This paper provides a comprehensive evaluation of data fusion techniques used to estimate and recalibrate Network Macroscopic Fundamental Diagrams (NMFDs) by leveraging heterogeneous traffic data sources. We assess the strengths and limitations of integrating Loop Detector Data (LDD), Floating Car Data (FCD), and Public Transport Data (PTD) for accurate NMFD estimation. Our longitudinal analysis, centered on traffic patterns in Athens, Greece, examines the repeatability, temporal variability, and loading/unloading patterns within NMFDs. Our findings indicate that unimodal NMFDs derived from LDD often mismeasure speeds due to positional biases in LDD, as they typically capture higher speeds near the downstream of intersections. In bimodal NMFDs, free-flow car speeds are significantly higher than average link speeds. These differences diminish under congestion, while public transport speeds are consistently lower. Utilizing K-means clustering, we successfully identified longitudinal demand clusters in the network. Our application of K-means clustering revealed that loading states consistently lie above unloading states in critical density regions, whereas free-flow conditions exhibit distinct trends across NMFDs. Weekday NMFDs show similar patterns with overlapping confidence intervals, while weekends exhibit a narrower critical density region and an earlier decline in flow. Our clustering approach effectively distinguishes between peak-hour upper-bound states and off-peak lower-bound states, demonstrating the potential of temporal clustering techniques for more precise NMFD estimation.