SWAT+ v3.2.0 Sobol Sensitivity Analysis Error and Unreasonable Sensitivity Results

[Khan muzafary]

Dear SWAT+ User Group,

I am currently working with SWAT+ v3.2.0 and I am experiencing problems with the parameter sensitivity analysis, and I would greatly appreciate any suggestions from SWAT+ experts who may have encountered a similar issue.

I am using the Sobol sensitivity analysis with observed monthly river flow (Channel 1 Monthly River Flow) as the observation. In some runs, the analysis stops at approximately 17% during the "Running Parameter Samples" stage and returns the following error:

"Sensitivity Analysis Cannot Continue"

The error message indicates that SWAT+ encountered a problem while running the model and suggests checking that:

• No output file is open;

• Parameter values and their ranges are valid; and

• Objects assigned to parameters are valid.

For example, in one of the failed runs, the parameters in use included:

• cn2 = 68.047 (Percent), range: 35–95

• esco = 0.145 (Replace), range: 0–1

• epco = 0.355 (Replace), range: 0–1

• canmx = 0.879 (Replace), range: 0–100

• perco = 0.840 (Replace), range: 0–1

• lat_ttime = 20.834 (Replace), range: 0.5–180

I have also tried different combinations and subsets of parameters, including hydrological and snow-related parameters, but I continue to encounter problems.

Interestingly, some Sobol analyses do finish successfully, but the resulting first-order sensitivity values are sometimes extremely large and clearly unrealistic (e.g., values on the order of 10^16–10^18), while in other runs the sensitivity values are exactly zero for all parameters. For example, when I tested the snow-related parameters:

• snofall_tmp

• snomelt_tmp

• snomelt_max

• snomelt_min

• snomelt_lag

• snow_lte

the analysis completed, but all first-order sensitivity values were 0.000000.

In another run with a smaller parameter set, the analysis produced reasonable-looking values (e.g., approximately 0.0064 for cn2, 0.0064 for esco, and 0.0039 for canmx). However, changing the parameter combination or number of samples can again result in zero or extremely large values.

I have also experimented with different parameter change types (Replace, Percent, and Relative) and different numbers of parameters and samples. The problem does not appear to be restricted to one particular parameter.

I have attached screenshots showing the error messages and several examples of the sensitivity analysis results.

For reference, I am using SWAT+ v3.2.0, Sobol sensitivity analysis, one parallel process, and observed monthly river flow for the sensitivity objective function.

Any suggestions, troubleshooting steps, or examples from similar cases would be highly appreciated.

Thank you very much in advance for your help.

Best regards,

Khan