<p>Ecosystem productivity governs terrestrial carbon exchange. However, its extent and dynamics are unpredictable, making it controversial to generalize key drivers. Therefore, this study assessed the spatiotemporal dynamics and drivers of carbon sink capacity using satellite and ground-measured datasets, Hurst exponent (<i>H</i>), geostatistical models, Moran’s, and vegetation indices. Theil–Sen, Mann–Kendall (MK) tests, Kruskal–Wallis test (KWt), SHAP (SHapley Additive exPlanations), and Pearson correlation analysis were performed. Random forest (RF) outperformed seven other machine learning models (<i>r</i>² = 0.81). The regional carbon sink capacity exhibited an increasing trend (<i>p</i> &gt; 0.05) (<i>β</i> = 11.79 gCm⁻²&#xa0;year⁻¹). Elevation, relative humidity (RH), soil moisture (SM), evapotranspiration (EVT), and leaf area index (LAI) showed positive correlations. Anthropogenic factors, wind speed (WS), greenhouse gases (GHGs), and temperature were negatively correlated. Shrublands, broadleaf, and mixed forests had a strong carbon sink capacity. Grasslands, wetlands, deciduous needleleaf forests, and barelands are carbon sources. Similarly, vegetation in the continental (32.6 gCm⁻²&#xa0;year⁻¹), temperate (−23.1 gCm⁻²&#xa0;year⁻¹), and dry (−82.2 gCm⁻²&#xa0;year⁻¹) climate types acts as a carbon sink and source, respectively (<i>p</i> &lt; 0.0001). RF and SHAP models ranked explanatory factors as (%): temperature (66.25), WS (19.77), EVT (12.57), and RH (11.79). Croplands (2.01%), grasslands (0.75%), and wetlands (0.1%) have the highest permutation importance. In contrast, needleleaf types were less vulnerable. Our findings generally highlighted that the regional carbon sink capacity is primarily driven by climate variability, underscoring the need to protect ecosystems, enhance resilience, sustainability, and mitigate emissions. Accordingly, ecosystem restoration, monitoring, climate-smart agriculture, urban greening, and identifying early warning signs are suggested.</p>