Change Detection of Land Cover at Flood Potential Areas using Multitemporal Image Data in East Jakarta City

East Jakarta City is one of 6 cities / regencies in the. The city of East Jakarta as one of the cities in the Special Capital Province of Jakarta has a significant growing population. Changing of land cover might result on flood disaster in related with the growth of population. Main purpose of the study is to determine variable that influence the flood height of an area with imagery data in the period of four decades – 1990, 2000, 2020, and 2020. The first step uses land cover classification analysis to identify land cover in the City of East Jakarta. Then, land use analysis uses change detection methods to determine the changes that occur in each land class. The second step of spatial analysis with the ordinary least square method and geographical weighted regression covering 6 variables – flood height, rainfall, contour, river water level, land cover and population density, in order to find out the variables that influence flooding in the research area. The land cover in the East Jakarta City in the period of 1990-2020 had a significant changed, especially for increasing of built-up area. Since total of the area is fixed, the wider built-up area means the lessen areas for other land uses such as vegetation, water body and un-built-up areas. Then, variables that affect flooding in each district are differ for each district. For district of Cakung, Makasar, Kramat Jati, Pulo Gadung is related to contour of the region. Then, at district of Ciracas, Cipayung, Duren Sawit are due to area of land developed. Next, at Jatinegara and Pasar Rebo Districts are about population density, and at Matraman District is rainfall variable. Understanding cause of the flood might give a better approach on propose a flood management disaster strategy that it might conduct in the future reseach. Keywords— Flood , remote sensing, land cover classification, change detection

INTRODUCTION Rapid growth of population can derive a problem in the city. City of East Jakarta has annual growth population rate at about 4.2% in the period of 2010-2019 [1]. Changing land use has also changing groundwater compartments. At about 40% of the population in Jakarta depend on groundwater resources, wherein this percentage is higher than ground water supply of upstream cities such as Bogor City and Depok City, which is water deficit occurred due to imbalance between abstraction and aquifer replenishment [2]. Flood is events to be inundated by a ridge of land that usually dried. Flood can be caused by rising sea levels as well as the volume of water displaced due to the inability of the absorption of water into the ground [3].
The East Jakarta is the 1st city with the highest disaster risk index value and the 2nd city with the highest flood disaster risk index value in DKI Jakarta Province [4]. Jakarta has experienced many severe river flood events due to heavy rains, especially in 1996,2002,2007,2013 and 2014 [5]. In particular, the floods that occurred in January 2013 resulted with 40 deaths, 45,000 refugees, and substantial economic damages. Flooding in February 2007 also caused extensive economic losses, increasing between 4.1 and 7.3 trillion IDR [6]. The floods in 2013 caused a loss of 15 billion IDR, and the disasters which struck in 2014 reached 100 million IDR per day [7].
Potential flood management efforts include 4 important aspects, namely i) potential for philosophical, ii) structural, iii) non-structural, and iv) socio-cultural handling [8]. The potential for handling philosophy is housing settlement revitalization and housing relocation in the flood-prone areas. The potential structural treatments consist of normalization, build reservoir, control dams, retaining dams, absorption wells, and bio pores. The potential of nonstructural measures comprises conservation and rehabilitation of forests and land in watersheds, purchase of land to expand conservation land and forest colonies. The potential for handling socio-culture is strengthening of groups and community cadres who care about environment, community assistance in behave pro environmental conservation. Therefore, it is important for this research to identify and analyze land cover, detect land cover changes and investigate spatial relationship between land cover and potential flood areas in the East Jakarta City to develop a sustainable liveable city.

II.
METHOD There are two primary data for the study consist of i) Landsat 5 TM imagery data for year 1990, Landsat 7 ETM + in 2000, 2010, and ii) Landsat 8 OLI TIRS imagery for year 2020. Image data that is used is May imagery with a percentage of cloud cover 0%. The six variables are flood height, rainfall, area height, river water level, land cover area, and population density. Then, the analytical methods are i) land cover classification, ii) land cover change (change detection), iii) ordinary least square and iv) geographically weighted regression spatial analysis.

A. Analysis of Land Cover Classification
Land cover refers to natural and artificial objects found on the surface of the earth, all of which can be observed by remote sensing. The analysis that is used in the study is the supervised classification. The maximum likelihood method considers the average value and the inter-class and channel safety (covariance) [9]. Then, the Maximum Likelihood Method based on the normal distribution (Gaussian) which estimates the probability function of each class [10].

