Reply To: SAGA output grid size

Homepage Forums Discussion Forum SAGA output grid size Reply To: SAGA output grid size

mmPaul Alexander


Hope you’re well and thanks for your interest in WUDAPT. If I understand you correctly, I think you might be referring to what we call “level 1” data under the WUDAPT protocol. This is essentially what Jason outlined in his presentation at ICUC-9 (the geoportal and multiple resolution analysis) see Linda See’s and Jason Chings’s presentation

So just to recap the protocol at the moment:

In order to derive parameters in a consistent manner globally and without the need for proprietary data (e.g. LIDAR services, building footprints etc.) or expertise in GIS/RS we opted to use multi-temporal multi-spectral data from Landsat 8 which is 30m resolution and freely available worldwide. However, since most models (and indeed LOCAL climate zones) can’t resolve at this resolution, we resample landsat to a minimum mapping unit of 120m (ideally we would go to 100m, but of course 100/30 = 3.3333… pixels so 120 = 4 pixels).

The tool chain in saga then uses the training areas to classify at 120m BUT if we imagine an area which is say openset lowrise (LCZ 6) with one or two highrise buildings (LCZ1) within the area, well if you were to design a traverse experiment, you would sample air temperatures in the most “representative” areas not the isolated highrise buildings.

What this means is that by resampling to 300m we remove isolated “patches” of LCZ and instead identify relatively homogenous neighbourhoods – and that is level 0. We are discussion this all the time, and might decide to revert back to 120m i.e. remove the resampling to 300m.

Now, for level 1, we use the LCZ blocks for stratified sampling, which means users derived detailed parameters within each block, so LCZ6 mixed with LCZ1 will theoretically have different roughness heights/building cover fractions than another pure LCZ6 site. Once level 1 are derived, our geoportal will allow users to generate UCPs for ANY resolution grid mesh.

My own argument has been that if you need to go “finer” than 300m, you likely will only look at 1 or 2 areas rather than an entire city so why would you need WUDAPT? And anyway, local scale processes won’t apply <100 m so personally I prefer 300m.

If you need to resample to 1km and derive parameters this way, that can be done of course, but the procedure isn’t finalised for this yet. So depending on exactly what you need to do, we can advise on a way forward.

Hope that helps.

- Paul.