Estimating biovolume using drone technology and dGPS in the Rambla Honda

(1)

Bachelor Thesis Earth Sciences 02/07/2018

Estimating biovolume using drone technology and dGPS in the Rambla

Honda

By: Etienne de Jong (10758364)

Primary supervisor: Dr. L.H. Cammeraat Secondary supervisors: MSc J. Zethof, W.M. de Boer University of Amsterdam – Institute for Biodiversity and Ecosystem Dynamics (IBED)

(2)

2 Abstract

Unmanned Aerial Vehicles (UAV) are an upcoming technology that is becoming very

accessible at low costs. In this research an UAV and a dGPS were used in the Rambla Honda

Basin (Spain) to estimate biovolume by subtracting the obtained Digital Surface Model (DSM)

from the Digital Terrain Model (DTM). This could be used as a less time-consuming way to

estimate biomass within an area. Two plots were measured in the Rambla wherein the elevation

was measured by using a dGPS to construct the DSM. Also both plots where photographed

using an UAV, these photos were processed in Agisoft to construct a point cloud and from there

construct a DTM. The aim of the research was to subtract the two models to get an accurate

estimation of the biovolume within the plot. The research question being: How accurate is

biovolume estimation solely by taking the difference between the DTM and DSM?

After georeferencing of both horizontal axis and interpolating the y-axis the result of the

subtraction of the models was compared to the height of the vegetation measured by hand in

the field. This resulted in an average difference of 56.4 cm between measured and calculated

height, when linear regression was used a R

2

value of 0.0043 was calculated. Therefore, the

conclusion can be made that the estimation is not accurate enough to be meaningful. However,

improvements are suggested to better the accuracy of this method in future experiments.

(3)

3

Results of Plot 1 ... 14 Results of Plot 2 ... 15 Results DTM-DSM ... 16 Field data ... 17 Obtained data ... 18 Discussion ... 21 Recommendations ... 21 Georeference points ... 21 Flight height ... 22 Point density ... 22 Camera ... 22 Vegetation Filter ... 22 Conclusion ... 23 Literature ... 24 Appendix ... 26

(4)

4

Introduction and aim of research

Remote sensing is an important tool used in earth sciences to obtain data (Nijland, Addink, De Jong, & Van der Meer, 2009). In the last decennium UAV’s (unmanned aerial vehicle) have been playing a bigger role in gathering remote sensed data (Colomina & Molina, 2014; Everaerts, 2008). One of the applications of UAV’s has been creating Digital Elevation Models (DEM) of areas and the resolution of the images has been improved greatly (DiBiase & Dutton, 2017). In this research this innovative technology is used to create a quick and easier way to calculate biovolume of Macrochla tenacissima and Anthyllis cytisoides.

The data obtained will be part of the PhD project of Jeroen Zethof who will try to link biovolume to different soil properties and functions. During the fieldwork soil properties were also examined for the PhD research project.

Why is monitoring of biomass important?

Monitoring biomass is an important activity in observing the condition of an environment (Bonham, 2013; Launchbaugh & Kesoju, n.d.). In this article the focus will be on above-ground biomass, which is a good indicator of the volume of carbon stored in an ecosystem and thus it indicates the productivity of the land (De Jong, Pebesma, & Lacaze, 2003). This indication can be of significant importance to, for example farmers, but also for nature conservation or foresters. Aside from productivity biomass can be used to measure competition between species and measure the abundance of species in an area (Kent & Coker, 1992). The amount of canopy cover can even have an influence on erosion (L. H. Cammeraat & Imeson, 1999). Fluctuation in biomass over a course of time can give important information about an area but measuring these fluctuations can be difficult and time-consuming (Cunliffe, Brazier, & Anderson, 2016). Semi-arid areas are a good example of places were biomass is hard to map. These areas are sparsely vegetated and large changes in vegetation structure occur over short periods of time (Cunliffe et al., 2016). Monitoring these semi-arid areas is required because they cover 40% the of terrestrial area and they play a vital role in reducing the amount of CO2 in the atmosphere plus buffer

the effect of climate change functioning as a C-Sink (Ahlström et al., 2015; Cunliffe et al., 2016). More understanding about the processes in these areas will give more insight to almost half of the terrestrial area on earth and bring forth more knowledge about an important CO2 sink.

Why should we monitor biomass with drones?

Remote sensing is a form of collecting spatio-temporal data that has been used for several decades. Remote sensing gives a good overview of an area and can provide a lot of valuable data. Or as stated: “Remote sensing offers the most suitable tool to obtain spatially continuous datasets on vegetation

parameters for environmental monitoring and modelling”(Nijland et al., 2009, p. 779). With new

developments in the field of unmanned aerial vehicles (UAV) a new form of remote sensing is rapidly becoming popular. There are four categories of airborne platforms specified by Gallacher & Khafaga (2015) that are used to do research.

1. Orbital (160+ km height) devices: this category consist mostly of artificial satellites, these however are very expensive and fly at 160 kilometres or higher (Gallacher & Khafaga, 2015; Everaerts, 2008). Therefore, they are outside the budget of many researchers and might not yield accurate data or be appropriate for specific types of research (Gallacher & Khafaga, 2015; Nijland et al., 2009). 2. Stratospherical (~20 km) devices such as Solar UAVs and hot air balloons, these are very expensive

still and are not ideal for small areas (Gallacher & Khafaga, 2015; Everaerts, 2008).

3. Commercial Airspaces (2-15 km) are divided in two groups just like the near ground category we will discuss shortly. The first group is fixed wing planes, manned and unmanned. The second one is rotors, like helicopters. Fixed wing vehicles have the advantage of having longer flight times while rotor vehicles have no minimum speed which improves the quality of images as they are more stable

(5)

5 (Gallacher & Khafaga, 2015). The overall advantage of this category is that they are big enough to carry multiple sensors at once. However, they still are expensive and acquire an experienced pilot.

4. The last category is near ground vehicles (<1km), these are the UAVs that are rapidly developing and expanding their importance. They are relatively cheap and easy to control. By operating them near the surface they produce images with almost no atmospheric interference and the highest spatial resolution ground sampling. Furthermore, drones are light and can therefore be used in all sorts of terrain, also terrain that until now have limited image data because they are remote (Westoby, Brasington, Glasser, Hambrey, & Reynolds, 2012).

