domingue, 1988). These calculations could also be made using tools in ArcGIS software.

To quantify the differences between the Aerial photo DEM’s hydrological maps and the dGPS DEM’s hydrological maps, the values in the attribute tables of the maps were compared. The length and/or area of the drainage lines and the catchments were compared to see if there were significant differences between the two DEM methods. After the analysis using GIS software the results were interpreted and used to draw a conclusion that answered the research question. If it was possible to identify a pattern in the error made in DEMs due to vegetation, the results could be used for upscaling into other areas. This would create the possibility to automatically remove vegetation errors in DEMs that were photogrammetrically derived from aerial photos. This would be useful in areas with similar vegetation patterns.

(11)

IBED

Figure 6: Aerial photograph of plot 2 in the research area surrounded by red lines and dGPS measurements marked by blue points

(12)

IBED

Figure 7: Aerial photograph of plot 3 in the research area surrounded by red lines and dGPS measurements marked by blue points

(13)

IBED

Results

After interpolation, the measurements made with the dGPS equipment seen in figure 6 & 7 a Digital Elevation model was created using Kriging interpolation. The Digital Elevation Models of Plot 2 and Plot 3 can be seen in figures 8 & 9.

Figure 8: Digital elevation model created from dGPS measurements of plot 2 in the research area surrounded by red lines

(14)

IBED

Figure 9: Digital elevation model created from dGPS measurements of plot 3 in the research area surrounded by red lines

Using the DEMs created from the dGPS measurements a difference map was made for both plots. In these difference maps, visualised in figure 10 & 11, the DEM created by dGPS measurements was subtracted from the aerial photograph DEMs.

(15)

IBED

Figure 10: Difference between DEM from aerial photographs and DEM from dGPS measurements of plot 2. Areas in green show overestimated values in the aerial photograph DEM, while red areas show underestimated values in the aerial photograph DEM.

(16)

IBED

Figure 11: Difference between DEM from aerial photographs and DEM from dGPS measurements of plot 3. Areas in green show overestimated values in the aerial photograph

(17)

IBED

The same method was applied on the slope maps derived from the two different DEMs. Subtracting the dGPS DEM from the aerial photograph DEM created the difference maps shown in figure 12 & 13.

Figure 12: Difference between slope (degrees) from aerial photographs and slope from dGPS measurements of plot 2. Areas in green show overestimated values in the aerial photograph DEM, while red areas show underestimated values in the aerial photograph DEM.

(18)

IBED

Figure 13: Difference between slope (degrees) from aerial photographs and slope from dGPS measurements of plot 3. Areas in green show overestimated values in the aerial photograph DEM, while red areas show underestimated values in the aerial photograph DEM.

To quantify the smoothing effect of the interpolation in the aerial photograph DEM on the terraces present on the bare slope, statistical analysis was performed on several cross sections between the two plots. Using the Wilcoxon rank-sum test the difference between the two

(19)

IBED

Table 1: Wilcoxon rank-sum test results comparing cross section elevation of aerial photograph DEM and dGPS DEM. P-value for each cross section and whether the H0 hypothesis is rejected is given in the table.

Cross Section Location Plot P-value H0

47 – 57 2 7.3440e-04 Rejected 61 – 29 2 8.1478e-07 Rejected 92 - 89 2 0.0365 Rejected 153 – 152 2 0.2913 Not rejected 166 – 157 2 0.0047 Rejected 27 – 25 3 5.0926e-07 Rejected 104 – 83 3 0.3323 Not rejected 117 – 116 3 0.1567 Not rejected 119 – 95 3 0.2163 Not rejected 136 -135 3 0.4293 Not rejected

To compare the differences in hydrological patterns between the aerial photograph DEM and the dGPS DEM the length and/or area of both the catchments present in the DEMs and the drainage channels are compared. The differences between the DEM methods and the plots can be seen in table 2 for catchments and table 3 for drainage channels.

