Exponential decay in the habitat quality model

I’m doing tests on the impact of roads on habitat quality
I set 7 different threat distances for the roads using exponential decay, but other than that all other conditions are the same.
According to the model, the expected result should be that the greater the impact distance, the worse the quality of nearby habitat. If displayed on ArcGis, it should be an increasingly larger buffer.
But the results show that as the threat distance increases, the buffer zone will first become larger and then smaller.
Maybe my understanding of the model is wrong, so I’m asking for professional explanations.

The threat distances I set are: 0.5 1 2 3 5 7 10(Km)

Sensitivity.csv (440 Bytes)
Threat.csv (104 Bytes)

quality_c_500m.tif (3.8 MB)

quality_c_1km.tif (4.7 MB)
quality_c_2km.tif (6.2 MB)
quality_c_3km.tif (7.4 MB)

Hi @hongkongdoll , good question. This is a little bit counter-intuitive. Would you please upload the logfile from one of these model runs so that we can see which version of invest you are using, and other useful diagnostic information.

Thank you,

I uploaded the corresponding work logs of the four impact distances
As for the 5, 7, and 10km maps, they cannot be uploaded due to their large capacity.

InVEST-natcap.invest.habitat_quality-log-2023-09-27–20_45_44.txt (6.2 KB)
InVEST-natcap.invest.habitat_quality-log-2023-09-29–11_09_32.txt (6.2 KB)
InVEST-natcap.invest.habitat_quality-log-2023-09-29–11_24_55.txt (6.6 KB)
InVEST-natcap.invest.habitat_quality-log-2023-09-29–11_27_19.txt (6.4 KB)

Thank you for these logs. I see you are using version 3.12.0. Would you mind updating to the latest version? InVEST Downloads, User Guides, and Data Sources

This model had some significant changes added in version 3.13.0, specificically related to the distance decay functions. The changes are described as follows,

The model now uses an euclidean distance implementation for decaying
threat rasters both linearly and exponentially. Since InVEST 3.3.0 a
convolution implementation has been used, which reflected how
the density of a threat or surrounding threat pixels could have an
even greater, cumulative impact and degradation over space. However, this
was never properly documented in the User’s Guide and is not the approach
taken in the publication. The convolution implementation also produced
degradation and quality outputs that were difficult to interpret.

Could you please try the latest version and let us know if the results make more sense? Thank you,

Of course, I am happy to see this explanation, which means that my understanding of the model formula may be correct.
I will update the version and try again, and will get back to you if there are any results.

I updated the model and reran.
The results look as expected and very reasonable
You can use InVEST’s guide as explained
As the buffer zone expands, the scope of influence spreads like water waves.
thank you for your help

1 Like

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.