The following requirements went into the design of the CalMAPLab:

Compounds should be measured with Research, Federal Equivalence Method or Federal Reference Method grade instrumentation.

The CalMAPLab was designed to provide research grade measurements of major air pollutants, greenhouse gases and other atmospherically relevant tracers at 1 Hz resolution where available. The complete suite of instruments is listed in Table A1 and briefly described below.

The CalMAPLab is a retrofitted 2023 Ford Transit 250 delivery vehicle with a medium roof height (2.5 m + 0.7 m weather station) and extended 3.8 m (148^′′) wheel base specifications. A number of key considerations went into the purchasing of the vehicle. In general, a larger vehicle is preferred for increased instrumentation space, air circulation and weight capacity. However, for sampling location flexibility and driver availability, our needs dictated that no commercial driver's license would be required. In addition, the vehicle length dimension was limited by the size of our indoor parking space on the UC Berkeley Campus which was deemed necessary for security and 24/7 instrument operation. Multiple heights are available for the Ford Transit. The medium roof option was chosen as a compromise between maximizing space and minimizing the risk of contact with low-hanging cables or tree branches above many smaller roads in California. Although the minimum height is regulated at ∼ 4.6 m (15^′) for communication wires and branches, test driving encountered situations where this was not the case (Public Utilities Commission of the State of California, 2020). Weight was a further consideration and total payload weight was estimated prior to vehicle purchase. The decision to purchase the 250 model (gross vehicle weight rating: 4114 kg, payload capacity: 1635 kg) was made based on model availability and assumptions of a 1490 kg maximum payload (incl. two occupants and a full fuel tank). Finally, electric powered vehicles were considered and tested. Two clear advantages exist over fossil-fuel powered vehicles; there are no exhaust emissions that interfere with sampling and there is a pre-existing battery-electrical system to power instrumentation. While the electric model would have been sufficient for nearby urban sampling, at the time of purchase, the electric model mileage (2023 Ford E-Transit ∼ 187 km less instrumentation power) was deemed insufficient for sampling communities more than 1 h away. The gasoline model chosen has a 25-galllon fuel tank with an average range of 418 km under our operating conditions, which is generally dominated by urban stop/start driving.

The power demands of all instrumentation in the CalMAPLab, and the associated air conditioning (A/C) was estimated prior to design of the electrical system via laboratory measurements. The total continuous power consumption of the measurement platform is ∼ 3 kW and discussed in detail in Sect. 3.1. The majority of this power is delivered at 120 V AC, with ∼ 200 W at 12 V DC. All equipment, including the A/C, is left running 24/7 to minimize the labor demands and potential instrument drift associated with powering down and restarting instrumentation, as well as ensuring preservation of high vacuum inside the mass spectrometer.

These power requirements can be met by an inverter system tied to a battery bank, engine alternator, external generator or some combination thereof. To facilitate the goals of 8–10 h of stationary and/or mobile operation, a battery-electric system with additional engine alternator charging while driving was chosen. This enabled stationary sampling without the worry of “self-sampling” exhaust emissions, while extending drive time capacity at a lower weight and space cost than a larger battery bank.

A 12 V electrical system with eight 270 Ah LiFePO₄ deep cycle GC3 batteries (Battleborn, Reno, USA) connected to two 2 × 120 Multiplus II 3 kW inverters (Victron Energy, Almere, Netherlands) wired in parallel for up to 6 kW of continuous power draw and 240 A charging capacity (see Fig. 2 for a complete electrical diagram) was installed in the CalMAPLab cargo area. Although the inverter capacity greatly exceeds system power demands, inverter efficiency reduces with temperature (2.5 kW at 25 °C, 2.2 kW at 40 °C, max 93 %), and this buffer room is required when sampling in hot locations. A 12 V secondary, after-market alternator (280XP-2023-UP-PFDI, Nations Starter and Alternator, Cape Girardeau, USA) was installed on the engine with associated regulator (Model WS500, Wakespeed, Reno, USA) and monitoring shunt for ∼ 1.5 kW of continuous power while the engine is running. The options for secondary alternators that fit the 2023 Ford Transit PFDI engine are limited. Since overall system voltage is determined primarily by the second alternator, the electrical system design should be considered in tandem with the vehicle base choice. Higher voltage DC systems may be preferred for smaller diameter wiring. However, this comes at the expense of additional DC-DC converters for powering 12 V equipment.

