Refined NIDIS Application for the Upper Colorado River Basin
Andrew Barney
CEE 6440 – GIS in
Water Resources
December 3, 2010
Contents
- Introduction
- Background
- Objective
- Scope
- Method
- Data Collection
- Revised SWSI Method
- Interpolation
- Results
- Discussion
- Conclusion
- References
Introduction
Background
The National Drought Policy
Act of 1998 established a commission to ensure a collaborative effort to
address drought related issues. The National Integrated Drought Information
System (NIDIS) was envisioned to improve drought monitoring and forecasting
capabilities.
The U.S. Drought Portal
currently provides early warning about droughts and provides various
stakeholders with information to help prepare for them and mitigate negative impacts.
Warning systems currently predict drought conditions on a regional scale.
Figure 1: NIDIS Drought Monitor Portal
Objective
A more refined drought
forecasting system would greatly benefit the public by offering drought
forecasts at a much smaller scale. Less guesswork regarding drought conditions
in a particular area will allow stakeholders to make more informed decisions on
how to adjust activities based on these conditions. A user interface for
observing refined drought conditions would demonstrate the utility of higher
resolution drought forecasting.
Numerous drought forecasting techniques
and models exist and researchers continue to refine them in an effort to
predict droughts with greater reliability. An interface that provides relevant
spatial data and connections to appropriate web services would be an excellent
aid for such researchers.
Creating a complete user interface at a
scale that would be useful for drought forecasting research far exceeds the
scope of this project. A task of a much smaller scale but still significant
value involves creating a GIS framework of limited scope for visualization of
drought conditions. The project will demonstrate the general processes,
requirements and challenges involved with creating a more ambitious refined
NIDIS application for visualization of drought conditions.
Scope
Research is currently being performed on
a refined NIDIS application for the Upper Colorado River Basin (UCRB) at the
Utah Water Research Laboratory (UWRL). Thus, a great deal of data required for
drought forecasting has already been determined for this study area. Using a
subset of this area would take advantage of these previous efforts.
The revised Surface Water Supply Index
authored by David Garen is currently used by the
state of Colorado as the method of choice for their seasonal water supply
forecasts. This method is suited well for the mountain west where snowmelt is a
large component of runoff and frequent reservoirs dot the landscape.
A small area around Gunnison, Colorado
serves as a good location to focus on for this project. The area borders the
edges of the UCRB which provides a logical outer boundary for the study area.
The area also contains numerous stream gages and reservoirs collecting relevant
data. The region is a good example of the type of terrain, climate and
hydrologic conditions expected across the remainder of the UCRB. The total size
of the study area was reduced over the course of the project to reduce the time
needed for drought index calculations.
The spatial units used to display
drought index calculations for this analysis are Hydrologic Unit Code (HUC)
10-digit watersheds. This scale is far more refined than what forecasts are
currently shown at but not so small that scarcity of local data inhibits the
utility of the calculations, computational times become excessive, or
visualization of the spatial units becomes problematic.
Figure 2: Upper Colorado River Basin and Project Study
Area
Method
Data Collection
Both spatial and tabular data are
required for the analysis. Descriptions, sources and preparation of these data
are listed below.
Spatial Data
HUC 10 Polygons: The Watershed Boundary
Dataset (WBD) provides watershed boundary shapefiles
at different scales. These scales are referred to by how many digits of an ID
code is required to identify them. For example, the
Upper Colorado River Basin is represented by the 2-digic HUC code (HUC 2) 14.
The WBD has shapefiles at scales of HUC 12 and HUC 8
polygons. However, the HUC 12 shapefile contains a
HUC 10 watershed field. Using the Dissolve operation within ArcGIS a HUC 10
feature class is easily obtained.
River Network: The river network is
required for delineating which areas are associated with particular runoff
data. A flowline shapefile
was obtained from NHDPlus, a source for a large collection
of accurate hydrologic spatial data.
Catchments: Catchments are the
smallest watersheds delineated as part of the NHDPlus.
This shapefile was used in conjunction with the river
network to delineate areas associated with runoff data.
Stream Gage Locations: Streamflow
forecasts are required by the revised SWSI calculations. The Colorado Basin
River Forecast Center (CBRFC) produces streamflow
forecasts for this area and provided a shapefile of
gage locations. Upon further research, this list was reduced to include only
those stations with published streamflow forecasts.
Reservoir Locations: Reservoir storage is
required as part of the revised SWSI calculations. The US Bureau of Reclamation
(USBR) does not offer prepared GIS spatial data. Thus, a reservoir location shapefile was created using spatial data found at the USBR
website. Only reservoir locations currently reporting data were collected.
