Contacts:
- Primary: Rafa
- Contributors: Lisa, Adrian, Hector, Jesse, James, Stacie, Rafa, Doug, Emily
The SDR model is one of the most widely applied InVEST models (Mandle and Natural Capital Project, 2021). Yet, it comes with a number of assumptions and creates a number of output indicators that are deprecated, poorly documented, and do not meet the current needs. This has also led to a rising number of user forum inquiries about the functioning of the model and how to interpret the resulting indicators. This PEP results from an in-depth review of model assumptions done by a group of expert NatCap users, scientists, and software engineers, and aims at streamlining the SDR model by (1) revising some indicators to reflect the conceptual understanding of the science team about model processes, based on the current state-of-science and (2) reducing the number of indicators to those which are most useful to current users.
Two of the most widely appreciated benefits of sediment retention by vegetation are 1) improved water quality from avoided sediment export from hillslopes to waterways, and 2) maintenance of soil fertility from avoided topsoil erosion. The current SDR model lacks indicators that clearly and comprehensively represent these services, despite the demand for such indicators both in NatCap applications and from users on our forum. This PEP would introduce two new sediment retention indicators that more comprehensively capture the ecosystem service being provided, one from the perspective of water quality and one from the perspective of soil fertility. In the case of sediment retention for water quality, we propose an indicator for sediment retention services for water quality that captures the ability of vegetation to both keep sediment from eroding and to trap sediment originating upslope. We also propose dropping several current model outputs that do not address the service as directly and are either (1) deprecated (were kept in the model at some point to ensure backward compatibility), or (2) could be calculated by users with a scenario approach (e.g., ask the user to run a “bare-soil” scenario, rather than having it predetermined/fixed in the code), or (3) do not fully represent what the PEP authors agreed that they should be representing. These could still be calculated if desired from intermediate outputs or by running an additional scenario.
This proposal would additionally alter the current calculation of the deposition and flux indicators that track trapping of sediment that erodes onto a pixel in order to make them more consistent with the conceptual processes the SDR model represents. Beyond those code updates, this PEP would also update the User Guide to harmonize terminology and improve clarity for users (including some new visualizations and description of use cases).
We thus propose that these revisions of SDR are better tackled through a PEP (and joint effort of the science team), rather than a piece-meal approach to solving individual problems with indicators and the User Guide in response to user/staff requests.
The revised SDR model seems to be highly aligned with the general PEP guidelines:
- Actionable: The SDR model has been shown to be relevant to many different stakeholders, decision problems, and scales
- Generalizable: The tool has been run from hyper-local case studies to global scales
- Innovation: The revised model will produce new indicators which, based on our own experiences and questions received on the User Forum, will be very useful for decision support (e.g., Retention indicator) and that a normal user could not easily calculate. The new indicators will further respond to issues that have been repeatedly raised in the User Forum.
- Credibility:
- The proposed changes have been vetted by the science and software team, and seem to be more aligned with the model's intention (e.g., as described in the User Guide).
- Similar to current SDR, the revised indicators we propose are mostly conceptual and hard to validate. Yet, the revised indicators are more aligned with the intention of what is described in the User Guide, avoid violating sediment mass balances (e.g. that more sediment is retained than what is eroded) and allow users to track sediment mass balance through the model in a way that was not possible before.
- Generalizability: as per above, the model has been shown to be useful in many settings and decision contexts.
- Feasibility: Calculation of all indicators will be possible within the existing model framework (James has already implemented these changes in a fork of InVEST).
Additional criteria that might be relevant
- Demonstrated interest: There are many threads on the forum where users ask for clarification regarding existing indicators
- Novelty: The proposed changes will make the model more consistent compared to what we have and will produce indicators that are otherwise hard to obtain by users from “raw” SDR outputs.
- Maintenance: The models will be maintained as part of InVEST, just like previous versions of SDR. Usability: SDR has been proven to be usable across scales, geographies, and decision making contexts.
No additional support is needed beyond what is done for InVEST. Some additional help might be required for forum users inquiring about changes, although we expect this to be offset by the reduction in time required to respond to forum posts asking for clarification on the current set of model outputs/indicators.
This PEP proposes the following changes, which we outline here and specify in more detail below:
- Update the calculation of intermediate parameters R (deposition) and F (flux). Currently R and F are calculated such that sediment that erodes from a pixel (as calculated by the Revised Universal Soil Loss Equation or RUSLE) can then be trapped by vegetation on that same pixel. This is conceptually inconsistent: the role of vegetation for reducing erosion and sediment runoff from a pixel is already captured in RUSLE’s C factor (Wischmeier and Smith, 1978). By allowing for immediate sediment trapping on the same pixel, this amounts to double-counting the role of vegetation. We propose updating the calculation so that all sediment that erodes from a pixel goes to the next downslope pixel in proportion to the proportion of water flowing into the next downslope pixel, where it can either be trapped or continue flowing downslope. This change will not affect estimates of water quality for any given scenario relative to the current formulation of the model; it will lead to some change in the attribution of where sediment retention services are being provided on the landscape. Rename the indicator “R” (which is also used in the USLE equation) to “T” for “trapped sediment”, or “trapping”.
