Comparison of Holt Winters and Simple Moving Average Models to Identify The Best Model for Predicting Flood Potential Based on The Normalized Difference Water Index.

− Flood is a condition in which water cannot be accommodated in a drainage channel such as a river or river. An area is said to be flooded if the water in the area is inundated in large quantities so that it can cover all or most of a large area. Determining forecasting or prediction on a potential in the long or short term, especially changes in water content levels in an area, requires a method, model, or approach that must be well tested. The lower the error value in a model, the better the model for testing a forecast. One of the data that can be used for analysis of potential flood models is the use of remote sensing data with technology from Landsat 8. The advantage of sensing data from Landsat 8 is that it has data good history and allows to see changes in land cover from year to year in an area. The purpose of this study was to determine the best model for forecasting the potential for flooding in an area using the Holt Winters model and the Simple Moving Average. The result of this research is that the RMSE, MAE, MAPE, MSE values in the Holt Winters model are 0.03598683, 0.02748707, 0.13944356, 0.00129505 while the RMSE, MAE, MAPE, MSE values on the Simple Moving Average are 0, 09681483, 0.06338657, 0.53775228, 0.00937311. The Holt Winters model is the best model of the Simple Moving Average because the forecast error value has a low value.


I. INDTRODUCTION
The current rate of global warming is extremely concerning. Flooding is one of the many natural disasters that have struck Indonesia as a result of the effects of global warming. A flood is a condition in which a region is inundated by large quantities of uncontrolled water flow, which inundates or strikes a region or settlement because the river cannot accommodate the overflowing water. Two factors can contribute to the occurrence of floods: natural factors and human factors [1], [2]. Flood disasters can be predicted by monitoring precipitation and water flow. In addition to heavy rainfall, soil texture, land slope, and land use can also contribute to flooding. In general, floods frequently occur near watersheds. Sometimes, heavy precipitation can cause river reservoirs to exceed the volume of the river, leading to flooding.
Using remote sensing data from Landsat8 and the Normal Difference Water Index model to analyze the flood potential model is one of the data that can be used. Based on calculations from previous remote sensing research titled (Development of an Inundated Area Identification Model Using Landsat [3] , the NDWI variable has a good ability to detect flood inundation. The benefits of remote sensing data include accurate historical information and the ability to observe annual changes in a region's land cover. The area coverage of the extensive remote sensing data enables a comprehensive view and analysis of the area, allowing the primary causes of flooding to be identified. This information can also be used as input for modeling potential flood zones. Several previous studies related to forecasting using the Simple Moving Average [4] with the title (Prediction of PDAM Water Usage Using the Simple Moving Average Method) explained that this SMA was able to provide good predictive results, namely the MAPE value of 0.1712, where if the MAPE value is 10%, then the prediction results are categorized as good, as well as for related research for the Holt With values of = 0.1, = 0.001, and =0.5, the Holt-Winters method [5] yields better results than the SSA method with a MAPE of 13.469 percent.
Researchers are interested in developing research using Landsat 8 data with the NDWI (Normalized Difference Water Index) [6]feature and comparing the Simple Moving Average and Holt Winters models to determine the best model for making predictions based on previous research. In order to compare the two methods optimally, researchers will compare evaluation values including MAE, MAPE, MSE, and RMSE. Landsat 8's resolution is ideal for detecting, measuring, and analyzing changes in earth's surface objects at a granular level. Landsat 8 is utilized because it is able to produce images for flood detection data processing [7]. The Normalized Difference Water Index (NDWI) feature is used to detect bodies of water. NDWI is a remote sensing-based identification method that is sensitive to changes in water content, according to publications [8] stating that an NDWI value greater than or equal to 0 indicates the water surface, whereas a value less than or equal to 0 indicates non-water level. The data that will be collected and used to predict potential flooding is based on two years of historical Landsat 8 data.
II. RESEARCH METHODOLOGY According to a study by Sulistyanto [9] load cells are commonly used to measure force and torque. The load cell sensor will deliver accurate and dependable results when the load cell is used properly. The results of research and observations using the Level Two SMA Method and the Level Two WMA Method indicate that both methods have the ability to reduce sociation well; however, the Level Two SMA method is superior to the Level Two WMA method in determining the reference point, as the reference point search is based on the data value when the load cell is stable. When conditions are stable, the standard deviation of the Level Two SMA Method and the Level Two WMA Method is 0.015572.