The research proposed here on vegetation biomass will be done in relative small areas and therefore the near ground micro UAVs are the best option for monitoring biomass in this area.

Biomassa vs Biovolume

In this thesis aboveground-biovolume is measured, biovolume is the volume that a shrub or a tree takes up. It differs from biomass because biomass is the mass that the physical plant weighs, while biovolume includes the space in between branches, so is the absolute space the plant uses. For plants this biovolume could help to calculate the biomass through formulas made for a specific type of plant. In this research biovolume will be estimated through UAV made images. Biovolume can be an important indicator for biomass (Gallacher & Khafaga, 2015).

Knowledge Gap

The use of UAV’s and structure from motion (SfM) technologies are relative young and upcoming techniques it is important to test and validate these technologies. As mentioned before monitoring biomass can be very important for an indication on the health of an ecosystem, especially in the field of physical geography the usage of UAV’s can revolutionize the way data is captured (Fonstad, Dietrich, Courville, Jensen, & Carbonneau, 2013; Smith, Carrivick, & Quincey, 2016). The technique is rapidly developing and can become a fast and cheap way to monitor changes in biovolume.

Much research has already been carried out using UAV’s for a variety of topics, however most drones used for vegetation research were not equipped with a standard RBG camera but rather with multispectral or LiDAR cameras (Colomina & Molina, 2014; Cunliffe et al., 2016; Deng & Shi, 2013). Furthermore not much research has been done on biovolume of Marcochla tenacissima and Anthyllis

(6)

6

Research question

This research aims to develop and evaluate a (semi-) automated way to calculate the above-ground biomass in an area in the Southeast of Spain using a micro-UAV (unmanned aerial vehicle) and a Differential Global Positioning System (dGPS). A set of field observations of biovolume will be used to validate the method. A Digital Terrain Model (DTM) calculated by UAV images and a Digital Surface Model (DSM) calculated by dGPS will be compared to estimate above-ground biomass. The main research question of this project is:

How accurate can above-ground biovolume of vegetation be assessed by using a rotor UAV equipped

with a RBG camera and by a dGPS in a semi-arid area?

To answer the main research question 4 sub-questions are formulated:

How accurate is biovolume estimation solely by taking the difference between the DTM and DSM?

- What is the biovolume of the vegetation measured by hand?

- What is the biovolume of the vegetation calculated from the different DEMs?

- Can an accurate estimation of biovolume be made from DTM-DSM?

- Is there a difference in accuracy estimation of biovolume between Anthyllis cytisoides and

Macrochla

tenacissima?

With the answers to these questions the goal of this research is to make an accurate estimation of the aboveground biovolume of vegetation using upcoming drone technology.

(7)

7

Study area

The study area is in the Rambla Honda located in the South-East of Spain near the city of Tabernas (Fig. 1). The geology is mainly Mica Schist with slopes varying between 19 and 24 degrees with alluvial fans and floodplains in the valley bottom. The main vegetation types on this site are Marcochla tenacissima on the upper part of the slopes and Anthyllis cytisoides is the main vegetation on the upper part of the alluvial fans (Boer & Puigdefábregas, 2005). The climate in this area is semi-arid having warm summers and mild winters, whereby this part of Spain is the driest area of Europe with annual precipitation around 260 mm and mean annual temperature of 17.8 °C (E. Cammeraat, 2017). The exact location is visualized in the figures below.

Figure 1:Location of fieldwork area by Etienne de Jong

Esri, HERE, Garmin, © OpenStreetMap contributors, and the GIS user community, Source: Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AeroGRID, IGN, and the GIS User Community

Fieldwork area

¯

Legend

a

Fieldwork Area

(8)

8

Figure 2:Locations of plots in the Rambla Honda

Source: Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AeroGRID, IGN, and the GIS User Community

´

0 25 50 100 150 200 Meters

Legend

Plot 1 Plot 2

Plot areas

(9)

9

Methods

UAV Flights and RBG camera

The remote sensing data used in this research was obtained during fieldwork. Two consecutive days were used to capture data of the study area with an UAV (DJI 3 advanced). Flight patterns where set by using the application drone deploy for iPad mini. Flights were done at an altitude of 20m and 50m above take-off elevation, both with an overlap of 70% as recommended by the application and literature (Cunliffe et al., 2016). The application calculates a lawnmower pattern over the area to make sure the whole area will be photographed. Before take-off 4 reference points were placed in the planned fight area. These reference points were measured with dGPS so they could be used as georeference points in ArcMap. These images were used to create a DTM of the area using Structure from Motion (SfM) technology in Agisoft.

Structure from Motion

Structure from Motion (SfM) is a modern technique used in this research to create a point cloud database. This point cloud data is created to construct a DEM in Agisoft. SfM uses images taken from different points of the same object. Combining the images corresponding features out of different images. After it is calculated that there is a 95% chance that the combined features are of the same location, a 3D model is generated (Smith et al., 2016). While SfM has been proved to be a suitable tool in remote sensing it is still very new and might not be as precise as more conservative methods (Fonstad et al., 2013; Smith et al., 2016).

Measuring biovolume by hand

After all flights were completed, the biovolume of the plants was measured by hand. Plants that where standing free in the field were measured manually after the drone had obtained DTM data so there would be no interference while making the DTM. All plants that were measured were given a code and were located with the dGPS to obtain their coordinates and altitude. To determine biovolume the width of the plant was measured at 3 different levels of height. The average width multiplied by the height will determine the biovolume of the sample (Solé-Benet, Contreras, Miralles, & Lázaro, 2009). These heights were measured at 10cm above the soil surface, halfway the height of the vegetation and about 10cm underneath the top of the vegetation. The mean width times the height of the plant was used to calculate the biovolume. This volume was calculated in cm3_.

Differential Global Positioning System

For the construction of the DSM a dGPS was used. The data was obtained using a Topcon HiPer Pro dGPS. A reference station was placed in the middle of the fieldwork area and the mobile station was taken into the field to use real time dGPS. dGPS has better accuracy compared to GPS, for it has a base station added which is linked to the GPS by a radio link (Matosevic, Salcic, & Berber, 2006). This dGPS measured the geo-reference points used to geo-reference the aerial images, the coordinates of the locations and elevation of the plants that were measured by hand.

