Using NEXRAD and ArcGIS to quantify precipitation from debris flow producing storms in the canyons of the eastern Uinta Mountains, Colorado and Utah

CEE 6440 GIS-Water Resources Term Project
Isaac Larsen
Utah State University
Geology Department
Fall 2002

Table of Contents
Introduction and Project Objectives
Study Area Description
Data Sources
Methods
Results
Discussion
Acknowledgments
References
Appendix
 

Introduction and Project Objectives
        Debris flows are important geomorphic agents and are significant geologic hazards in areas of high-relief terrain. In the canyons of the eastern Uinta Mountains, debris flows strongly influence the geomorphic arrangement of the Green River. Debris flow deposits (Figure 1) constrict the river and form a series of fan-eddy complexes (Schmidt and Rubin 1995).  Rapids created by debris flows are responsible for most of the drop in the longitudinal profile of the river and eddies downstream from debris fan constrictions are primary areas of fine-sediment storage within the river channel.  Debris flows are also significant sediment sources in canyon rivers (Melis 1997).  Despite the important linkage between river morphology and debris flows, the watershed conditions associated with debris flow initiation in the Green River canyons of the eastern Uinta Mountains are poorly known, especially the characteristics of the precipitation events that produce debris flows.
        Between September 1997 and August 2001 thirteen debris flows initiated from tributaries of the Green River in Dinosaur National Monument.  These thirteen debris flow events resulted from four separate rainstorms.  The nearest rain gages to the tributaries that produced debris flows are between 5.9 and 12.5 kilometers away, depending on the individual tributary. The spatial variability of rainfall from the storms makes it difficult to characterize precipitation based on data from these gages.  Through the use of NEXRAD (NEXt generation doppler RADar) data it may be possible to better characterize the precipitation events that produced the 13 debris flows.
        Rainfall intensity and duration have been shown to influence the initiation of debris flows and rainfall intensity/duration thresholds have been developed for some areas (Weiczorek 1987).  However, due to climatic and geologic dissimilarities, rainfall intensity/duration thresholds derived for one region cannot be applied to different areas (Wilson 1987).  The first objective of this project is to determine total rainfall, rainfall duration and rainfall intensity for each storm that produced a debris flow. Although a number of 13 is probably too small to develop robust rainfall intensity/duration thresholds, it is a first step towards understanding the hydrology of the debris flow producing storms.  The second objective is to determine if rainfall intensity thresholds for catchments burned by recent wildfires are lower than those for unburned catchments. This is to determine if hydrologic changes associated with wildfire could potentially be related to debris flow initiation.  The third objective is to determine the aerial extent of the storms in order to determine if localized or large frontal storms produce the debris flows.  The fourth and final objective of this project is to evaluate the accuracy of the NEXRAD data in order to determine if rainfall values derived from the radar data are reliable.
 

Study Area Description
        The Green River cuts through the eastern end of the Uinta uplift, forming three major canyons: Canyon of the Lodore, Whirlpool Canyon, and Split Mountain Canyon (Figure 2). A variety of Precambrian and Paleozoic sedimentary rocks are exposed in the canyon walls, including resistant sandstone and limestone units and more easily weathered shale units. The area has a semi-arid climate that supports a pinion-juniper forest. There have been several severe forest fires in tributary catchments to the Green River in the last eight years and the resulting change in hydrology associated with these fires could potentially be related to debris flow initiation.  Debris flows initiate during the summer and fall when thunderstorms produce runoff in the upper portions of tributary catchments.  The runoff cascades down the steep canyon walls, inundating colluvium stored at the base of the cliffs.  The colluvium fails, initiating debris flows that continue to entrain material as they travel to the Green River.
 
 

                                                                         Figure 1. Lodore Canyon debris flow

                                                                    Figure 2. Map showing location of study area.
                                                                    The red area is the outline of the DEM that is shown in the following figures.