Figure 3: Study Area Showing HUC 10 Watersheds, River
Network, USBR Reservoirs and Forecast Stream Gages
Tabular Data
Streamflow Forecasts: Streamflow
forecasts were obtained from reports found through the CRBFC.
Historic Streamflow
Data: Historic monthly streamflow data
(required to find non-exceedance probabilities
required by the revised SWSI method) were obtained from the US Geological
Survey (USGS).
Reservoir Storage Data: Reservoir storage data
were obtained through a USBR portal.
Revised SWSI Method
Background
The revised SWSI method was developed by
David Garen as an attempt to fine tune the SWSI
method which relied greatly on subjective weighting coefficients and did not
facilitate the comparison of conditions in different watersheds. The revised
SWSI is intended (as was the original SWSI) to reflect drought conditions in
snow dominated regions indicative of the American Mountain West. Drought
forecasting using this method uses total runoff as the sum of streamflow and water captured in reservoirs. Snowpack is
indirectly figured into calculations through the use of streamflow
forecasts.
While streamflow
and the change in reservoir storage reflects actual
runoff conditions, streamflow forecasts represent
expected runoff volumes. The CRBFC uses snowpack, previous streamflow
and other hydrologic and climate data to make these predictions. Forecasted streamflows represent total natural runoff, not the flows
recorded at stream gages. This is important as it means that changes in
reservoir storage do not need to be forecasted.
Process
The calculations required to determine
the revised SWSI index value are relatively straightforward. Streamflow and reservoir storage data represent known
runoff data and are used to determine non-exceedance
criteria at each gaging station. This is a common statistical process in the
field of hydrology, the steps of which are outlined below.
Streamflow forecasts for the gages in the study area give predictions of
total streamflow volume in acre-feet for April through
July. Monthly statistical data downloaded from USGS is converted from cubic
feet per second to a total volume in acre-feet over the four summer months for
the period of record. Reservoir storage data downloaded from USBR is used to
find the change of storage of all reservoirs associated with the streamflow gage (see section on interpolation) between
April 1 and July 31 for the period of record. The sum of total streamflow and change of reservoir volume represents the
total runoff for each forecast gage location (note that records are only used
for years when the streamflow gage and all reservoirs
associated with it are reporting data).
Table 1: Sample of Total Runoff Results for Taylor (below Taylor Reservoir) Streamflow Gage
|
Year |
Apr - Jul SF (kaf) |
Δ S Reservoirs (kaf) |
Total Runoff (kaf) |
|
2010 |
32.922 |
25.2 |
58 |
|
2009 |
84.888 |
25.1 |
110 |
|
2008 |
119.172 |
16.3 |
135 |
|
2007 |
66.618 |
13.7 |
80 |
|
2006 |
55.122 |
18.2 |
73 |
|
2005 |
61.458 |
22.6 |
84 |
|
2004 |
48.726 |
16.6 |
65 |
|
2003 |
39.996 |
36.6 |
77 |
|
2002 |
42.912 |
-8.02 |
35 |
|
2001 |
55.2 |
15.4 |
71 |
|
2000 |
64.068 |
12.9 |
77 |
|
… |
… |
… |
… |
Total runoff values for the period of
record are ranked, and non-exceedance
probabilities calculated. The SWSI index is calculated using the formula below,
where P is the probability of non-exceedance of the streamflow forecast value (found by interpolation of the
forecast runoff using the ranked values). The formula is designed to produce an
index between -4.1 (extreme drought) and 4.1 (extremely wet).
Equation 1: Revised
SWSI Equation
Table 2: Table Translating Total Runoff to SWSI Value for Taylor (below Taylor Reservoir) Streamflow Gage
|
Rank |
Q tot (acft) |
Non-exceedance (%) |
SWSI |
|
… |
… |
… |
… |
|
25 |
86.116 |
47.16981132 |
-0.23585 |
|
26 |
90.725 |
49.05660377 |
-0.07862 |
|
27 |
92.729 |
50.94339623 |
0.078616 |
|
28 |
93.125 |
52.83018868 |
0.235849 |
|
29 |
94.394 |
54.71698113 |
0.393082 |
|
30 |
96.317 |
56.60377358 |
0.550314 |
|
31 |
101.089 |
58.49056604 |
0.671666 |
|
32 |
101.416 |
60.37735849 |
0.86478 |
|
33 |
103.422 |
62.26415094 |
1.022013 |
|
34 |
105.2111 |
64.1509434 |
1.179245 |
|
… |
… |
… |
… |
The table below shows a sample of results for several calculations of revised SWSI indexes, including calculations using actual, instead of forecast, values. Drought index calculations are generally made when streamflow forecasts are published at the beginning of each month from January to June.
