Spatio-Temporal Analysis of Ground Movement Using Unmanned Aerial Vehicle Photogrammetry in Gampong

Ground movement is one of the most frequent disasters causing major damages in Indonesia. Unmanned Aerial Vehicle (UAV) has been widely used as a rapid observation method to obtain detailed characterization of ground movement. Often, active landslide area is difficult to access. This hinders close monitoring and observations of the ground movement. This study aims to demonstrate the use of UAV as tools for monitoring and observations on active ground movement area and to validate the results. For this purpose, the study was conducted at Gampong


Introduction
It is well understood that ground movement or landslide cause significant damage to the lives of people and environment in Indonesia (BNPB, 2021). Throughout 2021, a total of 1.321 landslides occurred in Indonesia due to high rainfall intensity (BNPB, 2021). Understanding the spatial and temporal phenomenon through ground movement detection and monitoring as its base is the starting step of landslide hazard and risk management (Shi & Liu, 2015). However, studies on employing spatial analysis for ground movement/landslide in Indonesia using unmanned aerial vehicle are still limited. Furthermore, in many cases, direct measurement are still difficult to perform. Safety is one the most related to the situation. In fact, understanding the volume of the ground movement and it rates will contribute to a better understanding on ground movement characteristics.
Certain remote sensing data were used to obtain detailed characterization of ground movement. The interferometric synthetic aperture radar (InSAR) technique provides the possibility to measure the deformation of landslide in long-term and large areas with wide coverage, high spatial resolution, and ability to operate under all weather condition (Kang et al., 2017;Squarzoni et al., 2020;Wang et al., 2019). ASTER Satellite imagery enables low costs landslide assessment with medium resolution satellite imagery tools (Alkevli & Ercanoglu, 2011).
Airborne and terrestrial geodetic Light Detection and Ranging scans (LiDAR) are powerful tools for rapidly collecting high densities of precise and high-resolution 3D surface point coordinates, which enable better analyzing surface topography of landslide (Jones & Hobbs, 2021). However, sufficient resolution, repeat rate and low cost are vital aspect to resolve the evolution of small landslide features and provide significant information on landslide dynamics.
Nowadays, the use of Unmanned Aerial Vehicle (UAV), unmanned aircraft has been widely used as a practical earth observation method (Kyriou et al., 2021). This is aimed at obtaining information about landslides by minimizing direct human contact at the landslide site (Afif et al., 2019). This method certainly saves time and cost (Giordan et al., 2020) and is suitable for relatively small area (Joyce. E. et al., 2009) The UAV also has more flexibility in its operation and simultaneously produces high image resolution (Rau et al., 2011). Several remote sensing data that being used in ground movement research in Aceh are (1) LiDAR and satellite imagery to obtain topography maps with 3D visualization for slope stability analysis (Purnama et al., 2022); (2) 30 meter of DEM resolution from USGS to analyze the slope (Amri et al., 2023); and (3) elevation, slope, and slope direction data from DEM by TerraSAR X satellite (Pamela et al., 2018). However, the use of remote sensing data from UAV in ground movement research in Indonesia are still rare.
This study was carried out in Gampong Lamkleng, Aceh Besar. Ground movement in Gampong Lamkleng first occurred on January 11, 2021 after heavy rain (76.2 mm/day) formed a ground fracture. A preliminary team from Universitas Syiah Kuala (USK) on January 13, 2021 found that the depth of the landslide was only approximately 1 meter and on January 20, 2021 the height of the landslide had reached around 2.8 meters with a landslide fracture length exceeding 300 meters (USK, 2021). The movement of the land that occurred required 14 families from 14 houses to evacuate because the position of the house was close to the location of the movement. Ground movement occurs on the slopes adjacent to the Krueng Aceh river with a steep river slope morphology (20⁰ -35⁰). Figure 1a shows the condition of the affected area captured on January 20, 2021. Meanwhile, Figure 1b shows the same location as indicated in a yellow circle captured on February 27, 2021. The geology of the Gampong Lamkleng area belongs to alluvial deposits with blackish brown characteristics, clay to silt grain sizes and not well consolidated (ESDM, 2021). This study is aimed at demonstrating the use of UAV in identifying ground movement rate using UAVs, which are the DEM accuracy from the UAV, movement rate, change in soil volume, and estimated losses. This study also attempts to investigate the accuracy levels of images produced by a rather inexpensive UAV that can be of references for particular geohazards research in the future.

Methods
To conduct this research, a study area was chosen where active ground movement was reported and monitored. This research was conducted at Gampong Lamkleng, Kuta Cot Glie District, Aceh Besar District of Indonesia (see Figure 1). Primary data collected in this research were aerial photographs of ground movement taken using Unmanned Aerial Vehicle (UAV) and Independent Check Point (ICP) data taken using Real-Time Kinematic Global Positioning System (RTK GPS). Furthermore, secondary data were damage and loss assessment parameters referred to the National Agency Disaster Management (BNPB). The area of active ground movement is shown in Figure 2.