After all this data was obtained the dGPS was used in two plots within the study area. As seen in Fig. 7 & 9 the two plots were taken on different sites of the study area. These plots were measured with the dGPS to obtain an accurate DSM. Underneath every plant in the plot the elevation was measured (also the ones that were not measured by hand for biovolume).

Using this equation, the biovolume is estimated.

(10)

10

Data processing

Afterwards all pictures were sorted per fight and put in separate files before they were processed in Agisoft (processing reports can be found in the appendix 4). Agisoft was used to generate both a DTM and an orthophoto of the area photographed. The program makes it easy to generate a DTM and orthophoto using the workflow option. This workflow function gives the order of steps that need to be taken to get the sought-after result.

First photos are aligned and when this is done a dense cloud is built using SfM technology. With this dense cloud a mesh and a DEM can be constructed and finally an orthomosaic can be made using all this data. For the flight done at 20m altitude the dense cloud could be formed at high quality to get a significant DTM. The quality level of the construction of cloud determines the amount of points that form the cloud. For the flight at 50m altitude the very high quality dense cloud was used.

ArcMap

In ArcMap, all orthophotos and DTM’s were converted from WGS 1984 to WGS 1984 UTM 30N. This was done because the accurate coordinate system of the study area was WGS 1984 UTM 30N, also the dGPS measured the location in WGS 1984 UTM 30N. After the projected coordinate systems were corrected the dGPS georeferenced points were projected on the orthophotos to georeference the orthophoto. Once the orthophoto was georeferenced the DEM would also be referenced. Once the DEM, orthophotos and dGPS georeferenced points are all corresponding the orthophotos were merged into one orthophoto. This was done so the total orthomap could be georeferenced once more and an 2nd_order

polygon adjustment could be shaped. This resulted in more accurate making of the images.

The data points from the dGPS plots was interpolated to form a raster full of data points using the kriging tool in ArcMap. This data was put in the ArcMap document and the DTM was clipped to the DSM so they had the same size. Finally using the raster calculator, the DSM could be subtracted from the DTM and give an approximation of the height of the vegetation. The figure below shows an overview of the workflow that was executed.

Shifting the data

When overlaying the DTM and the DSM it became clear that, not only the x and y-axis were not corresponding, also the z-axis was not compatible despite the georeferencing points. In ArcMap it was not possible to counter this problem, therefore Matlab was used to make an interpolation. For plot 1 the difference in height between DTM and DSM was varying around 6 meters while, for plot 2 it was varying around 16 meters. The data was not only lifted but also tilted for both plots. To make sure the data matched the bare ground points measured with dGPS were interpolated with the bare ground points on the DSM because these should have the same altitude (figure 3). The bare patch locations were stored as points and used to correct the shift and tilt. For this research the precise method can be found in the bachelor thesis of Adriaan Dekkers (2018).

(11)

11

(12)

12

Figure 4:Work flow image analysing.

After the data was obtained Agisoft and ArcMap were used for image processing. Agisoft produced an orthophoto and a DTM, using ArcMap both where georeferenced. In ArcMap the DSM was also processed and they were clipped. The clip was not perfectly fitting, so interpolation was done in Matlab and after it was corrected, the DTM – DSM was calculated.

(13)

13

Results

Orthophotos (Made with UAV)

The orthomosaic was made using Agisoft. Plot 1 was photographed from an altitude of 20 meters above take-off position while plot 2 was photographed from an altitude of 50 meters above take-off position. The green patches are vegetation Marcochla and Anthyllis among other types while the grey patches are bare soil.

Figure 5:Orthophoto of plot 1, images taken at 20 meters above take-off altitude. Red dots indicate georeference points used to georeference the image

Figure 6:Orthophoto of plot 2, images taken at 50 meters above take-off altitude. No red dots are in the images because they were placed at locations not visible in this image.

Plot 1 Orthophoto

´

0 2 4 8 12 16 Meters Legend Georefrence Points Border

Plot 2 Orthophoto

´

0 2 4 8 12 16 Meters Legend Border

(14)

14

Results of Plot 1

DSM of plot 1 obtained using the dGPS and the kriging tool in Agisoft.

Figure 7: DSM of plot 1 with an altitude between 676.6 meter and 693.6 meter as measured by the dGPS.

DTM of plot 1 obtained using UAV images and Agisoft software. It is clearly visible in this image that the altitude of the DSM and DTM don’t match and differ about 6 meters. Some vegetation patches do show.

Figure 8: DTM of plot 2 with an altitude between 670.9 meter and 688.2 meter measured with the UAV.

DSM Plot 1

´

Legend Border DSM Plot 1 Value High : 693.627 Low : 676.611 5 2.5 0 5 10 15 Meters

´

5 2.5 0 5 10 15 Meters

DTM Plot 1

Legend Elevation (m) Value High : 688.212 Low : 670.888 Border

(15)

15

Results of Plot 2

DSM of plot 2 obtained using the dGPS and the kriging tool in Agisoft.

Figure 9:DSM of plot 2 with an altitude between 687.9 meter and 671.9 meter measuerd by the dGPS

DTM of plot 1 obtained using UAV images and Agisoft software. The higher flight height resulted in less visible vegetation patches compared to the DTM of plot 1. For plot 2 the difference in measured altitude between DTM and DSM are even greater with a difference of 16 meters.

Figure 10:DTM plot 2 with an altitude between 654.7 meter and 670.6 meter measured by the UAV.

±

DSM Plot 2

5 2.5 0 5 10 15Meters Legend Border DSM Area2 Value High : 687.942 Low : 671.894

±

DTM Plot 2

5 2.5 0 5 10 15Meters Legend Border DTM Area2 Value High : 670.561 Low : 654.692

(16)

16

Results DTM-DSM

The results of subtracting the corrected DSM from the corrected DTM. This gives the height of all canopy roofs and other dissimilarities on top of the surface in meters.

Figure 11:Results DTM - DSM plot 1 the elevation is given in meters.

Figure 12:Result DTM - DSM plot 2 the elevation in given in meters.

