AZMET Weather Station - Location Suitability Analysis

Travis Zalesky

As part of UA GIST 602B

Abstract

The Arizona Meteorological Network consists of 28 weather stations, owned and operated by the University of Arizona, across Southern Arizona. While the utility of these stations is significant, particularly as their data is made available to the public for free, prior analysis has found that there may be gaps in coverage, and that these weather stations should not be relied upon for data interpolation across the whole of Arizona. Using both Boolean, and ranked choice site selection methods, six World Meteorological Organization guidelines, as well as several additional project specific selection criteria have been evaluated to identify potential locations for future (hypothetical) AZMET weather stations.

Introduction

Site selection via suitability analysis is one of the key roles of any Geographic Information System (GIS). Site selection in GIS is a modern evolution of the idea of sieve mapping, typically credited to McHarg in 1969, but widely utilized prior to its formal publication (McHarg, 1969; O’Sullivan et al., 2010). The basic idea is to selectively remove areas from consideration based on a number of criteria. Early sieve mapping efforts were done by hand, through use of transparent overlays, which when stacked, blacked out inappropriate sites, leaving the candidate sites transparent. Today, suitability analysis can be done much faster, on larger scales, and with more selection criteria using computers, but the basic idea remains the same.

The Arizona Meteorological Network (AZMET) is a network of meteorological stations run by the University of Arizona, College of Agriculture and Life Sciences (Arizona Meteorological Network, 2024). As seen in Geospatial Interpolation — Wind Speed, this is an invaluable source of weather data across Arizona (AZ), which is open to the public, and can be used for any number of applications. However, there is a large region in N. AZ without any weather stations, which can not be reasonably interpolated from existing stations (Fig. 1). In this project, I will use World Meteorological Organization (WMO) guidelines to identify several suitable locations for some (hypothetical) additional AZMET weather stations using both a binary and a ranked choice suitability analysis.

WMO guidelines call for weather stations to be (1) on a 0-3% slope, (2) in an open area, preferably in an agricultural or urban area, depending on the intended purpose, (3) accessible by road (within 500 m), (4) at least 9 Km from existing weather stations, (5) at least 10 Km from water bodies, and (6) within 9 Km of a host institution (responsible for maintenance and data collection; Alejo, 2018). Because AZMET weather stations are automated and upload their data via cell towers, the sixth criteria (within 9 Km of a host institution), has been relaxed, to be instead, within a 2 hours driving window of a host institution. Additionally, the station location can not be within a National Park or National Forest, and must be within 20 Km of a cell tower, with a clear line of sight, to facilitate data uploads.

Methods

Data was collected from a variety of publicly available sources. Digital Elevation Model (DEM) data was acquired from US Geological Survey (USGS) 3D Elevation Program (3DEP). Forty-eight, contiguous, 1/3 arc second resolution (approximately 10m) DEMs (from 32°N, 109°W to 38°N, 115°W), fully covering the extent of AZ, were downloaded via the National Map (U.S. Geological Survey, 2024). Land cover data consisted of the International Geosphere-Biosphere Programme (IGBP) Land Cover Classification map, and was obtained from USGS as part of the North America Land Cover Characteristics Data Base Version 2.0, downloaded via Earth Explorer (Belward et al., 1999). AZMET weather station location data was downloaded from the AZMET API utilizing R v4.3.2 (programming environment; R Core Team, 2021) and the "azmetr" package (Weiss & Scott, 2024). All other data was obtained through ArcGIS Online, and all data metadata, including sources, has been summarized in Table 1.

All data was projected in a NAD 1983 StatePlane AZ Central FIPS 0202 (Intl. Feet) (EPSG:2223) projection. This is an N. to S. Transverse Mercator projection, and is the most suitable compromise projection for data spanning the width of AZ. This projection will minimize distance distortions for this dataset.

All analysis and cartography was performed in ArcGIS Pro v3.2.2, unless otherwise stated.

Table 1

Preprocessing

AZ was initially selected from the US Census States layer and was saved to a new layer. All subsequent layers were clipped to the extent of AZ. Additionally, the ArcGIS Pro project processing extent was set to AZ, using the Analysis Environments setting.

