*3.3. Case Study Results*

In order to demonstrate the abilities of the EDSS, we choose a decision-making problem in which the decision maker needs to decide on the pump flow rate that supplies the water from the aquifer in the most upstream section (point A in Figure 5) to achieve the desired goal of water quality. More specifically, the algorithm needs to quantify the impact of changes in the pump flow rate on the downstream section water quality, given the additional inputs from WWTPs in the middle of the stream and the natural processes occurring along the stream's 28 km path (Figure 5). This decision-making problem represents a constant debate among stakeholders on the amount of freshwater allocated to the stream to meet water quality targets.

A scenario of excess ammonia (NH4-N) concentration downstream was simulated. As shown in Figure A3 in Appendix A, the EDSS input file defines a possible range of pump flows (*col\_name* = q) between 0.2 m3/s (*min\_val*) and 0.8 m3/s (*max\_val*). The flow changes as defined in the steps were set for two recursive runs, one with 0.1 m3/sec and the second with 0.01 m3/s. The "qin\_br1.csv" (*name*) is the input file that needs to be changed. The NH4-N concentration target of 0.57 g/m3 (*target*) was set. The weight is not relevant in this case, as only a single target parameter was used. The simulated NH4-N concentration is extracted from the output file of "tsr\_1\_seg42.csv" (*name*). The defined *score\_step* was set to 0.01 g/m3. The pump flow that minimizes the score was found by the EDSS as shown in Figure A4 in Appendix A: (1) id—Each EDSS execution has a unique string; (2) status—This field is changed from "RUNNING" to "COMPLETED" once the EDSS publishes the results; (3) result—This shows the results of the best run. In this case, as two recursive runs were defined, we can see two sets of results with different identification strings (*best\_run*). The first is for the search between 0.2 m3/s and 0.8 m3/s with 0.1 m3/s interval, in which a flow of 0.6 m3/s (*params*) had the best relative score of 0.525 out of the seven runs. The second results are for the run between 0.55 m3/sec and 0.65 m3/s with a 0.01 m3/s interval. In this second search, the best score was 0.025, corresponding to a flow of 0.55 m3/s; (4) A link to download the zip output files from the model executions is available, as shown in Figure A4.

In this example, seven parallel simulations were conducted on cloud computing resources in the first round. In the second round, eleven simulations were conducted in parallel. Altogether, eighteen simulations were conducted in order to reach the result. A more cloud-computing-expensive path could have been taken if a "step" of 0.01 m3/s was solely defined with no recursive rounds. In that case, 61 simulations would have been conducted, resulting in a shorter run time (in case these runs are performed in parallel), but with more cloud computing charges.