The batteries are wired as four separate battery banks, each consisting of two batteries with its own on/off switch. This allows the removal of some battery capacity should space or weight requirements change in the future. The DC loads are combined and fed into the inverters via busbars within Lynx Distributors (Victron Energy, Almere, Netherlands), either side of a Lynx Shunt (Victron Energy, Almere, Netherlands) fitted with a 500 A mega fuse. An additional 12 V DC fuse box (Blue Sea Systems, Menomonee Falls, USA) is wired to the output distributor for power to interior lighting and 12 V instrumentation. All other equipment runs off 120 V AC power generated by the inverters. Four 20 A duplex receptacles are wired to all corners of the van via a 120 V fuse box (Square D, Andover, USA), in addition to a 20 A line for the A/C. It is noted that some efficiency gains could be made by wiring DC instrumentation (e.g. Licor CO₂, CAPS NO₂, 2BTech O₃) directly into the DC supply rather than use supplied adapters. With this payload, the savings are negligible (∼ 10 W) compared to the total power draw. However, should the payload contain a large fraction of DC powered equipment, this is worth consideration. Shore power is provided via a 50 A split phase circuit and NEMA 14–50 plug. This high-capacity infrastructure was chosen to facilitate both overnight battery charging with simultaneous instrument operation, in addition to its common availability at RV parks for field campaign flexibility. The transition from pass through shore power to inverter power is slower on the second hot leg. To prevent power loss to sensitive instrumentation, back-up UPS's (Model 1500, APC Schneider Electric, Rueil-Malmaison, France) are installed on those receptacles. Communication between all electrical components is handled by a Cerbo GX (Victron Energy, Almere, Netherlands) which is connected to the van internal WIFI. All measured parameters are displayed in an online portal for continuous remote monitoring in the front of the vehicle.

The platform design incorporated heat management and vibration dampening in order to improve instrument performance. The CalMAPLab was designed to be able to operate in locations with > 30 °C ambient temperatures. Exterior solar heating, especially when stationary without the cooling effect of passing air, will increase the temperature in the van cargo area and cause the instruments to overheat. Additionally, the high thermal load produced by the continuous operation of the inverters and instruments within the cargo space would increase temperatures beyond many instruments' operable ranges when sealed shut without any cooling. To minimize exterior solar heating, the exterior of the van is painted white to increase albedo and the van walls and ceiling are lined with ∼ 5 cm of insulation and a DuraTherm liner (Legend Fleet Solutions, Tillsonburg, Canada). The interior is maintained at ∼ 22 °C with a 15 kBtu A/C (RP-AC3800-P-W-KT, RecPro, Bristol, USA) in combination with two whole-room circulation fans (Vornado, Andover, USA). Thermal regulation performance is evaluated in Sect. 3.1.1.

For vibration dampening, most instrumentation is mounted within two off the shelf 24U 19^′′ server racks each bolted to a shock-rope isolation plate system. The Vocus-PTR-ToF-MS was built into a custom steel rack with similar shock rope isolators for relevant components and mounted to the floor via an L-track locking system. This enables easy removal for major maintenance and payload flexibility. L-track lines were also placed along the floor, walls and ceiling at strategic locations for mounting and securing non-rack-mountable equipment. Particle instruments were placed in the driver's side rack to be in line with the roof inlet hole and minimize the number of bends and resulting particle loss. Both server racks are rotated such that access to the front and rear of each instrument is granted.

The heavy electrical system was placed over the rear wheel base so as to not exceed the front gross axle weight rating (GAWR), and for stability/improved vehicle handling. Other heavy equipment like the Vocus, UPS's and compressed gas cylinders are situated on the opposite side of the vehicle to the battery bank to balance L-R weight distribution. Overall, the weight was measured as 1733 and 2132 kg for the front and rear axles respectively, below GAWR limits of 1873 and 2502 kg.