Table 2: Difference between catchment area and length from aerial photographs and catchment area and length from dGPS measurements of plot 2 and 3

Plot 2 Old Plot 2 New Plot 3 Old Plot 3 New

Length (m) Area (m 2_{) Length} (m) Area (m 2_{) Length} (m) Area (m 2_{) Length} (m) Area (m 2₎ Count 508 508 405 405 239 239 397 397 Minimum 0.2292 0.003283 0.2294 0.003289 0.2296 0.003295 0.2294 0.003289 Maximum 22.2648 6.08184 30.1836 5.084182 34.8888 14.71236 30.872 6.065286 Sum 3782.6 563.1713 4216.645 565.1608 2228.654 456.8852 3894.727 469.2693 Mean 7.44606 3 1.108605 10.41147 1.395459 9.32491 1.911654 9.810396 1.182039 Standard Deviation 3.44528 7 0.825048 5.040076 0.926228 4.567201 1.683261 5.346835 0.929473 Nulls 0 0 0 0 0 0 0 0

(20)

IBED

Table 3: Difference between drainage line length from aerial photographs and drainage line length from dGPS measurements of plot 2 and 3

Plot 2 Old Plot 2 New Plot 3 Old Plot 3 New

Length (m) Length (m) Length (m) Length (m)

Count 505 405 238 397 Minimum 0.0286 0.0287 0.0286 0.0286 Maximum 6.932531 9.233281 12.7866 9.210499 Sum 610.6942 655.8751 359.7599 679.9269 Mean 1.209295 1.619445 1.511596 1.712662 Standard Deviation 1.137287 1.590251 1.652246 1.706425 Nulls 0 0 0 0

(21)

IBED

Discussion

Vegetation can cause problems in creating accurate Digital Elevation Models. As it may cause a higher amount of errors in elevation, which will result in a higher standard deviation, and less accuracy (Fabris & Pesci, 2005). In this research, efforts were made to improve this accuracy by creating a DEM based on dGPS measurements. When looking at the created DEMs a smoothing of the elevation is visible, compared to the DEM that was created from aerial photographs. When comparing the elevation data in the difference maps created for the two plots the improvement of the DEM is evident. The areas where the elevation in the aerial photograph DEM show higher elevation than the dGPS DEM correspond with the vegetated areas in the two plots. However, in most bare areas in the plots the elevation is close or equal in both DEMs. The removal of the vegetation error in the new DEM derived from dGPS measurements shows that this method is a useful tool in increasing DEM accuracy. In previous research from Gomes Costa et al. (2010) a vegetation error assessment using dGPS was conducted on a semi-arid catchment in Brazil. Gomes Costa et al. (2010) concluded that a vegetation error was present and found the average error to be 1.8 metres. This is significantly more than in this research, but in the semi-arid catchment in Brazil the vegetation cover had a height between 3 and 6 metres, which is significantly higher than the average height of both Stipa tenacissima and Pinus Halepensis, as the maximum height of vegetation in the difference elevation map was 1.23 metres.

Secondly, the improved DEM showed to be useful in interpreting microtopography in the research area. First, the difference in the elevation showed an improved visualisation of the topography of the plots. Next, creating slope maps for both DEMs showed an improved visualisation of the microtopography of the research area, as the slope maps showed the influence of vegetation on the slope of the two plots. The dGPS DEM revealed that the elevation decreases most directly underneath the Stipa grass tussocks. Using the aerial photograph DEM, the interpretation of the slope would mean that the elevation would decrease after the Stipa grass tussocks. The shift in slope angle can be seen in figure 12 & 13. The pattern seen in the dGPS DEM slope map follows the theory around sediment deposition in previous research. Puigdefábregas (2005) and Cammeraat & Imeson (1999) concluded that sediment accumulates in front of the Stipa grass tussocks, creating micro terraces. The fact that the sharp increase in slope angle is observed right underneath the vegetation is therefore logical.

Furthermore, the presence of terraces on the slopes of the two plots is visible in the dGPS DEM. The fact that these terraces are present in the new DEM and not on the old DEM, even though the terraces are present on bare slopes can possibly be explained by the smoothing effect of interpolation methods. In previous research from Yamamoto (2005) the smoothing effect of kriging interpolation is explained. In this interpolation, low values are often overestimated, while high values are often underestimated. Due to the lower point density of the aerial photograph DEM it is possible that these terraces disappeared due to the smoothing effect, while in the new DEM they became visible due to the higher measurement density. To test this theory 5 cross sections were taken from both plots (Appendix C) and

(22)

