Water Usage Savings and Xeriscaping Conversion in Beverly Hills, CA
Object-Based Supervised Classification of NAIP Imagery in 2009 and 2020
Executive Summary
Beverly Hills is known for its lush grass, a plethora of trees, and prestigious residents. As the climate changes and drought rages on, is there any hope for a region such as the 90210 to make effective progress toward a more sustainable future? What does the success or failure of Beverly Hills mean for similar and not-so-similar regions of the country seeking to use government entities to enact change?
A survey of water saved between the enactment of the city’s 2009 restriction and 2020 will evaluate the success of the region’s efforts. This analysis was conducted through object- based supervised classification and calculations of irrigation water required for natural grass land cover. These calculations were then used to estimate the total savings achieved over the eleven years by the many efforts of the city and its residents. Results showed a large shift to less water- intensive ground cover. Residents and commercial establishments switched large amounts of grass to synthetic turf, eco-friendly xeriscape, and bare ground. The study proves that the city’s conservation efforts are not in vain and provides ample evidence that government entities' incentives, protocols, and restrictions can make a large dent in the over-consumption of water in regions experiencing drought or seeking greater sustainability.
Purpose
Water consumption trends positively correlate with patterns of socioeconomic wealth and prestige (Harlan et al., 2009). Humans attempt to adapt to and sustain their lifestyles in rapidly changing climates, with as many as 2.3 billion people residing in regions of extreme drought (Ellerbeck, 2022). This study seeks to determine if socioeconomically advantaged communities can implement imperative water conservation efforts using Beverly Hills, CA as a case study.
Introduction
Despite the common misconception of Los Angeles as a desert biome, the region is classified as a split containing both a Mediterranean and a Semi-arid biome (Kauffman, 2003). The City of Beverly Hills lies directly on the dividing line between the two biomes. As a result, the region experiences mild, wet winters and warm, dry summers. Drought and water conservation are not novel concepts in Southern California, as the area has been battling a lack of water since at least 1895. However, the exceptional drought periods have become more intense with longer durations in the past decade (NOAA & NIDIS, 2022). To combat these climate patterns, Southern California Cities have been stepping up their game to address and enact potential solutions. The City of Beverly Hills, alongside the Metropolitan Water District, started restricting water use in 2009 for the first time in eighteen years (Epstein & Di Renzo, 2009). Since then, the city has made efforts to restrict water usage when appropriate and incentivize the conversion of irrigation-dependent vegetation to other substitutes (Beverly Hills, 2022b, 2022a; Braslow, 2022; Figoni, 2019; Figoni et al., 2019).
A literature review of popular studies regarding water conservation provided by the removal of grass has evidenced that few straightforward analyses for the topic exist (Chen et al., 2015; Johnson & Belitz, 2012; MacLachlan et al., 2017; Marx, 2021; Miller, 2020; Mini et al., 2014). This study seeks to provide a reasonable methodology for conducting a city-scale analysis of the effects and benefits of grass removal in favor of more sustainable alternatives.
Methods and Data
Beverly Hills Boundary Layer. The County of Los Angeles (2022) file “LA_County_Unincorporated_Boundaries” was sourced from the County of Los Angeles Enterprise GIS ArcGIS Hub. The Beverly Hills segment of the file was inspected and compared to the current boundary maps provided by the City of Beverly Hills for accuracy (CBH IT/ GIS, 2021). Downloaded as SHP file and then loaded into QGIS to SQL query for only the Beverly Hills attribute using the query:
Imagery Selection and Acquisition. The three studies included in the Literature Review utilized Quickbird, Landsat 5, Sentinel-2, and NAIP imagery. NAIP imagery was determined to be the most appropriate due to the spatial resolution of the images ranging between 0.6 and 1.0 meters (Waterman, 2020). One consideration was the inability to obtain imagery with a higher resolution than NAIP for accuracy assessment. However, it was determined that the original NAIP images would be used for ground truth comparison as the level of detail required for this classification was outside the scope of the potential provided by Landsat 7’s 30-meter resolution.
