Cellular Broadband Mapping in the California Central Valley

This project tested the viability of the use of drones for collecting data on cellular connectivity and service speeds in the rural San Joaquin Valley. Juniper Unmanned (a private company partner) and IGIS conducted low altitude drone flights (30 to 60 m above ground level) at 5 sites covering 2,772 acres in the San Joaquin Valley between Nov 1st and Nov 4th 2022. Flights were made using a DJI Matrice 600 drone equipped to carry 3 cellular phones measuring upload and download data speeds (in megabits per second) for three cellular service carriers: ATT, Verizon and T-Mobile.

Results from this project suggest that drones can be used to map cellular data speeds; however, with two very significant limitations: (1) Broadband speed data collected at altitude may not appropriately represent data rates observed on the ground. In order to safely operate drones for broadband data collection, they must be flown at an altitude sufficiently high enough to avoid any obstacles on the landscape that may otherwise obstruct or attenuate cellular signals at ground level. (2) Broadband speeds fluctuate significantly over a short period of time, often more than they vary spatially. The collection of broadband data using drones does not occur instantaneously, but rather over a period of time along a predetermined flight plan. Temporal fluctuations in speeds during these flights can contribute to a false impression of fine scale spatial variability in broadband speeds, when the underlying cause of the variability may more appropriately be attributed to the timing of when the data was collected along the flight path, rather than the position of the drone during the flight.

We conclude that drones equipped with cellular devices can be used to map broadband speeds at the scale of individual fields, but are probably not as efficient nor as cost effective as collecting similar data at a broader scale using conventional ground-based methods (i.e. cars equipped with a means of data collection). We also conclude that the existing mobile app technology for measuring and recording spatially explicit broadband speeds is not ready to support crowdsourcing or enterprise collection efforts due to its data access and management limitations.

To more appropriately and efficiently map cellular broadband speeds, we propose that this data should be collected multiple times over a study area, via conventional ground transportation, and using a cellular application that provides an efficient data pipeline for transferring broadband data into a collaborative spatial database.

The lack of quality broadband connectivity and cellular service in rural San Joaquin Valley impacts education (inability of students to do school work at home and connect to their instructors); telemedicine (inability to communicate with health professionals); and agriculture (inability to do business from home/office and failure to collect, share, or receive data from the field).

Extending fiber optic lines, cellular wireless, and other means for “middle mile”, “last mile”, and the “mile beyond last mile” so households and growers in the Central Valley have service is expensive and will require grant funds to complete.

Though residents are well aware of the lack of connectivity, granting agencies require proof that these areas are underserved. Current coverage maps ¹  used by the FCC to determine broadband access come directly from cellular companies and are not based on actual data collected. Rather they have been created by marketing departments to promote extensive, though highly inaccurate availability “data”. Documented proof of coverage to identify underserved areas is needed by granting agencies.

The overarching goal of this project was to test the viability of drones for collecting data on cellular service connectivity and service data speeds in rural San Joaquin Valley. Specific objectives included:

• Collect cellular network speed data using drone and cellular app technologies, and efficiently transfer this data to a mapping application;
• Map and analyze the cellular speed data in ArcGIS Pro (a desktop application) to assess the data’s quality and collection methodology, prior to recommendations for broader scale implementations;
• Visualize the cellular data analysis results via interactive web mapping; and
• If an efficient system of data collection, analysis, and visualization is discovered, develop a protocol to facilitate similar collection efforts, which may be used to collaboratively crowdsource data collection for broader scale data speed mapping.

Juniper Unmanned, accompanied by IGIS, conducted low altitude drone flights (30 to 60m above ground level) at 5 sites between the dates of November 1st and November 4th 2022, collectively covering 2,772 acres. Lower altitude flights were not possible due to obstructions at the sites, including trees, power lines and towers. The flights were made using a DJI Matrice 600 drone equipped to carry three cellular phones, each utilizing the RantCell ²  cellular application to collect spatially explicit upload and download data speeds (in megabits per second or mbs) in timed intervals over the study areas. Three cellular service carriers were selected for this effort: ATT, Verizon and T-Mobile.

T-Mobile was found to not have coverage at any of the test sites, therefore is not included in any further reporting. Verizon failed to function at the TerraNova farm sites because the network connection couldn’t be established at ground level prior to launching the drone. AT&T download speeds (6,230 records) and upload speeds (6,238 records) were captured via the drone. Verizon download speeds (1,866 records) and upload speeds (1,934 records) were also captured. These data were downloaded from RantCell’s online dashboard, restructured for GIS mapping, and interpolated using the kriging operation in ArcGIS Pro to help visualize the data over the study sites. The data were then analyzed to assess service variability over space and the time needed to complete the drone flights.

