Confluence Retirement

In an effort to consolidate USGS hosted Wikis, myUSGS’ Confluence service is scheduled for retirement on January 27th, 2023. The official USGS Wiki and collaboration space is now SharePoint. Please migrate existing spaces and content to the SharePoint platform and remove it from Confluence at your earliest convenience. If you need any additional information or have any concerns about this change, please contact myusgs@usgs.gov. Thank you for your prompt attention to this matter.
Skip to end of metadata
Go to start of metadata


The page is initially focusing on the work of Dr. Sinan Abood (USFS), who has been developing ideas and software for delineating riparian areas. He's happy to share ideas and software, so please feel free to post ideas and questions in the "Comments" section and we'll try to get a dialog going.
One of the main ideas of this work is that a fixed or really even a per-segment Euclidean distance buffer around streams is a poor way to approximate "riparian" areas. There are a variety of interesting questions in here, starting with exactly what the definition of "riparian" should be. Regardless, various users might see ways to leverage the delineations for their own scientific studies. It might be interesting to see if this type of approach would also be useful for defining "wetlands" that are either on- or off-stream.

 

An image taken from Sinan's presentation is great food for thought.

His approach relies on a flood height (usually the 50 year) that the user specifies. The image below shows an example of the impact of this choice. 


For more information on the tool see here. The tool requires the following minimum inputs for this analyis:

  1. 10 meter DEMs.  Acquired all CONUS 10m DEMs from Curtis Price along with a mosaic-ing and projecting tool.

  2. NHDPlus Version 2 flowlines.  NHDPlus version 2 from horizon

  3. Sub catchments of area of interest. HUC8's were used in this analyis.

  4. National Wetland Inventory. Acquired CONUS layer from USFWS.

  5. NHD High Res Waterbodies
  6. 50 Year flood heights

The tool basically walks up the river course and searches out a specified extent and samples the DEM based on a specified flood height.  This tool then finds areas along the river course that would correspond to this specified flood height and essentially does a cut and fill operation.   It also looks for water bodies, wetlands and specific soil types (poorly drained soils) that are tangent to the water course which may also indicate a riparian area. We have all the input data needed nation-wide and have the tool up and running.  We were able to get some test results but the question I am posing to this group concerns the estimation of flood heights. 

 

The tool allows for calculating flood heights based on NWIS Field Measurment Data.  These data are compiled from NWIS.  The idea is to transfer 50 year flood heights to the NHDPlus flowline layer for input to the tool. Before running the model, a determination of an appropriate 50-year flood height for each stream order for all flow lines is necessary for input into the tool. To estimate flood heights, data is compiled from NWIS sites which occur within or near each of the areas where riparian areas are being delineated. 

 

To calculate 50 year flood heights, the following NWIS data is used: the annual average stream flow, channel velocity, channel area, and channel width. The annual average flow rate measurements are organized by year and sorted from fastest (peak) to slowest for each stream gage location. After sorting, the annual flow rate measurements are ordinally ranked, so the fastest (peak) flow rate receives a value of 1. To calculate the recurrence interval, the rank number is divided by the number of measurements. The flow rate is plotted against the logarithmic recurrence interval to develop a flood occurrence regression (Bedient and Huber, 2002). As an example, an individual site regression is shown in Figure 2a below. The cross-sectional area (flow rate divided by velocity) is plotted against flow rate measurements (Figure 2b). Figure 2c shows the regression of width versus cross-sectional area. The width and cross-sectional area are determined from the previous regressions and the stream height calculated by dividing the cross-sectional area by the width (Mason, 2007). Using the regression equations for each site, 50-year flood heights are determined.

                   

 


This 50-year flood height calculation is repeated for all available gages within or in a close proximity of the area of interest. The next step is plotting calculated (if there are more than one gage on the same stream order an arithmetic average is used)  50-year flood height values vs. streams order then fit a second order polynomial equation to estimate 50-year flood height values for streams order with no gages (graph below). Then these values are attributed to the NHDPlus flow line layer as an input to the riparian tool.

 


note that there are also comments attached to the Trial Areas page, as well as in the forum.

  • No labels

11 Comments

  1. Pete VanMetre wrote: 

    I took a look at your riparian tool site (https://my.usgs.gov/confluence/display/cdi/Riparian+Tool+Trial+Areas) and have a couple of thoughts. First, in spite of uncertainties, I think this is a big step in the right direction compared with using a fixed interval buffer (50 m). The two examples you show on there look good.