This study aims to predict monthly rainfall in Ambon City using the Holt-Winter Exponential Smoothing method [10]. This study makes use of monthly precipitation data from January 2005 to December 2016. The data is the result of observations made at the Ambon Meteorological, Climatology, and Geophysics Agency's Meteorological Station. According to data analysis, monthly precipitation in Ambon City follows a seasonal pattern. The results of this study indicate that the Holt-Winter model based on the seasonal multiplication method is a suitable model for forecasting monthly rainfall data in Ambon City due to its low SSE/RMSE value. The Holt-Winter model predicts that precipitation in Ambon City will increase next year.
In a study to compare the forecasting accuracy of the Holt Winters method and the Singular Spectrum Analysis model (SSI) [11],the difference in methods undoubtedly affects the accuracy of the predicted outcomes. The distinction between the three methods is determined by comparing the accuracy and dependability of the forecast results. Mean Absolute Percentage Error (MAPE) is used to measure the accuracy of forecasting, and tracking signal is used to measure the dependability of forecasting results. The application concerned the production of shallots in Indonesia from January 2006 to December 2015. Forecasting the two SSA methods using L=39 and r=8 grouping windows. With values of = 0.1, = 0.001, and =0.5, the Holt-Winters additive method produced better results than the SSA method with a MAPE of 13,469 percent.
According to research [12], the Simple Moving Average is a predictive model. The moving average technique is employed to forecast demand by calculating the average value and actual demand value from a predetermined number of previous periods. As implied by the method's name, each new prediction is fixed to the previous period and used with requests from the new period, so the data used in the calculation changes over time. The simple moving average method is utilized for unstable, trendless, and unweighted data. With the simple moving average method, which is one method for predicting system models based on time series with computerized characteristics, regarding predicting carpets, it is anticipated that the carpet laundry company will be able to predict the monthly income of the carpets in order to calculate the monthly profit ratio. In order to develop a user-friendly system for predicting the profit of a successful carpet cleaning business. In a study authored by Laurensz [13] Floods are the most prevalent natural disasters in nearly all regions, and they can cause destruction, loss, and even loss of life. Volcanoes, archipelagos, and severe environmental degradation characterize this region. Standardized Precipitation Index (SPI), Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Soil Adjusted Vegetation Index (SAVI), and Inverse Distance Weighted (IDW) are the analysis methods used in this study for data exploration with Quantum GIS ( QGIS). In Manado during the rainy season, the NDVI value or greenness level of the vegetation index ranges from 0.451 to 0.639, indicating a high greenery vegetation index. And during the dry season, the index is between 0.44 and 0.61, indicating a high level of green vegetation. Using the NDWI method, it was determined that the level of wetness in Manado was moderate. According to the SAVI method, Manado has a range of values between -0.103 and 0.153, which indicates an abundance of puddles. According to the discussion, the subdistricts with the greatest potential for flooding in the city of Manado are Tuminting, Singkil, Paal Dua, Tikala, Wanea, Sario, Weneang, and Malalayang. Medium potential exists in the Bunaken District, while Mapanget District has little potential. The results of the review of the five journals have some bearing on the similarities and differences in research [13]. In this study, remote sensing data or Landsat 8 technology were used to conduct research on an object or area. The benefits of sensing data from Landsat 8 include the availability of accurate historical data and the ability to observe changes in land cover from one year to the next. The NDWI (Normal Difference Water Index) variable has a strong ability to detect flood inundation [14].

III. RESEARCH METHODOLOGY A. Analysis Techniques
In this analysis technique, researchers attempt to explain how the model process for solving a problem flows, so as to facilitate understanding of a model to be created. The analysis process is supported by collected data such as observations, literature studies, and Landsat 8 images of land objects [15]. In order to obtain the necessary specifications for this study, the several steps of the analysis procedure are as follows: B. Design Techniques At this stage, a system with the concept of an information retrieval system that can forecast the potential for flooding in an area and determine the best model from the Holt Winters model and the Simple Moving Average through evaluation tests will be developed. Figure 2 depicts an overview of the system that will be constructed.