IBED

tested for significant differences. The results shown in table 1 do not indicate a clear conclusion on this theory for both plots. When testing the theory of the smoothing effect of kriging on plot 2, four out of five cross sections show a significant difference with a p-value exceeding the alpha value of 0.05. However, in plot 3 four out of five cross sections do not show a significant difference with p-values staying above the alpha value of 0.05. This difference between plots can be explained by the difference in topography. When looking at the elevation difference maps (figure 10 & 11) and the difference slope maps (figure 12 & 13) the terraces are more easily spotted in plot 2 by sharp lines compared to plot 3. This might lead to a more significant difference between the old and the new DEM elevation in comparison with plot 3. Moreover, the change in aspect between the two DEMs is also visible (Appendix A1-A4). Due to the peaks in the aerial photograph DEM caused by vegetation there are some northern faces areas on the map, while the plots both have south facing slopes. When comparing this to the aspect of the dGPS DEM the north facing areas almost entirely disappear and have become south facing areas, which is in line with the slope of the plots.

Furthermore, the use of both DEMs in hydrologic analysis resulted in opposing findings. When comparing the area and length of the catchments in the two plots for both DEMs contradictory results were found. For plot 2, there is a decrease in catchments count, but an increase in average length and area. This is the same for the drainage line length for plot 2, even though the drainage line count decreases. However, for plot 3 there was an increase in catchment count and catchment length, but a decrease in catchment area. As for the catchments for plot 3, the drainage line count increases and the length increases. The fact that catchments and drainage line follow the same pattern is to be expected as drainage lines are based on the catchment delineations. Furthermore, the different patterns in increase or decrease in catchment and drainage lines count still show some similarities. This can be seen by the fact that in the new DEM the count for both plots approaches approximately the same number of catchments and drainage lines, as both counts are around four hundred. A possible explanation for this pattern can be found in the increase in measurement density in the new DEM in comparison with the aerial photo DEM, which leads to a higher accuracy in showing the microtopography in the area. However, previous research of Thomas et al. (2016) concluded that point density change in a DEM has no significant influence on the microtopography accuracy in a DEM. Differences in hydrological patterns in previous research (Buakhao & Kangrang, 2016, Osorio et al., 2007) have mostly been due to DEM resolution. However, this is not applicable to this research, as both DEMs have the same resolution. A likely cause of the differences in the hydrologic pattern is due to the error in the primary dataset. Research from van Niel et al. (2004) states that errors in primary datasets, in this case the aerial photograph DEM, cause propagated errors in secondary layers. Layers such as slope and aspect contain errors caused by errors in the DEM. Hydrologic patterns are derived from the DEM as well, which means the error propagates to the catchment and drainage lines layers. The large change in the count and mean in area and length can therefore most likely be caused by this propagation.

(23)

IBED Upscaling

With the results from this analysis there are also methods for upscaling the results to the catchment scale. With the difference map created between the two DEMs for elevation it is visible that the elevation difference is almost exclusively present in the areas covered with vegetation. A possible method to automatically remove this error could be excluding measurements made on vegetated areas or deducting the average height of the dominating vegetation. This could be done by using UAV photograph derived DEMs and then classifying the area in a binary vegetation map with supervised classification. In the master thesis from Jan Timmerman (2017) this method was applied to create a binary vegetation map. If the average height of the vegetation covered in the area is known, raster values located on vegetation areas can be decreased by this height. If the average height is unknown, the points in vegetated areas could be excluded from interpolation to prevent vegetation errors. If this is done, the subsequent layers created from this DEM are fixed automatically. This new method will eliminate dGPS measurements in order to increase the accuracy of an analysis. Moreover, this research also shows that the use of dGPS measurements is an accurate method to create a DEM. This could be applied to other areas and even larger areas, albeit that the method is time intensive.

Recommendations

For further research, there are several recommendations to improve upon this research. First, the upscaling possibilities should be further explored. In this research, a possible upscaling method is suggested. To ensure the upscaling is possible, this method should be tested. Furthermore, uncertainties should be eliminated from the research. The emergence of terraces on bare areas can likely be explained by the effect of smoothing in interpolation, as seen in the analysis for plot 2 using the small cross sections. However, this cannot be said with absolute certainty, as plot 3 did not show these significant differences. Therefore, it should be further explored why these terraces are significantly different in the DEM created from dGPS measurements for plot 3. Moreover, the different hydrologic pattern changes between the two plots can be further researched. If more plots are checked to see if there is a certain pattern when analysing differences in DEM methods, it could be discovered if one of these two changes is an outlier.

