| Research History |
| Software Structure |
| Specie Sensitivity Distribution |
| BAYESIAN Inference |
| MCMC Simulation |
| DIC Optimization |
| Ecorisk & Uncertainty |
| Joint Probability Curve |
| Exergy SSD |
| Main Function Lists Panel |
| BMC-SSD Panel |
| Models Optimization Panel |
| JPC Panel |
| ExSSD Panel |
| Work Path & Output Results |
| Installation & Initialization |
| Folder & File Extraction |
| SSD Models & Ecorisk |
| JPC & Its Indicators |
| Models Optimization & Parameters |
| ExSSD Models & ExEcorisk |
Links
| College of Urban and Environment Science |
| Peking University |
Ecorisk & Uncertainty
After the optimization model is chosen, the eco-risk of a chemical can be reflected by the potential affected fraction (PAF). It can be calculated by inputting chemical concentration value into the SSD model with decided parameters. During our development of the BMC-SSD software, the SSD models based on average values of the parameter distributions are found to have bigger deviations with the original data points compared with models based on mean values (He et al., 2014), while the SSD models based on Monte Carlo simulation mean values have the smallest deviations. However, the deviations between the two mean value models are not significant.
Since model parameters have their ranges, eco-risks calculated by parameter average values (or other statistical values) can only be the reflection of the average risks. However, uncertainty analysis can provide information about the highest and lowest risks, thus, it will be helpful to avoid over or under evaluation. 95% confidence intervals are normally chosen for uncertainty analysis (Grist et al., 2006; Ramin et al., 2011; Dietzel and Reichert, 2012; Jeremiah et al., 2012).
In initial versions of our software, 2.5th and 97.5th percentiles of the posterior distributions of parameters were chosen as arithmetic progression boundaries, then 20 or 10 sets of parameter arrays were obtained by interpolation within the boundaries. By doing that, 400 SSD curves (2 parameters) or 1000 SSD curves (3 parameters) were determined. For each concentration, an eco-risk value was read in each SSD curves, and the mean value of the 400 or 1000 eco-risk values was calculated. Then, a new SSD curve (eco-risk-mean value curve) was established by different concentrations and their corresponding eco-risk mean values. The deviations between the new curves and the original data were proved to be quite small, therefore, the new curve better reflected the original data compared to the parameter-mean value curves (as mentioned in last paragraph). Also, the highest and lowest eco-risks of a certain concentration were determined from the 400 or 1000 curves, and they were set as the boundaries of 95% confidence intervals (uncertainty boundaries). These two values were marked as the high risk value of 95% confidence intervals and the low risk value of 95% confidence intervals for each concentration.
In our released version (v 1.1), the calculation method of uncertainty boundaries is slightly adjusted. The parameter dataset after MCMC samplings are all used to calculate the uncertainty boundaries. For example, if 5000 iterations were assigned separately in 3 MC chains, the dataset with 3*5000 groups of parameters will be calculated. Then, those parameters will be plugged into the SSD models. 15000 SSD curves, 15000 groups of eco-risk results based on exposure data, and 15000 hazardous concentrations at pth level (HCp) will be calculated. Finally, the statistical analysis will be performed for those data. The 2.5th percentile and 97.5 percentile will form the 95% credible interval, and median (50th percentile) will be used to reflect the average level of SSDs, eco-risks, and HCp values.
The improved uncertainty analysis method considers all data in the parameters’ posterior distribution. Although it will be time-consuming, the development of computer facilitate calculating speed and lead to more accurate uncertainty analysi