West Hills College (i.e., “Farm of the Future”) (36.17303°N and -120.35839°W) is located in the Pleasant Valley, an area directly east of the Diablo Mountains and surrounded by foothills. 208 of the facility’s 239 acres were mapped for broadband speeds, which included row crop fields, pistachio orchards, farm infrastructure, and educational buildings. Chevron is interested in an assessment of broadband coverage in this area, as they are responsible for a large area of oil wells which requires connectivity for crew communications, data collection, and safety.

West Side Research and Extension Center (36.33896°N, -120.112536°W) is located near Calflax and West Side, California. 307 acres of the facility’s 485 acres were mapped for cellular broadband speeds. This facility supports research in agronomic, vegetable, vine and tree crops conducted by researchers from the University of California as well as faculty and staff from other research groups including several California State University campuses and USDA-ARS.

TerraNova Ranch is located west of Fresno, California. Two locations at TerrNova Ranch (North: 36.57612°N, -120.08953°W and South: 36.55388°N, -120.04250°W) were mapped for cellular broadband speeds. Data were collected over a 557 acre area in a northern portion of the farm, and 1,405 acres were mapped at a site further south at the farm. Don Cameron, the Vice-President and General Manager of TerraNova Ranch has been a proponent of extending broadband coverage in the Central Valley. TerraNova Ranch encompasses several thousand acres of non-contiguous fields, which will require a different approach for data collection.

Kearney Agricultural Research and Extension Center (36.59894°N, -119.50875°W) is located in Parlier, California. 295 acres of the facility’s 309 acres were mapped for cellular broadband speeds. Scientific research is conducted on production and environmental management related to fruits, vegetables and field crops. Kearney currently has extended WiFi capabilities, but comparative data on cellular connectivity is needed.

A statistical analysis of the data speed records, and the subsequent kriging interpolations, highlighted important limitations for mapping of the cellular broadband speeds. The first of these limitations is that broadband speeds vary greatly temporally, confounding confidence in the relatively fine scale mapping results produced by this methodology. Maps of TerraNova’s south site show that download speeds and upload speeds, collected simultaneously, did not correlate spatially. The incongruity of these data transfer speeds indicate that independent of each other, upload and download speeds can vary greatly over a short period of time due to a lack of stability in network speeds, potentially giving a false impression of spatial variability that is better attributed to temporal variability.

AT&T upload speeds varied from 1.7 mbs to 8 mbs, with the upload speeds slowest at the beginning and end of the drone flights, which were conducted in a lawnmower pattern from west to east, and showed a major spike in speed throughout the middle portion of the drone flight operations (Figure 1). Conversely, download speeds varied from 0.8 mbs to 7.9 mbs, in a splotchy pattern, presumably again due to the timing of the observations rather than the actual positions of the observations (Figure 2).

Click the arrow on the right of the map to see the AT&T upload speeds for other sites

Click the arrow on the right of the map to see the AT&T download speeds for other sites

Click the arrow on the right of the map to see the Verizon upload speeds for other sites

Click the arrow on the right of the map to see the Verizon download speeds for other sites

Mapping of data transfer speeds at all five study areas likewise showed considerable variability, albeit with upload and download speeds loosely coinciding, with sites to the southwest being faster than speeds observed to the northeast (See Figures 3 through 6, and Tables 1 and 2).

The use of drones for collecting cellular broadband speed data may not be an ideal way of properly mapping cellular data speeds. There are four primary reasons for this:

• Broadband speeds fluctuate significantly over time, often more than they vary spatially.
• The scale of broadband mapping using drones as a means of carrying cellular devices may not be as efficient as simply collecting similar data at a broader scale using conventional ground transportation.
• Broadband speed data collected at altitude (e.g., 30 to 60 m above ground level) may not be appropriately representative of data rates observed on the ground.
• The RantCell cellular app’s online data dashboard does not have an application programming interface (API), which is necessary to develop an automated process of data retrieval needed for larger scale or crowdsourced implementations of a similar data collection methodology. No other publicly available app appears to be more capable of supporting the desired multi-user, spatially explicit data speed recordings, with a more feasible pipeline for data access. RantCell would need to develop this new functionality into their system or a new custom application would need to be developed to fulfill this set of requirements.