D

D D

DTM-DSM Plot 1

Legend

D

Location of plants Plot 1 Border Elevation Value High : 1.42499 Low : -0.608521

±

10 5 0 10Meters

D

DTM-DSM Plot 2

5 2.5 0 5 10 15 Meters

±

Legend

D

Location of plants Plot 2 Border

Elevation

Value

High : 2.05249 Low : -1.82886

(17)

17 The DTM-DSM data clearly shows patches of vegetation. The crosses indicate the location were samples of vegetation were taken. The modification done in Matlab has reduced the difference in elevation between DTM and DSM from meters to less then a meter for plot 1. When compared to the orthophoto it shows almost all vegetation is represented in the DTM-DSM model.

Field data

Data of the biovolume in the field was obtained and documented for the two species. All raw data collected in the field are presented in appendix 1 2 and 3. This data also contains vegetation samples that were not taken inside one of the two plots but scattered over the entire hillslope. The average biovolume of Anthyllis cytisoides measured by hand (n=23) is 264062 cm3 _{with a range between 81597}

cm3 _{and 918956 cm}3_{. The average biovolume of Marcochla tenacissima measured by hand (n=25) was}

(18)

18

Obtained data

Table 1:Table of difference between biovolume measured by hand and estimated via DTM-DSM

sample_ID other_remarks height measured by hand(cm) height measured by DTM-DSM(cm) Absolute difference in measured height (CM)

stipa7 1 Plot 2 78 55 23 stipa6 2 Plot 2 75 34 41 anthyllus6 3 Plot 2 58 30 28 anthyllus4 4 Plot 2 60 9 51 stipa5 5 Plot 2 80 75 5 anthyllus5 6 Plot 2 47 18 29 anthyllus3 7 Plot 2 57 8 49 stipa4 8 Plot 2 55 23 32 anthyllus7 10 Plot 2 77 95 18 stipa8 11 Plot 2 96 76 20 stipa9 12 Plot 2 56 5 51 anthyllus8 13 Plot 2 56 8 48 anthyllus9 14 Plot 2 80 27 53 anthyllus10 15 Plot 2 80 21 59 stipa10 16 Plot 2 46 85 39 stipa11 17 Plot 2 105 -1 106 stipa12 20 plot 1 51 0 51 stipa13 21 Plot 1 99 -20 119 anthyllus13 22 Plot 1 80 55 25 average 70 30 47 STD 18 33 28 Min 46 -20 5 Max 105 95 119

(19)

19

Figure 13: Scatterplot of height measured by hand versus height measured by DTM-DSM. Anthyllis cytisoides labeled with an A Marcochla tenacissima is labeled with a M

As clearly visible in the plot there is now linear similarity between the height that were measured in the two different ways with R2 _{being 0.0043. The negative numbers in the plot are due to the negative}

altitude given by DTM-DSM DEM. Because the points are far apart no standard adaption can be made to get a more accurate estimation by DTM-DSM. Had all points been on one line an equation could have been made.

M A A M A A M A M M A A A M M M A R² = 0,0043 y = 0,1196x + 23,327 -40 -20 0 20 40 60 80 100 120 0 20 40 60 80 100 120 H eight me asured b y DT M -DSM( cm )

Height measured by hand (cm)

(20)

20

Figure 14: Scatterplot of height measured by hand versus height measured by DTM-DSM only for Anthyllis

Figure 15: Scatterplot of height measured by hand versus height measured by DTM-DSM only for Marcochla

The results for height assessment by DTM-DSM and by hand are shown in the scatterplots of figures 13, 14 and 15. In figure 13 all results are plotted this results in a R2_{or 0.0043 which does not indicate}

correlation. Figure 14 displays the results for only Anthyllis, with a much higher correlation of 0.32. Figure 15 shows the results for only Marcochla with a correlation of 0.023. If the outliers in figure 16 in the upper right corner and lower right are removed the correlation increases 0.85.

(21)

21

Discussion

In this research the objective was to estimate biovolumes of grass tussocks of Marcochla tenacissima

and Anthyllis cytisoides using an UAV equipped with a RBG camera and a dGPS. Two plots of

approximately 25 by 15 meters were selected in the South-East of Spain. A dGPS and an UAV were used to collect data on elevation. The datasets were used to create a DTM and a DSM of the plots. In the plots vegetation biovolume was measured by hand to get an estimation of the actual present biovolume of the two species. By comparing the DTM with the DSM an indication of biomass was made that is going to be compared with the data measured by hand.

The use of UAV images together with the SfM algorithm is a straightforward method to produce detailed DEMs and Orthophotos for small study sites. Georeference points in the plots collected by a dGPS are required to project the products into the right coordinate system. The use of a dGPS might be laborious but can also be used in a straightforward way to calculate a DSM by using the individual measured points and interpolate them using kriging. The study will compare now the DSM and DTM to assess vegetation biovolumes of the vegetation.

The first thing that was noticed after reviewing the DTM and DSM was that the DTM had on average elevation offset of about 6 meters in plot 1 and 16 meters in plot 2. The origin of this offset is not completely clear, but the cause is probably the lack of georeferenced points in the study area. This difference could not be corrected for using ArcMap or Agisoft alone. The georeferencing tool in ArcMap did resolve the mismatch on the x and y-axis. On the z-axis however, the difference in altitude remained more than 5 meters for plot 1 and up to 16 meters for plot 2. To counter this problem interpolations were made using Matlab focussing on the bare ground patches. This might have had minor influence on the obtained data.

After reconfiguring the DTM and DSM, in such way that they had the same surface elevation, the DSM was subtracted from the DTM. Giving an estimation of the volume of vegetation. As clearly visible the crosses, that show the coordinates where the samples were taken, are mostly on top of elevations, this indicates that the vegetation is measured as an increase in altitude by the UAV data. Therefore the vegetation is noticed by the model. After calculations on the difference in elevation the volume measured by hand does not approach the volume measured by DTM – DSM. The margin of error differs from 5 cm to 119 cm with a mean margin of error of 44 cm. This margin of error clearly shows that in this research there was no success measuring biovolume using a UAV images and dGPS data. The difference is too significant to get meaningful estimations of biovolume. When separating the vegetation types the correlation does improve for Anthyllis. For Marcochla this was not the case but this relation is negatively influenced by three outliers.