The 48 DEM files were initially imported into R v4.3.3 and were merged into a large geotif raster file using the terra package, v1.7-71 (Hijmans, 2023). The merged raster was then clipped to the extent of AZ, before being imported into ArcGIS Pro v3.2.2 for further analysis. The DEM layer was then evaluated for sinks, and hydrologically conditioned (see Watershed Modeling). Unfortunately, the spatial resolution and large extent of this DEM was found to be excessive for this analysis, requiring untenable computation times. Therefore, the hydrologically conditioned DEM was resampled to a 100 m spatial resolution (approximately 3 arc seconds), a 100x reduction in resolution. All subsequent raster analysis and raster data creation was scaled to match this 100 m resampled DEM.

The Transportation layer group consisted of a variety of roads layers from Interstates to local roads, optimized for dynamic mapping at variable scales. These layers were flattened into a single roads layer using the Merge, Data Management tool.

Boolean Site Selection

Slope

For ease of site maintenance, as well as to prevent wind, and other anomalies, associated with steep slopes, the WMO recommends locating weather stations on flat ground. The percent slope was calculated across AZ from the hydrologically conditioned resampled DEM, using the Slope, Spatial Analysis tool. The slope raster layer was then reclassified as ≤ 3% slope = 1, and >3% slope = 0 (Fig. 2).

Canopy Classification

To ensure accurate weather instrument readings, as well as to enable automated data uploads, the weather station must be located in an open area, not under a closed canopy or dense forest. Additionally, while any chosen weather station sites will be cleared of vegetation and planted with low-growing ground cover, the University of Arizona is not interested in cutting or clearing numerous trees for the proposed weather station. The land cover classification was reclassified into a binary classification for open/closed canopy classification. IGBP land cover types determined to be closed canopy include “Evergreen Needleleaf Forest”, “Evergreen Broadleaf Forest”, “Deciduous Broadleaf Forest”, “Mixed Forest”, “Closed Shrublands”, as well as “Urban and Built”. All other classifications were considered to be open canopy. The reclassified layer was converted to polygons using the Raster to Polygon, Conversion tool, with the simplify polygons option.

Proximity to Roads

Using the flattened roads layer, two roads buffers were generated. The first buffer was set to 10 m. This represents an exclusion area to keep weather stations off of the road surface and shoulder. However, this does not account for actual road width, and sites along major roadways, such as multi-lane highways, may need to be further evaluated prior to final site selection. The second buffer was set to 500 m, ensuring that the location is at least, somewhat, accessible by road, and that installation and maintenance will not be prohibitive.

Proximity to Existing Weather Stations

To ensure maximum coverage of weather data, the WMO recommends a minimum distance of at least 9Km between weather stations. Therefore, a 9 Km exclusion buffer was created around the 28 currently operating AZMET weather stations. This is not an exhaustive list of AZ weather stations, however, at a minimum, weather stations should not overlap within the AZMET network.

Proximity to Water Bodies

Large water bodies can generate microclimatic conditions, which may not be representative of the surrounding area. Therefore, the WMO recommends that weather stations be kept at least 10 Km away from water bodies, unless the weather station is specifically designed for costal conditions. The land cover classification was reclassified into a binary classification for water/land. As before, the reclassification layer was converted to polygons using the Raster to Polygon, Conversion tool, with the “simplify polygons” option. A 10 Km buffer was created around all AZ water bodies.

Proximity to Host Institution

As weather stations and weather instruments will require regular site maintenance, calibration, and repair, the WMO recommends that weather stations be within 9 Km of a qualified host institution. Because the University of Arizona only has a limited number of campus locations in AZ which have relevant Agricultural or Environmental Studies departments, this guideline has been relaxed. Fortunately, a colleague at Northern Arizona University (NAU) has agreed to oversee any potential weather stations in their area, massively expanding the search area in the North. A 2-hour driving time, service window from either the UA main campus (Tuscon, AZ), the UA Yuma satellite campus, or the NAU Flagstaff campus, has been offered as a substitute site requirement (Fig. 3). This service area was calculated using the ESRI Network Dataset, with generalized area polygons, and was provided for a nominal fee.

National Parks and Forests

National Parks and Forests were excluded from consideration to preserve the natural beauty of such areas. Parks were selected from the USA Parks layer using a simple attribute query, and were saved to a new layer.

Cell Tower Coverage

AZMET weather stations transmit their data via cell towers, operating at approximately 800 MHz. While cell tower communications at 800 MHz can reliably cover 28-30 Km, to ensure seamless data uploads, the site should be located within 20 Km of a cell tower, and with a clear line of site. Using the tower heights data provided with the cell towers layer (converted to meters), a veiwshed analysis was performed, with a 20 Km outer perimeter (Fig. 4). The resulting viewshed raster was converted to polygons using the Raster to Polygons, Conversion tool, with the “simplify polygons” option.