The sampling inlet was designed to minimize slip stream and self-sampling, without extending vehicle height to a level that would impact route flexibility. As such, air intakes for the instruments are positioned out the front of the mobile lab via connection to a t-slot metal stability beam secured to two roof racks. Inlet lines then enter through a hole in the roof through bulkhead adapters (Swagelok, Solon, USA) fitted to custom machined plates secured on either side of the aluminum roof. Sample air is then distributed to the different racks and instruments via the flow paths shown in Fig. 3.

There are three gas phase sample lines each made from PTFE to minimize wall losses. The gas rack and Vocus lines are 1/4^′′ OD to minimize lag time. The Licor-7200RS samples from a separate 3/8^′′ OD line through a large HEPA filter (Pall Life Sciences, Port Washington USA) to accommodate the internal pump sensitivity to larger pressure drops. Each instrument has a capped t-piece and 2-way Teflon-valve for connection to a calibration line and isolation from all other instruments. In addition, each instrument has its own filter (DIF-BK60, United Filtration Systems, Sterling Heights, USA) to prevent particle contamination. Flow to the gas rack (∼ 5 slpm) and LI-7200 (∼ 10 slpm) is provided through the internal pumps of the instruments and for a response time of ∼ 1 s (see Table A1 for breakdown by instrument). An additional bypass pump (Model DOA-V751-FB, Gast, Benton Harbor, USA) and needle valve is used on the Vocus inlet plate to increase flow from 100 sccm to 5 slpm for similar response time.

The particle inlet consists of a 1/2^′′ OD stainless steel line fitted with an aerodynamic inlet cone (Brechtel, Hayward, USA) and in-house machined silica bead cartridge dryer. Drying efficiency is continuously monitored with an in-line humidity sensor (HMP 60, Vaisala, Vantaa, Finland). Three flow splitters (Brechtel, Hayward, USA) accurately divide the sample stream and distribute the sample air to each particle phase instrument via short (< 50 cm) sections of conductive tubing. Here, use of non-stainless steel is minimized while maintaining enough flexibility that vehicle motion does not damage the inlet itself. Flow is provided by the instrument's internal pumps for laminar flow (Re = 1400 at 0.43^′′ ID tube diameter, 11 L m⁻¹, 20 °C and 1000 mbar with an entrance length of 0.76 m) up until the flow splitters. Particle loss was estimated for different particle diameters as a percentage loss through the inlet (von der Weiden et al., 2009) with results shown in Fig. S4. When stationary, loss ranges from 6 % for PM_2.5 up to 40 % for the smallest size bins in the sub 10 nm range. As driving speeds increase larger particles are preferentially sampled, increasing sampling efficiency above ∼ 300 nm. This increase can be as high as 40 % for PM_2.5 when driving at speeds of 25 m s⁻¹ (90 km h⁻¹). Under these conditions, an isokinetic inlet is clearly desirable. However, the majority of driving undertaken by the CalMAPLab is carried out at 10 m s⁻¹ (36 km h⁻¹) where the effect is minimal. As such, the additional expense of such an inlet could not be justified. All four sample inlets have a water trap installed to prevent instrument contamination in heavy rain.

The design requirements for the CalMAPLab data acquisition system were to (a) acquire and synchronize measurements from a variety of instrumentation at the finest time resolution supported by each instrument, (b) accurately geolocate each timestamp with GPS data, (c) efficiently store and back up all data, (d) provide real time visualisation of select data both as time series and spatially on a map, both locally and remotely, (e) facilitate valve actuation for calibrations and (f) be flexible enough to support changes in instrument payload without code changes. To accomplish that, data-streams from each instrument are collected with a custom data acquisition system referred to as VanDAQ, the architecture of which is shown in Fig. 4a. The hardware running VanDAQ consists of 2 separate Linux-operated computer systems: a low-power mobile server (Vault Pro VP6670-6, Protectli, Vista, USA) in the vehicle and a stationary server located on the UCB Campus. RS232 serial, USB, and Ethernet data connections link each instrument to the mobile server and a number of Python software processes facilitate data transfer into a local PostgreSQL database. The data is also packaged as one-minute files and transmitted over a cellular-WIFI data link via a 5G router (MOFI6500-5GXeLTE, MoFi Network, Toronto, Canada) to the central server which ingests data to the central database. While this feature was not used during testing, the architecture supports multiple mobile servers running the same software, all reporting data to the central server with a unique platform identifier. All processes in VanDAQ are written in open-source software (Python, PostgreSQL) and maintained on a publicly available repository (https://github.com/CalMAPLab, last access: 22 June 2026).

