IDENTIFYING SENSITIVITY THRESHOLDS IN ENVIRONMENTAL MODELS: WHEN DOES A MODEL BECOME INSENSITIVE TO CHANGE?

Sensitivity of environmental models to changes in variable parameters can be measured in a variety of ways. One simple, yet effect way to judge model sensitivity is to calculate a sensitivity index (Hamby, 1994). Standard sensitivity indices compare the standardized percent change in a parameter threshold of interest and the model's state variable from the default model to an altered model run, resulting in a normalized dimensionless index value (Lenhart et al., 2002; Millington et al., 2009). Although traditional sensitivity indices are good at estimating a model's sensitivity to small changes above or below the default model parameter thresholds, they do a poor job of gauging model sensitivity with larger changes in parameter thresholds and indentifying thresholds at which models becomes insensitive to change. To better explore model sensitivity away from the default model parameter thresholds, and identify thresholds where a model becomes insensitive to changes in a particular variable, I have developed the relative sensitivity index. The relative sensitivity index is able to calculate a model’s sensitivity relative to a previous model run, allowing the user to track changes in model sensitivity away from the default model run. To demonstrate the ability of the relative sensitivity index in exploring model sensitivity, the Tsetse Ecological Distribution (TED) Model's (a spatially explicit dynamic model that predicts tsetse fly distributions in Kenya) sensitivity to three parameters (NDVI, maximum temperature, and minimum temperature) was analyzed using two standard sensitivity indices and the relative sensitivity index.