B. Analysis of Land Cover Change (change detection)
Land cover change is the process of changing the area of an area either increasing or decreasing on a land cover and use [11]. This can be influenced by natural processes such as influence of climate, volcanic eruption, sea-level changes, environment, and others [12].
Among the global thresholding methods, the Otsu variant class model is the most widely used. This model looks for the optimal threshold by maximizing variance between classes. This model is nonparametric, without supervision and can choose the threshold automatically. Change detection can be divided into two forms, namely in the form of increase and in the form of reducing/decreasing [13]. Information on land cover change is used as one of the data to determine the characteristics of an area so that the direction of planning to be carried out is better.

C. Ordinary Least Square and Geographically Weighted Regression Spatial Analysis
The most widely known and applied regression technique is ordinary least squares (OLS). In the spatial regression realm, the OLS technique produces a global regression model from the area of observation. The global regression model is an introduction to spatial regression that produces local models [14].

III.
RESULT AND DISCUSSION

D. Overview of East Jakarta
The East Jakarta City is one of the cities in DKI Jakarta Province which has 10 districts, namely Cakung, Cipayung, Ciracas, Duren Sawit, Jatinegara, Kramat Jati, Makasar, Makasar, Matraman, Pasar Rebo, and Pulo Gadung as can be seen in Figure 1. Administrative boundaries are as follows: North Side : Central Jakarta City and North Jakarta City Southern Side : Bogor Regency, West Java Eastside : South Jakarta City Westside : Bekasi City, West Java  Figure 1 depicts the height of altitude of the area that can be categorized into three groups. Firstly, at about 20,98% (3.944,9 Ha) lies at the height of 1 -7 meter above seal level, and there is the smallest percentage area lies at the height of 65 -74 meter above sea level (2,3% or 432,5 Ha). Then, majority of it lies at 1 -18 meter above sea level (52,33% or 9.839,6 Ha).
Total number of populations of the area are 2.937.859 inhabitants within total area of 18.803 Ha [15]. The average population density is about 15,624 people/Ha, with the most densely populated is Matraman (31,112 people/Ha). Into more detail, it can be seen in Table 1.

E. Analysis of Land Classification
Combination of color composites (composite bands) in the study is Red 7, Green 5, Blue 3 with a basic reason for facilitating researchers to analyze classification of land cover in satellite imagery. The land cover analyzed is divided into 4 land classes, specifically developed land, vegetation, water bodies, and undeveloped land, as can be seen in Figure 2 and Figure 3. The classification method that is used is supervised classification maximum likelihood. Vol. 9 Issue 07, July-2020 Table 2 indicates land cover changing within 4-decade from 1990 -2020. Amongst four variables of the land cover, it is only one variable which shows increasing development area in the significant valuethe built-up area. Since the total are is fixed, in general we may see that the others three variable have been decreasing, with one exceptional condition whereby the water body in the period of 1990 -2000 had increased at about 23%. Drastically changing has been occurring at the vegetation total area between decades from 1990 -2020, at about (-) 32,7%, (-) 36,8%, (+) 7,47%, respectively.
Value the producer's and user's accuracy of each class can be seen in Table 3 with values >70%. Based on the calculation of the accuracy test, the value obtained for overall accuracy is 94,62% and the kappa coefficient value is 0,8999 or 89,99%. The minimum value for receiving a remote sensing-based mapping is 85%, so that the data from the guided classification conducted in the study can be accepted or used [16].  Detection of change can be divided into two forms, namely, increase and decrease. Data entered to compare land cover changes are two years of data, for example, 1990 and 2020. The auto-thresholding method is Otsu's the results of land change analysis can be seen in Figure 4. In general, it is indicated that total land cover of built-up area instead of changed to another land use, its area has been getting wider and wider occupied the others. Based on Table 4, land cover of the built-up area from 1990 to 2000 has been increasing significantly at every decade. Even though the number of percentages per decade has been lessening, but the number of built up area has been getting bigger and bigger with percentage of changing at about 18,53%, 10,24%, 8,24%, respectively. It might assume that land-use change occurred due to continuous growth of population, hence to support their activity the needs of housing settlements and infrastructures also increase. . Increases and reductions are influenced by seasonal conditions of the annual image that has been used, it caused total area of the water bodies that mostly in the form of lakes/reservoirs and rivers will also be different. In one hand, decreasing land cover of water body in the period of 2000-2010, it might cause by the changing of reservoir at Banjir Timur Kanal that is covered by vegetation, so that the satellite analysis read the data of 2000 as a vegetation area. On the other hand, within the period of 2000 -2010, Conversely, due to rain pouring there were some puddle points on the road or outside rivers, reservoirs/lakes, that the satellite analysis read it as the water body.

