Accurate transport demand forecasting can benefit from multimodal data, yet practical challenges arise when different institutions hold separate datasets and cannot share them directly. While institutions may not share data directly, they may share models trained by their data, where such models cannot be used to identify exact information from their datasets. In this context, we propose a Knowledge Adaptation Demand Forecasting (KADF) framework that leverages pre-trained models from one transport mode (source) to forecast demand for another (target), without direct data sharing. The framework captures shared travel patterns across modes through a knowledge adaptation strategy, separating target-mode data into individual and shared components. A pre-trained source model transfers generalized knowledge to improve target-mode predictions. Experimental results on real-world datasets show that KADF outperforms baseline and state-of-the-art models, demonstrating the effectiveness of knowledge transfer without compromising data privacy. This approach supports collaborative forecasting in a decentralized data environment. .