Transpiration (T) is pivotal in the global water cycle, responding to soil moisture, atmospheric stress, climate changes, and human impacts. Therefore, establishing a reliable global transpiration dataset is essential. Different global transpiration products exhibit significant differences, necessitating the evaluation of errors. Collocation analysis methods have been proven effective for assessing the errors in these products, which can subsequently be used for multisource fusion. However, previous results did not consider error cross-correlation, rendering the results less reliable. In this study, we employ collocation analysis, taking error cross-correlation into account, to effectively analyze the errors in multiple transpiration products and merge them to obtain a more reliable dataset. The results demonstrate its superior reliability. The outcome of this research is a long-term daily global transpiration dataset at 0.1° resolution from 2000 to 2020. Using the transpiration after partitioning at FLUXNET sites as a reference, we compare the performance of the merged product with input datasets. The merged dataset performs well across various vegetation types and is validated against in-situ observations. Incorporating non-zero ECC considerations represents a significant theoretical and proven enhancement over previous methodologies that neglected such conditions, highlighting its reliability in enhancing our understanding of transpiration dynamics in a changing world.