Arctic coastal sea ice phenology is a critical climate indicator, yet sub-kilometer long-term observations remain scarce, limiting our understanding of complex local freeze-thaw dynamics. This study investigates the long-term evolution and thermodynamic drivers of coastal sea ice using a continuous, 20-year (2003–2023) high-resolution dataset derived from ground-based GNSS Interferometric Reflectometry (GNSS-IR) at Tuktoyaktuk in the Beaufort Sea. By employing a physics-based Amplitude Integration Factor (AIF) method, we successfully bridged decadal hardware discrepancies to extract an uninterrupted, thermodynamically consistent climatological record. Trend analysis of the 20-year record reveals a statistically significant shortening of the continuous ice season by 4.63 days per decade (p = 0.04). This climatological decline is profoundly asymmetric, driven primarily by a substantially delayed autumn freeze-up (+3.40 days per decade) rather than an advanced spring breakup (−1.42 days per decade), underscoring the dominant influence of enhanced summer oceanic heat uptake and thermal memory. The physical reliability of these localized observations is corroborated by their strong coupling with accumulated Freezing Degree-Days (R² = 0.74). Crucially, cross-scale comparisons demonstrate that GNSS-IR detects autumn freeze-up onset 5.5 ± 3.7 days earlier than 4-km gridded satellite products (IMS). This systemic lead time confirms the unique capability of GNSS-IR to resolve initial nearshore frazil ice formation – a critical sub-grid thermodynamic process typically diluted in coarse-resolution remote sensing. Ultimately, this work provides an essential high-resolution baseline for validating regional climate models.