Closed Manangka closed 10 months ago
In GitLab by @luitjansl on Nov 16, 2023, 15:12
tests were done with the twri model. It had anisotropy added to it (K22 was set to 20*K in the npf package). It was partitioned in 4 different ways, and it was partitioned using flopy and imod-python. FIrst it was established that without partitioning, flopy and imod-python give the same results.
Then the difference between the unpartitioned head and the partitioned head was computed for the last timestep in the simulation.
The following table gives the max and mean difference for the head results. Note that these results are exactly equal for flopy and imod-python, except for the "four squares"partitioning. There, the flopy error is larger than the imod-python error. So the story was considered validated. | Partition boundary | Flopy split max | Flopy split mean | Imod python split max | Imod split mean |
---|---|---|---|---|---|
Diagonal_1 | 0.5191836798861 | -2.220987858 | 0.5191836798861 | -2.220987858 | |
Island | 4.494228506469 | -0.02097133 | 4.494228506469 | -0.02097133 | |
intrusion | 0.86444 | 0.0078825572697 | 0.86444 | 0.0078825572697 | |
Four squares | 12.096101364 | -20.828593 | 12.096101364 | 3.88502683 | |
Note: when imod-python creates a partition file, it describes the exchange from the partition with the highest partition number to the partition with the lowest partition number. So from partition 2 to partition 1. Flopy does this the other way around and gives the partition from 1 to 2. we changed imod python to do the same thing as flopy, to facilitate comparisons.
In GitLab by @luitjansl on Nov 16, 2023, 13:51
The following is a starting point for a test with anisotropy in k. After partitioning the simulation and running the simulation, we should get the same results as when running the simulation without partitioning. But this is not the case. We should investigate if the same happens in flopy and if not, why.