Data Sources
    The main components of Stage III NEXRAD data are the Digital Precipitation Array Products (DPA).  The DPA product used in this study is the operational hourly rain gage data, which are hourly radar-only estimates of precipitation collected on an approximately 4x4 kilometer grid.  This grid is referred to as HRAP (Hydrologic Rainfall Analysis Project), and in this study the grid is represented as a series of points. Each point represents the center of an HRAP grid cell and can be thought of as an individual rain gage.  For more details on NEXRAD data and information regarding its accuracy please follow the link below:
        http://www.nws.noaa.gov/oh/hrl/dmip/stageiii_info.htm

    NEXRAD data are available from 1996 to the present and are arranged in monthly increments in .tar files with the form "SiiiMMYYRFCID".  Within each monthly .tar file are daily .tar files with the form "SiiiMMDDYYRFCID" and each daily .tar file contains hourly zip files with the naming convention "xmrgMMDDYYhhz" The two digit codes refer to the month, day, year, hour, and RFC codes. The data for this project are for the Colorado Basin Forecasting Center (CBRFC).  Archived NEXRAD Stage III precipitation data can be downloaded from the National Weather Service River Forecasting Center at the following site:
        http://dipper.nws.noaa.gov/hdsb/data/nexrad/nexrad_data.html

    DEMís are available from the EROS Data Center at:
        http://seamless.usgs.gov/

    TauDEM, an ArcGIS toolbar plugin used for terrain analysis of DEM's is available at:
        http://moose.cee.usu.edu/taudem/taudem.html

    Digital Raster Graphics (DRG) of topographic quads were purchased from the link below but may be avaiable at no cost for other areas:
        http://www.mapmart.com

    Hourly rainfall data collected by Remote Automated Weather Stations (RAWS) are available at two locations near my study area. These stations are maintained by the National Interagency Fire Center and data can be downloaded from the following site courtesy of the Western Regional Climate Center:
        http://www.fs.fed.us/raws/

    The make_raindat_v5 software used to extract NEXRAD data was developed by Ross Woods (NIWA, New Zealand).  This program decompresses the monthly, daily, and hourly .tar and .gzip files for the number of months selected by the user. The program reads the hourly xmrg files and writes a space delimited .dat file with the naming convention rain_071997_081997_CB.dat.  This file contains the hourly rainfall data for each HRAP cell.  Make_raindat_v5 also creates a "latlong.txt" file that contains the lat/long coordinates and the HRAP identifier of each HRAP cell.
 

Methods
    The methods outlined below are adapted from instructions prepared by Christina Bandaragoda and Dr. David Tarboton (Civil and Environmental Engineering Department, Utah State University).  The methods are written in a step by step fashion in order to aid someone wishing to do a similar study.

-Step 1  Using DEM analysis to delineate contributing areas in order to determine where to focus precipitation estimation efforts.
    After installing TauDEM and loading the toolbar, add the DEM to the view. Then in order to do basic DEM processing such as pit filling, computing flow directions, and contributing areas choose the TauDEM operation Basic Grid Analysis<Do All to prepare the DEM.   The purpose of step one is to determine specific locations where rainfall from recent storms has contributed to debris flows. The upslope contributing area for each debris flow is where the runoff that leads to debris flow initiation is generated. This area can be defined as a watershed using the Network Delineation Tools in TauDEM.  Field mapping in the summer of 2002 determined the location of each debris flow initiation site and these locations were transferred from paper maps into a point shapefile and treated as 'outlets'. This was done in ArcCatalog and ArcMap with the the following  steps:
In ArcCatalog:
    1) Right click directory and select New<Shapefile...
    2) Select the name and set the feature class to point.
    3) Click Edit to edit the coordinate system. I worked in UTM Zone 12 with the NAD27 Datum.  Click Add, then OK twice.
In ArcMap:
    4) Add the DRG's for the study area
    5) Add the point shapefile to ArcMap using File<Add Data...
    6) With the Editor Toolbar select Editor<Start Editing.  Select the folder that contains the initiation site shapefile, and set this as the target layer.
    7) Use the create new feature task and the create new feature button to create points that correspond with the mapped initiation sites.
    8) Choose Network Delineation<Select outlets shapefile. This is the point file of initiation sites and the choose Do All. This Delineates the stream network and watershed upslope of
    each initiation site. This is the area where rainfall contributed to debris flow initiation and where rianfall quantification will be done (note: the location of initiation site
    'outlets' may have to be adjusted so that they fall on the delineated stream network). Similar methods were used to delineate
    each tributary catchment that produced a debris flow and to delineate a larger catchment that included all of the tributary catchments of interest.
 

