Surface drifter trajectory prediction is essential for applications in environmental management, maritime safety, and climate studies. This survey paper reviews research from the past two decades, and systematically classifies the evolution of methodologies into six successive generations, including numerical models, data assimilation, statistical and probabilistic approaches, machine learning, deep learning, and hybrid or AI-based data assimilation (1st–5.5th Generation). To our knowledge, this is the first systematic generational classification of trajectory prediction methods. Each generation revealed distinct strengths and limitations. Numerical models ensured physical consistency but suffered from accumulated forecast errors in observation-sparse regions. Data assimilation improved short-term accuracy as observing networks expanded, while machine learning and deep learning enhanced short-range forecasts but faced challenges such as error accumulation and insufficient physical constraints in longer horizons. More recently, hybrid frameworks and AI-based data assimilation have emerged, combining physical models with deep learning and traditional statistical techniques, thereby opening new possibilities for accuracy improvements. By comparing methodologies across generations, this survey provides a roadmap that helps researchers and practitioners select appropriate approaches depending on observation density, forecast lead time, and application objectives. Finally, this paper highlights that future systems should shift focus from deterministic tracks toward credible uncertainty estimates, region-aware designs, and physically consistent prediction frameworks.

Multi-retriever Retrieval-Augmented Generation (RAG) integrates retrievers of different modalities, such as text and image, to capture richer retrieval signals than a single retriever. A key challenge, however, lies in score fusion: fixed-weight methods like Reciprocal Rank Fusion (RRF) fail to adapt to query-dependent variations in modality utility, while LLM-driven Adaptive Weighting (AW) remains vulnerable to confidence miscalibration. This miscalibration often produces erroneous weights that severely degrade retrieval quality. To address these limitations, we propose Adaptive Weighting with Anchor Retention (AWARE). AWARE pairs per-query LLM-assigned weights with Anchor Retention (AR), a systemic safeguard that unconditionally preserves the top-j candidates from each retriever, ensuring high-quality results survive even under catastrophic weight misallocations. Experiments on a RAG benchmark of 9,666 Korean artworks across five LLM backbones (Gemma3 4B/12B/27B, Qwen2.5-VL 3B/7B) demonstrate that, averaged over four metrics (Recall@5/10, NDCG@5/10), AW improves upon RRF by 21.0%, while AWARE further extends this improvement to 27.0%, consistently outperforming baseline AW. Notably, AWARE coupled with the compact Qwen2.5-VL 3B model matches the performance of the Gemma3-27B model (within ±0.01), proving that robust multi-retriever retrieval does not require resource-heavy large LLMs.