Watershed scale - choosing the right scale for my analysis

Dear Community,

I am presently engaged in assessing three ecosystem services using the INVEST model: Sediment Delivery Ratio (SDR), Seasonal Water Yield (SWY), and Urban Cooling. My work encompasses multiple locations, but I am encountering challenges in determining the appropriate scale of analysis for two specific case studies: Natal and Osasco, both cities in Brazil.

Osasco’s urban footprint is extensive, covering nearly 94% of its area, and the spatial resolution of my land use and cover (LUC) raster is approximately 30 meters. While I have acquired raster data for a substantially larger territory, I am currently refining the analysis to fit within the confines of Osasco’s municipal boundaries. Given Osasco’s high degree of urbanization, significant results are emerging exclusively in non-urban spaces. This observation prompts me to question whether I should expand the analytical boundary to encompass adjacent regions outside the city limits, thereby capturing the full spectrum of ecosystem services these areas provide.

What are the potential limitations and considerations of broadening the study area beyond Osasco’s municipal borders? Similarly, Natal’s situation mirrors that of Osasco. I would greatly appreciate insights into the implications of scaling up the area of analysis for urbanized contexts such as these and any guidance on how to proceed.

Thank you for your time and expertise.

Best regards,

Nathan Debortoli

Hi Nathan -

It’s a little hard to mix these models, so good to think about how to do it. SDR and SWY were designed for more of a larger, mixed landscape, where urban is a relatively small part compared with forest, agriculture and other land uses. But Urban Cooling is the opposite, it was designed for specifically urban centers, not the larger, mixed landscape that SDR and SWY are.

What is the reason that you’re interested in water yield and sediment? If you’re analyzing water and sediment flow to something like a water treatment plant for the city, then I would recommend running these two models on the entire watershed that drains to the water treatment plant (or whatever might be of interest). Then run Urban Cooling in a separate study area that only includes the city.

If you’re interested in water yield mainly for flooding in the city, there is a separate Urban Flood Risk model that you could consider using. In general, urban areas function very differently than more natural landscapes, so if you need to focus on the area within the city itself, I’d say to use the urban-specific models.

Let me know what your interest is in sediment and water yield, and we can think about the best approach.

~ Stacie

1 Like

Hi Stacie,

Thank you for your prompt and insightful response to my previous inquiry. We are currently delving into the potential contributions of urban and peri-urban agriculture to ecosystem services within our urban environments, and your input is highly valuable to our research.

I found your perspective on the suitability of applying the Sediment Delivery Ratio (SDR) and the Seasonal Water Yield (SWY) models at scales larger than the municipal level particularly enlightening. It resonates with my understanding that these considerations often necessitate a broader scope. However, given our specific focus on urban and peri-urban agricultural zones that sit on the edges of municipal boundaries, I am seeking further advice on the appropriate extent to which we should apply the watershed scale in our analyses.

The challenge we face pertains to the fact that multiple significant watersheds intersect many of the municipalities we are studying. It would seem impractical to limit our scope to a single watershed in such cases. Nonetheless, expanding our analysis to encompass the full areas of these watersheds would extend several kilometers beyond municipal borders, which might dilute the relevance to our urban agricultural context. Could you share any strategies from your experience on how to effectively approach this complex spatial delineation?

Additionally, in reviewing our SDR findings, I observed that the estimated values for avoided erosion are exceedingly high, reaching into the billions of tons. This is especially true for municipalities with extensive territories, such as Manaus in the Amazon region and Brasília in the Cerrado, which also happen to be sources of major rivers in Brazil. In your experience, are such large figures for avoided erosion commonplace in tropical zones, or should this prompt a reassessment of our modeling parameters? I played a bit with the L values between 122-300, it did not change the results considerably. I also tried the flow accumulation between 100-500 and also did not see much changes.

I look forward to your valuable insights on these matters, which will undoubtedly enhance the precision and applicability of our study.

Any help on this regard would be very welcomed.

Thanks a lot!

Kind Regards,

Hm, that’s an interesting situation. If you do include large natural areas around the cities, then yes, they are likely to really dominate the results, which is often not useful.

What are the particular aspects/services/disservices of the agricultural zones that you want to analyze? And how large are they, relative to the urban centers? (It might be useful to see a map).

If your study area is mostly urban, with a fringe of agriculture, then I’d advise using the Urban models, which will work on both. However, there isn’t an urban sediment model, and the Urban Flood Risk and Stormwater Retention models are rather different than SWY in the information that they provide.

Honestly, my work is mainly not in urban centers, so it would be good to hear from others with more experience.

As for the SDR results, yes, the numbers can be quite high, especially in mountainous areas, and especially if you’re aggregating over large areas. Which version of Workbench are you using? The latest version (3.14) has updates that have tempered the very high values quite a bit. Also, we always advise calibrating model results against observed data before trusting the absolute values.


1 Like