(24)

IBED

Conclusion

In short, this study used dGPS measurements to create a detailed DEM. From this DEM microtopography and hydrologic patterns were derived. These were then compared with microtopography and hydrologic patterns created from the aerial photograph DEM used in the master thesis of Jan Timmerman (2017). With the results from this analysis, an answer to the question: ‘‘How does improvement of Digital Elevation Model accuracy help interpret microtopography in the Alquería catchment?’ can be given.

Firstly, the improved DEM created from dGPS measurements removed most of the errors caused by vegetation to create a more accurate display of the plots’ elevation. This allowed for a more precise visualisation of the areas microtopography. The emergence of terraces on bare slopes and the movement from high slopes directly after the vegetation to directly underneath the vegetation, are indications of this change. The change in aspect of the slopes also changed from a mix of all aspects to an almost exclusively southern aspect, this shows that the new DEM gives an improved visualisation of the plots’ microtopography.

Secondly, the improved DEM created from dGPS measurements showed a change in the hydrological patterns compared to the original DEM created from aerial photographs. In the original DEM, the catchments and drainage lines were influenced by peaks in the DEM caused by the vegetation errors. In the improved DEM, the change for the two plots was different, with a decrease in catchments and drainage lines in plot 2 and an increase in catchments and drainage lines in plot 3. However, in the improved DEM the catchments and drainage lines in both plots moved towards the same number.

Hence, from this research it can be concluded that dGPS measurements improve the accuracy digital elevation models, which helps interpret microtopography and hydrologic patterns, through a more detailed and correct depiction of the research area.

(25)

IBED

Acknowledgements

I wish to thank my supervisor Dr. Cammeraat for his guidance during this research. His advice and constructive suggestions have helped greatly during this research. Next, I would like to express my gratitude to Dr. Seijmonsbergen and Drs. Mulder for helping me with questions I had during the Bachelorthesis. I would also like to thank Jan Timmerman for providing the starting materials for the analysis. Finally, I would like to thank my group members: Olaf de Haan, Marle de Jong, Cynthia van Leeuwen and Bart ter Mull, for helping me during the fieldwork in southeast Spain and my research.

(26)

IBED

References

Buakhao, W. & Kangrang, A. (2016). DEM Resolution Impact on the Estimation of the Physical Characteristics of Watersheds by Using SWAT. Advances in Civil Engineering, vol. 2016, 1-9. Cammeraat, L.H., Imeson, A.C. (1999). The evolution and significance of soil-vegetation patterns following land abandonment and fire in Spain. Catena 37, 107–127.

Cammeraat, L.H. (2002). A review of two strongly contrasting geomorphological systems within the context of scale. Earth Surf. Process. Landforms 27, 1201–1222.

Cammeraat, L.H. (2004). Scale dependent thresholds in hydrological and erosion response of a semiarid catchment in southeast Spain. Agriculture, ecosystems & environment, 104(2), 317-332.

Chaplot, V., Darboux, F., Bourennane, H., Leguédois, S., Silvera, N., & Phachomphon, K. (2006). Accuracy of interpolation techniques for the derivation of digital elevation models in relation to landform types and data density. Geomorphology, 77(1–2), 126-141.

DiBiasi, D., Dutton, J.A. (2017). The Nature of Geographic Information: An Open Geospatial Textbook. PennState College of Earth and Mineral Science. Retrieved 3 May 2017, from: https://www.e-education.psu.edu/natureofgeoinfo/

ESRI (2012). Understanding drainage systems. Retrieved 13 April 2017, from: http://webhelp.esri.com/arcgisdesktop/9.3/printBooks_topics.cfm?pid=6050

Fabris, M. & Pesci, A. (2005). Automated DEM extraction in digital aerial photogrammetry: precisions and validation for mass movement monitoring. Annals of Geophysics. 48(6). 973-988.

Gomes Costa, C. A., Teixeira, A. d. S., de Andrade, E. M., Pereira de Lucenas, A. M., & Holanda de Castro, M. A. (2010). Analysis of the influence of vegetation in the altimetry of SRTM data in semi-arid watersheds. Revista Ciencia Agronomica, 41(2), 222-230.