Imagery Suitability: Determining the Ideal Month of Analysis. To determine the ideal month of analysis, several factors were considered. First, to ensure that the grass being evaluated was likely irrigated lawn, an analysis to determine California’s driest month must be completed. This analysis was conducted using the North America Climate – Monthly Mean Precipitation GRID file downloaded from the USGS ScienceBase-Catalog (USGS, 2011). The file was loaded into ArcGIS Pro alongside the previously generated Beverly Hills boundary layer. The Monthly Precipitation layers were clipped to the Beverly Hills boundary layer extent by creating and implementing a Batch Extract by Mask tool. Within the Batch Extract by Mask tool, a rasterized layer of the Beverly Hills boundary polygon for the Cell Size and Snap Raster parameters. Symbology was applied to the masked Monthly Precipitation layers by creating and implementing a Batch Apply Symbology From Layer tool to match masked layers to the original precipitation layers.
-
![]()
January
-
![]()
February
-
![]()
March
-
![]()
April
-
![]()
May
-
![]()
June
-
![]()
July
-
![]()
August
-
![]()
September
-
![]()
October
-
![]()
November
-
![]()
December
Monthly Mean Precipitation in Beverly Hills, CA.
Statistical Analysis of Monthly Precipitation. Batch summary statistics using clipped precipitation layers as means of obtaining mean, minimum, max, and range values were conducted. The results were batch-exported tables to Excel, and a chart was created from the sheets. In Excel, all tables were merged into one table listing all months and corresponding values. Conditional formatting was applied to highlight cells in each column with the lowest and highest values to determine the month with the lowest variation, mean, minimum, maximum, and range of precipitation values (i.e., the month where vegetation will likely only appear on NDVI if irrigated). The results show that July is the most appropriate month to conduct analysis (see figure).
July is the prime summer season in California, characterized by the state’s hottest and driest weather (UC Santa Cruz, 2020). This is significant for the analysis of water usage as the distinction of an irrigated lawn compared to natural, unirrigated vegetation will be enhanced by eradicating natural rainwater. The second lowest statistical values for mean precipitation occur in June. This will also ensure that any unirrigated vegetation is unlikely to be as bright as irrigated vegetation due to minimal amounts of natural water occurring in the month before the study month. Similarly, May also has relatively low precipitation, but consideration must be given to the higher precipitation value in the month prior.
Statistics of Monthly Precipitation (mm) in Beverly Hills, CA (USGS, 2011).
Note. Red conditional formatting highlights the highest values, while green formatting highlights the lowest values.
Data availability must be taken into account alongside the ideal months upon which to conduct analysis. NAIP imagery was the ideal candidate for this study as it seeks to use object- based supervised classification to determine the area of Beverly Hills occupied by grass in both the 2009 and 2020 imagery.
NAIP imagery for 2020 was downloaded from the NOAA Data Access Viewer, and the acquisition date was determined to be May 15th, 2020, using metadata alongside the NAIP collection data for 2020 (Biediger, 2021; NOAA, 2022). For 2009 two rasters were downloaded from USGS Earth Explorer and then mosaiced together (USGS Earth Explorer, 2022). The acquisition date for the 2009 NAIP images was June 26th, 2009. Extract by mask was used for both rasters to limit the extent to the Beverly Hills Boundary layer.
NDVI vs. Color Infrared. The 2009 and 2020 NAIP images were altered to Color Infrared (CIR) using band combinations 4, 1, and 2. A Normalized Difference Vegetation Index (NDVI), using bands 4 and 1, was performed on both the 2009 and 2020 images. The resulting NDVIs were then compared to their respective Color Infrared layers to determine which layer would be most appropriate for classification. The NDVI of both images does an excellent job of differentiating vegetation from all other land cover types. All urban and non-vegetation features are given similar values. All vegetation present within the images is given similar values, thus removing the issues of shade misclassification; the NDVI images will allow for more accurate classification. However, the NDVI image was ruled out for classification due to the issue presented in trees and grass land cover. The CIR image was ideal for classification due to the inclusion of the NIR band and the promotion of distinguishing vegetation from other land cover types.
Color Infrared and NDVI Comparison: Beverly Hills, CA.
Filtering. Using the ArcGIS Pro Raster Functions toolbox, the convolution More Sharpening high-pass filter was applied to the 2009 image to aid in creating accurate training data. The 2020 image did not need sharpening as it was already much sharper than its 2009 counterpart.
Classification. Object-based classification provides many advantages over pixel-based classification (ArcGIS Pro, 2022a). The classification method segments the image through the algorithm’s accounting of neighboring values and shapes alongside user-input parameters. Segmentation aids in the identification of homogenous land cover regions. The ArcGIS Pro Classification Wizard was utilized for the supervised, object-based classification of both NAIP CIR images. The segmentation parameters that produced the best results were a spectral detail level of 18, a spatial detail level of 18, and a minimum segment size of 1. The classes used were Water, Shadow, Barren, Urban, Grass, and Trees.
