Multifaceted characteristics of arctic and boreal evapotranspiration in a warming world from multi-source remote sensing

Sammanfattning

Total evapotranspiration (ET) trends remain highly consistent across FLUXCOM-X-BASE and GLEAM v4.2a, indicating a robust growing-season ET change in northern high latitudes. However, the mechanisms behind this change are less consistent than the total flux suggests. This thesis examines whether vegetation greening explains ET changes, whether ET components compensate for one another, and whether dry-wet ecosystem classifications represent process-level ET controls. Growing-season ET north of 45°N from 2001 to 2020 was analyzed using FLUXCOM-X-BASE and GLEAM v4.2a, together with MODIS leaf area index (LAI), land surface temperature (LST), albedo, and the GLWD v2.0 dry-wet classification. The analysis focused on total ET trends, component partitioning, nonlinear temperature responses, and pixel-wise driver attribution. Although total ET trends are broadly consistent between the two products, their inferred mechanisms differ substantially. Dry ecosystems show a near-compensatory substitution among ET components, whereas wet ecosystems show weak or nearly absent component changes. The direction of this substitution is broadly consistent with vegetation greening, but the relationship between pixel-level Ec/ET trends and LAI trends is very weak, with an R² of approximately 0.02. This suggests that greening alone cannot explain the magnitude of ET component changes, and that non-vegetation processes are also important. ET responds nonlinearly to LST anomalies. A cold-end breakpoint near -9°C likely reflects constraints from snow cover and vegetation dormancy rather than water limitation. When stratified by latitude, an inverted-U warm-end response appears in the 45-55°N band for both dry and wet pixels, suggesting that the transition from energy limitation to water limitation follows a latitudinal gradient rather than the GLWD-defined dry-wet split. Pixel-wise partial correlations further show that dry and wet ecosystems within the same latitude band have similar driver structures, with the largest dry-wet difference reaching only 0.08 absolute units for any single driver. Latitude separates ET controls more clearly: LST and soil moisture correlations strengthen toward higher latitudes, while LAI weakens slightly. Finally, driver attribution differs strongly between products. The LAI-dominant pixel share drops from 75% and 65% in dry and wet pixels under FLUXCOM-X-BASE to 34% and 29% under GLEAM v4.2a, whereas LST correlations remain relatively stable. This divergence likely reflects predictor circularity in FLUXCOM-X-BASE, whose machine-learning framework uses MODIS-derived variables that overlap with the driver set. Overall, total ET trends are robust across products, but mechanistic attribution is highly product-dependent. Future large-scale ET studies should therefore validate findings across multiple datasets and critically examine internal model assumptions before drawing causal conclusions.