Best methodology to detect small spatial/temporal variation?

[p_monk]

I am analysing the effect on Land Surface Temperature (LST) of renewable energy (initially large solar farms). 

To calculate LST I am using a process (attached) that uses Landsat 8 scenes.  While I explore the data I am sampling a Landsat scene in Google Earth Engine (I've done this at 100 and 30m intervals), extracting band values to a CSV, calculating LST in Excel, then mapping the data using ArcGIS Pro 2.8.1.

To evaluate temperature variations caused by large solar installations I'm looking at data prior to construction, and post-construction.  It looks like any effect is small, and possibly limited to areas within 200m of the edge of the solar installation, but to evaluate this further I'd like to improve upon an existing method that uses circular buffers at 100m from the boundary of the solar installation. 

Within each buffer - effective a 100m wide ring around the solar farm - problematic ground is masked out (e.g. asphalt, disturbed ground, slopes etc.) - and an average LST is calculated.  ANOVA is used to compare before and after to detect a statistically significant difference.

I'd like to evaluate variation with more fidelity - preferably at 30m intervals that relate to the Landsat 8 resolution.  There are obviously several things that could influence LST - surface type, aspect, slope, vegetation, wind speed/direction etc.  I can get data for some of these easier than others! 

Can you suggest any tools or approaches that will allow me to analyse LST variation?  For example, looking at how spatial variation has changed pre/post-construction, and determine the magnitude of these changes, or how to pull in other variables (I'm familiar with regression analysis but not in this context) to explain observed variation?

I have a basic understanding of ArcGIS Pro.  I can implement built-in analysis tools (e.g. hotspot analysis), though probably don't know how to use them with nuance.  I have a basic (but improving) understanding of Python/Javascript.

In case it is relevant, in due course I anticipate scaling this (using Earth Engine) and running the analysis using Python. 

At this stage, I'm looking to explore approaches in ArcGIS Pro, though without much experience of ArcGIS Pro a few pointers would be really helpful. 

Thanks for your time.

Philip

[Elijah]

Dear Philip,

I have to state that I am not sure I did fully understood what exactly you're doing but it sounds like this may apply.

If you have all your dataset as rasters, you can organize them into dates, looking much like time series. Mask out areas of no interest as no data. Convert the data into multidimensional raster data. Try to do a change detection using the time series rasters you have compiled. Also, obtain a trend raster from the workflow...which will likely indicate if LST has been negative, constant or positive over time and space.

Hope this helps.

Elijah.