Training Samples Table.
Note. The number of training samples used for each class for both the 2009 and 2020 NAIP images of Beverly Hills. Significantly more training samples were required for the 2020 image, which contained a greater variation in spectral characteristics of surface materials present due to the change from lawn to various turfs, sports courts to various synthetic alternatives, roofing material changes, seasonal tree color, and the installation of solar panels.
Support Vector Machine Classification with a maximum number of samples parameter of 1500 was used due to the flexibility regarding sample distribution and resistance to errors of noise present in the 2009 NAIP image (ArcGIS Pro, 2022b). The classified rasters were then converted to multipart polygons, excluding the simplification of polygons parameter, using the Raster to Polygon Tool.
The total area of Beverly Hills classified as grass was calculated for the 2009 and 2020 NAIP images. This was conducted by adding another field in each layer’s attribute table. The Calculate Geometry option was utilized to produce the total area of the grass class in US acres.
The total water saved by removing grass in Beverly Hills was estimated by first calculating the maximum amount of water potentially used for irrigation of natural lawns in 2009 to avoid the potential of underestimation. The City of Beverly Hills (2018) provides the equation for Maximum Applied Water Allowance (MAWA) as follows:
Maximum Applied Water Allowance (MAWA) Equation.
Note. This equation produces the allotted water allowance in gallons per year based on several variables (The City of Beverly Hills, 2018). ETo refers to the evapotranspiration rate in inches over a year. ETAF refers to the applicable adjustment to the evapotranspiration rate defined by the City of Beverly Hills (0.55 for residential dwellings and 0.45 for commercial establishments). LA refers to the square footage of the landscape, including Special Landscape. The conversion to gallons per square foot is facilitated by including 0.62. SLA refers to the square footage of the landscape, which falls under the “Special Landscape” classification, defined by Beverly Hills as “dedicated solely to edible plants and/or recreational areas” (Figoni et al., p.2; The City of Beverly Hills, 2018). Since the calculation already includes the special landscape area in the LA variable, the SLA specification will be excluded for accuracy due to the lack of official documentation of land designated as a Special Landscape area.
Results
2009 Supervised Classification.
2020 Supervised Classification.
Grass Class Changes Between 2009 and 2020.
Acreage of Grass Classification for 2009 and 2020.
MAWA Calculations
2009
MAWA = 348,575,078.5 gallons/year
2020
MAWA = 276,351,873.7 gallons/year
Note. ETAF is assigned a value of 0.5 as it is the average of potential values 0.45 and 0.55 since both residential and commercial areas are included in the study area
The calculation for total water saved between 2009 and 2020 is the difference between the MAWA values of the two years or 72,223,204.8 gallons/year. The amount saved per year cannot be applied to or multiplied by the eleven years between 2009 and 2020 as it has yet to be known when or how much change occurred each year, and the shift was most likely gradual. To convert gallons/year to acre-feet, the following calculation was conducted:
325,851 gallons = 1 acre-foot
72,223,204.8 gallons = 221.645 acre-feet
Discussion
Changes. During the training data creation process for 2020, the magnitude of change that had occurred since the 2009 image was acquired was evident. There was a great change in not only the removal of grass in favor of xeriscaping, artificial turf, or bare ground but also the number of sports courts and fields replaced with more sustainable materials. Additionally, the growing trend of solar panel installation was evident, suggesting that water conservation efforts can go hand in hand with other sustainability efforts to compound the impact of such measures. Another change in the eleven years between the two NAIP images was the drastic increase in construction and development in previously barren or tree-covered regions.
Considerations and Sources of Error. Some considerations regarding this study must be made. First, it is notable that the City of Beverly Hills may have a greater conversion rate if there were not such strict regulations regarding the permitting of artificial turf and other artificial ground coverage (Beverly Hills Permit Center, 2022). The second consideration was the potential and impact of dormant grass in the summer season when the NAIP imagery was obtained. Additionally, the results of reduced irrigation and unirrigated grass due to drought cannot be analyzed independently. The NAIP image quality improvement between 2009 and 2020 undoubtedly impacted the classification accuracy. The 2009 image increased the difficulty for both the training sample creator and the classifier in distinguishing grass from tree cover. Most notably, the confusion between the tree and grass land cover classes occurred in both images and required manual correction during reclassification, no matter how homogenous, spectrally varied, or large the class training samples were. A final consideration for a study such as this one is the potential for a wide variety of results from different methods or even the same method used here. The user’s defining of classification parameters, classes, training samples, level of reclassification, band combination used for classification, and experience distinguishing land cover types play big roles in the product of this analysis. It is also significant the limitations of this study in determining how much grass was replaced with Xeriscape and how much was left to the elements as bare ground (in this study, they were treated as one and the same).