Site Selection

To find the intersection of all the individual site selection parameters, first the exclusion criteria were erased from the AZ study area using the Pairwise Erase, Analysis tool. In order of removal, the exclusion criteria were, not in a National Park or Forest, > 10 Km from a water body, not in a closed canopy landcover class (including Urban), > 10 m from road centerline, and >9 Km from AZMET weather station (Fig. 5). Next the intersection of the 500 m roads buffer, the cell tower coverage polygon, the 2-hour driving time polygon, and the remaining AZ study area (not erased in the prior step) was calculated using the Intersect, Analysis tool. The resulting intersection polygons were converted to a raster using the Polygon to Raster, Conversion tool, with all potential locations raster value = 1. The final site selection analysis was performed by multiplying the site criteria intersection raster by the slope reclassification raster to remove any sites with a > 3% slope from consideration. The resulting binary raster was then converted back into polygons.

Figure 2. Slope reclassification.

Figure 3. Two-hour drive time from window from University of Arizona, Tucson, University of Arizona, Yuma, and Northern Arizona University, Flagstaff.

Figure 4. Cell tower viewshed map with maximum radius of 20 Km.

Figure 5. Exclusionary layers sequentially erased from Arizona study area.

Figure 6. Multiple ring buffer surrounding existing AZMET weather stations.

Figure 7. Calculating rank field for number of cell towers in range. All ranked choice suitability criteria utilized similar field calculations to generate rank.

Ranked Choice Site Selection

For the ranked choice site selection slope, canopy classification, proximity to roads, proximity to water bodies, and National Parks/Forests layers were all kept as in the boolean site selection process. However, to further categorize and rank sites the proximity to existing weather stations, proximity to host institution, and cell tower coverage layers have all been updated with a categorical 1-4 ranking scheme, where 1 = best, and 4 = worst.

Proximity to Existing Weather Stations

In order to expand the AZMET as far across AZ as possible, locations furthest away from existing weather stations are preferred. Three AZMET station buffers were generated using the Multiple Ring Buffer, Analysis tool, at 9, 15, and 20 Km (Fig. 6). The multi-buffer layer was merged with the AZ study area using the Union, Analysis tool to incorporate regions beyond the 20 Km buffer perimeter into the analysis layer. Locations within 9 Km were given a rank of 4, while locations beyond 20 Km were given the highest rank of 1 (Fig. 7). Given weight for this criteria = 0.3.

Proximity to Host Institution

While driving distance from the nearest host institution remains a high priority concern, project managers may consider the possibility of longer drive times for an otherwise excellent location. Driving times were calculated, as before, using a 1, 2, and 3-hour driving window. Locations less than an hour away were given the highest rank of 1, while locations more than 3 hours away were given a rank of 4 (Fig. 7). Given weight for this criteria = 0.4.

Cell Tower Coverage

Project managers do not want to extend the cell tower criteria beyond the prior 20 Km limit, but would like to prioritize regions as close to cell towers as possible. Additionally, some locations may be able to connect to multiple cell towers. A multi-buffer layer was created, as with the AZMET multi-buffer layer, at 5, 10, and 20 Km. The multi-buffer layer was combined with the prior viewshed analysis using the Pairwise Intersect tool with the existing cell tower viewshed polygon layer. Locations less than 5 Km were given the highest rank of 1, while locations more than 20 Km away were given a rank of 4. Similarly, locations with 0 cell towers in view were given a rank of 4, and those locations with 1, 2, and 3 or more cell towers in view were given a rank of 3, 2, and 1 respectively (Fig. 7). Given weight for these criteria = 0.1, and 0.2 respectively.

Site Selection

All the five classified ranking criteria were converted to a raster using their respective rank classifications as the raster cell value. The overall location ranking was calculated by a weighted sum raster calculation (Fig. 8). The sum of all weights, as detailed in the previous paragrahs, is equal to 1, therby resulting in a scaled output from 1 to 4. Additionally, the binary slope classification layer was multiplied by the weighted sum such that areas with a suitable slope remaned unchanged (multiply by 1), and those with an unsutable slope were multiplied by 0. Then, the Set Null tool was used to modify the output raster, such that 0 values (unsuitable slope) were converted to No Data. This overall ranking raster was then combined with the boolean exclusionary criteria by clipping the raster to the extent of the previously generated erase polygon, not including the AZMET proximity erase. The resulting raster then had to be reclassified to the nearest integer value before being converted back into polygons.