Each instrument in the mobile platform is serviced by an acquirer process. The acquirer is a Python script that is configurable via a text file to acquire data based on the data link style, protocol and format unique to each instrument. Therefore, there is only one acquirer code body that is launched into multiple processes configured to each instrument. The acquirer is the only software component in the system that understands the unique communication needs of the instruments, converting the data into a common-currency exchange format consisting of a Python dictionary for each individual data value with metadata describing the platform, instrument, type, parameter, and unit of the data item. Configuration file format, including examples and how to set up a new instrument, are detailed in the GitHub repository documentation. The system is SNTP-disciplined and synchronized by system-timesyncd with time alignment across instruments established from deployment of acquirers on a single acquisition computer. Each acquirer process timestamps its records using the shared host wall clock with a 1 s resolution. Possible skew between acquirers is bounded by serial latency, process scheduling and queue handoff. We measure a maximum timestamp skew between pairwise observations of 1 s from acquirers stamping adjacent seconds 0.3 % to 0.5 % of the time. The unified and packaged measurement data are loaded by the various acquirer processes into a single POSIX interprocess communications (IPC) queue (max messages = 50, max data per message = 8 kb) for further processing downstream. Overall, this feature of the VanDAQ architecture facilitates the incorporation of various diverse instruments and changing payloads without change to the rest of the system downstream of the acquirer.

In the van the measurement queue is unloaded by the Collector process which (1) loads the data items into the mobile server's PostgreSQL database via batched row INSERT and (2) packages the measurement dictionaries into one-minute submission data files for transmission to the central server. As the Acquirer is the only process that knows how instruments communicate, the collector is the only component which deals with the intricacies of loading the PostgreSQL database. Ingestion performance was quantified from production collector logs with an average sustained throughput of ∼ 1300 rows s⁻¹ compared to an average arrival rate of ∼ 160 rows s⁻¹ from roughly 160 1 Hz instrument-parameter pairs (including some engineering data). This indicated significant operational headroom for continuous acquisition with the current instrument payload. Durability of the system is primarily provided by the PostgreSQL database. Following power loss, only data within the IPC and therefore uncommitted to the database is lost.

The computer system runs an operator dashboard website which is hosted on an Apache HTTP server and available via the van's internal WIFI network. The dashboard displays graphical time series of measurement parameters produced by all instruments, alarm states, and a real-time street map with data tracks and wind rose plotting (see Figs. S5–S9). Shape files can be uploaded to the server to highlight areas of interest on the map during a drive. In addition, a control panel is set up to send various serial commands for valve triggering and calibrations.

The submission files are transferred via cellular or WIFI link to the central server by the van's Submitter process. Submission of data files to the central server is asynchronous and can deliver data for ingestion into the central database in near-real-time (1+ min latency) or, depending on the state of the mobile network link, will store data files and submit when possible. Another collector process on the central server ingests data from the submission files and inserts measurements into the central database, whose schema is a duplicate of the mobile database(s). The central server also hosts an analysis dashboard website allowing researchers to browse historical data, monitor the current state of the van, and observe mapped data collection in near-real-time remotely.