G. Spatial Analysis 1)
Ordinary Least Square (OLS) Independent variables (X) that are estimated to have positive or negative linear relationships to the flood height of an area have been analyzed by OLS regression. The height of flood is placed as dependent variable in the regression analysislabelled as Y. Independent variables that is allowed into the model are only two because in the ArcGIS application requires value of the degree of freedom of the model have to be greater than 2. OLS diagnostics of the best model is as described in Table V. OLS regression analysis results in Table 5 show significant values marked by probability values below 0.05. The significant value plays a role in testing the results of subsequent regressions.  Table 6 OLS results show that the height of the region has a negative relationship, while the variable area of builtup has a positive relationship. This means that the greater value of the height of a certain area (X1) indicates the lower the height of the flood. In contrary, the greater value of builtup area (X2) indicates the higher the height of the flood. The results of heteroskedasticity can be seen in Table 7 that the BP Koenker value is above 0.05. Meaning, in the model, there is no heteroskedasticity and the value of stationary independent variables in each unit of analysis. The significance of the independent variable is seen from the probability value in the model results. However, it should be noted also the figures from Koenker (BP). Values from Koenker (BP)> 0.05 then Robust_Pr is not used. The probability value of the variable height of the area and the area of built land has a probability value above 0.05 then Robust_Pr is used because it has a value below 0.05 which is equal to 0.035655 * for the height of the area and 0.020644* for the area of the developed land. The multicollinearity value in the model can be seen from the VIF of each variable. In Table 8, it can be seen that the VIF value of the variables in the model is less than 10, so there is no multicollinearity among the independent variables [13]. If there is multicollinearity, the variable can be said to be unfavorable because it does not explain anything. The results of multicollinearity can be seen that the VIF values of the two independent variables are below the value of 10. So, there is no multicollinearity in each variable and the resulting model can be used.
Residual is the difference between the existing value and the predicted value of the model. Residual values from OLS modeling must be normally distributed. Because the more the residual value away from the number 0, the resulting model is not good. The test can be performed with Bell Curve or Jarque-Bera values.  Table 9 depicts that results of the model residual normality have a Jarque-Bera value with above a significant value of 0.05 [14]. Meaning that the model residuals are normally distributed and there is no bias in the resulting model. Model performance is indicated by adjusted R2 and AICc values. Model performance test results in Cipayung District are as illustrated in Table 10.  The performance of the model can be seen in Table 10, the adjusted R2 value of the model is 40.65%. It means that 40.65% of the model has explained the phenomenon of flood height in Cipayung District. The AICc value of the model is 79.30. The greater the value, the better the model will be produced. The coefficient value of each variable shows how much influence the variable has on the flood height in Cipayung District. From the OLS results, it was found that the greatest influence to minimize the height of flooding that occurred was the height of the region, which amounted to 1,196955. The relationship between the area of built-up land and flood height is directly proportional. Thus, the higher the value of the area of built-up area, the higher the height of the flooding that is occurred in Cipayung District.

2)
Geographically Weighted Regression (GWR) The global regression model that has been obtained from the OLS process can be broken down in to local regression model using the GWR method. Calculation of the local model is based on an assumption that the effect of independent variable toward dependent variable will be different at each location [17]. The local model will also able to describe the condition of each village in a better way and it will cover inaccuracy of the global regression model.  Table 11 shows the predicted value of flood height that is obtained by entering the observation value of the independent variable into the model. The difference between the observed value and the predicted value is the residual value which is then standardized into the residual standard value. Residual standard values must be randomly distributed. If the residual standard values appear to be clustered, spatial autocorrelation occurs and the resulting GWR model is not good.  Table 12 shows that there are some differences in the value of the coefficient of altitude in each sub-district. Meaning that the results give an idea of which areas are contributing more to the high and low floods of an area. In summarize, the greater the coefficient value, the lower the height of the flood. The spatial autocorrelation test of the GWR model in Figure 5 uses the formula from Moran's I. The desired result of the test is a randomly distributed residual standard value. When it is viewed from the results of the spatial autocorrelation test presented in the figure, it appears that the standard residuals in the GWR model have been spread evenly with a significance value of 0.148 and a z-score of -1.444. The data from the spatial autocorrelation test concluded that the resulting model is valid and can be used.

IV. CONCLUSION
Due to the growth of population, the condition of land cover in the East Jakarta City in the period of 1990-2020 had a significant changed, especially for increasing of the land cover of built-up area. Since total of the area is fixed, the wider built-up area means the lessen areas for other land uses such as vegetation, water body and un-built-up areas.
The results of OLS and GWR analyzes might give a specific note of the variables that affect flooding at each district, as follows. Knowing each cause of flooding due to research of change detection of land cover might give a better approach on dealing with flooding in a certain area, so that the planning process might heading to the correct direction. Hence, for the future research, result of the research gives a quite significant input in order to develop a more appropriate flood management disaster strategy in the level of both district as well as city level.