Sociohydrological Impacts of Water Conservation Under Anthropogenic Drought in Austin, Texas

Sociohydrological Impacts of Water Conservation Under Anthropogenic Drought in Austin, Texas

Water Resources Resarch

Supporting Information for

Sociohydrological impacts of water conservation under anthropogenic drought in Austin, Texas

Betsy Breyer1, Samuel C. Zipper2,3, Jiangxiao Qiu4

1Department of Geography and Geographic Information Science, University of Illinois at Urbana-Champaign, Champaign, IL USA, Email: , ORCID: 0000-0001-6355-1203

2Department of Earth & Planetary Sciences, McGill University, Montreal, QC, Canada. Email: , ORCID: 0000-0002-8735-5757

3Department of Civil Engineering, University of Victoria, Victoria, BC, Canada

4School of Forest Resources & Conservation, Fort Lauderdale Research and Education Center, University of Florida, Fort Lauderdale, FL, USA. Email: , ORCID: 0000-0002-3741-5213

Contents of this file

Meteorology data

Streamflow data

Water availability and use data

Urban vegetation data

Socio-demographics data

Structural equation model: Path diagram

Hierarchical linear model development

Introduction

Data collection and pre-processing procedures are described below for five types of data sets: meteorological data (drought severity, precipitation, and temperature), streamflow data (for the Colorado River and urban streams), water availability and use data (reservoir levels, urban water withdrawals, and household water use estimates), urban vegetation data (the Enhanced Vegetation Index), and socio-demographic data (demographic composition as well as various measures of wealth and poverty). We first indicate data sources, data extent, data resolution, and intended scale of analysis. We then describe processing steps we followed to produce inputs to data analysis, making note of any relevant data gaps or idiosyncrasies.

Meteorology

Two different meteorological datasets were collected for different scales of analysis. To quantify upstream drought for the watershed-scale analysis, we downloaded gridded datasets of the 6-month Standardized Precipitation Evapotranspiration Index [SPEI; Vicente-Serrano et al., 2009], a multiscalar drought index which considers both water supply (precipitation) and demand (evapotranspiration) [Santiago Beguería et al., 2010; S. M. Vicente-Serrano et al., 2010]. This is a 0.5° resolution gridded monthly product spanning the period 1901-2013 calculated using input data from the CRU TS 3.22 dataset [Climate Research Unit, 2014]. Reference evapotranspiration is calculated using the Penman-Monteith FAO-56 method [Allen et al., 1998]. For each month from 2000-2013, we calculated the average SPEI value for all pixels in the contributing area to the Colorado River upstream of the City of Austin.

For the submunicipal analysis of urban water use and urban subwatershed streamflow, we used local meteorological data obtained from the NOAA GHCN-Daily network [National Climatic Data Center, 2015]. Specifically, we downloaded all available historical data from the Camp Mabry station (30.32°N, -97.76°E; GHCN #USW00013958) in northwest Austin, spanning the period 6/1/1938 - 12/31/2015. This included daily maximum/minimum temperature and precipitation, which were aggregated to monthly means (temperature) and totals (precipitation). These were then used to calculated potential evapotranspiration using the Droogers-Allen modification to the Hargreaves approach [Droogers & Allen, 2002]. We then calculated a monthly precipitation deficit and monthly SPEI using the R package SPEI for 1-12, 18, and 24 month timescales [Beguería & Vicente-Serrano, 2013].

Streamflow

Municipal water availability was quantified using the reservoir level of Lake Travis, and downstream water flow was quantified using the discharge of a closest gage downstream of city Austin. At the submunicipal scale, urban subwatersheds flow was aggregated based on the discharge data at the outlet of each urban subwatersheds. We obtained daily streamflow and water level data from the USGS National Water Information Service (NWIS) web portal [United States Geological Survey, 2015].

For the first and second approaches to subwatershed analysis (outlined in subsequent paragraphs), we use a group of five subwatersheds of the Colorado River shown in Table S1 These sites were selected due to an uninterrupted record of > 30 years and for being contained all or partially within the City of Austin limits. For the third analysis, in which we are comparing to water use data since 2012, we select a larger group of 8 subwatersheds (Table S1). These were selected for being headwatersheds in which at least 75% of land is served by the Austin Water Utility and having a continuous data record for 2012-2015, to coincide with our zip code-level water use data.

Water availability and use

At the municipal scale, we used a monthly time series of municipal water withdrawals from Lakes Travis and Buchanan provided by Austin Water Utility staff [Breyer, 2015]. A monthly time series of reservoir levels was also obtained through LCRA Historical Lake Levels records [Lower Colorado River Authority, 2016]. At thesubmunicipal scale, we used a panel dataset on mean monthly single-family residential (SFR) water use 2012-2015 by zip code (n=44) obtained through the City of Austin data portal [Austin Water Utility, 2016].

Urban vegetation

We quantified urban vegetation by using the Enhanced Vegetation Index product available from MODIS 16-day satellite imagery, 250 meter resolution [National Aeronautics and Space Administration, 2016]. This product provides atmospherically corrected surface reflectance data derived from the blue, red, and infrared bands, and is calculated for each pixel as follows:

This product allowed for consistent comparisons of vegetation greenness across space and time [Heute et al., 2002]. Because we were interested in residential vegetation, parks, commercial and industrial centers, and other non-residential land uses were masked from each image using a 2012 land use inventory shapefile obtained from City of Austin data portal [City of Austin, 2012]. A mean value of 196 MODIS pixels fell within each zip code after masking. Masked data were visually compared to imagery to assess how well the mask removed non-residential areas. The mask performed well in including all residential areas but, because of the relatively large pixel size of MODIS data, removal of non-residential areas was imperfect. Following masking, EVI values were aggregated to a mean value for each zip code at a monthly time step.

Socio-demographics

Zip code level socio-demographic and housing characteristics data were obtained from the U.S. Census American Community Survey, 2009-2013 5-year estimates (Tables DP04, DP05, S1701, and S1903) downloaded for Travis County, Texas, from the US Census data portol, American FactFinder[United States Census, 2014]. The US Census provides median household income data in 2013 US dollars, correcting for inflation over the enumeration period.

Structural equation model: Path diagram

In structural equation modeling, the analysis typically begins by constructing a path diagram to map conceptual linkages among components of an interdependent system. Below, we present our path diagram for the relationships water supply and use components in the integrated urban-regional system of City of Austin (city scale) and the Texas Colorado River basin (regional scale). Variables in parentheses are measureable and used as inputs for the structural equation models (SEM). Arrows connecting boxes indicate that we hypothesize there will be some sort of relationship between the proxy variables, as detailed in the main text. We constructed SEMs of this configuration before and during water use restrictions in order to evaluate the nature of these relationships both before and during water use restrictions, as well as any changes between the two time periods. We use the term reversal to refer to a change in sign of a relationship, the term decoupling for a relationship changing from significant to not significant, and the term recoupling to refer to a relationship changing from not significant to significant.

Hierarchical linear modeling: Model development

Hierarchical linear modeling (HLM) is a form of regression appropriate for data that exhibit clustering, or statistical dependence within groups (here, groups are formed by zip codes) [Gelman & Hill, 2007; Zuur et al., 2009]. In our case, a panel dataset of monthly zip code-level residential vegetation estimates (2012-2014) comprise the dependent variable for HLM. These EVI data exhibit not only temporal autocorrelation (dependence over time due to seasonality) but also within-zip code clustering—each zip code’s vegetation time series will fluctuate over the seasons around a mean level of ‘greenness’ that itself varies across zip codes. HLM controls for, and even leverages, statistical dependence through a composite error term that is partitioned variance across two levels, the zip code and the individual observation. A key step in HLM development is the specification of fixed and random effects. Fixed effects are explanatory variables that have the same effect on the dependent variable regardless of group membership—estimated fixed effects coefficients are equivalent to coefficients in ordinary least squares regression. Random effects, by contrast, vary across groups in the panel dataset, with that variation following a Gaussian distribution. Random effects may apply to the intercept term or to the estimated coefficient for an explanatory variable, in which case a random ‘slope’ is specified. Below we present the functional form for a ‘random slopes’ HLM, which we developed to explain variation in the dependent variable.

Here, EVI in time period i in zip code j is given as a function of:

  • A fixed intercept term γ0
  • The fixed effect (γ1) of TMAX (mean maximum monthly temperature) in time period i
  • The fixed effect (γ2) of time (given in months from the first time period) in time period i
  • The fixed effects (γ3) of SFR water use for zip code jin time period i
  • The fixed effects (γ3) of median household income in zip code j
  • The fixed effect (γ5) of an interaction of time and income on EVI
  • The random effect (γj)of 3 month urban SPEI in time period i for zip code j

Note that two error terms are included: U0j and εij. U0j reflects the effect of ‘group’ (zip code) membership and follows a normal distribution centered on mean greenness for zip code j. ε is an independent and random error term that adheres to the usual assumptions of linear regression (ε~N(0, σ²)).

Table S1. Streamflow monitoring stations used for urban hydrological analysis.

Station Number / Station Name / Latitude [°N] / Longitude [°E] / Drainage Area [km2] / BFI
Long-Term Subwatersheds for Drought Sensitivity and Baseflow Analysis
08156800 / Shoal Creek @ W 12th St / 30.28 / -97.75 / 31.9 / 0.09
08154700 / Bull Creek @ Loop 360 / 30.37 / -97.78 / 57.8 / 0.38
08158600 / Walnut Creek @ Webberville Rd / 30.28 / -97.65 / 132.9 / 0.28
08155300 / Barton Creek @ Loop 360 / 30.24 / -97.80 / 300.4 / 0.33
08159000 / Onion Creek @ US Hwy 183 / 30.18 / -97.69 / 831.4 / 0.27
Short-Term Headwatersheds for Municipal Water Use Analysis
08156910 / Waller Creek @ Koenig Ln / 30.32 / -97.72 / 2.8 / -
08158030 / Boggy Creek @ Manor Rd / 30.29 / -97.71 / 4.3 / -
08155541 / W Bouldin Creek @ Oltorf Rd / 30.25 / -97.77 / 4.6 / -
08158380 / Little Walnut Creek @ Georgian Dr / 30.35 / -97.70 / 13.5 / -
08156675 / Shoal Creek @ SilverwayDr / 30.35 / -97.74 / 14.5 / -
08158920 / Williamson Creek @ Oak Hill / 30.24 / -97.86 / 16.3 / -
08158927 / Kincheon Branch @ William Cannon Blvd / 30.21 / -97.83 / 17.4 / -
08158200 / Walnut Creek @ Dessau Rd / 30.38 / -97.66 / 67.9 / -

1