Optimal approach to optimisation

[rarygit]

A suggestion made by Nils during the recent online seminar was to run several tree planter scenarios, each run having a specific tree and canopy dimension. This occurs because Tree Planter cannot automatically change these in the simulation.

What would be an optimal approach to optimisation in circumstances where

• the planting polygon is large,
• the site has high paved surfaces & building %, with lower vegetation % and
• the maximum Tmrt values are very high?

My question relates to the reasoning one should use to efficiently & effectively combine the scenarios described by Nils into a small number of "most probable solution sets" to run in SOLWEIG. Then to compare the magnitude of heat mitigation for each solution set. I am interested in the reasoning towards "efficient & effective", or "optimal approach" to conducting the tree planter optimisation; i.e something to support the choices made.

Has anyone come across literature that describes an optimal approach to designing such scenarios? I would like to avoid a plug and chug approach, manually spinning scenarios and then guessing the best match (aka. trial and error).

Intuitively I could begin by rezoning the planting area into a gradient of Tmrt max values, and then subdivide the scenarios hierarchically, e.g. highest Tmrt max values --> plant larger trees required. Exclusion criteria would be inherent in the planting polygon dimensions, e.g. building distance < largest canopy width of trees.

[rarygit]

For example Zhou etal. 2017 outlined some straightforward landscape design criteria for their siting of residential shade trees:

"Because infinite potential tree locations exist, the simplification of potential tree siting location set is necessary. Potential tree placement on the residential parcel is summarized based on landscape design guidelines. In the northern hemisphere, landscape design guidelines suggest that trees should be planted on the south, west, or east of structures. Because of the space limitation on the west and east side of the house, we limited tree placement to the south of the building. To avoid tree crown overlap, the potential tree siting locations within the existing tree crown is excluded. Further, to avoid unnecessary tree shade coverage on the rooftops, a minimum distance of 3 m between the tree and the building is predefined. We locate two trees because this is the most common number of trees to be planted in the desert city considering the water usage and landscape regulation, but in general, the spatial optimization method can be used to locate any number of trees in the 3D environment."

[biglimp]

To do what you are proposing requires some python coding I presume. Have a look at the new tutorials we just added on how to make your own script. That you potentially can do and examine what you want in various combinations...

I move this over to the Discussions section

https://umep-docs.readthedocs.io/projects/tutorial/en/latest/Tutorials/PythonProcessing1.html#pythonprocessing1 https://umep-docs.readthedocs.io/projects/tutorial/en/latest/Tutorials/PythonProcessing2.html

[rarygit]

Thanks for that!

For the moment the optimal approach has been a combination of environmental and engineering praxis (e.g. first lay out the water retention areas), run the Tree Planter scenarios both inside and outside those area, then input the candidate cdsm's in SOLWEIG and check the level of mitigation through shading. At the same time to use the pergola approach where optimum tree size (e.g. 15 mH, 10 mW) won't be achieved until Yr 15 (for example). You have to also model some of the in-between years; in this case an allometric approach helps with sizing your species accordingly. Zhao et al 2017 was quite helpful in remembering that the constraints in tree location are often social-ecological, rather than simply the physics of heat mitigation.

In relation to pre-defining the water retention areas, this approach seems to be working well in support of the decison-making.

An interesting option is to constrain the number of trees as a cost constraint and see where Tree Planter prioritises the planting e.g. reduce the number of trees from 80 to 50. I have a sub-area in the west of an urban site, which I thought would be a candidate for tree planting. But the optimisation consistently rules out placing any trees there. I have to check whether the Tmrt intensity and shading duration are the reason,

At the moment I am using the greedy algorithm because the areas & tree numbers are larger than the Tree Planter test data; I need quick feedback on the design/mitigation approach; and the genetic algorithm alone can be unstable, even crashing with small planting polygons.

[rarygit]

OK, I saw it in the Discussion section, so this can be closed.

[rarygit]

Your links to the UMEP Python scripting tutorials are helpful; setting it up now. Thanks!