Electronic Theses and Dissertation

Fenomena upwelling di danau maninjau merupakan salah satu permasalahan lingkungan yang berpotensi menyebabkan penurunan kualitas air dan kematian massal ikan pada keramba jaring apung. data yang digunakan dalam penelitian ini merupakan data sekunder berupa data meteorologi harian yang diperoleh melalui nasa prediction of worldwide energy resources (nasa power). penelitian ini bertujuan untuk mengidentifikasi kategori potensi upwelling menggunakan metode fuzzy clustering, membandingkan kinerja model machine learning dalam mengklasifikasikan kategori upwelling, serta menentukan kombinasi metode yang paling optimal dalam mengidentifikasi dan memprediksi kategori upwelling berdasarkan variabel meteorologis. data yang digunakan berupa data harian periode 2001-2025 yang meliputi radiasi matahari, temperatur, kecepatan angin, dan tekanan permukaan. metode fuzzy clustering yang digunakan terdiri dari fuzzy c-means (fcm), fuzzy gustafson kessel (fgk), dan fuzzy c-shells (fcs), sedangkan model machine learning yang digunakan meliputi random forest, xgboost, dan lightgbm. hasil clustering menghasilkan tiga kategori, yaitu berpotensi upwelling, transisi, dan tidak berpotensi upwelling, yang selanjutnya digunakan sebagai label pada proses klasifikasi. hasil penelitian menunjukkan bahwa seluruh model machine learning mampu mengklasifikasikan kategori upwelling dengan performa yang sangat baik berdasarkan nilai accuracy, precision, recall, dan f1-score. kombinasi metode fgk dan xgboost menghasilkan performa klasifikasi paling optimal dibandingkan kombinasi metode lainnya. hasil uji beda menggunakan kruskal-wallis menunjukkan bahwa seluruh variabel meteorologis memiliki perbedaan yang signifikan antar kategori cluster, yang kemudian diperkuat oleh uji lanjut dunn yang menunjukkan adanya perbedaan signifikan pada setiap pasangan cluster. hasil feature importance menunjukkan bahwa kecepatan angin merupakan variabel paling dominan dalam membedakan kategori potensi upwelling, diikuti oleh temperatur, tekanan permukaan dan radiasi matahari. selain itu, confidence analysis menunjukkan bahwa model memiliki tingkat keyakinan prediksi yang tinggi dan relatif stabil. penelitian ini menunjukkan bahwa integrasi metode fuzzy clustering dan ensemble machine learning efektif digunakan untuk mengidentifikasi pola potensi upwelling berbasis variabel meteorologis dan berpotensi dikembangkan sebagai dasar sistem identifikasi dini (early warning system) pada ekosistem danau tropis. kata kunci: upwelling, fuzzy clustering, machine learning, confidence analysis, feature importance, danau maninjau

Upwelling events in Lake Maninjau represent a significant environmental issue that can lead to water quality degradation and mass fish mortality in floating net cage aquaculture. The data used in this study are secondary data in the form of daily meteorological data obtained through NASA Prediction of Worldwide Energy Resources (NASA POWER). This study aims to identify potential upwelling categories using fuzzy clustering methods, compare the performance of machine learning models in classifying upwelling categories, and determine the most optimal combination of methods for identifying and predicting upwelling categories based on meteorological variables. The dataset consists of daily observations from 2001 to 2025, including solar radiation, air temperature, wind speed, and surface pressure. The fuzzy clustering methods employed are Fuzzy C-Means (FCM), Fuzzy Gustafson–Kessel (FGK), and Fuzzy C-Shells (FCS), while the machine learning models include Random Forest, XGBoost, and LightGBM. The clustering process produced three categories: potential upwelling, transition, and non-potential upwelling, which were subsequently used as labels for the classification process. The results demonstrate that all machine learning models achieved excellent classification performance in terms of accuracy, precision, recall, and F1-score. Among the evaluated combinations, the FGK–XGBoost model achieved the best overall classification performance. Furthermore, the Kruskal–Wallis test revealed statistically significant differences in all meteorological variables across the identified cluster categories. These findings were further confirmed by Dunn's post hoc test, which indicated significant differences between every pair of clusters. Feature importance analysis identified wind speed as the most influential variable in distinguishing potential upwelling categories, followed by air temperature, surface pressure, and solar radiation. In addition, confidence analysis showed that the model produced highly confident and relatively stable predictions. Overall, the results indicate that the integration of fuzzy clustering and ensemble machine learning is effective for identifying potential upwelling patterns based on meteorological variables and has strong potential to serve as the foundation for an early warning system in tropical lake ecosystems. Keywords: Upwelling, fuzzy clustering, machine learning, confidence analysis, feature importance, Lake Maninjau.