Data Collection
Aerial photos were taken using the DJI Phantom 4 UAV with flight planning carried out in navigation mode based on the path created using the DroneDeploy with flights altitude was 67 meters (220 ft) above ground level, forward overlap was 75%, sidelap was 65%, and flight speed was 6.7 m/s. The shape and area of the flight path is adjusted to the area to be studied, which is ± 11 ha. UAV flights were carried out 10 times starting from January 15 to February 27, 2021 for every 3 or 5 days interval. Figure 3 show the UAV flight path and check point (CP) distribution for DEM validation. There are 19 check points measured using RTK GPS to obtain reference elevations for accuracy assessment of DEMs. Check point measurements are only carried out once on the last day of measurement.

Data Processing
Aerial photos are processed based on the photogrammetric method using the Agisoft Metashape software through four stages, which are camera alignment, generating dense point clouds, generating surfaces: Mesh and DEM (Digital Elevation Model), and generating Orthomosaic (Agisoft, 2020). The main products of photogrammetry are Orthomosaic and DEM (Wolf et al., 2014).

Data Analysis
Orthomosaic and DEM were analyzed using the Quantum GIS (QGIS) and ArcGIS applications with the following steps: 1.
DEM accuracy assessment Accuracy of DEMs was evaluated by means of the root mean square error (RMSE) estimator. RMSE is the square root of the average squared difference between the data coordinate values and the coordinate values from independent sources with higher accuracy (BIG, 2020). RMSE was computed comparing x and y coordinates measures on DEM against reference x and y coordinates for nineteen validation point. For every validation point, the reference x and y coordinates were subtracted from the corresponding DEM x and y coordinates. RMSE are measure in the same units as the response. In this study the unit used is meter (m). RMSE was calculated using Eq.
(1) (ASPRS, 2015): Where, is the coordinate in the specified direction of the i th check point in the data set, is the coordinate in the specified direction of the i th check point in the independent source of higher accuracy, n is the number of check points tested, and i is an integer ranging from 1 to n.

Elevation regression analysis
A simple regression analysis was performed on the elevation data to find out how much influence the DEM elevation data accuracy test generated by the UAV had on the RTK elevation data. The dependent variable (X) used is the RTK elevation data with higher accuracy and the independent variable (Y) is the DEM elevation measured on the same day as the RTK measurement. Dependent variable is a variable that measure and predicted to be dependent upon the independent variable, while independent variable is a variable that is manipulated to investigate whether it consequently brings change in another variable. Simple regression analysis was calculated using Eq.
(2) (Sugiyono, 2006): constants a and b were determined using (Eq). 3 and 4: Where: Y is dependent variable, X is independent variable, a is constanta (intercept), intersection with vertical axis, and b is regression coefficient.

Rate and direction analysis
Rate and direction of ground movement were analyzed using digitized ICP data from Orthomosaic. The x and y coordinate values obtained from the ICP, then the average speed value was calculated using Eq. (5) (Syaifullah, 2014): Here, DAB is distance between point A and point B, XA and XB are for X coordinate value at point A and B, respectively, YA and YB are for Y coordinate value at point A and B, respectively.
Ground movement rate was calculated by dividing the distance on each measurement day based on the measurement duration shown in Table 1. The direction of movement is analyzed using the overlay method on orthomosaic imagery on each measurement day to obtain the movement direction vector.

Determination of cross profile and volume area
The orthomosaic and DEM images were analyzed using the QGIS software to depict the cross-elevation profile of each ground movement grid. Grids are drawn with transverse lines based on reference lines defined as boundaries. Grid lines are made with 10 meters distances between the lines. Furthermore, the boundaries of the analysis area are made in accordance with the conditions of ground movement from field observations. Grid lines are then analyzed using terrain analysis tools -cross profiles to obtain elevation values for each specified grid. The elevation value that has been obtained is then exported into an excel file to calculate volume and obtain a graph of changes in movement. The grid lines and the volume calculation area are shown in Figure 4.
Volume is calculated using the average cross-section method (Muda, 2008).The average cross-section is the calculation of volume by means of regular vertical slice at certain intervals. Volume is the result of multiplying the distance or interval to the average area of the cut area.
Where V is the volume of the ground movement, A1 is first cross-section area, A2 is second cross-section area, and L is length from the first to the second cross-section area.