Table 3: Sample of Final SWSI Values for Gage Watersheds
|
USGS ID |
Name |
Year |
Month |
Forecast Runoff (kaf) |
Actual Runoff (kaf) |
Forecase SWSI |
Actual SWSI |
Difference |
|
9109000 |
Taylor Blw Taylor Park Res |
2005 |
Jan |
100 |
83 |
0.68 |
-0.25 |
0.93 |
|
Feb |
110 |
83 |
1.70 |
-0.25 |
1.95 |
|||
|
Mar |
110 |
83 |
1.70 |
-0.25 |
1.95 |
|||
|
Apr |
105 |
83 |
1.16 |
-0.25 |
1.41 |
|||
|
May |
100 |
83 |
0.68 |
-0.25 |
0.93 |
|||
|
Jun |
90 |
83 |
-0.06 |
-0.25 |
0.19 |
|||
|
9110000 |
Taylor @ Almont |
2005 |
Jan |
160 |
100 |
2.53 |
-0.40 |
2.93 |
|
Feb |
177 |
100 |
2.60 |
-0.40 |
3.00 |
|||
|
Mar |
177 |
100 |
2.60 |
-0.40 |
3.00 |
|||
|
Apr |
168 |
100 |
2.24 |
-0.40 |
2.64 |
|||
|
May |
160 |
100 |
2.13 |
-0.40 |
2.53 |
|||
|
Jun |
138 |
100 |
1.60 |
-0.40 |
2.00 |
|||
|
9112500 |
East @ Almont |
2005 |
Jan |
190 |
188 |
0.81 |
0.61 |
0.19 |
|
Feb |
220 |
188 |
2.00 |
0.61 |
1.38 |
|||
|
Mar |
215 |
188 |
1.95 |
0.61 |
1.34 |
|||
|
Apr |
205 |
188 |
1.31 |
0.61 |
0.70 |
|||
|
May |
160 |
188 |
-0.71 |
0.61 |
1.32 |
|||
|
Jun |
192 |
188 |
0.91 |
0.61 |
0.30 |
|||
Interpolation
Concept
The most difficult consideration involved in the development of this GIS framework involves interpolation of data. The spatial units chosen to visualize drought index values (in this case, HUC 10 watersheds) do not generally exhibit an obvious association with gages reporting the data necessary for calculations. Some thought and study must go into an interpolation process which accurately associates the correct data (or combination of data) with each spatial unit.
Streamflow at any point along a network can be described as the accumulation of all runoff upstream from that point minus losses and storage. Thus, streamflow data is relevant as an indicator of conditions upstream from a gage. In other words, the closest downstream gage gives the most relevant data for a particular area. Likewise, reservoir storage should be associated with the closest downstream streamflow gage. Reservoirs create a reduction in runoff that would otherwise be attributed to this gaging station.
Process
A watershed delineated between gaging stations represents an area over which a particular drought index value (calculated using data from a particular gaging station) is the most relevant. To create polygons over these areas the following process was followed:
- Establish connectivity of river network
- Add field to river network table called “Gage”
- Select section of river network using Utility Network Analyst (place a flag at a gage station, barrier flags at all gaging stations upstream and select “trace downstream” option) – works best going upstream to downstream
- With editing enabled, use field calculator to assign appropriate gage ID value to “Gage” field for selected features
- After entire river network over the study area has been assigned to gages, join the catchments feature class to the river network (both have common “comid” field)
- Perform dissolve operation using “Gage” field
Figure 4: Gage Watershed Creation - Trace Upstream Using Utility Network Analyst & Assign Gage ID to Stream Network Elements
Figure 5: Catchments and River Network
The result is a polygon feature class or gage watersheds (GWS) with each polygon associated with a stream gage (Figure 6). Note that reservoirs associated with a gage are contained within the corresponding GWS.
Figure 6: Delineated Gage Subwatersheds
Next, SWSI values (calculated using excel) are assigned to each GWS and interpolated using area weighted averages so they can be associated with the HUC 10 watersheds. Because this process must be performed many times, a simple model was created to accomplish this step.
