InVEST Urban Cooling Model - Green Area Maximum Cooling Distance


I recently used the InVEST Urban Cooling Model (@chris ) to evaluate the heat mitigation ability of the City of San Diego as part of a remote sensing capacity building program called NASA DEVELOP. While I discovered this forum too late in the program to post questions and incorporate feedback I did some work to explore the Green Area Maximum Cooling Distance that I thought might benefit others. Here’s a discussion on our work, hope it helps!

While there appears to be thorough documentation on the green area threshold for a substantial cooling effect (Zardo et al., 2017) of 2 ha there was little we could find on selecting a GAMCD. We were particularly interested in GAMCD since the project was aimed at building a decision making tool for the City. Quantifying the cooling of green areas could provide the needed data to support increasing green spaces. Since we were also mapping urban heat we used the empirical data from the land surface temperature (LST) data to see if the cooling from green spaces was visible. LST was derived from Landsat 8 Surface Reflectance Tier 1 product using Google Earth Engine (GEE) . The general steps of this process (in ArcGIS Pro) were 1) create raster of green areas that are under 2 ha 2) generate Euclidean distance raster 3) reclassify (bin by 100m) 4) run zonal statistics as table 5) create chart.

We repeated this as we refined model inputs and saw a number of interesting albeit somewhat intuitive outcomes. Our initial landuse input used the National Land Cover Dataset (NLCD) with a resolution of 30m acquired using GEE. First we looked at the effect of green spaces vs blue spaces (water) to see if we should include open water as a green space in the biophysical table. Given that we are using surface temperatures and not air temperatures we didn’t expect to see much. Does the grass in the park really affect the temperature of the nearby pavement? (note that the scales and sizes of these charts are not the same):

Two things we noted 1) the temperatures closer to water were much cooler than that of green spaces 2) temperatures rise much faster as you move away from water vs land. Lastly this highlighted a major challenge to using the model for San Diego, the cooling effect of the Pacific Ocean was not accounted for. It was especially clear when comparing the LST maps to the Air temp intermediate layer, while LST had a clear trend decreasing towards the ocean Air temp did not show a similar pattern. This proved to be a major area of uncertainty in our results.

Next we changed our landuse raster to use one provided by the City, converting from a polygon to a raster we chose a resolution of 15m, compared to 30m of NLCD.

The increase in resolution provided a smoothing effect and eliminated the low values present in the 1-100m bin from the NLCD data. Upon inspection of the NLCD raster we saw cells that overlapped green areas were classified as something else. The result being that green area temperature values were ending up in the 1-100 m bin.

Lastly we looked at how green area cooling changes at nighttime.

The result shows a much more subtle effect which makes sense since shade is no longer a factor in cooling.

So how did this information change our inputs? Since dcool is described in the documentation as “the distance over which a green space has a cooling effect” we played around a little with looking at where the slope was zero and looked at a few modeled curves and eventually…. Stuck with the default (400m).


I hope this helps anyone else who is interested in refining this input. While we did not end up altering the GAMCD it did help us better understand the model behavior and the parameter. This was all a learning process and could certainly be improved on so feel free to ask questions!