-Step 2  Preparing the NEXRAD rainfall Grid
    The purpose of step two is to select the HRAP cells of interest from the larger data set.
    1) After downloading the NEXRAD files run the make_raindat_v5 program for the first month of interest. This will generate a latlong.txt file that can be opened in Excel and saved as a
    dbase file.
    2) Add the latlong table the view in ArcMap using Tools<add XY data... For the X Field choose Long and for the Y Field choose Lat and choose edit to define the coordinate
    system.  This displays the location of all the HRAP cells for the particular RFC as shown in Figure 3. (note: due to limitations in the size of an excel spreadsheet the eastern extent
    of the Colorado River basin is not covered. One possible way to remedy this may be to use a different program, such a Sigma Plot, that a higher row capacity).


   Figure 3.  HRAP points for the Colorado Basin Forecasting Center with U.S. counties and rivers.

    3) The next step is to add point data for the location of the RAWS rain gages. The latitude and longitude for each station are saved in a dbase table and added to the view as XY
    data the same way the HRAP cells were added.
    4) The next step is to select the points of interest from the large array of HRAP cells.  In my case the radar points of interest include all points within the larger
    study watershed, points within a buffered zone around the watershed (so that interpolation between points will produce values for areas at the edge of the watershed), and the radar
    points surrounding the RAWS stations.  Attempts to clip the points of interest using the Geoprocessing Wizard caused ArcMap to crash, so I selected the points by hand and created a
    comma delimited file with columns for lat, long, and HRAP identifier.  This 'radarpoints' file is used in Step Three to tell the make_raindat_v5 program what HRAP cells to extract data
    for.  This table was added using the Tools<add XY data... option.  At this point the view looks something like Figure 4.


Figure 4. Layout of field area showing locations of HRAP points, RAWS stations, study catchments and contributing areas (in blue).

- Step 3  Preparing the NEXRAD data
    I selected four months of data (September 1997, June 1998, July 1999, and August 2001) to analyze, in order to investigate four separate storm events (September 19-23, 1997, June 16-17, 1998, July 29-30, 1999, and August 15, 2001).  The data was extracted using the make_raindat_v5 program.  Each month of interest was selected both as the beginning and ending months and the comma delimited file of lat, long, and selected HRAP points was chosen as the "radarpoints" file. The program extracted hourly data for each HRAP cell for the entire month. Hourly data for the dates of interest were copied from the .dat file into Excel.  In Excel the sum of the rainfall at each HRAP cell was calculated and a comma delimited file with columns for total rainfall and the HRAP identifier was created and added to ArcMap using File<Add Data...

-Step 4 Viewing the rainfall data
    The first step in viewing the rainfall data is to join the data table with the rainfall total and HRAP identifier to the 'radarpoints' table that contains each HRAP identifier and the latitude and longitude of each HRAP cell.  This is done using the joins and relates option. After the tables are joined rainfall can be interpolated between HRAP points using the Spatial Analyst>Interpolate to raster>Spline. In practice the spline function would not work on the joined tables and caused ArcMap to crash, so I created a dbase file that contained the latitude and longitude, HRAP identifier, and rainfall total for each cell. Then using ArcToolbox I turned this file into a point coverage using Conversion Tools>Import to Coverage>Table to Point Coverage. I then performed the spline function on the point coverage, keeping the default settings. I then manually classified the spline of precipitation in order to create a larger number of categories for the values that fell on contributing areas.  For example if the 12.3-26.2 mm precipitation isohyet fell over a contributing area I changed the classification so that there would be separate rainfall isohyets for the 10-15 mm, 15-20 mm, 20-25 mm, and 25-30 mm values. This made a more detailed display of the range of rainfall values that fell on each contributing area.  For the September 1997 storm the divisions were made at 5 mm intervals and for the July 1999 and August 2001 storm the divisions were made for 0.5 mm intervals. This resulted in a layout similar to Figure 5.


Figure 5.  Spline interpolation of rainfall data from HRAP cells for September 19-23, 1997. Note that precipitation values increase from east to west. This is likely due to an orographic effect, as elevation also increases westward. This trend is apparent in all of the storm events. Negative precipitation values area for interpolated areas without HRAP cell values.






