Abstract

Coral reefs have been termed as one of the most biodiverse marine ecosystems, but are now becoming threatened by climate change, acidification of the ocean, and anthropogenic disturbances. It is required to identify effective and scalable monitoring for the conventional manual surveys. The research will build a unified system that incorporates remote sensing and artificial intelligence in improving long-term monitoring of the health of corals. It used a multi-source dataset, including satellite imagery, the high-resolution data provided by UAVs, and in-situ measurements, and processed these data through a data fusion architecture and analyzed them with the ResNet-50 based convolutional neural network optimized with Adam (learning rate 0.001, batch size 32). The data have been divided into 70% training, 15% validation, and 15% testing, and preprocessing measures like normalization and augmentation have been used to enhance the generalization of the models. The findings reveal the major statistical increases whereby the average coral cover classification accuracy improved to 98% in 2020, and species identification accuracy improved to 95% compared to 2015, which was 75%. The effectiveness of the model is further justified by performance measures that have a precision of 96, a recall of 95, and an F1-score of 95, which implies that the model has great reliability and minimizes false detections. The results validate that AI use, along with the multi-platform sensing, is a significant improvement in the accuracy of monitoring and predictive analysis of ecology. The paper concludes that the suggested framework offers a scalable, data-driven solution to the problem of preserving coral reefs, yet there are still issues of noise in the environment, a lack of data, and the availability of technologies that need further development in the adaptive learning models and worldwide monitoring partnerships.

Keywords: Coral reefs, Remote sensing, Long-term monitoring, Satellite imagery, Coral bleaching, Reef health, Machine learning, and Environmental monitoring.