Coordinated estimates of 4-day 500 m global land surface energy balance components

Accurate estimations of global land surface energy balance components [including net radiation (Rn), latent heat flux (LE), soil heat flux (G) and sensible heat flux (H)] are crucial for quantifying the exchange of heat and water between the land surface and atmosphere. In this study, a novel and practical model for Coordinated estimates of 4-day 500 m global land Surface Energy Balance components (CoSEB) was developed, using the multivariate random forest technique and a synthesis of remote sensing and reanalysis datasets, as well as the extensive observations at 336 eddy-covariance sites worldwide. The CoSEB model effectively balances the estimates of Rn, LE, H, and G by learning the physics of energy conservation embedded in these components within the training datasets. Its advantages include 1) the accurate estimation of the four energy components and their ratios [i.e. evaporation fraction, defined as $\mathrm{LE}/\left(\mathrm{Rn}-G\right)$, which reflects the proportion of surface available energy that is allocated to LE instead of H], and 2) the ability to maintain energy balance among the four energy components, i.e. $\mathrm{Rn}-G-\mathrm{LE}-H=0$. With the 10-fold cross-validation at 286 sites, the CoSEB model achieved the root mean square error (RMSE) of 16.42 W/m², 16.40 W/m², 16.49 W/m², 4.79 W/m² and 0.22, and the coefficient of determination (R²) of 0.92, 0.86, 0.83, 0.55 and 0.59, respectively, for estimating Rn, LE, H, G and evaporative fraction, which were comparable or superior to those estimated by other typical data-driven uncoordinated and coordinated models. In the validation with test datasets at another 50 eddy-covariance sites, the CoSEB model, with the RMSE of 18.23 W/m², 17.98 W/m², and 19.17 W/m², and the R² of 0.87, 0.72, 0.71 in estimating Rn, LE, and H, respectively, outperformed all six state-of-the-art algorithms/products, i.e. FLUXCOM, BESSV2.0, MOD16A2, PML_V2, ETMonitor and GLASS. Besides, the CoSEB model successfully maintained energy balance, exhibiting an energy imbalance ratio [EIR, defined as $100\%\times \left(\mathrm{Rn}-G-\mathrm{LE}-H\right)/\mathrm{Rn}$] of 0, in comparison to other coordinated and uncoordinated models/products presenting the EIRs of as high as 50%. The new model also produced consistent global patterns with the six state-of-the-art products for each of the Rn, LE, and H estimates at 8-day, monthly, and yearly scales. The CoSEB model is promising for operationally generating high-accuracy global balanced land surface energy components and their ratios, which is essential for rationally partitioning surface available energy into H and LE, and for revealing the spatiotemporal variations of the energy components and their attributions and mechanisms at regional and global scales.