-Step 5  Precipitation data analysis
    The NEXRAD data were analyzed in the following manner. The range of precipitation values for each contributing area were determined by examining the spline value(s) overlaid on the contributing areas. Making the spline of precipitation transparent with the Effects tool allows rainfall values and contributing areas to be seem simultaneously.  Based on the range of spline precipitation values, a maximum and minimum rainfall total were determined. Ideally, I would have averaged the values of the splined rainfall values covering the contributing areas and used those values, but I was unable to determine the steps involved in that process. The duration of the rainfall for each storm was determined based on the length of time between the first and last hour of precipitation recorded by all HRAP values in the study area. This requires the assumption that rainfall begins and ends at the same time across the field area and in general the NEXRAD data support this.  The maximum and minimum rainfall total for each contributing area were divided by the storm duration to determine a rainfall intensity with values of mm/h.  During a single storm precipitation can be intermittent, so I also calculated a maximum hourly rainfall intensity for each contributing area using the maximum hourly rainfall recorded at the nearest HRAP point. The nearest HRAP point was determined using the measure tool and the maximum hourly rainfall was found in the .dat file.

-Step 6  Comparison of NEXRAD and RAWS data
    In order to evaluate the accuracy of the NEXRAD data the precipitation totals recorded by the RAWS stations were compared to the precipitation spline value at the same location.  Plots precipitation versus time were also made for each RAWS station and the two nearest HRAP points in order to compare precipitation throughout each storm.
 

Results
Results # 1  Minimum and Maximum Rainfall Totals, storm duration, and minimum and maximum rainfall intensities at each contributing area

Table 1: Total Rainfall, storm duration, and rainfall intensity for each contributing area.

 

Results # 2 Maximum hourly precipitation at nearest HRAP point

Table 2: Maximum hourly rainfall at nearest HRAP cell center.

 

Results # 3 Comparison of total rainfall from RAWS station and interpolated NEXRAD precipitation

Table 3: Comparison of NEXRAD and rain gage precipitation estimates.

 
 

Results # 4 Comparison of precipitation vs. time plots of RAWS data and two nearest HRAP cells
 


Figure 6: (a) Cumulative precipitation comparison between Lodore RAWS gage and two nearest HRAP cell centers for September 19-23, 1997.
                (b) Cumulative precipitation comparison between Yampa Plateau RAWS gage and two nearest HRAP cell centers for September 19-23, 1997.
 
 
 


Figure 7: RAWS precipitation June 16-17 1998, there is no precipitation recorded by NEXRAD anywhere in the field area during this time.
 
 
 


Figure 8:(a) Cumulative precipitation comparison between Lodore RAWS gage and two nearest HRAP cell centers for July 29-30, 1999.
               (b) Cumulative precipitation comparison between Yampa Plateau RAWS gage and two nearest HRAP cell centers for July 29-30, 1999.
 
 
 
 


Figure 9: (a) Cumulative precipitation comparison between Lodore RAWS gage and two nearest HRAP cell centers for August 15, 2001.
               (b) Cumulative precipitation comparison between Yampa Plateau RAWS gage and two nearest HRAP cell centers for August 15, 2001.
 