Calculating losses
The loss values were calculated using the Damage and Loss Assessment method for the affected sector determined by (BNPB, 2011), which are the infrastructure sector, the productive economic sector, the social sector, and cross-sectors. Losses calculation is determined by the degree of damage and the economic impact of each individual element. The indicator for calculating losses (inaSAFE, 2018) is described in Table 2.  The losses value was calculated using Eq. (7): Here, Nk total is value of total loss, A total is total damage area, Nk per m2 is value of loss per square meter, and Fk is multiplier factor (%).

DEM accuracy
Out of a total 25 check points scattered at the ground movement location, only 19 check points can be seen from Orthomosaic data. The 19 check points are used as ICP, namely point to assess the accuracy of data from aerial photo processing.  GCP (Ground Control Point) are number of points in a location that have coordinates for the purposes of correcting data and improving the overall image in the rectification process (Natar et al., 2020). ICP (Independent Check Point) is a point on the ground whose coordinates are known and used as quality control from the resulting product (aerial photograph) (BIG, 2020). GCPs are used when processing data, while ICPs used when the data has become a product to obtain the horizontal accuracy from aerial photograph.
GCP plays an important role in bringing coordinates from the original orthophoto grid into the desired coordinate system and for assessing the accuracy of the final Geo-reference (Persia et al., 2020). GCPs are essential for ensuring the accuracy and consistency of geospatial data by aligning the data to a specific coordinate system and correcting geometric distortions. The number and distribution of GCPs affect data accuracy (Tahar, 2013). GCPs that are not evenly distributed in the mapping area will produce Orthophoto data and topographic contours that are not precise and accurate (Nugraha and Putrasakti, 2019). One of the factors that influence the good accuracy of DSM and orthophoto resulting from photogrammetric processing is the number and distribution of GCPs, namely by placing GCPs around the edges of the study area and within the research area with a multilevel distribution to optimize vertical accuracy (Martínez-Carricondo et al., 2018). Research conducted by (Mancini et al., 2013) concluded that the RMSE value decreased when the number of GCP was increased from 5 to 10 points.

Elevation regression value
The results obtained from the linear regression test on DEM elevation from aerial photo processing and elevation from RTK measurement on the same day are shown in Figure (5). It is found that the coefficient of determination (R 2 ) between 19 elevation points are very high, namely 0.999, which is almost close to 1. It can be stated that the accuracy of drone data and RTK measurement data is 99% accurate.
Despite the high R 2 value, there is a difference elevation between RTK and DEM measurement. The difference of RTK and DEM elevation is due to position of the checkpoint (ICP) in the study site. ICPs are placed on steep areas and right at the ground movement location, and some ICPs are placed close to trees and shrubs. In the aerial photograph processing, the filtering and smoothing processes on land cover also effect the DEM elevation.

Ground movement rate and direction
The result of measuring the ground movement rate from January 20 to February 27, 2021 are shown in Fig. (5).
The measurement results are correlated with daily rainfall data obtained from Aceh Besar Climate Station which is 9.2 km away from the study site. The distance between the rainfall stations to the study site is acceptable for regional rainfall purposes. Figure 6 shows the rate of ground movement in Gampong Lamkleng and its correlation with daily rainfall. The most unstable conditions on the ground are in the 3 rd measurement, namely January 20-23, 2021 with a maximum movement rate of 0.69 meter/day. After that, slowly the soil begins to find its stability, marked by the maximum and minimum range of ground movement rates that are getting closer to the median value. Rainfall has an influence on ground movement with the coefficient of determination (R 2 ) is 0.95.  Figure 7 shows the direction of ground movement in 1 measurement period. The most noticeable change in horizontal position was in the 2 nd to 3 rd measurement with movement distance was 0,5 -2 meters. Changes in horizontal position are relatively the same in subsequent periods.
Rain causes changes in groundwater surface and dynamics which reduce stability cause landslides (Guzzetti et al., 2022). There are two types of rain that trigger landslides, namely heavy rain reaching 71-100 mm per day and less heavy rain but lasting for several hours or several days which is the followed by heavy rain (Hidayat and Zahro, 2018). Based on rainfall data collected through BMKG throughout January 2021, rain with an intensity of 70-100 mm per day started occurring form January 5 to January 20, 2021 accompanied by less heavy rain that lasted a long time.