The Postgres database schema (see Fig. 4b) on both the mobile and stationary servers follows a star data model, with two central fact tables for instrument measurement values and alarms, and dimension tables for alarm level and type, mobile platform ID, instrument, acquisition type (e.g. environmental measurement, instrument engineering value, or geolocation), parameter name, measurement unit, timestamp, and geolocation. The dimension table contents are dynamically built from incoming measurement metadata, so that new measurement types can be accommodated without changes to database structure or requirements for direct dimension table maintenance. The geolocation dimension table is assembled from incoming GPS instrument data identified by the “gps” acquisition type (which otherwise are treated the same as all instrument measurements), with columns for platform, instrument (GPS unit), and timestamp IDs. Multiple GPS receivers can be accommodated on the mobile platform(s), with the most accurate or relevant geolocation stream selectable for the given output. Instrument data queried for map display or geolocated raw data files are joined to the geolocation table by sample time, filtered by platform and selected GPS receiver.

Acquired data are visualized via a Python web application built on the Dash framework (Plotly Technologies Inc.) (Carson Sievert, 2020). The application queries data through a SQLAlchemy interface (Bayer, 2012), and provides last-5 min time series plots of data acquired by all instruments for diagnostic purposes, a dynamic table of alarms, a live map display, and a page of instrument controls. The app is hosted on both mobile and stationary servers by the Apache2 web server (Apache Software Foundation).

Post-drive, raw files are generated from the database based on pre-determined criteria (default is all time stamps where GPS is not equal to NA). These files are then post corrected with a custom adaptable workflow. First, data from each instrument is lag corrected for sample and instrument response delays such that data points line up with the correct geolocation. This is done monthly by burning a lighter at the inlet and documenting the time each instrument's concentrations change and is validated by calculating covariances between different species across different lags (see Fig. S10). Data from instruments which sample at frequencies above 1 Hz is also averaged to 1 Hz.

Next, the raw GPS coordinates are cleaned to correct positional inaccuracies and gap-fill missing timesteps. A Kalman filter is applied to smooth the coordinates before they are linearly interpolation to 1 Hz. This step is necessary since the u-blox GPS output is slightly less than 1 Hz (average 0.8 Hz) and NA removal would lead to loss of valid chemical measurements. Coordinates are then snapped to road geometries to standardize locations, with preferential matching to road segments with directions within 60° of the GPS bearing. These road linestrings are created from the OpenStreetMap (OSM) database by grouping continuous roads of the same name and direction, then splitting these grouped geometries into short road segments which are identified by centerpoint coordinates (OpenStreetMap contributors, 2017). Each 1 Hz sample is matched to these smaller road segments, typically 30 m lengths, and H3 indices for spatial aggregation. Individual drive pass statistics are generated based on a time threshold exceedance (default 30 s) between visits to a 30 m segment ID. Aggregation at these coarser temporal and spatial scales is then made depending on the visualisation required. Each of these transformational steps is shown in Fig. 5.

Following GPS processing, calibrations are linearly interpolated and then applied. Finally, data is flagged based on engineering parameters to remove periods of suspect data and calibration times.

Live visualization of geolocated data was essential for VanDAQ's data acquisition system, serving two primary functions: tracking drive progress when sampling all roads in a neighborhood, and enabling on-the-fly routing decisions when tracing emission plumes based on spatial concentration gradients. Figure 8 demonstrates both capabilities. Panel (a) shows how real-time visualization of spatial gradients enabled efficient mapping of a large CH₄ emissions plume through West Oakland, CA. While the time series data alone does indicate elevated concentrations, the visual cues within the map provide intuitive tracing of the impacted area and clear guidance on where to sample next. Panel (b) highlights the importance of this visualization for neighborhood-wide sampling, where the sheer number of roads makes routing from memory impractical. Here, we see that same emissions plume fully mapped out with a clear impacted area in the north. This particular example can be attributed to emissions from a wastewater treatment plant (WWTP) to the west through the chemical speciation provided by CalMAPLab, whereby the lack of C₂H₆ in the CH₄ plume rules out natural gas, but the presence of odorous sulfur compounds like methanethiol (CH₄S) that are commonly associated with WWTP processes is compelling evidence for that facility (Li et al., 2021). Without real-time mapping, such semi-transient emission events could easily be missed or misattributed, particularly in areas without prior knowledge of sources. VanDAQ's live visualization allows users to identify unexpected features in any incoming data stream and adapt drive plans accordingly.