Remote-sensing carbon maps reveal no single winner in southwest China’s karst

Gross primary productivity, or GPP, is the amount of carbon captured by plants through photosynthesis before subtracting the carbon used in plant respiration. Because GPP links atmospheric carbon, vegetation productivity and ecosystem functioning, reliable estimates can improve regional carbon-cycle assessments and climate monitoring. A new evaluation of five satellite-derived GPP products finds that no single product is best for every purpose across southwestern China’s karst landscapes, where fractured limestone, thin soils, seasonal water stress and patchy vegetation make carbon estimates especially difficult.

The study examined the BESS, GOSIF, GLASS, MOD17 and VPM products for 2001–2018, evaluating each one through site-scale comparisons, mapped spatial patterns, seasonal and interannual changes, and estimates of uncertainty. The researchers also separated karst terrain from non-karst areas, allowing them to ask whether familiar relationships between vegetation, climate and productivity remained valid in one of China’s most distinctive ecological regions. Karst landscapes cover extensive parts of southwest China and are visually dramatic, but their subsurface drainage, exposed rock and limited soil development can make surface vegetation appear more disconnected from regional water and climate conditions than it does in conventional ecosystems.

To test performance against direct field observations, the study compared the remote-sensing products with GPP derived from flux towers. Such towers measure exchanges of carbon dioxide and other gases between an ecosystem and the atmosphere, providing an independent check on large-scale satellite estimates. GOSIF generally produced the strongest correlations with the flux-tower-derived values, making it the most consistently aligned product with the tower record. VPM, however, showed relatively low estimation errors at several individual sites. These contrasting results demonstrate why a product that tracks temporal observations well may not necessarily minimize absolute error everywhere on the map.

The distinction is important because satellite GPP products are not direct photographs of photosynthesis. They combine remotely sensed vegetation and climate variables with assumptions about how plants respond to light, temperature, moisture and carbon availability. The algorithms also differ in the ecosystems and sensors used during calibration, and they may represent uncertainty differently. A product can therefore perform well against one validation metric while falling short against another, particularly in an environmentally heterogeneous region where vegetation types, soil depths and rainfall regimes change over short distances.

All five products agreed on a broad regional signal: annual mean GPP was lower in karst areas than in non-karst areas. This consistency provides a useful indication that the products are capturing a real spatial contrast rather than an artefact that appears in only one dataset. The result is also ecologically plausible, because exposed bedrock, limited soil water and relatively low plant-cover fractions can constrain productivity. The agreement does not mean that every product estimates the magnitude of that difference correctly, but it strengthens the case that lower productivity is a characteristic feature of the karst landscape in the mapped region.

Seasonality was another point of agreement. Mean monthly GPP was generally highest during summer and lowest during winter, reflecting the warm growing season and the strong influence of seasonal plant activity on the land carbon cycle. The gap between karst and non-karst areas was smaller in summer than in other parts of the year. One possible interpretation is that peak-season productivity benefits from the greater availability of water and energy, allowing vegetation in karst systems to approach the productivity of neighbouring landscapes. The available results do not identify a single mechanism, so the finding is best treated as an observed pattern rather than a complete explanation.

The long-term trends were less uniform. BESS, GOSIF, MOD17 and VPM showed significant increases in annual mean GPP across the study period, pointing toward a regional rise in estimated photosynthetic activity. GLASS also indicated an increase, but its trend was weak and not statistically significant. The contrast matters for scientists studying whether southwest China’s ecosystems are becoming more or less productive, because a product with an unstable or uncertain trend could change the apparent strength of a multi-decade change. It also demonstrates why analysts should examine both the direction of remote-sensing signals and the statistical evidence supporting them.

Uncertainty was evaluated using the three-cornered hat, or TCH, method, which is designed to separate errors among related remote-sensing products and provide an estimate of their relative uncertainty. MOD17 had the lowest mean relative uncertainty under this measure, followed by VPM. That result is not the same as declaring either product universally superior. TCH assesses uncertainty within the ensemble and metric, whereas site comparisons assess agreement with observed fluxes; rankings can change when the objective, places sampled and evaluation period change. A product with low modelled uncertainty may still disagree with a tower, while a product with higher uncertainty may capture observed variation well.

Together, the results argue against searching for a universal champion among GPP maps. GOSIF may be attractive when the priority is correlation with flux-tower observations. VPM can offer relatively low errors at some sites, and MOD17 showed the smallest TCH-derived mean relative uncertainty. BESS, GOSIF, MOD17 and VPM also detected significant long-term increases that GLASS did not resolve confidently. Product choice should therefore be tied to the scientific question: a regional trend analysis, seasonal profile, site-level flux study or broad carbon budget may require different performance criteria.

Subject of Research: Comparative evaluation of satellite gross primary productivity products and their uncertainty in southwest China’s karst region.

Article Title: Evaluation of spatiotemporal patterns and uncertainty of multi-source remote sensing GPP products in the Southwest karst region of China

Article References: Dong, H., Wu, L., Li, X., Zhang, Y., Luo, A., & Tian, F. (2026). Evaluation of spatiotemporal patterns and uncertainty of multi-source remote sensing GPP products in the Southwest karst region of China. Scientific Reports. https://doi.org/10.1038/s41598-026-76893-z

Image Credits: AI Generated

DOI: 10.1038/s41598-026-76893-z

Keywords: gross primary productivity, remote sensing, karst ecosystems, carbon cycle, China, uncertainty analysis, flux towers, vegetation productivity, satellite products, climate monitoring, Evaluation, spatiotemporal