Figure 8. Raster calculator expression, with weights, used to calculate the ranked choice site selection output, prior to application of boolean exclusionary criteria.

Results

Boolean Site Selection

Each of the eight boolean site selection criteria reduced the search area by between 2% to 72% (Table 2). The site selection process left a search area of 17,000 Km2, a reduction of 94%. There are more than 22K qualifying sites remaining across AZ (Fig. 9).

Ranked Choice Site Selection

Using the ranked choice site suitability selection, the entire area of Arizona was ranked on a 1 to 4 scale from “Highly Suitable”, to “Not Suitable” (Table 3). Less than 2% of the state was considered highly suitable, with roughly 15% each suitable, and less suitable, and nearly 70% ranking as not suitable (Fig. 10).

Table 2

Figure 9. Qualifying locations identified via boolean site selection.

Table 3

Figure 10. Locations across Arizona, ranked by suitability for AZMET weather stations.

Conclusions

While it is not surprising that tens of thousands of qualifying sites remain from this analysis, the search area for qualifying locations has been massively narrowed down. Using Boolean (T/F) analysis, I was able to remove 94% of the state from consideration, while a ranked analysis approach narrowed the search area even further, with less than 2% of the state qualifying as highly suitable.

While much of the initial interest lay in finding locations in Northern AZ, particularly around the Flagstaff area (see Geospatial Interpolation — Wind Speed), there appears to be a very limited number of sites in that region. This is largely the result of the numerous, large National Forests surrounding Flagstaff, namely, Coconino, Kaibab, and Prescott National Forests. Additionally, much of this region is forested and was eliminated from consideration by the IGBP land cover classification criteria. Despite this fact, there are still several locations in Northern AZ that would qualify, or are even highly suitable sites for an additional AZMET weather station, under the current analysis.

Additionally, while AZMET weather station density is relatively high in Southern Arizona (compared to the North), there were still many qualifying locations to be found. Highly suitable locations remain surrounding Phoenix, Tuscon, and Yuma, with a particularly large qualifying region between Tucson and Phoenix. This region may be of particular interest due to its proximity to the UA main campus.

While this analysis has effectively limited the search area by eliminating large swatches of the state from consideration, final site selection would have to be carried out by further analysis. These site suitability maps could be further compared to maps of existing university property, state or federal property, or land for sale, etc.

Suitability analysis is a simple idea, eliminate areas one at a time until only the qualifying locations remain. However, depending on the chosen criteria and size of the data set, it can quickly become an overwhelming endeavor. This analysis, using 8 selection criteria, required 7 distinct data sets, and 48 separate processing steps just for the Boolean analysis alone. Despite the simple concept, large or comprehensive suitability analysis can be a massive undertaking, requiring careful analysis, and, potentially, substantial computational resources.

References

Alejo, L. (2018). Suitability analysis for optimum network of agrometeorological stations: A case study of Visayas Region, Philippines. Journal of Agrometeorology, 20. https://doi.org/10.54386/jam.v20i4.564

Arizona Meteorological Network. (2024). Arizona Meteorological Network (AZMet) Data. https://azmet.arizona.edu

Belward, A. S., Estes, J. E., & Kline, K. D. (1999). The IGBP-DIS Global 1-Km Land-Cover Data Set DlSCover: A Proiect Overview. Photogrammetric Engineering & Remote Sensing, 65(9), 1013–1020.

Hijmans, R. J. (2023). terra: Spatial Data Analysis. https://CRAN.R-project.org/package=terra

McHarg, I. L. (1969). Design with Nature. Natural History Press.

O’Sullivan, Unwin, Unwin, D., & Corporation, E. (2010). Geographic information analysis (2nd ed.). John Wiley & Sons.

R Core Team. (2021). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. https://www.R-project.org/

Weiss, J., & Scott, E. (2024). azmetr: Access Arizona weather data from the AZMet API. https://doi.org/10.5281/zenodo.7675685

Disclaimer

This project has been conducted for educational purposes only. It represents a hypothetical site selection exercise. While I am currently a student at the University of Arizona, this project has not been commissioned by UA, AZMET, NAU, or any other organization, public or private. I have done my best to source and cite relevant WMO weather station site selection criteria, however these selection criteria, and associated weights, represent my own educated opinion, and should not be considered as authoritative. Neither do the selection criteria (and weights) represent the official position of AZMET. Please enjoy this hypothetical exercise as a learning tool only, and feel free to contact me with any questions or concerns regarding this analysis.

Page updated

Google Sites

Report abuse