C. Data Collection Method
During the research phase, data research becomes a critical element. Therefore, every researcher must comprehend the appropriate data collection techniques for the type of study being conducted [16].
1. Observation Observations were made through direct observation of the research object. In this study, researchers used Bekasi as a single example of a location area. Using coordinates 6°14′06′′S 106°59′32′′E, it was determined that the research was conducted in the province of West Java. Figure  3 depicts the urban region of Bekasi.

Figure 3. Area of Bekasi City
In general, the Bekasi City region has a tropical monsoon climate (Am) with a high humidity level of 78 percent. The daily climate conditions are extremely warm. This is particularly influenced by increased land use, particularly industry or commerce and human settlements. The estimated daily air temperature ranges between 24 °C and 33 °C. Bekasi City experiences two seasons due to its tropical monsoon climate: the rainy season and the dry season. The dry east-southeast monsoon influences the dry season in Bekasi City from early May to September, with August being the driest month. Meanwhile, the rainy season in Bekasi is influenced by the wet and humid southwest-northwest monsoon, which typically blows from November to March, with the peak of the rainy season occurring in January with more than 300 mm of rainfall per month (wikipedia).

Secondary Data
Secondary data is supplementary data obtained from research data sources through intermediary media or indirectly in the form of books, records, existing evidence, or archives, whether published or unpublished. This study's secondary data is comprised of literature studies from national and international journals and news articles from related websites. There are numerous supporting evidences. Researchers make use of precipitation data as secondary data. May 2019 to  [17] applied to the dataset (Mean Squared Error). to obtain the ideal model.

A. Data Collection
The primary data used by the author in this study are satellite images obtained by downloading data from https://www.usgs.gov between 1 May 2019 and 1 April 2021.

B. Preprocessing
After retrieving Landsat 8 image data, the data preprocessing stage is carried out. From the previous stage, it did not accurately represent the case study area, so it required further adjustments using clipping techniques; QGIS was used for the clipping stage. as displayed in image 4 below, When the clipping process is complete, the modeling stage, namely the transformation stage to obtain a water index with the NDWI (Normal Different Water Index) feature, is executed. Using QGIS, satellite images that have undergone the clipping procedure are processed. Can be seen in Figure 5, After preprocessing, the Normalized Difference Water Index (NDWI) method is used to extract features. The level of validity in the research area is determined by transforming QGIS-processed raster data using the NDWI algorithm. Two spectral colors, namely red and blue, are assigned to raster data that has been modeled using NDWI. As shown in Figures 6, 7 and 8 below, the blue color represents land that has an ari value or a high level of water value validity, while the red color represents land that does not have a valid value. After obtaining the image from the Normal Difference Water Index extraction feature, it is necessary to rasterize 103 the NDWI image using Python. Figure 9 demonstrates how data can be incorporated into a model, as demonstrated. After completing the raster, a.csv file will be generated, as shown in Figure 10. From May 2019 to April 2021, a total of twenty-four data periods will be utilized. 1 period consists of 517 columns and 830 rows, with a total of 12,408 columns and 19,920 rows in each excel file containing 248,167,360 data points. The obtained data consists of a set of numbers representing each color pixel that has undergone the raster data process. Therefore, it is necessary to calculate the data in order to determine the average value of each period prior to processing the data.
After obtaining the raster's output data, search for the average value for each period of the raster's data. The data must be calculated to determine the average value for each period before they can be processed. Figure 11 illustrates this. Following the completion of the hold preprocessing, the data for the Simple Moving Average and Holt Winters models must be prepared. Following are data from each period based on the outcomes of the preprocessing procedure detailed in table 2. This is a forecasting model for data with trend and seasonal patterns. The initial step in forecasting is determining the initial value (the overall smoothing value) or calling the level. This procedure involves substituting the value from the previous year. So as to: After making a forecast for the period 1 to 24, the next step is to make a forecast for the next three months. The resulting value is as follows: 25 period = ( 0,11613379 + (1 × −0,0002188) × 0,50472699 = 0,05850542 26 period = ( 0,11613379 + (2 × −0,0002188 ) × 1,08147029 = 0,12559524 27 period = ( 0,11613379 + (3 × −0,0002188 ) × 0,54681018 = 0,06350314 In this study, 1 period's worth of image data is sufficient for 1 period's size, which is 900 MB. In this study, 24 periods (2 years) of image data were used, resulting in an image data size of 21,6 GB.