Conclusion
The results of this study evidenced that a large amount of water can be conserved yearly by government efforts to incentivize, promote, and restrict water usage. The City of Beverly Hills made great progress between 2009 and 2020, saving 221.645 acre-feet of water per year. The results of this study prove that there is hope for regions wishing to find the same level of success regardless of socioeconomic factors present in their populations. However, great care must be taken with the results and assumptions made by analyses such as this, as variation, inconsistencies, and errors can occur too easily.
Further Research
Further research is recommended for further exploration and removal of sources of error and uncertainty. Potential avenues include the inclusion of vegetation height in classification; a Normalized Difference Turf Index to differentiate xeriscaping and turf replacement of grass; land use classifications for the distinction of how much grass cover is being removed in residential and commercial land; edge detection for better identification of trees and distinction from grass; and water delivery records to more accurately determine actual water usage in 2009 and 2020 (Chen et al., 2015; Coleman et al., 2020; Hjelmstad et al., 2019; Johnson & Belitz, 2012; MacLachlan et al., 2017; Marx, 2021; Miller, 2020; Mini et al., 2014; Sovocool et al., 2006).
References
ArcGIS Pro. (2022a). Train Support Vector Machine Classifier (Spatial Analyst)—ArcGIS Pro | Documentation. ArcGIS Pro. https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-analyst/train-support-vector-machine-classifier.htm
ArcGIS Pro. (2022b). Image Classification Wizard—ArcGIS Pro | Documentation. ESRI. https://pro.arcgis.com/en/pro-app/latest/help/analysis/image-analyst/the-image-classification-wizard.htm
Beverly Hills Permit Center. (2022). CD-PermitCenterPlanReviewPermittingGuide-0922.pdf.
Beverly Hills. https://www.beverlyhills.org/cbhfiles/storage/files/217620531524563657/CD-PermitCenterPlanReviewPermittingGuide-0922.pdf
Beverly Hills. (2022a). Watering Schedule. Beverly Hills. https://www.beverlyhills.org/departments/publicworks/recyclingandconservation/wateringschedule/
Beverly Hills. (2022b). Water Conservation. Beverly Hills. https://www.beverlyhills.org/departments/publicworks/recyclingandconservation/waterconservation/
Biediger, J. (2021, January 19). NAIP Image Dates Data Hub. USDA Online ArcGIS Hub. https://naip-image-dates-usdaonline.hub.arcgis.com/items/3c550e2869cb40f1b12244666b0f7c5d
Braslow, S. (2022, June 15). Beverly Hills Enacts New Water Conservation Measures. Beverly Hills Courier. https://beverlyhillscourier.com/2022/06/15/beverly-hills-enacts-new-water-conservation-measures/
CBH IT/ GIS. (2021). GIS Maps and Apps for Public Access—City of Beverly Hills, CA. GIS Maps and Apps for Public Access - City of Beverly Hills, CA. https://gis.beverlyhills.org/VBHApps/
Chen, Y., Mcfadden, J. P., Clarke, K. C., & Roberts, D. A. (2015). Measuring Spatio-temporal Trends in Residential Landscape Irrigation Extent and Rate in Los Angeles, California Using SPOT-5 Satellite Imagery. Water Resources Management, 29(15), 5749–5763. https://doi.org/10.1007/s11269-015-1144-2
Coleman, R. W., Stavros, N., Yadav, V., & Parazoo, N. (2020). A Simplified Framework for High-Resolution Urban Vegetation Classification with Optical Imagery in the Los Angeles Megacity. Remote Sensing, 12(15), Article 15. https://doi.org/10.3390/rs12152399
County of Los Angeles. (2022, August 9). LA County Unincorporated Boundaries. County of Los Angeles Enterprise GIS Hub - ArcGIS. https://egis-lacounty.hub.arcgis.com/datasets/lacounty::la-county-unincorporated-boundaries/explore?location=33.780046,-118.302300,8.36
