The brain exhibits rich rhythmic activity generated by synchronous neural firing. These neural rhythms, which possess a variety of spatiotemporal patterns, are linked to cognition and vigilance. Neural rhythms are a defining feature of arousal states: in humans, alpha (8-12Hz) is a key marker of quiet wakefulness, and is associated with attentional processes, while delta (1-4Hz) is the most prominent rhythm during non-REM sleep, and is tied to memory consolidation and brain waste clearance. Prior work sought to localize the oscillatory generators of specific rhythms, but it is not clear what brainwide network activity patterns (which need not oscillate at the same frequency) underlie the rise and fall of neural rhythms. To investigate this question, we collected accelerated fMRI (temporal resolution<400ms) simultaneously with EEG in 32 humans drifting in and out of sleep, and analyzed these data with novel approaches to determine which brainwide fMRI patterns predict fluctuations in neural rhythms. We demonstrate that continuous variations in neural rhythms can be predicted from hemodynamics in held-out subjects, with predictive information present even in individual brain regions. Alpha and delta rhythms were predicted by substantially distinct patterns encompassing cortical and subcortical areas. We then determined how these areas separate into rhythm-specific networks by probing for shared information that predict each rhythm. We trained and tested models on each pair of brain regions, and calculated ‘performance benefits’---defined as the pair performance minus the highest performance of its individual parts---as a metric of unique or redundant information. After clustering the performance benefits, we found that alpha-predictive information is highly separable in two networks linked to arousal control and visual cognition. Conversely, delta information is not clusterable, and is instead diffusely distributed primarily across frontal cortical regions. These results identify the brainwide network patterns that underlie fluctuations in arousal-related EEG rhythms.