    On your question of how to group similar gauges for estimating the 50 y flood height, my call is to use hydrologic landscapes. They factor in most of the information in other options you mention (physiographic regions, climate). There is a question of what level to use, but at our RSQA scale, I'd use one of the more detailed levels of delineation. I pulled this off the web and looking at SESQA, it divides up mostly into four HLRs, probably based on slope going from coastal plain to mtns. Although Daren is better qualified than me to answer this question...

    I also think this is safer to use on the RSQA regions than nationally, at least not without some testing. Our regions are all pretty tame compared with the country as a whole (no deserts, Gulf coastal plains, high mountains, etc.). Our two western regions are in lower, wetter, and more level areas than a lot of the west, so it might work OK.

    Having spent a lot of time in small streams the past 5 years (eco surveys), I would incorporate a 10 m minimum buffer width (assuming that is measured from a line feature that in theory is channel center). There are a lot of places where small streams have high banks (incised streams in particular) that probably exceed the 50 y flood, so those will go to zero buffer with the new approach and our GIS overlays will not capture what's going on there. The difference between forest on that high bank or the parking lot for an apartment complex is important to us.

  2. Greg Schwarz wrote: 

    Took a quick look at your link referenced in the email and I'm a bit confused. I would think the 50 year flood height would be simple to compute from the stage data at a station - using standard statistical methods for computing percentiles from a sample. Why all the incorporation of channel velocity, width, height, etc. And this sounds like a flood plain analysis - not a riparian zone analysis.  And, why do we need to group gages? You can estimate a 50 year stage at an individual gage, and this estimate for the particular spot on the network will likely be more reliable than an estimate based on a grouping of gages. Having a lot of red flags popping up on this.

  3. Sinan Abdood wrote: 

    I think the easiest way to answer your colleague is to tell him to read my paper. It has all the details regarding why we chose using 50 year flood height and why we calculate it 50 year flood height that way. I don’t know about calculating 50 –year flood height using stage data and standard statistical methods for computing percentiles but I am very interested to learn, so it would be great to follow up with your colleague regarding that and maybe we could compare both methodologies.
    Why we group gages because most gages shows a range of 50-year flood heights at a given stream order. also sometimes there not enough data to calculate 50-year flood heights for all streams order within our network so in this case we need to include more gages located far from our target HUC12, or HUC10, or HUC8, or even HUC6 in order to have enough measurements to plot 50-year flood heights values vs. stream orders plot. Hope this helps.

  4. Ken Eng wrote: 

    There are some concerns about how the 50-yr flood heights are being transferred to ungaged locations by using the poly regr (predictand=50 yr flood hts and predictors=stream order). 1) the example provided shows a single point for each stream order, but when you use dozens to hundreds of sites across a large area you will have a lot more points for each stream order and they will have a lot of variation that will significantly increase the error of your predictions and reduce the fit of your poly regr. 2) the distribution of points will also be disproportionately heavy on the high-stream orders and lighter on the low-order streams because of the way the USGS monitors streams, which will give you different confidence in your predictions at different stream orders. 3) the method also ignores the impacts of human alteration, such as land use and dams, by making this relationship only between flood heights and stream order (floods are substantially impacted by land use and dam operations). A lot of the ungaged locations with different mixtures of human alteration will often not be represented well by the limited network of streamflow gauges that monitor natural to altered waterheds, so predictions will likely be erroneous/unreliable. 4) the error of the method needs to be tested and documented better (bootstrapping the error distribution would help). Grey areas should be identified and mapped where the flood hts are within the error of the poly regr error.

  5. Mike


    My gut tells me that, even if you could accurately locate the 50-yr flood height at gaged locations and then extrapolate those data to all stream segments (which seems extremely difficult by itself), the underlying premise (that riparian extent is related to the 50-yr flood height) is probably not applicable across large swaths of the CONUS.  I can see how a 50yr flood height would create a terrace or strip of riparian vegetation in UNDISTURBED areas in MOUNTAINOUS/HILLY terrain.  But across most of the central and southern plains, streams are incised and/or channelized.  So a 50yr flood height has nothing to do with local geomorphology and vegetation.  

    I realize that the 50yr flood ht is just 1 of 6 ingredients in the model, but I fear the massive effort required to obtain this ingredient for the CONUS will in the end produce estimates that are not much better than a wild guess for non-mountainous areas.

    1. maybe get some example results from potential problem areas? Would it be possible to pick areas w/other sources against which to check the results (beyond visual interpretation)?

