Abstract ID: 142

Application of a Monte Carlo Global Sensitivity Analysis For Model Calibration at a LBA Flux Tower Site

The calibration procedure of Land Surface Models (LSM) is often very complex and expensive, as it could involve the optimization of tens to hundreds of parameters. To simplify the procedure and isolate parameters with insignificant influence on the model output, a preliminary sensitivity analysis (SA) of the inputs should be performed. Local SA methods, like partial derivatives methods, are frequently applied for this purpose, but they cannot detect interactions or nonlinearities in parameters. In this work we used the Morris SA method to screen and rank the LSM parameters. The method performs a Monte Carlo global SA, efficiently exploring a large region of the input space. Two modifications in the Morris procedure are suggested, to adequate the method for the LSM parameters characteristics. The method was applied to the IBIS model at the LBA Flona Tapajós (km 67) site, for the period of January to December of 2003. We performed sensitivity analysis of three model outputs (LE, H and NEE) to 32 input parameters. The method was effective and efficient in ranking all parameters in order of decreasing importance, at a low computational cost. Of the 32 analyzed parameters, 10 showed to be influent to H, with 8 presenting some interaction or nonlinearity. For LE, 7 parameters are influent, with 6 showing interactions or nonlinearity. And for the NEE, 9 parameters showed some importance, with 6 presenting interactions or nonlinearity in their behavior. Based on this classification of the input parameters, one could develop more objective and efficient calibration procedures for LSMs.

Session:  Carbon - The LBA Model-Intercomparison Project.

Presentation Type:  Poster