The model (shown below) completes these steps:
- Joins table of SWSI values to GWS feature class
- GWS feature class is converted to a raster (with SWSI as the raster value)
- Zonal Statistics as Table tool used to create a table that gives the mean SWSI value from the raster using the HUC 10 watersheds as zones
- The table is joined to the HUC 10 feature class
With some symbology changes, SWSI drought index values can be visualized at the level of HUC 10 watersheds (Figure 7).
Results
Figure 7 shows an example of the visualization expected from a refined NIDIS application.
Figure 7: HUC 10 Catchments with Symbology Showing SWSI Values for Jan 2005
As would be expected, variance from actual conditions decreases the later in the year the SWSI value is calculated (Figure 8).
Figure 8: Variance from Actual Conditions for January and June forecasts
The process performed to achieve these results (gathering data, preprocessing, interpolation and visualization) sheds light on many techniques that would be useful for developing a fully functional NIDIS GIS framework. Also, many issues were uncovered that require further attention in order for a larger scale project to proceed.
Discussion
The manner of interpolation presented in this report is a good case study for interpolation of other data values. For example, if someone wanted to use Thiessen polygons to assign precipitation values to HUC 10 watersheds in preparation for a drought index calculation, much the same process as above could be followed. With some adjustments other interpolation techniques such as inverse distance weighted, nearest neighbor, or elevation weighted interpolation (appropriate for snowpack data) could be used to assign appropriately weighted values to HUC 10 catchments. Whether calculations are performed before or after interpolation of variables depends on what the calculations call for. For the revised SWSI method, index values were found for the GWS areas then interpolated to the spatial units because there was essentially only one set of values to interpolate. If there are many difference variables requiring different types of interpolation, it would be necessary to interpolate the values separately then compute index values using the field calculator. Another option for this situation would be converting different variables interpolated over the study area to raster datasets, compute index values using raster calculations, then interpolate to the spatial units.
Data collection was one of the most problematic and time consuming parts of this project. An initial set of forecast streamflow gages provided by the CBRFC proved to be deceptive, with just over half of these gages providing publicly available river forecasts. Reservoir data is likely not complete (see below), and even the data available is very difficult and slow to obtain. More streamlined webservices for providing the data necessary for drought forecasting analysis is essential in order to create an effective system.
As shown in Table 3, forecast streamflows almost always exceeded calculations of actual runoff (forecasts on average predict 20% more runoff than actual conditions show). Overestimating water supply is not erring on the side of safety, and thus there must be some discrepancies in the data. There is likely a large discrepancy because streamflow forecasts add “adjustments” to make up for diversions from the stream. Although the large reservoirs were accounted for in the calculations presented in this report, there are many other adjustments made to the runoff calculations of which the general public is not aware. The process for determining streamflow forecasts should be made more transparent to allow data to be used correctly in drought forecast analysis.
Conclusion
Many of the difficulties and encountered during the completion of this project were associated with data collection and tedious, time-consuming computations. In the context of a larger NIDIS project, however, these two issues do not fall within the scope of a GIS framework. An understanding of drought forecasting techniques, interpolation, hydrologic processes and relevant variables is necessary for completing this portion of the project. Some techniques demonstrated above will form the foundation for a larger scale NIDIS project. The process of completing this project was valuable in developing a more in depth plan of action for a larger project and for realizing the biggest barriers and most important dimensions to focus on.
References
Alcorn, Brenda. “Water Supply Forecasting Tools and Processes”, Colorado River Commission Technical Workshop. 5 Dec, 2008.
“CBRFC Conditions”, National Weather Service Colorado Basin River Forecast Center. < http://www.cbrfc.noaa.gov/>
Garen, David C. 1993. “Revised Surface Water Supply Index (SWSI) for Western United States.” Journal of Water Resources Planning and Mangement 119, no. 4 (July/Aug): 437-454.
“National Hydrography Dataset Plus” Horizon Systems Corporation. < http://www.horizon-systems.com/nhdplus/>
“Upper Colorado Region Reservoir Operations”, Bureau of Reclamation <http://www.usbr.gov/uc/crsp/GetSiteInfo>
U.S. Drought
Portal, National Integrated Drought Information System.
<http://www.drought.gov/portal/server.pt/community/drought_gov/202>
“USGS Real-Time Water Data for the Nation” USGS. < http://waterdata.usgs.gov/nwis/rt>
“Water Supply Forecasting Tools”, Colorado Basin River Forecast Center. <http://www.cbrfc.noaa.gov/wsup/doc/cbrfc_ws_tools.pdf> 29 Nov, 2010.