- Create a new output and indicator for sediment retention services for water quality called "Avoided Export". The raster output represents vegetation’s contribution to both avoided erosion from a pixel and trapping of sediments originating on upslope pixels, e.g., the service of a pixel from the perspective of downstream water user.
- Create a new output and indicator for sediment retention services for soil fertility called "Avoided Erosion". This raster output represents avoided erosion benefits to that pixel, e.g., from the perspective of maintaining soil fertility.
- Drop 2 legacy sediment retention indices currently included in the model. One of the frequently asked questions on the User Forum is what metric people should use for the sediment retention “service”, and we have not historically had a solid answer, because neither of the “retention” outputs are very useful. They are indices only (not quantities), compared against a hypothetical entirely barren landscape (which is unrealistic), and one of the indices has no explanation at all given in the User Guide, so no current NatCapper actually knows what it means. Removing these, and adding clear guidance about calculating retention to the User Guide (see #5) would significantly reduce confusion.
- Updates to the User's Guide to reflect the above changes, harmonize terminology and variable names and provide a conceptual figure for visualizing the modeling steps and related outputs. We will also more clearly explain what’s happening in the different modeling steps and present options and use cases for quantifying the retention “service” from model outputs, informed by common questions on the User Forum.
The required technical changes to address those 5 points are detailed below:
-
Update the calculation of the intermediate parameters R (deposition) and F (flux).
-
Update the
$R$ (deposition) calculation so that it does not include export from pixel i. Basically removing the$E_i’$ from the equation below so that a pixel cannot retain the sediment that erodes directly from itself (note: equation numbers are matched to the User Guide). Thus from$$ R_i = dR_i \cdot \left(\sum_{j \in {pixels\ that\ drain\ to\ i}}F_j \cdot p(i,j)\right) + E_i' $$
Revise to
$$ R_i = dR_i \cdot \left(\sum_{j \in {pixels\ that\ drain\ to\ i}}F_j \cdot p(i,j)\right) $$
Rename the indicator from “R” (which is used in the USLE equation) to “T” for “trapped sediment”, or “trapping”.
-
Update
$F$ (flux) calculation so that all export/erosion originating from that pixel goes on to the next downslope pixel or pixels (rather than just a fraction). Thus from$$ F_i = (1 - dR_i) \cdot \left(\sum_{j \in {pixels\ that\ drain\ to\ i}}F_j \cdot p(i,j)\right) + E_i' $$
Revise to
$$ F_i = (1 - dR_i) \cdot \left(\sum_{j \in {pixels\ that\ drain\ to\ i}}F_j \cdot p(i,j)\right) $$
In addition, we would rename variable
$R$ (deposition) to$T$ (for sediment trapping), to avoid confusion with the$R$ variable name already used previously in the calculation of the$RUSLE$ , and update variable names in the User Guide for consistency and clarity.
-
-
Create new ecosystem service indicator output (raster) called “ Avoided Export” representing vegetation’s contribution to avoided erosion from a pixel and trapping of sediments originating upslope (the service from the perspective of a water user):
$$ TR_i = (RKLS_i - USLE_i) \cdot SDR_ + R_i $$
Where:
-
$TR_i$ is Total Retention -
$RKLS$ is basically$USLE_i$ minus the$C$ and$P$ factors which represent the benefits of vegetation and good management practices.
The watershed-level outputs of the model would also be updated to include this indicator as follows: remove "sed_retent" and add the sum of "Total Retention" per polygon and the sum of "Local Erosion Control" per polygon.
-
-
Create new ecosystem service indicator output (raster) called “Avoided Erosion” representing avoided erosion benefits to that pixel, from the perspective of soil loss:
$$ AE_i = RKLS_i - USLE_i $$
note (in the User Guide) that
$R_i$ ($T_i$ in the new nomenclature) could be added separately if trapping of sediment originating upslope could mitigate erosion on a downslope pixel (e.g., if high quality soil is eroded from upslope) -
Drop 2 legacy sediment retention indices currently included in the model Described in the section "Legacy" in the user's guide.
-
Updates to User Guide (can happen separately from the code changes requested above):
- Update the User Guide to reflect changes in equations 78 and 79 (and associated, more explanatory, text)
- Add use cases to the User Guide for applying different outputs
- Add a figure showing conceptually what the different variables in the model are and how they relate to a toy landscape and which variables are output as the retention indicators given above.
There is an additional issue with the LS factor, which James will address separately. The current LS factor values are very high compared to what is obtained from other software, e.g., SAGA. The SAGA LS factor also results in more realistic results when compared to sediment observations.
Mandle, Lisa and Natural Capital Project. (2021). Database of publications using InVEST and other Natural Capital Project software. Stanford Digital Repository. Available at: https://purl.stanford.edu/bb284rg5424
Wischmeier, W. H., & Smith, D. D. (1978). Predicting rainfall erosion losses - a guide to conservation planning. Predicting Rainfall Erosion Losses. United States Department of Agriculture.