Only the vegetation sampled within the plots that were entirely measured with the dGPS could be used. Because of the shift that had to be done using Matlab, the vegetation outside of plots could not be shifted accurately because there was no data on the bare patches outside the plot and the z-axis offset could not be adjusted. This meant that a lot of sampled vegetation could not be used to estimate biovolume. The samples within the plot did not show any explained variance of height with an R2_{value of 0.0043.}

Therefore we did not get to estimate the biovolume.

Recommendations

The results of this research might not be accurate enough to use on the estimation of biovolume, however a lot can be learned from the problems encountered during this research.

Georeference points

The first improvement that can be done for further research is using more georeference points (Fonstad et al., 2013; Gallacher & Khafaga, 2015; Smith et al., 2016). When using the georeferencing tool in ArcMap more points are needed to create a 2nd_{order polynomial fit. When fitting the DTM with both}

(22)

22 the dGPS data and the orthophoto it was necessary to use a 2nd_{order polynomial fit because the data did}

not only shift but also curved. To create a 2nd_{order polynomial fit ArcMap needs eight at least}

georeference points but the more points the more accurate the data. When using the georeferenced points, they should be evenly spread out over the entire area (Fonstad et al., 2013; Gallacher & Khafaga, 2015; Smith et al., 2016). This is one of the reasons why plot one was easier to georeference.

Flight height

Another finding that can be clearly noticed in the results is that the accuracy of the orthophoto of plot 1 is much higher than the one of plot 2. This has to do with the difference in flight height, the flight for plot 1 was executed at an altitude of 20 meters above take-off elevation, while the flight for plot 2 was executed at an altitude of 50 meters above take-off elevation. For further research it is recommended to fly on as a low an altitude as possible, this will make sure all irregularities on the surface will be captured by the camera and be used in the creation of the point cloud. Another problem is the constant flight height. This experiment was done on a slope but with the constant flight height a distortion occurs between the upper and lower part of the slope.

Point density

If this point cloud has a higher density the DSM – DTM calculation would also be more accurate with a higher density of points in the dGPS data. For this research many dGPS points were measured on bare soils and however this was helpful when shifting the z-axis, the vegetation was not always measured accurate enough (Dekkers A., 2018). This might have resulted in a lack of accuracy of surface altitude underneath canopy.

Another improvement for obtaining biovolume data using drones might be to experiment on flat surfaces before using this technology on a hillside. The steep hillslope made it more difficult to georeferenced not only a shift but also a tilt had to be calculated out of the data. If the research for this article would have been carried out in a flat landscape the problem with the tilted DEM’s would have been easier to resolve, as then in that case all borders of the plot would have been at the same altitude.

Camera

The aim of this research was to find a quick and cheap way to measure biomass using UAV technology. Even though a RBG camera is the cheapest option it might not be useful enough for the purpose of measuring biovolume to estimate biomass. There are other camera systems that might be more appropriate for mapping biovolume or even biomass such as LiDAR, multi-spectral cameras or even infrared camera’s (Colomina & Molina, 2014; Enayati, Veissy, & Rahimpour, 2015; Fonstad et al., 2013).

Vegetation Filter

Maybe the best way to improve the outcome of this study is to not use dGPS data at all but use a vegetation filter on the aerial images. Enayati et al. (2015) used solely aerial images to create a DEM of a site in Iran with similar vegetation. Next, they filtered the vegetation out of the DTM using the point cloud they created using the aerial images and made a DSM. This technique could also create a DTM and a DSM which do not have to be shifted or tilted because they are from the same dataset this could prevent an offset between datasets (Deng & Shi, 2013; Enayati et al., 2015).

(23)

23

Conclusion

To conclude, the use of UAV-based DEMs and dGPS based DSM is in principle a fast and straightforward way to map biovolume of vegetation. Technical constraints such as the 6 meter offset between two datasets during data processing do not allow to provide enough evidenc on the obtained accuracy of biovolumes. The main question asked in this research was: How accurate is biovolume

estimation solely by taking the difference between the DTM and DSM? This question is answered by

these sub questions.

What is the biovolume of the vegetation measured by hand?

Even though the data could not be used for this research, data of the biovolume in the field was obtained and documented. The average biovolume of Anthyllis cytisoides measured by hand (n=23) is 264062 cm3 _{with a range between 81597 cm}3 _{and 918956 cm}3_{. The average biovolume of Marcochla tenacissima}

measured by hand (n=25) was 759952 cm3 _{with a range between 71083 cm}3_{and 2452267 cm}3_.

What is the biovolume of the vegetation calculated out of the different DEMs?

This question could not be answered because the difference in height of the vegetation that was measured by hand and the height of the vegetation measured by the UAV is so dissimilar that there would not be an accurate estimation and the difference in height. Is already proof of the failure of the experiment.

Can an accurate estimation of biovolume be made from DTM-DSM?

In this research it has not been managed to construct an accurate estimation of biomass using DTM – DSM method. The difference between the estimated biovolume and the by hand measured biovolume is too great to get an accurate indication. Improvements on the gathering of the data have to be made to get an accurate indication.

Is there a difference is accuracy estimation of biovolume between Anthyllis cytisoides and Marcochla tenacissima?

Due to a lack of reliable field data and technical constraints it is not possible to give an indication of the accuracy of biovolume estimates for the two species. The field data show a difference in average height, but this information is not sufficient to provide estimates of biovolumes. So, no significant difference in the accuracy of the estimation between Anthyllis cytisoides and Marcochla tenacissima can be presented here and further study is required.

In answer to the main question, this research was not successful in calculating an accurate estimation of the biovolume solely by taking the difference between the DTM and DSM. However, with some improvements on the method to both fieldwork and data analysis it might still be possible to measure biovolume using a more time saving method.

Acknowledgement

First of all, I would like to thank dr. Erik Cammeraat for being my supervisor during the writing of my thesis and giving valuable support during the writing process. Next I would like to thank Thijs de Boer for his guidance during the fieldwork and his help and trust using the UAV in the field and giving a helping hand when using Agisoft and ArcMap in the GIS-Studio. Furthermore I would like to thank Jeroen Zethof for all his help and patience during the fieldwork and for giving feedback and advice on writing my thesis. Lastly, I would like to thank my fellow students Adriaan Dekkers, Annabel Isarin, Niels Verweij and Maartje Wadman for giving support during the fieldwork and working with the image analysis software.

