Marc Los Huertos
Environmental Analysis, Pomona College
2026-06-26
This SOP describes the assembly, deployment, data retrieval, and post-processing of low-cost particulate matter (PM) sensor networks built around Raspberry Pi single-board computers and Plantower PMS5003/PMS5004 laser-scattering sensors. These sensors estimate PM1, PM2.5, and PM10 mass concentrations and report particle counts in six size bins (>0.3, >0.5, >1.0, >2.5, >5.0, >10.0 \(\mu\)m).
Low-cost PM sensors (LCS) provide spatial and temporal resolution that is not achievable with Federal Reference Method (FRM) or Federal Equivalent Method (FEM) monitors alone (Snyder et al., 2013). However, LCS data require quality screening, relative humidity correction, and calibration against reference instruments to produce scientifically defensible results (Ardon-Dryer et al., 2023; Barkjohn et al., 2021). When properly corrected and quality-controlled, LCS networks can complement regulatory monitoring by filling spatial gaps, supporting community exposure studies, and providing near-real-time information during wildfire smoke events (Morawska et al., 2018).
| Role | Responsibility |
|---|---|
| Lab PI | Approve deployment plan, oversee data quality |
| Field Lead | Coordinate site access, logistics, and safety |
| Trained Personnel | Assemble, deploy, retrieve, and process data |
| New Students | Complete training before independent deployment |
The Plantower PMS5003 and PMS5004 are optical particle counters that use a laser (680 \(\pm\) 10 nm) and a photodiode to detect light scattered by individual particles drawn through the sensing chamber by a small fan. The sensor reports particle counts in six size bins and uses a proprietary algorithm to estimate PM1, PM2.5, and PM10 mass concentrations in \(\mu\)g/m\(^3\) (Barkjohn et al., 2021).
The sensor outputs data in two formats: CF=1 (“standard particles,” factory-calibrated against Chinese urban aerosol) and CF_atm (“atmospheric environment,” with an additional adjustment at higher concentrations). For ambient monitoring in the US, the CF=1 output is used as the starting point for correction (Barkjohn et al., 2021).
| Item | Qty per Unit | Notes |
|---|---|---|
| Raspberry Pi Zero+ (or 3+/4+) | 1 | With microSD card (32 GB min) |
| Plantower PMS5004 (or PMS5003) | 1 | With ribbon cable |
| BME280 sensor module | 1 | Temperature, RH, pressure |
| USB power supply (5V, 2.5A) | 1 | Or solar panel + battery |
| Weatherproof enclosure | 1 | Ventilated; inlet tube for PM |
| microSD card (32 GB) | 1 | Pre-loaded with sensor software |
| Ethernet or Wi-Fi dongle | 1 | If Pi Zero (built-in on Pi 3+/4+) |
| Item | Notes |
|---|---|
| Zip ties, mounting brackets | For securing enclosures |
| Extension cord or solar panel | Power supply for remote sites |
| GPS unit or smartphone | For recording coordinates |
| Field notebook | Site metadata |
| Laptop with SSH client | For on-site diagnostics |
Before field deployment, collocate all sensor units at a single location for a minimum of 48 hours:
Retrieve data at regular intervals (weekly to monthly):
scp or rsync.Raw PM2.5 readings from PMS5003/5004 sensors overestimate FRM/FEM concentrations by approximately 40% under typical US ambient conditions (Barkjohn et al., 2021). The EPA US-wide correction reduces root mean square error from approximately 8 \(\mu\)g/m\(^3\) to 3 \(\mu\)g/m\(^3\) (Barkjohn et al., 2021).
library(tidyverse)
raw <- read_csv("data/raw/site01_20260101_20260201.csv")
raw <- raw %>%
filter(
pm25_cf1 >= 0,
pm25_cf1 < 1000,
rh_pct >= 0,
rh_pct <= 100,
temp_c > -40,
temp_c < 60
)
# Flag and remove sensor freeze periods
raw <- raw %>%
mutate(frozen = pm25_cf1 == lag(pm25_cf1) &
pm25_cf1 == lag(pm25_cf1, 2) &
pm25_cf1 == lag(pm25_cf1, 3)) %>%
filter(!frozen | is.na(frozen)) %>%
select(-frozen)\[ \text{PM}_{2.5,\text{corrected}} = 0.524 \times \text{PM}_{2.5,\text{cf1}} - 0.0862 \times \text{RH} + 5.75 \]
where PM2.5,cf1 is the CF=1 output (\(\mu\)g/m\(^3\)) and RH is relative humidity (%) (Barkjohn et al., 2021).
If your sensors are PMS-U (post-mid-2021), apply the additional linear correction from Ardon-Dryer et al. (2023) before the EPA correction. PMS-U sensors can be identified by a count0.3/count0.5 ratio approximately 3 times lower than PMS-O sensors.
hourly <- raw %>%
mutate(hour = floor_date(timestamp, "hour")) %>%
group_by(hour) %>%
summarize(
pm25_mean = mean(pm25_corrected, na.rm = TRUE),
pm25_sd = sd(pm25_corrected, na.rm = TRUE),
rh_mean = mean(rh_pct, na.rm = TRUE),
temp_mean = mean(temp_c, na.rm = TRUE),
n_obs = n(),
.groups = "drop"
) %>%
filter(n_obs >= 45)
daily <- hourly %>%
mutate(date = as_date(hour)) %>%
group_by(date) %>%
summarize(
pm25_daily = mean(pm25_mean, na.rm = TRUE),
n_hours = n(),
.groups = "drop"
) %>%
filter(n_hours >= 18)calc_aqi <- function(pm25) {
breakpoints <- tibble(
lo_pm = c(0.0, 9.1, 35.5, 55.5, 125.5, 225.5),
hi_pm = c(9.0, 35.4, 55.4, 125.4, 225.4, 325.4),
lo_aqi = c(0, 51, 101, 151, 201, 301),
hi_aqi = c(50, 100, 150, 200, 300, 500)
)
bp <- breakpoints %>% filter(pm25 >= lo_pm, pm25 <= hi_pm)
if (nrow(bp) == 0) return(NA_real_)
with(bp, lo_aqi + (hi_aqi - lo_aqi) / (hi_pm - lo_pm) * (pm25 - lo_pm))
}
daily <- daily %>%
mutate(aqi = map_dbl(pm25_daily, calc_aqi))| Flag | Condition | Action |
|---|---|---|
| F1 | RH >95% for >4 consecutive hours | Flag; consider excluding |
| F2 | PM2.5 >500 \(\mu\)g/m\(^3\) for >1 hour | Verify against regional monitors |
| F3 | Sensor freeze (identical readings >10 min) | Exclude |
| F4 | Hourly data completeness <75% | Exclude hour |
| F5 | Daily data completeness <75% of hours | Exclude day |
Re-collocate sensors with a reference instrument (or with each other) at least once per year and after any sensor replacement. Document drift and update correction factors if needed (Malings et al., 2020).
data/raw/ with the naming convention: siteID_YYYYMMDD_YYYYMMDD_raw.csvdata/processed/data/metadata/When reporting LCS data in publications or reports, include (Morawska et al., 2018):
| SOP | Title | Relationship |
|---|---|---|
| F001 | Field Safety | Required prerequisite |
| F101 | Electrical Power in the Field | Power supply guidance |
| F501 | Air Temperature Monitoring | Companion: temp sensors on same Pi |
| F503 | PM Filter Sampling | Reference-quality gravimetric companion |
| F504 | Picarro GHG Analyzer | Companion field instrument SOP |
Before deploying air quality sensors independently, personnel must: