Estimating Water Demands for Irrigation Districts on the Lower Colorado River
David R. Kracman
December 7, 2000
The Lower Colorado River Authority (LCRA) provides surface water diversions for four major rice-growing irrigation districts near the Gulf of Mexico. Since these diversions make up nearly 80% of all surface water supplies from the lower Colorado River, forecasting water demands in the districts is important for basin planning and reservoir operation purposes. Geographic Information System (GIS) applications were used in conjunction with a regression developed by Dr. Quentin Martin at LCRA to estimate and depict climactic influences on water demands in the four districts. New data will be incorporated into the regression to better predict agricultural water demands, and the results can be incorporated into a larger planning model of the Colorado River to forecast reservoir levels in the Highland Lakes system and maximize benefits accrued by meeting water demands.
The Colorado River basin encompasses a large portion of central Texas, as shown in the figure below:
The State of Texas, and the Colorado River Basin, in Albers Conical Equal Area Projection
In order to manage the surface waters in the lower Colorado River basin, the Texas Legislature created the Lower Colorado River Authority (LCRA) in 1934. LCRA has the responsibility of providing for municipal, agricultural and manufacturing water demands, controlling flooding, generating hydroelectric power, providing flood control, and meeting recreational needs along the lower Colorado. LCRA regulates the flows in the Lower Colorado through the operation of the Highland Lakes, a series of six manmade lakes and embankments located in central Texas.
Highland Lakes
Only two of the six reservoirs, Lake Travis and Lake Buchanan, have significant storage capabilities, with nearly 2 million acre-ft of combined storage. The other reservoirs operate as "run of the river" impoundments, in which outflows equal inflows and reservoir levels remain nearly constant. By storing and releasing reservoir water in the Highland Lakes system, the LCRA can regulate reservoir levels and downstream flows to meet the different water demands.
In order to ensure that the lower Colorado River is operated in an efficient manner, the University of Texas at Austin has proposed to work with the LCRA in designing a scenario-based stochastic programming model to optimize storage and release decisions in the Highland Lakes. This model, currently under development, should maximize the expected revenue from the sale of irrigation water while providing adequate municipal and industrial water supplies, maximize recreational benefits, and maximize hydroelectric power generation revenues.
In order to achieve these goals, it is important to understand the particular demands that must be met by the LCRA. Currently, around 80% of all surface water diversions from the lower Colorado River are used to irrigate rice farms near the Gulf of Mexico. Most of these farm lands are located in four irrigation districts: LCRA Lakeside Division, LCRA Gulf Coast Division, Garwood Irrigation Company and Pierce Ranch.
On average, the four districts divert around 500,000 acre-ft of water for irrigation each year, which is applied on around 80,000 acres. Typically, delivery losses through the canal systems can result in a 20% loss of diverted water before it reaches the field.
Because these irrigation districts play such a crucial role in determining the methods of water distribution in the Lower Colorado River, I chose to use Geographic Information Systems (GIS) in order to gain a better understanding of the water demands in the districts. I also used a regression developed by Dr. Quentin Martin at LCRA to examine the link between hydrologic variables in the Colorado River basin and water demands at the irrigation sites.
The four irrigation districts considered in my study area are LCRA Lakeside Division, LCRA Gulf Coast Division, Garwood Irrigation Company and Pierce Ranch. All of these districts are located near the Gulf of Mexico, not far from the mouth of the Colorado River. An Autocad file with the approximate boundaries of the districts was obtained through the LCRA, and added as a new theme using the CAD Reader extension. It is important to note that these district boundaries are APPROXIMATE, and should only be used to obtain a frame of reference for the districts.
For the GIS applications in my project, I chose the Albers Conical Equal Area projection as the standard projection for all subsequent shapefiles. Maps were created in ArcView 3.2, and projections were done through the ArcView Projection Utility. Details on the projection used are shown below:
In order to orient the irrigation districts, I added the Counties shapefile from Exercise 3. Since the Counties shapefile was originally in geographic coordinates, and the district boundaries were in Lambert Conformal Conic, I used the ArcView Projection Utility to project both themes into Albers Conical Equal Area. I then added a USGS shapefile of HUC Region 12 and an EPA Rf1 river reach shapefile for Region 12. By performing a query for the portions of these shapefiles that are within the Colorado River basin, converting them to new shapefiles, and adding them to the Counties and district boundary themes, I obtained the maps below. Note that the Garwood Irrigation District has been highlighted in yellow.
Below is a map showing the names of the counties in the study area.
A table showing some of the important properties of the districts is shown below. Note that two separate crops of rice are grown each year. The first is planted in March or early April and harvested in July. The plants regrow and are harvested again in October and November:
In April of 1988, Dr. Quentin Martin, Chief Water Resources Planner at LCRA, developed a regression to predict water diversions in the four irrigation districts based on rainfall, total rice acreage, gross lake evaporation and the time of initial water diversion for rice irrigation. Regressions were developed for both monthly and annual time steps, and the results were able to explain between 46% and 81% of the water diversion variations for the monthly data, and between 85% and 88% of the variation in annual data. Because the optimization work currently under development will probably use a monthly time step, I chose to focus on the monthly regressions rather than considering annual variations. The relationships used for the regressions are listed below:
Coefficients were determined for each time step and each irrigation district in the LCRA study. Depending on the magnitude of the calculated coefficients and the resulting coefficient of determination (R2), one of the four relationships was chosen to approximate water diversions for a given district. The data used for the regressions were from 1968 to 1986. 1968 was picked as the earliest date because this was approximately when the adoption of two-crop rice farming in the area became widespread, replacing the traditional single crop harvest.
For my project, I wanted to test the regressions, using updated data, while utilizing Geographic Information System tools to establish and evaluate the connection between hydrologic variables and water demands in the districts.
GIS Applications and Data Acquisition
The first step in my project involved acquiring the necessary data for the regression formulation.
Delay Factor
For the delay factor, F, I had to obtain records concerning the timing of water diversions in the district.
Chris Riley, Associate Hydrologist at LCRA, provided a spreadsheet with data on the first and last date of water diversion from each district for the period of 1968 to 1999. A few gaps exist in the data, particularly for earlier dates, but for most years in the period, planting delay factors were calculated.
Gross Lake Evaporation
Next, I obtained information on Gross Lake Evaporation from the Texas Water Development Board (TWDB). TWDB maintains a site on surface water in which data can be found on Gross Lake Evaporation for 1 degree grids across the state of Texas.
Data for each grid was downloaded for the 1968 to 1997 period, and set up in an Excel spreadsheet. In addition, a shapefile of the 1 degree grid mesh was downloaded from the Texas Natural Resources Information System (TNRIS) through the TNRIS Data Catalog.
Texas Counties with TWDB 1 Degree Mesh
The 1 degree mesh was originally in geographic coordinates, and was projected into Albers Conical Equal Area.
In order to represent the Gross Lake Evaporation data graphically, I used the Excel spreadsheet to calculate average monthly evaporation for each grid from 1954 to 1997 (the years for which evaporation data is available from TWDB), and saved the results as a DBF file. I then added this file to ArcView, and joined the table, using the TWDB id numbers as the common field, to the attribute table for the 1 degree mesh. I was then able to project the evaporation data graphically for the entire state.
For the original LCRA regression, TWDB quadrangle 811 was used in determining gross lake evaporation data for Garwood, Pierce Ranch and Lakeside Irrigation District, while quadrangle 912 was used for the Gulf Coast Irrigation District. For my preliminary work, I followed the methods used by LCRA. However, since the Gulf Coast Irrigation District spans four quadrants, with significant acreage in quadrants 912 and 911, it may be more appropriate to take an average using more than one quadrant.
Rainfall
For data on rainfall, I received assistance from Bob Rose, meteorologist at LCRA, who directed me to the National Climatic Data Center (NCDC) site for monthly precipitation data. I chose to follow the methods used in the LCRA regression by selecting the four National Weather Service stations closest to the study area: Columbus (NWS 00001911), Bay City (NWS 0000569), Pierce (NWS 00007020) and Danevang (NWS 00002266). The latitude and longitude for each station were listed in an associated NCDC text file. To display the locations of the stations, I designed an excel spreadsheet, and saved it as a DBF file. After adding it as a table in ArcView, I created a new view and added an event theme (under the View options) to display the stations in geographic coordinates. I was then able to project the stations into the Albers Conical Equal Area projection.
