Soil Moisture Initialization for Climate Prediction: Characterization of Model and Observation Errors

Current models for seasonal climate prediction are limited due to poor initialization of the land surface soil moisture states. Passive microwave remote sensing provides quantitative information on soil moisture in a thin near-surface soil layer at large scale. This information can be integrated with a land surface process model through data assimilation to give better prediction of the near surface and deep soil moisture states than model predictions or remote sensing observations alone. To achieve this, it is necessary to have a good understanding of both the model and observation errors. We have characterized the model error in the catchment-based land surface model(CLSM) used by the NASA Seasonal-to-Interannual Prediction Project (NSIPP) and the observation error of the near surface soil moisture from Scanning Multifrequency Microwave Radiometer (SMMR) data by comparing them with long term in-situ measurements of soil moisture collected in Russia, Mongolia and China. We found that in dry climate areas, such as Mongolia and central China, central Russia or when the soil is frozen, (e.g. fall and winter in Russia), the CLSM has a dry bias. In wet climate areas, such as the east coast of China and western Russia, the catchment model has a wet bias. The model error in Eurasia is typically less than 0.12(v/v). We also found that SMMR-derived soil moisture data has a wet bias in China, and dry bias in Mongolia and throughout most of Russia. The satellite observation error is large in wet and densely-vegetated regions and small in dry region. This indicates that if correct model error and observation error are used, data assimilation will give better soil moisture in China and Russia. Our analysis also shows that the SMMRderived soil moisture data show a over-projected seasonal change than the in-situ measurements, while the modeled soil moisture shows a depressed seasonal change than the in-situ measurements. This indicates that assimilating remote sensed soil moisture with larger seasonal change into the model producing depressed s easonal change will give better soil moisture seasonal change than the soil moisture estimate either pure model output or remote sensing satellite data alone. Lastly, the model predicts the rootzone soil moisture very close to in-situ measurements, indicating the assimilating surface soil moisture into the catchment model will give improved rootzone soil moisture which is important for climate prediction. Our error analysis has many implications for data assimilation and currently we are developing different assimilation algorithm to best take into account the model and satellite observation errors.