A heterogeneous land surface model initialization study

Abstract

Accurate energy and water storage initialization is critical for skillful land surface model prediction. A properly initialized model has equilibrium land surface states via spin‐up runs. The present study investigated aspects of spin‐up runs for the Noah Land Surface Model (LSM) at point stations using the 1 km grid of the Land Information System. The model is run repeatedly through a single year until a predefined equilibrium is achieved. Nine different model initialization methods were tested and compared in these spin‐up runs at twelve southern Great Plains surface observation stations in the Midwestern United States. Soil moisture is used as the primary land surface state to evaluate the spin‐up initializations. The model outputs are compared to evaluate the discrepancies over the annual cycle of repeated runs to learn how the model attains equilibrium land surface states. The results indicate that the climatological average state does not necessarily return the most efficient LSM initialization. Among the various tested methods, the spin‐up runs initialized with the spatially heterogeneous states averaged over a short period are found to perform better than others. Heterogeneous land surface conditions are also found to play a vital role in the spin‐up response. More stable land surface states are obtained through longer spin‐up runs, which also produce more similar coefficient of variance, suggesting that the longer spin‐up runs could yield similar heterogeneous fluxes irrespective of the initialization method. Comparing the results from different methods, a computationally economic technique for the single‐year spin‐up is proposed.