In deciding which NWS station should be used for each of the four districts, I followed the procedures used in the LCRA regression: Bay City precipitation data were used for LCRA Gulf Coast Division, Pierce data were used for Pierce Ranch, and Columbus data were used for LCRA Lakeside Division and Garwood Irrigation Company. Because data was missing for several months at some of the stations, data from nearby stations were used to fill the gaps. For Bay City, missing data were replaced by taking the average of Pierce and Danevang precipitation. For Pierce data gaps, I used Danevang data. For missing information at Columbus, Pierce rainfall was used.
Crop Acreages
Total acreages for first and second crops were obtained from LCRA. Average values are shown in the table above.
Water Use
Data on water use for all districts were also obtained from LCRA. These data were combined with the delay, evaporation and rainfall data to form a comprehensive spreadsheet for all irrigation districts.
The data contained in the spreadsheet above were then used in a test regression. Lakeside Irrigation District was chosen as the example district, and the month of June was chosen as the month for the regression. For the original regression developed by LCRA, the second of the regression equations above was chosen:
I chose the same form for my regression, using acreage and the product of rainfall and acreage as the independent variables, and water demand as the dependent variable. This data was then pasted into SPSS, a statistical package capable of developing regressions. A linear regression was used to develop the following results:
The results show a calculated coefficient of determination (R2) of 0.787, compared to a value of 0.49 for the original LCRA regression. The standard errors for the coefficients decreased from 0.5896 and 0.1386 to 0.275 and 0.113 respectively, another sign that the "goodness of fit" for the regression improved. The results for the regression, compared with the regression developed by LCRA, are shown below:
In the graph above, the purple "actual" dots show actual water demand in the Lakeside Irrigation District for the month of June over the period of 1976 to 1997. The green line shows predicted water demand over time with LCRA's earlier regression, and the dark blue line shows predicted water demand with the new coefficients. Both regressions do a reasonable job predicting water demand, but the new regression gives a slightly general fit for most years.
While these preliminary results are very promising, regressions must be performed for the other districts, for each month that irrigation occurs. In addition, statistics should be obtained for all four regression equations to ensure that gross lake evaporation and delay variables are not omitted in instances where their corresponding coefficients are significant.
GIS can be a powerful tool for constructing and interpreting linear regressions for water demand in rice-growing irrigation districts along the lower Colorado River. By using ArcView to project district boundaries, weather station locations, and hydrologic conditions, the variables that determine water use become more clear. By using new data, older regressions may be improved in order to better estimate future water demands. These relationships could be incorporated into basin management models to maximize benefits from water use, given different hydrologic scenarios. GIS should also be helpful in displaying the results of these analyses.
I will continue to attempt to fill data gaps for the important variables, seeking the assistance of LCRA staff and members of the individual irrigation districts. In the LCRA work, data adjustments were made in some months where TWDB evaporation rates seemed excessive. I will try to determine whether adjustments are necessary for the new regression. For the Gulf Coast Irrigation District, average evaporation rates over several of the TWDB quadrangles might give a more accurate representation for the district area, and I will explore whether changes should be made in this respect. In addition, I will determined if some significant rainfall events that occurred late in the month should be subtracted from that month's precipitation total and added to the subsequent month's rainfall. LCRA did this for a handful of months where demands for the following month were significantly reduced after a large, late-month rainfall event. This will require obtaining daily rainfall data for the study area. The regression results can then be incorporated into a larger stochastic optimization model to help determine beneficial storage and release policies for Lake Travis and Lake Buchanan.
Special thanks to the following individuals and organizations for their assistance and support:
Dr. Quentin Martin, Chief Water Resources Planner, LCRA
Chris Riley, Associate Hydrologist, LCRA
Dr. Jobaid Kabir, LCRA
Bob Rose, Meteorologist, LCRA
Jill Land, LCRA
Dr. David Maidment, Director, CRWR
Dr. Daene McKinney, CRWR
Sarah Stone, Graduate Research Assistant, CRWR
Tim Whiteaker, Graduate Research Assistant, CRWR
Vicki Samuels, Graduate Research Assistant, CRWR
Jordan Furnans, Graduate Research Assistant, CRWR
Melissa Figurski, Graduate Research Assistant, CRWR
Terri McKinley, Gulf Coast Irrigation District
Ruben Solis, Texas Water Development Board