UTILIZATION OF SATELLITE DATA AND MACHINE LEARNING FOR FLOOD INUNDATION MAPPING IN MAJALAYA WATERSHED

• Main Author: Siti Burnama, Nabila
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/71856
• Institution: Institut Teknologi Bandung

Majalaya is one of the metropolitan cities in West Java, which plays an important role in increasing the regional economy with industry, agriculture, and animal husbandry in the area. Unfortunately, Majalaya is one of the areas experiencing annual flooding, with approximately six incidents recorded yearly. This resulted in losses due to the flood disaster, and until now, Majalaya is still trying various efforts to prevent flooding in the area. And to study and analyze a flood event, one crucial aspect is sufficient hydrological data. Recording rainfall is an essential aspect of hydrological analysis. Some of the problems encountered in the rainfall data, especially for the Majalaya watershed, there are only three rain stations in this watershed, namely Cisanti, Cibeureum – Kertasari, and Paseh stations. Then, the recording of rainfall is only on a daily scale. Therefore, this thesis uses satellite rainfall data to model flood inundation maps. This thesis aims to determine whether GSMaP rainfall can be used as surrogate rainfall data from gauge rainfall. The second purpose is to model Machine Learning to predict inundation height quickly. The variables used to predict the inundation height are rainfall data, distance from the point to the nearest river, distance from the point to the river inflow, and elevation. This thesis is divided into two steps, the first is modeling the flood inundation map manually using the HEC – RAS application, and the second is modeling the inundation maps using Machine Learning. Both models use satellite rainfall data that has been calibrated using Quantile Mapping. Furthermore, the selection of floods event is based on the Citarum river discharge, which is above 100 m3/s because the capacity of the Citarum river in the Majalaya Watershed is 100 m3/s. sixteen events were suspected of causing flood inundation. And these events were simulated using HEC – RAS to make an inundation map. The inundation maps from HEC – RAS simulation will be extracted for Machine Learning modeling. The HEC – RAS simulation results will be used as the desired output. So, this Machine Learning model has input variables to predict the height of this inundation, namely rainfall, the distance from the points to the nearest river, the distance from the points to river inflow, and elevation, with the output of this model being inundation height. Several methods have been tried in this model: Regression and Neural Networks. In predicting inundation height, the evaluation method used is Multiple R – Square (R2), where the prediction results are considered correct if the R2 value is close to one. Predictions of inundation height using the regression method have not yielded promising results because the value of R2 is 0,052, which is still far from one. Furthermore, the Logistic Regression method also does not give promising results, with R2 being 0,0381. And the method Neural Network model still does not give good results, with the value of R2 for training data being 0,43 and for the testing data being 0,49. Because the Neural Network still does not yet close to one, Normalization is added to the model. Normalization is used to improve the Neural Network model, so the results obtained the R2 for training data being 0,9212 and the R2 for testing data being 0,977. The inundation height modeling with GSMaP satellite data was successful. This thesis focuses on making a predictive model by utilizing satellite data. From this research, many things can be developed and be developed in various aspects of the management and development of water resources. Such as developing FEWS (Flood Early Warning System) and predicting and calculating a hydrology or hydraulics analysis quickly.