Satellite-based rainfall estimation is evolving rapidly. Most studies use data, which is spatially fine, but poorly regarding time. On the other hand, availability of verification data is also quite rare. This study used Hillman Form B report that was corrected by ME-48 from Malang Climatological Station. 2009-2016 IR1 satellite data were used in hourly temporal resolution (only less than 3% data missing). Four estimation methods were compared: Auto Estimator, CST, mCST, and Quantile Analysis Equation. Data processing was carried out using Python and R statistic as a quality control. The analysis was done by creating a graph that combines False Alarm and Miss Information for each rainfall intensity. Binary transformation was done for enabling information to be plotted. All rainfall estimation methods have a high false alarm (more than 74% at 1 mm) but quite low miss (less than 0.03%). By taking into account its error pattern, satellite data can be used in rainfall observation. The Quantile equation is slightly superior to other methods. This study is relatively inexpensive to be duplicated so it can be used as an evaluation tool for rainfall estimation best practice for Meteorological and Climatological Agency’s network.
Prakiraan cuaca sangat penting untuk mendukung segala kegiatan aktivitas masyarakat. Untuk menghasilkan prakiraan cuaca yang akurat dibutuhkan pengetahuan dan pengalaman dari prakirawan cuaca yang didukung dengan teknologi pemodelan cuaca. Pada penelitian ini, dilakukan sebuah pemodelan curah hujan menggunakan artificial neural network (ANN) di Stasiun Meteorologi Kemayoran. Pada proses pembuatan model ANN, dibutuhkan pelatihan data menggunakan kondisi cuaca di masa lalu. Data yang digunakan untuk pelatihan dalam membuat model ANN adalah data cuaca harian periode Januari 2011 s.d. Desember 2019 yang selanjutnya diuji dengan menggunakan studi kasus selama periode Januari s.d. Agustus 2020. Variasi model dibuat berdasarkan jenis input dan jumlah hidden layer untuk mengetahui perbedaan penggunaan data prediktor yang digunakan. Kemudian model ANN dibuat dengan menggunakan pendekatan 3 – lapisan yang terdiri dari lapisan input, lapisan tersembunyi, dan lapisan output. Selanjutnya perbandingan model tersebut diuji menggunakan nilai koefisien korelasi (R) dan rata – rata kesalahan absolut (MAE) untuk mengetahui model yang terbaik. Berdasarkan hasil penelitian, prediksi hujan menggunakan data parameter input kondisi cuaca harian berupa suhu udara, kelembaban udara, dan durasi penyinaran matahari memiliki nilai koefisien korelasi (R) sebesar 0.4 – 0.5 dan rata – rata kesalahan absolut (MAE) sebesar 9.7 – 9.8 mm. Sedangkan jika model dibuat dengan parameter input hujan di hari – hari sebelumnya, nilai koefisien korelasi (R) hanya 0.1 – 0.3 dengan nilai rata – rata kesalahan absolut (MAE) sebesar 11.3 – 12.3 mm. Kondisi tersebut menunjukkan bahwa prediktor yang lebih baik digunakan dalam memprediksi hujan harian berdasarkan artificial neural network adalah dengan menggunakan parameter input kondisi cuaca permukaan.
Increase in frequency and strength of cumulonimbus is one of the impacts of climate change. The presence of cumulonimbus usuallyy causes extreme weather. Cumulonimbus can produce heavy rainfalls, tornadoes, turbulences, and other extreme weather events. Upper air conditions have a great effect on the process of cloud growth. Radiosonde observations can be used to predict the presence of cumulonimbus in the short-term period of weather forecast. This study aimed to predict the occurrence of cumulonimbus using radiosonde data based on the machine learning approach. In this study, indices data from upper-air observation were used. The model prediction of radiosonde data was trained using machine learning to predict the presence of cumulonimbus. Based on data processing results, the prediction of cumulonimbus events using radiosonde indices data is good enough when implemented in new test data. The influence of the Convective Available Potential Energy (CAPE) index in the predictor index predicts cumulonimbus. Machine learning model can predict cumulonimbus incidence by 80% in one month testing period when adding the CAPE index. Meanwhile, when not using CAPE, cumulonimbus events’ predicted results only reach 72% of events. The false alarm rate when adding CAPE was 17% and without CAPE was 21%. Based on these results, it can be concluded that the prediction of cumulonimbus cloud events using radiosonde data based on the machine learning approach is sufficiently reliable to be used.
Indonesia has diverse topographical conditions that result in Indonesia having a unique climate. One of the unique climate elements to be studied is rainfall, because rainfall has a different pattern in each region, this different rainfall pattern is caused by several climate phenomena factors that affect the rainfall pattern, including El-Nino Southern Oscillation (ENSO), Indian Ocean Dipole (IOD) and Madden Julian Oscillation (MJO). Medan City is the capital of North Sumatra province which is one of the areas in the flood-prone category in North Sumatra, where the factor of flooding is due to rainfall events in a long period of time, so the author wants to know which climatic phenomena factors can affect rainfall events in Medan city by using Machine Learning technology through the Matlab application, where in this study has a method by forming four combination models, namely the combination of the influence of IOD, SOI and MJO; second combination of IOD and SOI; third combination of SOI and MJO; and fourth combination of MJO and IOD, these four combinations will be the rainfall value of the four models. Furthermore, the rainfall value of the model is compared with the observed rainfall value and verification test using Mean Absolute Error (MAE) and correlation. Then the calculation of the comparison between the four rainfall models with the observed rainfall obtained the lowest MAE value during the SOI and MJO phenomenon of 15.0 mm and the highest correlation value during the IOD and SOI and SOI and MJO phenomena. So it is concluded that the combination of SOI and MJO has the best verification value. This shows that based on Machine Learning modeling, the model shown as the best predictor in Medan city is when the model combination consists of SOI and MJO.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.