(24)

24

Literature

Ahlström, A., Raupach, M. R., Schurgers, G., Smith, B., Arneth, A., Jung, M., … Zeng, N. (2015). Carbon cycle. The dominant role of semi-arid ecosystems in the trend and variability of the land CO₂ sink. Science (New York, N.Y.), 348(6237), 895–9. http://doi.org/10.1126/science.aaa1668 Boer, M., & Puigdefábregas, J. (2005). Effects of spatially structured vegetation patterns on hillslope

erosion in a semiarid Mediterranean environment: a simulation study. Earth Surface Process

Landforms, 30. http://doi.org/10.1002/esp.1180

Bonham, C. D. (2013). Measurements for Terrestrial Vegetation Measurements (second). Wiley-Blackwell. http://doi.org/10.1002/9781118534540

Cammeraat, E. (2017). Fieldwork manual Southeast Spain.

Cammeraat, L. ., & Imeson, A. . (1999). The evolution and significance of soil–vegetation patterns following land abandonment and fire in Spain. CATENA, 37(1–2), 107–127.

http://doi.org/10.1016/S0341-8162(98)00072-1

Colomina, I., & Molina, P. (2014). Unmanned aerial systems for photogrammetry and remote sensing: A review. ISPRS Journal of Photogrammetry and Remote Sensing, 92, 79–97.

http://doi.org/10.1016/J.ISPRSJPRS.2014.02.013

Cunliffe, A. M., Brazier, R. E., & Anderson, K. (2016). Ultra-fine grain landscape-scale quantification of dryland vegetation structure with drone-acquired structure-from-motion photogrammetry.

Remote Sensing of Environment, 183, 129–143. http://doi.org/10.1016/j.rse.2016.05.019

De Jong, S. M., Pebesma, E. J., & Lacaze, B. (2003). Above-ground biomass assessment of

Mediterranean forests using airborne imaging spectrometry: the DAIS Peyne experiment, 24(7), 1505–1520. http://doi.org/10.1080/01431160210145560

Deng, S. S., & Shi, W. Z. (2013). Integration of Different Filter Algorithms for Improving the Ground Surface Extraction From Airborne Lidar Data, XL(June), 105–110. Retrieved from

https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XL-2-W1/105/2013/isprsarchives-XL-2-W1-105-2013.pdf

DiBiase, D., & Dutton, J. A. (2017). The Nature of Geographic Information | The Nature of

Geographic Information. PennState College of Earth and Mineral Science. Retrieved from

https://www.e-education.psu.edu/natureofgeoinfo/

Enayati, H., Veissy, M., & Rahimpour, F. (2015). Design a filter to detect and remove vegetation from ultra-cam-x aerial images’ point cloud to produce automatically digital elevation model. In

International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives (Vol. 40, pp. 159–162).

http://doi.org/10.5194/isprsarchives-XL-1-W5-159-2015

Everaerts, J. . (2008). The use of unmanned aerial vehicles (uavs) for remote sensing and mapping.

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XXXVII(Part B1), 1187–1192. Retrieved from

http://www.isprs.org/proceedings/XXXVII/congress/1_pdf/203.pdf

Fonstad, M. A., Dietrich, J. T., Courville, B. C., Jensen, J. L., & Carbonneau, P. E. (2013).

Topographic structure from motion: a new development in photogrammetric measurement. Earth

Surface Processes and Landforms, 38(4), 421–430. http://doi.org/10.1002/esp.3366

Gallacher, D., & Khafaga, T. (2015). Ecological monitoring of arid rangelands using fixed-wing micro- UAVs (drones) in the MENA region. Innovation Arabia 8, Dubai, UAE: Hamdan Bin

Mohammed Smart University, (February 2015), 181–197.

(25)

25 Kent, M., & Coker, P. (1992). Vegetation Description and Analysis, a Practical Approach (Third).

London: Belhaven Press.

Launchbaugh, K., & Kesoju, S. R. (n.d.). Why Measure Biomass? Retrieved June 19, 2018, from http://www.webpages.uidaho.edu/veg_measure/Modules/Lessons/Module

7(Biomass&Utilization)/7_2_Why Measure Biomass.htm

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. http://doi.org/10.1109/TIM.2006.880918

Nijland, W., Addink, E. A., De Jong, S. M., & Van der Meer, F. D. (2009). Optimizing spatial image support for quantitative mapping of natural vegetation. Remote Sensing of Environment, 113(4), 771–780. http://doi.org/10.1016/j.rse.2008.12.002

Smith, M. W., Carrivick, J. L., & Quincey, D. J. (2016). Structure from motion photogrammetry in physical geography. Progress in Physical Geography, 40(2), 247–275.

http://doi.org/10.1177/0309133315615805

Solé-Benet, A., Contreras, S., Miralles, I., & Lázaro, R. (2009). Organic Waste as Amendements for Limestone Quarry Restorarion in Semi-Arid Environments. 1st Spanish National Conference on

Advances in Materials Recycling and Eco-Energy. Retrieved from

http://www.eeza.csic.es/Documentos/Publicaciones/Solé-et-al(RECIMAT-2009).pdf

Westoby, M. J., Brasington, J., Glasser, N. F., Hambrey, M. J., & Reynolds, J. M. (2012). “Structure-from-Motion” photogrammetry: A low-cost, effective tool for geoscience applications.