  6. Allen Gellis wrote:

    Mike

    Really good work in developing this tool for NAWQA needs.  You are providing much needed data to help us  understand the role of floodplains in ecologic and sediment assessments.

     I have a couple of questions and thoughts on the tool

     1)    First of all I agree with Pete that extrapolating out using hydrologic landscapes (HLU) makes sense. But if I am correct, they did not use land use in the HLU delineation.  A floodplain in an urban stream vs forested in the same HLU could look completely different.  Perhaps subdividing the HLU by land use may help in extrapolation.   I guess we can look at the model output in different land uses under the same HLU to see if this is occurring.

    2)    Are you estimating floodplain width based on a 50 year flow?  If this is correct, is this what you are calling the active floodplain? In stable streams, flows start to go overbank in ~ 2 year recurrence intervals.  

    I emailed Sinan Abood , the creator of your tool some time ago about how he estimated the RI for floodplain width.  He told me: Flood height is what I calculate and not Bank height. 50 year flood height is the height of water that intersects with first terrace and upward sloping.


    Sinan is using the 50 year RI, because he thinks this is a terrace level. I still don’t know what this means from a geomorphic standpoint.

     I think caution should be observed when using this tool.  We might over-extending the floodplain width in stable channels with this method.  In incised, streams this could make more sense.   A 50-year flow is a large event.  Perhaps this is important ecologically, but for sediment this is a really large, infrequent event. Perhaps we can test other frequency of flows. i.e.d20 years and test if this has better statistical meaning.

    3)    You are calculating a 50 year flood height at the gaged stations using the following: the annual average stream flow, channel velocity, channel area, and channel width.  Have you compared your results to the program  the USGS uses "PEAK FQ"?  I think this would be useful in verifying the tool’s output

     

    4)   I also emailed Abood Sinan Ayad -FS to see if they have verified his model in the field.  He said, Yes we did accuracy assessment on Hiawatha National Forest --- which is in Michigan.   

    This is great that they tried to verify the model.  But a forested watershed in Michigan may only work for that region.

    Another suggestion would be to check some of your results with field data.  I know we are doing this for Smith Creek, VA, but we might want to think about other areas of the U.S.

    1. Using PEAK FQ is a good suggestion! Am going to have this installed and do a comparison. Will post results.

      Also, I plan on doing more of an intensive comparison with your Smith Creek work, will sit down with your group and do more of an analysis.

  7. Greg Schwarz wrote:

    I have a number of concerns with the stated approach. First, what is special about the 50-year flow event with regards to riparian areas? My understanding is that with respect to biological habitat in the stream the riparian condition primarily refers to vegetation adjacent to the stream. I don't immediately see how mapping an extended buffer area based on episodic flooding events relates to a characterization of the habitat.

    An extended buffer area based on episodic flooding events would be relevant to describing interactions between the stream and inundated land surface during flood events, and may be very important for identifying locations where sediment loads carried during flood events may be deposited or mobilized. So delineating these areas is certainly relevant to water quality, and maybe indirectly to habitat. But if that is the purpose, I don't see why it is useful to define a regional 50-year flood stage. Each stream has unique characteristics that will affect how flow is related to stage - which is why each streamgage is individually rated. Rather than define a regional 50-year flood stage, it would seem to make more sense to define the 50-year stage at each gage, and use this to define inundated area upstream of the specific gage (with some allowance for water-surface slope during flood events). For areas with no downstream gage, a regional approach may be necessary, but the vast majority of the landscape could be addressed with individual gage records.

    Lastly, I am aware of another approach being developed that uses the stage-discharge relation at a gage to determine the stage at which the stream is at a bankfull condition. The method then uses that condition to estimate the total flow in the stream in excess of bankfull flow, representing the flow that inundates the flood plain. An accumulation of that flow over the period in which stage exceeds bankfull stage gives an estimate of the total volume of water inundating the flood plain upstream of the gage. If the depth of water on the floodplain is the same everywhere upstream (a strong assumption), then an estimate of the upstream inundated area could be obtained by dividing the inundation water volume by the average depth, represented by the average stage above bankfull stage. The delineation of floodplain area would then proceed by determining the critical upstream elevation below which the total area equals the estimated inundation area - a non-trivial GIS calculation to be sure, but computers work cheap. This calculation differs from the one implemented by the tool in question in that it partially accounts for water surface slope during the flood event. It is only partially accounted for because the translation of inundation area into a delineated area presumes all areas of equal elevation are either inundated or not inundated. A more refined calculation would allow this elevation to vary with distance upstream of the gage.