Hengl, T., Gruber, S., and Shrestha, D. P. (2004). Reduction of errors in digital terrain parameters used in soil-landscape modelling, Int. J. Appl. Earth Obs. Geoinfo., 5, 97–112. Jenson, S.K., & Domingue, J.O. (1988). Extracting Topographic Structure from Digital Elevation Data for Geographic Information System Analysis. PHOTOGRAMMETRIC ENGINEERING AND REMOTE SENSING, 54(11), 1593-1600.

Lam, N.S.N., Quattrochi, D.A., Qiu, H.L., and Zhao, W. (1998). Environmental Assessment and Monitoring with Image Characterization and Modeling System Using Multiscale Remote Sensing Data. Applied Geographic Studies 2(2). 77–93.

(27)

IBED

Li, Z. (1994). A comparative study of the accuracy of digital terrain models (DTMs) based on various data models, ISPRS Journal of Photogrammetry and Remote Sensing, 49, 2–11. Lopez, C. (2002) An experiment on the elevation accuracy improvement of photogrammetrically derived DEM, Int. J. Geogr. Inf. Sci, 16, 361–375.

Matosevic, M., Salcic, Z., & Berber, S. (2006). A comparison of accuracy using a GPS and a low-cost DGPS. IEEE Transactions on Instrumentation and Measurement, 55(5), 1677-1683. Mokarram, M. & Hojati, M. (2015). Comparis of digital elevation model (DEM) and aerial photographs for drainage. Model Earth Syst Environ 1:46. 1-6.

van Niel, K. P., Laffan, S. W., & Lees, B. G. (2004). Effect of error in the DEM on

environmental variables for predictive vegetation modelling. Journal of Vegetation Science, 15(6), 747-756.

Osorio, J. , Wolfe, M.L. , Heatwole, C., Benham, B., Cardenas, A. (2007). Comparison of watershed boundaries derived from SRTM and ASTER digital elevation datasets and from a digitized topographic map. Written for presentation at the 2007 ASABE Annual International Meeting Sponsored by ASABE Minneapolis Convention Center Minneapolis, Minnesota 17 - 20 June 2007.

Puigdefábregas, J. (2005). The role of vegetation patterns in structuring runoff and sediment fluxes in drylands. Earth Surface Processes and Landforms, 30(2), 133-147.

Terrier, P. & Schutz, Y. (2005) How useful is satellite positioning system (GPS) to track gait parameters? A review J. Neuroeng. Rehabil. 21–11

Thomas, I. A., Jordan, P., Shine, O., Fenton, O., Mellander, P. -., Dunlop, P., et al. (2017). Defining optimal DEM resolutions and point densities for modelling hydrologically sensitive areas in agricultural catchments dominated by microtopography. International Journal of Applied Earth Observation and Geoinformation, 54, 38-52.

Timmerman, J. (2017). Comparing soil losses in heterogeneous vegetated semi-arid hillslopes by remote sensing of topography and vegetation patterns. MSc Thesis, University of Amsterdam.

Toz, G. & Erdogan, M. (2008). DEM (DIGITAL ELEVATION MODEL) PRODUCTION AND ACCURACY MODELING OF DEMS FROM 1:35.000 SCALE AERIAL PHOTOGRAPHS. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1.

Trimble Navigation Limited (2004). Why post process GPS data? Mapping & GIS. Retrieved 12 April 2017, from: http://www.esri.com/~/media/Files/Pdfs/partners/hardware/why-postprocess.pdf

(28)

IBED

Wechsler, S.P. (2007). Uncertainties associated with digital elevation models for hydrologic applications: a review. Hydrol. Earth Syst. Sci., 11, 1481–1500.

Wilson, J.P. & Gallant, J.C. (2000). Digital Terrain analysis. In Terrain analysis. Principles and applications (pp. 1-28). London, John Wiley & Sons.

Yamamoto, J. K. (2005). Correcting the smoothing effect of ordinary kriging estimates. Mathematical Geology, 37(1), 69-94.

(29)

IBED

Appendix A: Maps

(30)

IBED

(31)

IBED

(32)

IBED

(33)

IBED

(34)

IBED

(35)

IBED

(36)

IBED

(37)

IBED

(38)

IBED

(39)

IBED

(40)

IBED

(41)

IBED

Appendix B: Plot Cross Sections

Appendix B1: Cross section old DEM of plot 2. Height and distance given in metres.