(26)

26

Appendix

Appendix 1

Table 2:Raw field data measured by hand

sample_ID diameter_near_ground (cm) diamete_halfway (cm) diameter_top

(cm) height(cm) other_remarks biovolume (CM^3)

stipa1 107 77 77 83 628227 stipa2 92 92 83 85 673285 stipa3 125 125 79 76 914035 anthyllus1 62 71 57 74 296822 anthyllus2 77 70 58 50 233472 stipa4 35 45 45 55 8 95486 anthyllus3 56 81 55 57 7 233472 anthyllus4 47 57 40 60 4 138240 stipa5 94 98 70 80 5 610168 anthyllus5 33 51 41 47 6 81597 anthyllus6 42 61 42 58 3 135494 stipa6 72 81 68 75 2 407008 stipa7 140 95 90 78 1 915416 anthyllus7 43 80 81 77 10 356048 stipa8 133 120 105 96 11 1367082 stipa9 31 63 60 56 12 147566 anthyllus8 31 63 60 56 13 147566 anthyllus9 35 71 63 80 14 253875 anthyllus10 40 60 48 80 15 194702 stipa10 47 47 30 46 16 78588 stipa11 120 127 72 105 17 1187211 anthyllus11 60 90 60 74 18 362600

(27)

27 anthyllus12 66 87 72 67 19 376875 stipa12 40 43 29 51 20 71082 stipa13 114 102 108 99 21 1154736 anthyllus13 48 77 35 80 22 227555 stipa14 172 140 128 114 23 2452266 anthyllus14 37 72 61 67 24 215144 1HM 70 75 80 72 405000 1HA 60 65 70 65 274625 1FM 94 83 50 102 583995 1FA 40 70 90 85 377777

2AM 120 127 100 67 2 naast elkaar 896378

2FM 100 107 100 95 994850 2HM 100 92 79 47 383525 2HA 76 116 100 97 918956 3HM 90 90 80 85 638444 3HA 55 70 75 70 311111 4HM 70 90 100 85 638444 4HA 35 65 55 70 186861 3FM 130 130 110 90 1369000 3FA 50 60 50 70 199111 4FM 110 100 80 90 841000 4FA 40 60 35 90 182250 5FA 50 65 55 55 176611 5FM 80 85 60 75 421875 5HM 115 115 115 85 1124125 5HA 50 65 55 60 192666 average 73 82 69 74 522338

(28)

28 Appendix 2

Table 3:Data Measured by hand of _Marcrochla

sample_ID diameter_near_ground (cm) diamete_halfway (cm) diameter_top (cm) height(cm) biovolume (CM^3)

stipa1 107 77 77 83 628227 stipa2 92 92 83 85 673285 stipa3 125 125 79 76 914035 stipa4 35 45 45 55 95486 stipa5 94 98 70 80 610168.89 stipa6 72 81 68 75 407008 stipa7 140 95 90 78 915416 stipa8 133 120 105 96 1367082 stipa9 31 63 60 56 147566 stipa10 47 47 30 46 78588 stipa11 120 127 72 105 1187211 stipa12 40 43 29 51 71082 stipa13 114 102 108 99 1154736.00 stipa14 172 140 128 114 2452266 1HM 70 75 80 72 405000 1FM 94 83 50 102 583995 2AM 120 127 100 67 896378 2FM 100 107 100 95 994850 2HM 100 92 79 47 383525 3HM 90 90 80 85 638444 4HM 70 90 100 85 638444 3FM 130 130 110 90 1369000 4FM 110 100 80 90 841000 5FM 80 85 60 75 421875 5HM 115 115 115 85 1124125 average 96 94 80 80 759951 STD 35 27 25 18 523308 Min 31 43 29 46 71082 Max 172 140 128 114 2452266

(29)

29 Appendix 3

Table 4:Data measured by hand of Anthyllis

sample_ID diameter_near_ground (cm) diamete_halfway (cm) diameter_top (cm) height(cm) biovolume (CM^3)

anthyllus1 62 71 57 74 296822 anthyllus2 77 70 58 50 233472 anthyllus3 56 81 55 57 233472 anthyllus4 47 57 40 60 138240 anthyllus5 33 51 41 47 81597 anthyllus6 42 61 42 58 135494 anthyllus7 43 80 81 77 356048 anthyllus8 31 63 60 56 147566 anthyllus9 35 71 63 80 253875 anthyllus10 40 60 48 80 194702 anthyllus11 60 90 60 74 362600 anthyllus12 66 87 72 67 376875 anthyllus13 48 77 35 80 227555 anthyllus14 37 72 61 67 215144 1HA 60 65 70 65 274625 1FA 40 70 90 85 377777 2HA 76 116 100 97 918956 3HA 55 70 75 70 311111 4HA 35 65 55 70 186861 3FA 50 60 50 70 199111 4FA 40 60 35 90 182250 5FA 50 65 55 55 176611 5HA 50 65 55 60 192666 Average 49.26 70.74 59.04 69.09 264062 STD 13.05 13.70 16.64 12.83 164519 Min 31 51 35 47 81597 Max 77 116 100 97 918956

(30)

30 Appendix 4A

Processing Report Plot 1

28 June 2018

(31)

31

Survey Data

Fig. 1. Camera locations and image overlap.

Number of images:

Flying altitude:

Ground resolution:

Coverage area:

51 22.5 m

8.67 mm/pix

3.66e+03 m²

Camera stations:

Tie points:

Projections:

Reprojection error:

51 33,119

184,125

0.643 pix

Camera Model

Resolution

Focal Length

Pixel Size

Precalibrated

FC300S

(3.61mm)

4000

3000

x

3.61 mm

1.56 x 1.56

μm

No

Table 1. Cameras.

(32)

32

Camera Calibration

Fig. 2. Image residuals for FC300S (3.61mm). The colors

indicate from blue to red increasing values of camera

calibration residuals.

FC300S (3.61mm)

51 images

Type

Resolution

Focal Length

Pixel Size

Frame

4000 x 3000

3.61 mm

1.56 x 1.56

μm

Value Error Cx Cy B1 B2 K1 K2 K3 P1 P2 F 2311.25 Cx -21.1488 0.26 1.00 0.06 0.06 0.51 0.10 -0.10 0.12 -0.01 0.01 Cy 41.3016 0.4 1.00 -0.67 0.14 0.37 -0.05 0.05 0.00 0.09 B1 -11.4719 0.12 1.00 -0.04 -0.22 0.04 -0.01 -0.05 0.13 B2 0.744362 0.096 1.00 0.09 -0.05 0.06 -0.16 -0.09 K1 -0.00948153 5.7e-05 1.00 -0.80 0.78 0.05 0.05 K2 0.00542332 9.9e-05 1.00 -0.98 -0.04 -0.03 K3 0.00671422 6.5e-05 1.00 0.04 0.04

(33)

33 P1 -0.000966334 1.9e-05 1.00 0.10 P2 -0.000999079 1.9e-05 1.00

Table 2. Calibration coefficients and correlation matrix.

