16 th Engineering Research and Development for Technology Conference Sentinel-2-derived land cover map using pixel- based classification for hydrological applications Jonah Lee I. Bas 1, a * , Ricardo L. Fornis 1, b 1 Department of Civil Engineering, University of San Carlos, Talamban, Cebu City, Philippines a jonahleebas@gmail.com, b ricfornis@gmail.com Abstract—Hydrological studies require reliable land cover datasets to estimate model parameter values. This paper assessed the accuracy of the maximum likelihood classification algorithm of ArcGIS in classifying land cover from a Sentinel-2 image of the Upper Butuanon River Watershed. An overall accuracy of 90.60% and a kappa coefficient of 0.8815 show that the results of maximum likelihood classification represent real land cover scenarios in the study area. Pixel-based classification was established as a reliable method of generating land cover maps that may be used for hydrological applications such as curve number grid creation and percent impervious estimation. Keywords—land cover, pixel-based classification, maximum likelihood classification, Sentinel-2 I. INTRODUCTION Hydrological studies such as rainfall-runoff modelling and flood risk management require reliable land cover datasets to estimate model parameter values. Up-to-date land cover maps for many areas in the Philippines are scarcely available, if not unobtainable. Researchers often resort to manual digitization of aerial imagery that is both time-consuming and inefficient when working with large study areas. An easy alternative to this tedious process is through pixel-based image classification of open source satellite data. This process of image classification analyzes the spectral properties of every pixel within an area of interest [1]. Several classification methods discussed in [2] are native in some geographic information system applications. The aim of this paper is to assess the accuracy of the maximum likelihood classification algorithm of ArcGIS in classifying land cover from a Sentinel- 2 satellite image of the Upper Butuanon River Watershed. II. METHODOLOGY A. Study Area The study area is located within the jurisdiction of the city of Cebu encompassing the political boundaries of seven barangays (villages) having a total area of 35.363 km 2 . The elevation of the watershed ranges from 40 meters to 750 meters above mean sea level. Baguio clay loam and Faraon clay are the dominant soil types of the study area. B. Sentinel-2 Data Properties and Preprocessing The Sentinel-2 images acquired from the USGS Earth Explorer image database considered for this study were selected based on the following criteria: (a) cloud cover of less than 10 percent and (b) acquisition date beginning January 2018. Bands 2, 3, 4 and 8 (red, green, blue and near-infrared bands, respectively) of the images were added into the ArcGIS data frame for layer stacking. Fig. 1 Near-natural color representation of the study area derived from the channel combinations of bands 4, 3, and 2. Channel combinations using bands 2, 3 and 4 will give a near-natural color representation of the images as shown in Fig. 1. These images were clipped to include only the study area using raster processing tools in ArcGIS. By careful visual inspection of the images, the data acquired on February 9, 2018 was considered superior among the other datasets due to the absence of cloud cover. C. Land Cover Classification Scheme A standardized land cover classification scheme was presented in [3] to be used when working with remotely- sensed data. The classification scheme involves two levels of classification. For the purpose of this study, only the Level I categories were adopted. These categories are the following: urban/built-up land, forestland, agricultural land, rangeland and barren land. D. Maximum Likelihood Classification The entire process of image classification from satellite data was done within the ArcGIS environment. Training samples for each of the land cover categories were gathered. The channel combination of the near-infrared, red and green bands was used as aid in selecting training samples as land