    Lots of interesting attributes can be computed from the analysis just described. For example, by knowing the duration of the bankfull condition and the distance upstream included in the inundated area, it is possible to compute the average velocity of flow in the floodplain, that estimate being the ratio of the upstream inundation distance to the flood duration. It would be of further interest to see if this velocity responded to features mapped in the floodplain, such as forests. Presumably, velocity is slower in floodplains containing obstacles to the free movement of water. This, in turn might better identify locations where sediment is likely to deposit during flood events (the slower velocity areas), or where it is likely to mobilize (the higher velocity areas).

    The above describes the determination of floodplain area for a given flood event. It would be harder to determine the area for a hypothetical 50-year event (unless a 50-year event was recorded at the gage). The problem is in knowing the duration of the flood event. But it may be possible to make a statistical estimate based on extrapolation of observed durations and relating to known recurrence frequency.

    Sorry for the length of this comment but this topic is of considerable interest to me. Thanks for your patience.

  8. Peter McCarthy wrote:

    Mike,
    Thanks for putting together the documentation and requesting input. I think this is a really interesting and thought provoking method for defining the riparian zones and has a lot of merit to the methodology.  I have included some discussion points below that may either result in slight changes to the methods and/or to your terminology.

    • From a terminology perspective, if you continue to use the methods you've described I would be happier if you chose not to refer to your estimate in any flood related percentile or recurrence as these terms have very specific connotations and methods.  If you truly are looking for the 2-percent annual exceedance probability flood then I would recommend retrieving these values from USGS reports, from the StreamStatsDB for gaging stations, and/or solving equations for ungaged locations.  Unfortunately for your project, these data may not be available everywhere and nationally consistent in terms of data used.  We can get into a much longer conversation if this is the direction you chose.  
    • You may still want to use data from the peak flow file for estimating your recurrence interval discharge. My understanding of the methods used is that the streamflow measurements are representative of the population of streamflows and recurrence intervals which is not true (this is my opinion and a statistical liklihood, not a statistical fact). The goal of the streamgaging program is to measure at or near the peak Q, but this may not consistently happen for all gages, and probably occurs less often as the drainage areas decrease and peaks become flashier and more difficult to measure on the peak. Once you have Q, then you can use the stage-discharge curve to retrieve the stage. 
    • Stage is an interesting concept that I think you can continue to use, but I would like to share some additional thoughts regarding how this may be qualified in your description of methods.  
      • In theory, a stage of zero should be near a flow of zero, but in many cases this is not true.  I believe a systematic review of stage-discharge curves will provide you a better feel for how accurate this assumption is. My impression from what little streamgaging I worked around is that this is more likely to be true for smaller streams where the point of zero flow can more easily be identified and used to set the stage or gage datum of zero. Defining a gage datum of zero in large channels is more challenging and may not well represent the point of zero flow. 
      • The gage datum is supposed to be tied to NAVD88, however the quality of this relation varies greatly for gages.  Any active gages are supposed to have surveyed data referencing the elevation at a gage datum of 0; however I know that the elevation for many gages is still scaled from a topo map.  
      • I should know more about DEMs, but does a DEM cell over a body of water represent the channel bottom or the water surface, and how accurate would this be?
      • In combining these thoughts, I think that by adding the stage to the DEM could be largely inaccurate. This could potentially be negated by finding the difference in stage between a 50- or 100-year event and the stage at bankfull (~2-year event).  This difference would negate the issue of gage height at zero flow, and potentially negate the issue of the DEM accuracy in the channel. The flip side being that you would need a near channel overbank elevation.
    • Regarding your question on how these gages should be grouped, I am curious about relations between precip vs. gage heights and/or average basin or channel slope vs. gage height.  Also, I'm wondering if a regression equation could or should be developed using basin characteristics rather than stream order might be more accurate. 
    • What thoughts do you have on a lower limit to your methods for determining riparian zones?  With that question I am thinking about a lot of streams that dry up and don't have riparian zones, vs other streams that dry up, but still have riparian zones, and how the consistency of lower flows can impact the riparian zones.  

    Obviously I have lots of questions, many of which I'm sure you've already thought through and is not news to you.  As I started off with, I think this is a very interesting project and method for determining riparian areas in a consistent national model.  If you'd like to discuss any of my thoughts or questions feel free to give me a call.

    1. Mike, 

      Thanks for adding these comments. I finally logged in today to see you've already done it!


      Pete