Refining CASA model for robust carbon sequestration assessment: A case study of Northeast China

Carnegie–Ames–Stanford Approach (CASA) is a widely used light-use-efficiency (LUE) model for estimating terrestrial carbon sequestration, yet its accuracy remains limited in climatically heterogeneous regions. We present a regionally adapted CASA framework that (i) calibrates maximum LUE ( ε max ) for six vegetation types in Northeast China using site-level observations, (ii) incorporates a nitrogen-limitation scalar ( N ε ), (iii) couples the revised model (CASA-N-Opt) with GSMSR to quantify carbon sink strength and trends, (iv) improves the spatial resolution of the carbon-sink product from 500 m to 30 m via a mass-conserving refinement guided by high-resolution proxies. Key findings include: (1) ε max values (g C·MJ −1 ) range from 0.326 (deciduous needle-leaf forest) to 0.677 (cropland) with the N ε marginally increasing for natural vegetation but considerably decreasing for cropland. (2) CASA-N-Opt alleviates the systematic underestimation of the original CASA, reducing RMSE by 36% and increasing R 2 from 0.807 to 0.873. Forests, croplands, and wetlands saw RMSE drop by 22–46%; summer and autumn RMSE fall by 53% and 30%, respectively. Cross-validation confirmed robust spatial transferability, with extrapolation RMSE well-controlled for croplands and wetlands. (3) Mean annual regional sequestration is 147 Tg C·yr −1 (130.4 g C·m −2 ·yr −1 ), with forests (63.9%) and croplands (36.4%) as dominant contributors, while wetlands (0.2%) act as a weak sink and grasslands (−0.5%) as a weak source. From 2003 to 2020, forest sink strength declined, cropland remained stable, while wetlands and grasslands shifted toward stronger sequestration. The proposed CASA-N-Opt framework eliminates systematic bias and provides a reliable tool for regional carbon accounting and low-carbon planning in temperate-boreal landscapes. • An observation-constrained, scale-explicit CASA framework is developed for carbon-sequestration assessment. • Shrinkage-based calibration of ε max for six vegetation types improves parameter robustness and spatial transferability. • A nitrogen-limitation scalar (Nε) is incorporated to reduce systematic underestimation and improve bias correction. • Compared with baseline CASA, CASA-N-Opt raises R² from 0.807 to 0.873 and reduces RMSE by 36%. • Carbon-sequestration patterns are refined from 500 m to 30 m by mass-conserving redistribution using NDVI and land cover.