Satellite anomalies caused by space weather events such as geomagnetic storms pose serious risks to satellite operations, particularly in Low Earth Orbit (LEO). While previous studies have applied models like Wavelet Recurrent Neural Networks (WRNN), anomaly transformers, and autoencoders for anomaly detection, most focus solely on temporal prediction without spatial visualization. This research addresses that gap by implementing a Long Short-Term Memory (LSTM) model within the Satellite Anomaly Information System (in Indonesia, it is called Sistem Informasi Anomali Satelit (SIAS)) to predict both anomaly coordinates and spatial distributions using space weather data. Following the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology, the system was developed using telemetry and environmental data from POES, OMNIWeb, and NOAA GOES, processed into time-series format, and modeled using Python and Flask, with a Laravel-based web interface. The LSTM model provides predictions at 10-min and 1-day intervals and generates longitude-latitude coordinates and heatmap contours. The evaluation results show strong model performance with an MAE of 1.8%, indicating the effectiveness of the system in improving real-time spatial anomaly detection and offering significant advantages over previous approaches in terms of both precision and usability.