Camera Locations

Fig. 3. Camera locations and error estimates.

Z error is represented by ellipse color. X,Y errors are represented by ellipse shape.

Estimated camera locations are marked with a black dot.

X error (cm)

Y error (cm)

Z error (cm)

XY error (cm)

Total error (cm)

27.8298

17.9831

23.7394

33.1344

40.7609

Table 3. Average camera location

error. X - Longitude, Y - Latitude, Z -

Altitude.

10 m -70 cm -56 cm -42 cm cm -28 cm -14 cm 0 cm 14 28 cm 42 cm cm 56 cm 70 x 5

(34)

34

Digital Elevation Model

Fig. 4. Reconstructed digital elevation model.

Resolution:

1.73 cm/pix

Point density:

33.3 points/cm²

664 m 693 m

(35)

35

Processing Parameters

General Cameras 51 Aligned cameras 51 Markers 4

Coordinate system WGS 84 (EPSG::4326)

Rotation angles

Point Cloud Yaw, Pitch, Roll

Points 33,119 of 35,466

RMS reprojection error 0.282636 (0.642657 pix)

Max reprojection error 0.849616 (18.1836 pix)

Mean key point size 2.49147 pix

Effective overlap

Alignment parameters 6.00916

Accuracy Highest

Generic preselection Yes

Reference preselection Yes

Key point limit 40,000

Tie point limit 4,000

Matching time Dense

Point Cloud 8 minutes 49 seconds

Points

Reconstruction parameters 15,498,986

Quality High

Depth filtering Moderate

Depth maps generation time 2 hours 3 minutes

Dense cloud generation time

Model 9 minutes 27 seconds

Faces 72,989

Vertices

Reconstruction parameters 36,848

Surface type Height field

Source data Sparse

Interpolation Enabled Face count 90,000 Processing time DEM 1 seconds Size 5,670 x 5,481 Coordinate system

Reconstruction parameters WGS 84 (EPSG::4326)

Source data Dense cloud

Interpolation Enabled

Processing time

Orthomosaic 20 seconds

Size 7,110 x 7,550

Channels

Reconstruction parameters 3, uint8

Blending mode Mosaic

Surface Mesh

Enable hole filling Yes

Processing time

Software 1 minutes 21 seconds

Version 1.4.0 build 5650

(36)

36

Appendix 4B

Agisoft PhotoScan

Processing Report Plot 2

28 June 2018

(37)

37

Survey Data

Fig. 1. Camera locations and image overlap.

Number of images:

Flying altitude:

Ground resolution:

Coverage area:

41 75.2 m

2.94 cm/pix

0.0458 km²

Camera stations:

Tie points:

Projections:

Reprojection error:

41 34,311

154,905

0.549 pix

Camera Model

Resolution

Focal Length

Pixel Size

Precalibrated

FC300S

(3.61mm)

4000

3000

x

3.61 mm

1.56 x 1.56

μm

No

Table 1. Cameras.

(38)

38

Camera Calibration

Fig. 2. Image residuals for FC300S (3.61mm). The colors

indicate from blue to red increasing values of camera

calibration residuals.

FC300S (3.61mm)

41 images

Type

Resolution

Focal Length

Pixel Size

Frame

4000 x 3000

3.61 mm

1.56 x 1.56 μm

Value Error F Cx Cy K1 K2 K3 P1 P2 B1 -9.89662 B2 -4.15128 F 2253.83 1.3 1.00 0.1 5 -0.45 -0.25 0.17 0.29 -0.17 0.10 Cx -29.9276 0.23 1.0 0 -0.16 -0.10 -0.03 0.12 0.05 0.06 Cy 41.8905 0.31 1.00 0.34 -0.12 -0.08 0.02 0.10 K1 -0.00979011 6e-05 1.00 -0.78 0.63 0.01 0.04 K2 0.00492227 9.5e-05 1.00 -0.88 -0.07 0.01 K3 0.00572063 6e-05 1.00 0.01 0.04

(39)

39

P1 -0.00158542

2.3e-05 1.00 0.05

P2 -0.00160728

1.8e-05 1.00

Table 2. Calibration coefficients and correlation matrix.

Camera Locations

Fig. 3. Camera locations and error estimates.

Z error is represented by ellipse color. X,Y errors are represented by ellipse shape.

Estimated camera locations are marked with a black dot.

X error (cm)

Y error (cm)

Z error (cm)

XY error (cm)

Total error (cm)

19.9152

20.8308

28.9031

28.8191

40.8158

Table 3. Average camera location

error. X - Longitude, Y - Latitude, Z -

Altitude.

m 50 -80 cm -64 cm cm -48 -32 cm -16 cm 0 cm 16 cm 32 cm 48 cm 64 cm 80 cm x 20

(40)

40

Digital Elevation Model

Fig. 4. Reconstructed digital elevation model.

Resolution:

2.94 cm/pix

Point density:

11.6 points/cm²

Processing Parameters

General

Cameras 41

Aligned cameras 41

Rotation angles Point Cloud

Yaw, Pitch, Roll

Points 34,311 of 36,141

RMS reprojection error 0.283162 (0.548713 pix)

Max reprojection error 1.04391 (12.5308 pix)

Mean key point size 2.11371 pix

Effective overlap Alignment parameters

4.72765

Accuracy Highest

Generic preselection Yes

630 m 698 m

(41)

41

Reference preselection Yes

Key point limit 40,000

Tie point limit 4,000

Adaptive camera model fitting Yes

Matching time 5 minutes 2 seconds

Alignment time

Optimization parameters 13 seconds

Parameters f, cx, cy, k1-k3, p1, p2

Fit rolling shutter No

Optimization time

Dense Point Cloud 1 seconds

Points

Reconstruction parameters 70,265,894

Quality Ultra High

Depth filtering Aggressive

Depth maps generation time 4 hours 11 minutes

Dense cloud generation time

DEM 1 hours 26 minutes

Size 7,468 x 11,524

Coordinate system

Reconstruction parameters WGS 84 (EPSG::4326)

Source data Dense cloud

Interpolation Enabled

Processing time

Software 1 minutes 34 seconds

Version 1.4.0 build 5650

Estimating biovolume using drone technology and dGPS in the Rambla Honda