Volume changes
The result of ground movement volume changes measurement on January 15 to February 17, 2021 are shown in Figure 8. It is shows that the largest volume change in ground movement was found on 3 rd measurements 15 -20 January 2021 with a volume change value of 1411,063 m 3 and the smallest volume change on 10 th measurement 23 -27 February 2021. The volume change value is related to daily rainfall where the highest daily rainfall occurs on January 14, 2021 with a rain intensity of 123.1 mm/day. Heavy rain occurred on January 5, 11, 14, 16, 18, 20. Rain intensity began to decrease on January 21 until February 2021. Rainfall is one of the factors causing the disruption of slope stability.
Slope stability is affected by the following factors, namely (1) strength of soil and rock, (2) type of soil and stratification, (3) discontinuities and plane of weakness, (4) groundwater table and seepage through the slope, (5) external loading, and (6) geometry of the slope (Kumar et al., 2022). Based on study site, ground movement occurs on the slopes adjacent to the Krueng Aceh river with a steep river slope morphology (20⁰ -35⁰). The geology of the Gampong Lamkleng area belongs to alluvial deposits with blackish brown characteristics, clay to silt grain sizes and not well consolidated (ESDM, 2021). The slope angle is an essential component of slope stability analysis. As the slope angle increases, the shear stress in soil or other unconsolidated material generally increases as well (Yalcin, 2007). The ground movement volumes observed during this research were significantly decreased after about 1.5 months since its initial stage on January 15, 2021. A slightly increase was identified during another downpour event in early February 2023.

Loss estimations
The result of calculating the estimated value of losses incurred due to ground movement are described in Table  4. Based on Table 4, it was found that the total estimated losses incurred due to ground movement in Gampong Lamkleng amounted to IDR 1,055,854,000. Referring to Perka BNPB No 15 of 2011, the sectors for which damage and losses were calculated in Gampong Lamkleng were permanent houses and trunks with details of 7 units of heavily damaged houses (70%) and 20 km 2 of roads. All affected houses were in the category of heavily damaged because the house could not be reused due to the position of the house which was right at the location of the ground movement.
The ground movement that occurred in Gampong Lamkleng resulted in loss of homes for several families. They were required to move and rebuild their homes in areas that were considered safe from ground movement threats. During the trunk repaired process, the residents could still use alternative roads even though the distance was farther from the initial locations of their houses. In addition, residents who legally owned the land in ground movement area would suffer other losses due to the reduced selling value of the land. Some of the ground movement areas were previously used by residents to plant coconut and banana trees. The location of ground movement that separated the residential area and the farming area would make it more difficult for residents to farm because the landslide area is impassable.  Slope stability is closely related to ground movement. Slope stability is strongly influenced by the shear strength of soil to determine the ability of the soil to resist failure (Haryadi et al., 2019). Slope management is the most important effort in landslide mitigation to stabilize each slope, especially slopes that have the potential to trigger landslides (Sri Naryanto et al., 2020). Management for slope mitigation include the Vetiver system, which is a plant used to stabilize slopes, terracing, building slope protection structures, and closing soil cracks at the end of the dry season. Other landslide mitigation actions are drainage management, vegetation management, and community empowerment management (Susanti et al., 2021).

Limitations of the research
This research prove that the use of UAV did support monitoring on active ground movement area. A relatively affordable UAV provides moderate resolutions that enable researchers or disaster practitioners to observe the process of ground movement/landslide from a close distance. Notwithstanding with the results, apparently, there are some limitations of the results need to be acknowledged. First, the canopy of the vegetations show that the accuracy of the results is lower than an open area. However, since the RTK GPS we used also depend on satellite connections, it is difficult to know which has a more valid results between UAV or RTK GPS. It would be good if further research can be done to compare the UAV results with non-satellite dependent topography equipment. The absence of Ground Control Points in this research could also contribute the resolutions of UAV images.
Furthermore, a rather large RMSE show that the UAV and its camera we used in this research need further process to enhance accuracies of the ground movement observations. It is certain that higher specification of the camera and its geographical system could enhance the accuracies or lower the RMSE. Rainy day also hindered direct measurements since it could damage some sensors of the equipment. Latest development of UAV has been extended to overcome such situation. However, such UAV is certainly less affordable than what demonstrated in this research.

Conclusions
This research attempts to demonstrate the use of a relatively affordable UAV for monitoring active ground movement area. The study applied the UAV photogrammetry in identifying and analyzing ground movement that occur in Gampong Lamkleng of Aceh Besar Indonesia, a study are selected for this research. It was discovered that the accuracy of the DEM resulting from the processing of the UAV data was low because the GCP data was not used. However, this research proves the images could help researchers to identify the ground movement process in spatio-temporal modes. Therefore, for observations and monitoring ground movement, the UAV techniques and equipment were found to be useful. Furthermore, high rainfall has a major effect on ground movement. The effect of high rainfall on January 15 to January 23, 2021 on ground movement were the ground subsidence reaches 3 meters, the maximum movement rate was 0.69 meter/day, and the change in soil volume was 1.411,063 m 3 . The rate and volume of ground movement decreases as rainfall decreases. The estimated value of the losses for each sector affected by ground movement in Gampong Lamkleng is IDR 1.055.854.000. It is recommended to take aerial photos in the morning with clear sky conditions so that the result will not be interrupted by shadows and the use of GCP is required to produce an accurate DEM.