Land-cover classification stability tracks divergent productivity and water-use-efficiency trajectories across boreal woody vegetation in the Daxing’anling Mountains

Abstract

Annual land-cover labels are often treated as fixed ecological classes, yet boreal pixels near canopy-cover thresholds may repeatedly alternate between forest and woody-savanna categories. We tested whether classification persistence identifies distinct long-term carbon–water trajectories, and whether such trajectories are explained by tree-cover change or climate covariation, across 56,980 km² of persistent woody vegetation in the Daxing’anling Mountains (2001–2023). Stable IGBP forest, stable IGBP woody savanna, and mixed classification accounted for 7.2%, 45.5%, and 47.3% of the area, with mean tree cover of 57.0%, 43.2%, and 51.8%, respectively. Water-use efficiency (WUE) increased similarly in stable forest and woody savanna (+0.193 and +0.199 g C kg⁻¹ H 2 O decade⁻¹) but less in the mixed domain (+0.100), mainly because gross primary productivity (GPP) rose more slowly there while evapotranspiration (ET) trends remained small overall. In stable woody savanna, raw tree cover correlated with GPP (ρ = 0.634) and WUE (ρ = 0.454), but detrended associations weakened markedly (ρ = 0.237 and 0.011), indicating shared long-term trends rather than synchronous coupling. Detrended vapor pressure deficit (VPD) covaried positively with GPP and ET, whereas multivariable models linked higher VPD to lower WUE; precipitation–VPD collinearity precluded causal attribution. The anomalous 2003 and 2009 seasons were cool, wet, and low-VPD, with depressed GPP and ET, and 2003 also showed exceptional fire activity. Excluding both years preserved the principal stable-versus-mixed contrasts. Classification persistence is thus a useful stratification tool but does not by itself establish ecological mechanism or land-cover conversion.