(42)

IBED

Appendix B3: Cross section old DEM of plot 3. Height and distance given in metres.

(43)

IBED

Appendix C: Terrace Cross Sections

Appendix C1: Cross section between Stipa tussocks 47 and 57. Height and distance given in metres.

608.8 609.0 609.2 609.4 609.6 609.8 610.0 610.2 0.0 0 0.1 1 0.2 3 0.3 4 0.4 5 0.5 7 0.6 8 0.7 9 0.9 1 1.0 2 1.1 3 1.2 5 1.3 6 1.4 8 1.5 9 1.7 0 1.8 2 1.9 3 2.0 4 2.1 6 2.2 7 2.3 8 2.5 0 2.6 1 2.7 2 2.8 4 2.9 5 3.0 6 3.1 8 3.2 9 3.4 0 3.5 2

Cross Section Stipa 47 - 57 Plot 2

New Height Old Height

611.3 611.4 611.5 611.6 611.7 611.8 611.9 612.0 612.1 612.2 0.0 0 0.1 1 0.2 3 0.3 4 0.4 5 0.5 6 0.6 8 0.7 9 0.9 0 1.0 2 1.1 3 1.2 4 1.3 5 1.4 7 1.5 8 1.6 9 1.8 1 1.9 2 2.0 3 2.1 4 2.2 6 2.3 7 2.4 8 2.6 0 2.7 1 2.8 2 2.9 3 3.0 5

Cross Section Stipa 61 - 29 Plot 2

609.2 609.4 609.6 609.8 610.0 610.2 610.4 610.6 610.8 0.0 0 0.1 1 0.2 3 0.3 4 0.4 6 0.5 7 0.6 8 0.8 0 0.9 1 1.0 2 1.1 4 1.2 5 1.3 7 1.4 8 1.5 9 1.7 1 1.8 2 1.9 3 2.0 5 2.1 6 2.2 8 2.3 9 2.5 0 2.6 2 2.7 3 2.8 4 2.9 6 3.0 7 3.1 9

Cross Section Stipa 92 - 89 Plot 2

(44)

IBED

612.0 612.2 612.4 612.6 612.8 613.0 613.2 613.4 0.0 0 0.1 7 0.3 4 0.5 1 0.6 9 0.8 6 1.0 3 1.2 0 1.3 7 1.5 4 1.7 1 1.8 9 2.0 6 2.2 3 2.4 0 2.5 7 2.7 4 2.9 1 3.0 8 3.2 6 3.4 3 3.6 0 3.7 7 3.9 4 4.1 1 4.2 8 4.4 6 4.6 3 4.8 0 4.9 7 5.1 4

Cross Section Stipa 153 - 152 Plot 2

613.7 613.8 613.8 613.9 613.9 614.0 614.0 0.00 0.11 0.23 0.34 0.45 0.56 0.68 0.79 0.90 1.01 1.13 1.24 1.35 1.46 1.58 1.69 1.80 1.91 2.03 2.14

Cross Section Stipa 166 - 157 Plot 2

616.8 617.0 617.2 617.4 617.6 617.8 618.0 618.2 618.4 618.6 0.0 0 0.0 6 0.1 1 0.1 7 0.2 3 0.2 9 0.3 4 0.4 0 0.4 6 0.5 1 0.5 7 0.6 3 0.6 8 0.7 4 0.8 0 0.8 6 0.9 1 0.9 7 1.0 3 1.0 8 1.1 4 1.2 0 1.2 5 1.3 1 1.3 7 1.4 3 1.4 8 1.5 4 1.6 0 1.6 5 1.7 1 1.7 7 1.8 2 1.8 8

Cross Section Stipa 27 - 25 Plot 3

(45)

IBED

620.5 621.0 621.5 622.0 622.5 623.0 623.5 0.0 0 0.1 1 0.2 3 0.3 4 0.4 6 0.5 7 0.6 9 0.8 0 0.9 2 1.0 3 1.1 5 1.2 6 1.3 8 1.4 9 1.6 0 1.7 2 1.8 3 1.9 5 2.0 6 2.1 8 2.2 9 2.4 1 2.5 2 2.6 4 2.7 5 2.8 7 2.9 8 3.1 0 3.2 1 3.3 2 3.4 4 3.5 5 3.6 7