Figoni, D. (2019). Turf Replacement Rebate Program and Residential Landscape Irrigation. Beverly Hills Granicus. https://beverlyhills.granicus.com/MetaViewer.php?view_id=58&clip_id=6551&meta_id=395392
Figoni, D., Tse, M., & Borboa, G. (2019). CITY OF BEVERLY HILLS PUBLIC WORKS DEPARTMENT MEMORANDUM. Beverly Hills Granicus, 3. https://beverlyhills.granicus.com/MetaViewer.php?view_id=58&clip_id=6840&meta_id=413953
Hjelmstad, A., Garcia, M., & Larson, K. (2019). Effect of drought policies on Los Angeles water demand. World Environmental and Water Resources Congress 2019: Watershed Management, Irrigation and Drainage, and Water Resources Planning and Management - Selected Papers from the World Environmental and Water Resources Congress 2019, 239–250. https://www.mendeley.com/catalogue/98505bbf-e009-3027-80e8-1589c6cb1ab5/
Johnson, T. D., & Belitz, K. (2012). A remote sensing approach for estimating the location and rate of urban irrigation in semi-arid climates. Journal of Hydrology, 414–415, 86–98. https://doi.org/10.1016/j.jhydrol.2011.10.016
Kauffman, E. (2003). Climate and Topography. In Atlas of the Biodiversity of California. California Department of Fish and Wildlife. https://wildlife.ca.gov/Data/Atlas
MacLachlan, A., Roberts, G., Biggs, E., & Boruff, B. (2017). Subpixel land-cover classification for improved urban area estimates using Landsat. International Journal of Remote Sensing, 38(20), 5763–5792. https://doi.org/10.1080/01431161.2017.1346403
Marx, A. (2021). Quantifying the Multiplier Effect of Southern California’s Turf Removal Rebate Program with Time-Series Aerial Imagery. JAWRA Journal of the American Water Resources Association, 57(2), 344–355. https://doi.org/10.1111/1752-1688.12901
Miller, D. L. (2020). Remote Sensing of Urban Vegetation during Drought in Southern California [Ph.D., University of California, Santa Barbara]. In ProQuest Dissertations and Theses. https://www.proquest.com/georef/docview/2492950718/abstract/7C51649763864E93PQ/1
Mini, C., Hogue, T. S., & Pincetl, S. (2014). Estimation of residential outdoor water use in Los Angeles, California. Landscape and Urban Planning, 127, 124–135. https://doi.org/10.1016/j.landurbplan.2014.04.007
NOAA, & NIDIS. (2022). California. Drought.Gov. https://www.drought.gov/states/california
NOAA. (2022). NOAA: Data Access Viewer. Digital Coast: Data Access Viewer. https://coast.noaa.gov/dataviewer/#/imagery/search/
Sovocool, K. A., Morgan, M., & Bennett, D. (2006). An in-depth investigation of Xeriscape as a water conservation measure. Journal (American Water Works Association), 98(2), 82–93. https://www.jstor.org/stable/41312933
The City of Beverly Hills. (2018). WATER EFFICIENT LANDSCAPE WORKSHEET. Beverly Hills, 09, 5. https://www.beverlyhills.org/cbhfiles/storage/files/6069574471403443495/WaterEfficientLandscapeWorksheet08.14.2018.pdf
UC Santa Cruz. (2020). The California Seasons. Randall Morgan Initiative. https://randallmorganinitiative.ucsc.edu/collections/field-notes-and-writing/on-california/the-california-seasons.html
USGS Earth Explorer. (2022). EarthExplorer. USGS Earth Explorer. https://earthexplorer.usgs.gov/
USGS. (2000). Estimated Use of Water in the United States in 2000—Conversion Factors (USGS Circular 1268). USGS. https://pubs.usgs.gov/circ/2004/circ1268/htdocs/text-conversion.html
USGS. (2011). North America Climate – Monthly Mean Precipitation—GIS Data—ScienceBase- Catalog. USGS ScienceBase-Catalog. https://www.sciencebase.gov/catalog/item/4fb55169e4b04cb937751d9b
Waterman, R. (2020). Fast and Simple NAIP Imagery. ArcGIS Blog. https://www.esri.com/arcgis-blog/products/arcgis-living-atlas/imagery/fast-and-simple-naip-imagery/