Discussion
    Using the methods above it is possible to calculate rainfall intensity values for debris flow contributing areas using NEXRAD data, however, the reliability of those values appears to be questionable. This, along with the small number of events makes it difficult or impossible to develop intensity/duration thresholds. Melis (1997) reported that rainfall intensities of at least 25 mm/h produce debris flows in Grand Canyon. Although Grand Canyon and the Green River canyons are geographically and climatically distinct, the debris flow initiation mechanism is the same. Based on this, it seems that rainfall intensities required to produce debris flows in the canyons of the eastern Uinta Mountains should be roughly similar to those in Grand Canyon.  Determining rainfall intensities based on the entire duration of the storm produced rainfall intensity values that area much too low to produce debris flows (Table1).  Analysis of the NEXRAD rainfall data indicated that during a prolonged storm event, oftentimes most of the rainfall fell during a one hour period.  Therefore I calculated rainfall intensity based on the greatest hourly rainfall value recorded by the nearest HRAP cell.  Calculating rainfall intensity based on the greatest hourly rainfall value at the nearest HRAP cell produced higher values,
some of these values are greater than 25 mm/h, but most are not (Table 2).
       The variability associated with choosing the time interval over which to calculate intensity is probably much less important than errors due to inherent problems of NEXRAD data itself.  Table 3 compares cumulative rainfall values derived from NEXRAD data with values recorded at rain gages at the same location. There are very few cases where these numbers correspond, and in most cases precipitation values derived from the NEXRAD data are much lower than those at the RAWS gages.  Figures 6-9 compare the RAWS precipitation to the precipitation at the two nearest HRAP cells for the storms in question.  Again, these figures show that precipitation from the gages and NEXRAD precipitation do not correspond.  In one case, (June 16-17, 1998) a storm produced between 20 and 30 mm of rain at the two RAWS stations, but no precipitation was recorded by the NEXRAD system.
    These difficulties appear to be fairly common. In a comparison study of NEXRAD and rain gage precipitation estimates Mizzell (1999) found that the radar consistently underestimated precipitation, by a total of 61% over the entire study period. Although the degree of underestimation varied with storm intensity, duration, and site, the degree of underestimation was not consistent. This study, although much smaller in scope produced similar results. Table 3 compares NEXRAD and gage rainfall estimates and while the radar estimates area consistently lower there is no consistent trend in underestimation that can be discerned. One potential cause of the inconsistencies could be the distance between the nearest NEXRAD station and the study area. Dinosaur National Monument is located approximately 150 km from the nearest NEXRAD station in Grand Junction, CO. This is at the very edge of radar coverage for that station.  The large distance could potentially lead to weak radar pulse returns and difficulties sampling the atmosphere near the ground surface due to the sampling beam geometry.  Because problems with the data I was unable to determine reliable rainfall intensity thresholds for unburned or burned areas. I was also unable to determine how large the debris flow triggering storms were. However, examination of satellite images may be better suited for that task.
    Despite these  problems I was able to learn more about these storms than without the use of NEXRAD. The NEXRAD data did show that precipitation increases from east to west in this area (Figures 11-13). This corresponds with an increase in elevation across the study area, so it appears that there is an orographic effect on precipitation. The data can be used to make very general statements about the debris flow producing storms and the precipitation intensities can be viewed as minimum values.
 

Acknowledgments
I would like to thank Christina Bandaragoda for help implementing the make_raindat_v5 program.
 

References
Melis, T.S., 1997, Geomorphology of debris flows and alluvial fans in Grand Canyon National Park and their influence on the Colorado River below Glen Canyon Dam, Arizona [Ph.D.
    thesis]: Tucson, University of Arizona, 502 p.
Mizzell, H.P., 1999, Comparison of WSR-88D derived rainfall estimates with gauge data in Lexington County, South Carolina [M.S. thesis] Columbia, University of South Carolina, 39
    p.
Schmidt, J.C. and Rubin, D.M., 1995, Regulated streamflow, fine-grained deposits, and effective discharge in canyons with abundant debris fans: in Costa, J.E., Miller, A.J., Potter,
    K.W., and Wilcock, P.R., eds., in Natural and Anthropogenic influences in fluvial geomorphology: Washington D.C., American Geophysical Union, Geophysical Monograph 89.
Wiezorek, G.F., 1987, Effects of rainfall intensity and duration on debris flows in central Santa Cruz Mountains, California, in Larson, R.A. and Slosson, J.E., eds., Storm-induced
    geologic hazards: Case histories from the 1992-1993 winter in southern California and Arizona: Boulder, Colorado, Geological Society of America Reviews in Engineering Geology, v.
    XI.
Wilson, R.C., 1987, Broad-scale climatic influences on rainfall thresholds for debris flows: Adapting thresholds for northern California to southern California, in Larson, R.A. and
    Slosson, J.E., eds., Storm-induced geologic hazards: Case histories from the 1992-1993 winter in southern California and Arizona: Boulder, Colorado, Geological Society of America
    Reviews in Engineering Geology, v. XI.
 

Appendix

Catchment names and interpolated rainfall maps.
 


Figure 10. Map of study area showing names and locations of study catchments.
 


Figure 11. Map of interpolated rainfall for September 19-23, 1997.
 


Figure 12. Map of interpolated rainfall for July 29-30, 1999.
 
 


Figure 13. Map of interpolated rainfall for August 15, 2001.