Assessment of land surface model uncertainty: A crucial step towards the identification of model weaknesses

Abstract

Listen

Translate article

Summary Assessment of uncertainty in land surface models is complex, mainly because of the numerous sources of error from model parameters, initial states, input forcing data, and the model structure. These sources of uncertainty interact together to impact the simulated output generated from the land surface model. To account for these uncertainties, the goal in diagnostic model evaluation has been to determine the erroneous/inadequate components of the model structure that need improvement. However, the specification of inaccurate model components is not straightforward, requiring crucial steps to determine the uncertainty contributions from individual error sources. Also, the interaction between the uncertainty sources makes it difficult to assess the impact of the individual error sources on the simulated output. The approach undertaken in this study was to quantify the specific uncertainties linked with model parameters, states, input forcing variables, and spatial variations in landscape properties, such that the remaining model uncertainty is equivalent to the inadequacy/inaccuracy associated with the model structure. This study employed the Evolutionary Data Assimilation, together with multi-dimensional clustering, to quantify these individual uncertainties for the Community Atmosphere Biosphere Land Exchange (CABLE) model, in terms of soil moisture estimation for the Yanco area in south-east Australia. The findings showed that the updated soil moisture was more accurate than both the open loop and calibrated estimates. The minimum uncertainty for model components was found to reduce the original and updated bounds by 68% and 62% respectively. The estimated model pathway has accurately reproduced the updated estimates with less than 0.02 m 3 / m 3 error, and was found to be more accurate than both the calibrated and the updated estimates when evaluated against in-situ soil moisture.