Cross Section Stipa 108 - 83 Plot 3

614.0 614.5 615.0 615.5 616.0 616.5 617.0 0.0 0 0.1 1 0.2 3 0.3 4 0.4 5 0.5 7 0.6 8 0.7 9 0.9 1 1.0 2 1.1 3 1.2 5 1.3 6 1.4 7 1.5 9 1.7 0 1.8 1 1.9 3 2.0 4 2.1 5 2.2 7 2.3 8 2.4 9 2.6 1 2.7 2 2.8 3 2.9 5 3.0 6 3.1 7 3.2 9 3.4 0 3.5 1 3.6 3 3.7 4 3.8 5 3.9 7

Cross Section Stipa 117 - 116 Plot 3

616.4 616.6 616.8 617.0 617.2 617.4 617.6 617.8 618.0 618.2 618.4 0.00 0.11 0.23 0.34 0.46 0.57 0.68 0.80 0.91 1.02 1.14 1.25 1.37 1.48 1.59 1.71 1.82 1.93 2.05 2.16 2.28 2.39 2.50 2.62

Cross Section Stipa 119 - 95 Plot 3

(46)

IBED

615.0 615.5 616.0 616.5 617.0 617.5 618.0 618.5 0.0 0 0.1 1 0.2 3 0.3 4 0.4 6 0.5 7 0.6 8 0.8 0 0.9 1 1.0 3 1.1 4 1.2 5 1.3 7 1.4 8 1.6 0 1.7 1 1.8 2 1.9 4 2.0 5 2.1 6 2.2 8 2.3 9 2.5 1 2.6 2 2.7 3 2.8 5 2.9 6 3.0 8 3.1 9 3.3 0 3.4 2 3.5 3

Cross Section Stipa 136 - 135 Plot 3

(47)

IBED

Appendix D: Matlab Interpolation Script

%% I nt er pol ati on met hod co mp ari son % J er oen Kooi j man - 10718397 % Bac hel or pr oj ect Eart h Sci enc es cl ear cl c cl os e all addpat h(' t ool s') l oad I nt er pol ati on. mat

%% Dat a pr epr oc essi ng

i ndex = r andper m( 4711);

sti pa = [ Var Na me2, Var Na me3, Var Na me4] ; dat a1 = sti pa(i ndex( 1: 1413),:)

dat a2 = sti pa(i ndex( 1414: 4711),:)

%% I D W

i dwv al ues = i nt er pi dw( dat a2(:, 1), dat a2(:, 2), dat a2(:, 3), dat a1(:, 1), dat a1(:, 2), - 2)

%% Kri gi ng

X = [ dat a2(:, 1), dat a2(:, 2)]; Y = dat a2(:, 3);

D = vari ogr a m( X, Y,' nr bi ns', 20,' max di st', 1500000,' pl otit', tr ue,' t ype',' ga mma'); xl abel (' Di st anc e ( Lags)'); yl abel (' Vari anc e')

% Vari ogr a m

i ni t _r ange = 100000; i ni t _sill = 25; i ni t _nugget = 0; vari ogr a m_ mo del = ' gaussi an';

[r ange, si ll, nugget , vstr uct] = vari ogr a mf i t( D. di st anc e, D. val ,...

i ni t _r ange,i ni t _sill, D. nu m,...

' nugget',i ni t _nugget ,...

' model ', vari ogr a m_ mo del );

[ Zi , Zi var] = kri gi ng( vst r uct, dat a2(:, 1), dat a2(:, 2), dat a2(:, 3), dat a1(:, 1), dat a1(:, 2));

%% Co mp ari son of i nt er pol ati on met hods

Di ffI DW = dat a1(:, 3) - i dwv al ues; Di ff Kri g = dat a1(:, 3) - Zi ; Er r orI D W = su m( Di ffI DW) ; Er r or Kri g = su m( Di ff Kri g);

%Er r or esti mati on

anov a1([ dat a1(:, 3), i dwv al ues] ) anov a1([ dat a1(:, 3), Zi ]) anov a1( Di ffI D W) ; anov a1( Di ff Kri g);

%Ra nk s u m

i = r anks u m( dat a1(:, 3), i dwv al ues) k = r anks u m( dat a1(:, 3), Zi )