Satellite remote sensing reveals how efficiently terrestrial ecosystems convert water loss into carbon gain, but product choice shapes the answer. By benchmarking 64 satellite-derived estimates against flux-tower observations, researchers identified reliable approaches, mapped global patterns from 2001 to 2015.
Ecosystem water-use efficiency (eWUE), calculated as gross primary production divided by evapotranspiration, links the terrestrial carbon and water cycles. Flux towers provide valuable local measurements, yet their sparse distribution and limited footprints cannot represent global conditions. Satellite products offer wider coverage, but existing gross primary production and evapotranspiration datasets rely on different theories, inputs, and processing pipelines. Previous studies often evaluated only a few products, producing conclusions that could be inconsistent or even contradictory. Based on these challenges, in-depth research is needed to systematically compare remote-sensing eWUE products and clarify their global patterns, long-term trends, and dominant drivers.
A research team led by the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, with collaborators in China, France, and the United States, published the study in the Journal of Remote Sensing. The work addresses a practical obstacle in global ecosystem assessment: researchers can obtain markedly different estimates of carbon gained per unit of water lost depending on which satellite datasets they combine, complicating climate analysis, ecological monitoring, and water-resource planning worldwide.
The comparison showed that carbon–water coupled products, particularly PMLv2 and BESSv2, reproduced flux-tower observations more accurately than products estimating gross primary production and evapotranspiration independently. Most of the 64 estimates underestimated observed eWUE, while widely used MODIS, FLUXCOM, and GLASS products ranked below the best combinations. The nine strongest estimates nevertheless produced broadly consistent maps of long-term average eWUE. Their agreement provides a more defensible basis for global assessment and product selection, while divergence among popular products (MODIS, FLUXCOM and GLASS) warns against relying on one dataset when evaluating ecosystem performance, regional differences, long-term variability, or climate-related change across the world over time today.
Monthly root mean square errors across the 64 estimates ranged from 1.23 to 1.83 grams of carbon per kilogram of water, with 49 estimates showing negative bias. Annual errors ranged from 0.75 to 1.56, and 53 estimates underestimated tower observations. Trend results remained less consistent: five of the nine leading products showed significant global increases, one declined, and others showed no clear overall direction during the study period. Driver analysis was more stable. Leaf area index dominated eight of the nine best products, accounting for 39% to 64% of dominant-driver pixels; in the ensemble estimate, its share reached 70%, ahead of atmospheric carbon dioxide and shortwave radiation.
“Physically linking photosynthesis and water loss can improve satellite-based ecosystem assessment,” the research team said in a proposed quotation for author approval. “The findings show why product selection matters: even accurate datasets can disagree on long-term trends. Better observations, coupled modeling, and longer records will be essential for dependable global monitoring.”
The researchers cross-paired eight global gross primary production products with eight evapotranspiration products to generate 64 eWUE estimates. They validated monthly estimates at 67 eddy-covariance sites and annual estimates at 55 sites after land-cover and data-quality filtering. The nine lowest-error annual products, their ensemble mean, and three widely used products were analyzed from 2001 to 2015 across global, hemispheric, land-cover, and climate-zone scales. Root mean square error, correlation, bias, linear regression, and partial correlation supported evaluation, trend detection, and driver attribution.
The framework could help researchers select dependable satellite datasets for drought assessment, carbon-cycle studies, ecosystem restoration, agricultural water management, and climate adaptation. Future systems may improve through multivariate machine learning that estimates carbon uptake and water loss together while preserving process-based constraints. Longer flux-tower records, harmonized climate inputs, and better representation of vegetation structure are needed to reduce uncertainty in long-term trends. Robust global eWUE monitoring could strengthen ecosystem forecasting and guide decisions on water security, land management, and climate mitigation.