STUDY OF PHYTOPLANKTON COMMUNITIES WITH MACHINE LEARNING MODEL AND MANUAL ANALYSIS: CASE STUDY IN JAKARTA BAY

• Main Author: Salsabella Suwarno, Aulia
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/56870
• Institution: Institut Teknologi Bandung

Phytoplankton is the primary producer in marine ecosystem which plays a vital role in fisheries by providing food resource for fish, but in the other hand could cause mass death due to the harmful algae bloom phenomenon. To date, studies regarding phytoplankton still rely on manual procedures and are both time-consuming and labour-intensive. The application of machine learning could be a solution by using Convolutional Neural Netwoyk (CNN) algorithm for automating phytoplankton identification. In this research, four scenarios of phytoplankton identification model were built with different outputs based on phytoplankton image data from P2O-LIPI using the same model architecture, VGG16. Based on model testing, the scenario that does identification with a higher taxa level—genus—outperforms with accuracy level of 88,75%. This could be explained by how CNN groups data based on the similarities within a category, thus when identification is done on a lower taxa level like class which has a larger variation among data within the same category, this would obstruct the model in generalizing a pattern. Phytoplankton identification were done automatically using machine learning model and manually using microscope. The result from automatic identification shows inadequate accuracy. This is due to the limitation in the model which only recognizes five generas and when presented with other genera, it will force the identification result into one of the five generas. Moreover, development from this model by adding an “Others” category that accommodates generas outside the five aforementioned generas has worse performance. Thus, further analysis of phytoplankton is done using the results from manual identification. Phytoplankton abundance in Jakarta Bay on July 2019 ranges between 19.351-4,314 million cell/L with highest abundance found on coastal area and diminishes towards the open ocean. The composition of phytoplankton community in coastal area is dominated by Chaetoceros, Cylindrotheca, and Navicula, whereas the open ocean area is dominated by Bacteriastrum and Chaetoceros. Based on Pearson’s correlation coefficient between the phytoplankton abundance and eight environmental parameters, the abundance is found to have a significant positive correlation with temperature, pH, and phosphate ion,and a significant negative correlation with salinity.