The US National Park System encompasses diverse environmental and tourism management regimes, together governed by the 1916 Organic Act and its dual mandate of conservation and provision of public enjoyment. However, with the introduction of transformative science policy in the 2000's, the mission scope has since expanded to promote overarching science-based objectives. Yet despite this paradigm shift instituting "science for parks, parks for science", there is scant research exploring the impact of the National Park Science Policy on the provision of knowledge. We address this gap by developing a spatiotemporal framework for evaluating research alignment, here operationalized via quantifiable measures of supply and demand for scientific knowledge. Specifically, we apply a machine learning algorithm (Latent Dirichlet analysis) to a comprehensive park-specific text corpus (combining official needs statements -i.e. demand- and scientific research metadata -i.e. supply-) to define a joint topic space, which thereby facilitates quantifying the direction and degree of alignment at multiple levels. Results indicate an overall robust degree of research alignment, with misaligned topics tending to be over-researched (as opposed to over-demanded), which may be favorable to many parks, but is inefficient from the park system perspective. Results further indicate that the transformative science policy exacerbated the misalignment in mandated research domains. In light of these results, we argue for improved decision support mechanisms to achieve more timely alignment of research efforts towards distinctive park needs, thereby fostering convergent knowledge co-production and leveraging the full value of National Parks as living laboratories.