📄 earth science

Improving the measurement quality of surface meteorological observations through routine physical maintenance and shield cleaning practices

This study demonstrates that routine physical maintenance and shield cleaning of Automatic Weather Stations significantly improve data quality by reducing systematic biases and increasing agreement with manual observations for daily maximum temperature and relative humidity.

Original authors: Leonard Macharaga, Moven Manjowe, Tinetariro Wesley Benjamin Chikati, Rebecca Manzou

Published 2026-09-08

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Leonard Macharaga, Moven Manjowe, Tinetariro Wesley Benjamin Chikati, Rebecca Manzou

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the world of weather forecasting, the most critical data often comes from the quiet, unassuming corners of the landscape where instruments sit exposed to the elements. These devices, known as automatic weather stations, are the workhorses of modern meteorology, constantly measuring the air's temperature and moisture to feed the models that predict storms, droughts, and heatwaves. For these machines to be useful, their readings must be true reflections of the atmosphere, not distorted by the very environment they are meant to observe. A common enemy to this accuracy is dust. In dry, windy regions, fine particles can settle on the protective cages surrounding the sensors, acting like a layer of grime that blocks airflow and traps heat. When this happens, the instruments begin to drift from reality, recording temperatures that are too high or humidity that is too low. If these errors go unnoticed, they can ripple through the entire system, leading to flawed forecasts and missed warnings for communities that rely on them.

A team of researchers from the Zimbabwe Meteorological Services Department set out to test a simple, low-cost solution to this problem at a station in Chisumbanje. Located in a semi-arid region known for its hot days and wind-blown red dust, the station serves as a perfect laboratory for studying how environmental grime affects data. The scientists focused on two specific variables: the highest temperature reached each day and the amount of moisture in the air, known as relative humidity. To see if cleaning the equipment actually worked, they set up a controlled experiment comparing the automatic station against a traditional manual weather station located right next to it. The manual station, which uses liquid-in-glass thermometers inside a wooden, louvered box, served as the trusted reference point, representing the ground truth of what the weather was actually doing.

The researchers tracked the data for forty days, splitting the timeline into two equal halves. The first twenty days represented a period of heavy dust accumulation, where the automatic station's plastic shields had become caked with dirt. On August 20, 2025, the team performed a routine physical maintenance intervention. They took the automatic station's multi-plate plastic shields and thoroughly cleaned them with distilled water to remove the layers of dust. They also cleared the protective mesh caps on the humidity sensors, ensuring that air could flow freely to the sensitive parts inside. The following twenty days captured the performance of the station in its newly cleaned state. By comparing the data from before and after this cleaning, the team could isolate exactly how much the dust had been skewing the results.

The results showed that the simple act of washing the equipment made a measurable difference, though the impact varied depending on what was being measured. For the daily maximum temperature, the automatic station was already performing quite well even when dirty, but it still consistently read slightly lower than the manual station. After the cleaning, the agreement between the two stations became nearly perfect. The automatic station's readings aligned much more closely with the manual reference, with the average difference shrinking significantly. The data points, which had previously scattered a bit, now huddled tightly around the line of perfect agreement, showing that the removal of dust had restored the instrument's ability to sense the true heat of the day.

The improvement was even more dramatic for relative humidity. Before the cleaning, the automatic station was struggling significantly, consistently recording moisture levels that were far lower than what the manual station observed. The gap between the two was substantial, with the automatic station underestimating the humidity by a wide margin. Once the dust was washed away and the sensor vents were cleared, the automatic station's performance jumped up. While it did not become perfect, the error dropped sharply, and the readings began to track the manual station much more faithfully. The dust had been physically blocking the sensor from "feeling" the moisture in the air, and cleaning it allowed the instrument to breathe again.

Despite these clear improvements, the study also highlighted a limit to what a simple wash can achieve. Even after the cleaning, the automatic station still showed a small, consistent tendency to read lower than the manual station. This suggests that while the dust was a major culprit, there was also an underlying issue with the sensor's calibration that had drifted over time. The researchers noted that the humidity sensor, which relies on a delicate internal membrane, is particularly vulnerable. Dust can lodge deep inside the protective mesh, creating a tiny, stagnant pocket of air that the sensor cannot easily clear with a surface wipe. This means that while physical cleaning is a powerful tool for fixing immediate problems caused by the environment, it cannot fully replace the need for periodic, professional recalibration in a laboratory setting.

The findings from Chisumbanje offer a clear lesson for weather services around the world, especially in regions where dust is a constant presence. The study demonstrates that routine physical maintenance is not just a matter of keeping equipment looking tidy; it is a critical step in ensuring the accuracy of the data that drives weather forecasts and climate records. By simply cleaning the shields and sensors, the researchers were able to recover a significant amount of lost accuracy, proving that low-cost, hands-on care can keep high-tech instruments reliable. However, the lingering bias in the humidity readings serves as a reminder that cleaning is only one part of the puzzle. To maintain the highest quality of data, weather stations need a combination of regular field cleaning and scheduled, professional checks to catch the subtle drifts that dust alone cannot explain.

Technical Summary: Improving Measurement Quality of Surface Meteorological Observations

Problem Statement
Automatic Weather Stations (AWS) are increasingly replacing manual observation networks, yet their data quality is frequently compromised by sensor degradation over time. In regions like Zimbabwe, characterized by semi-arid climates and significant wind-blown dust, this degradation is exacerbated by environmental factors. Unlike traditional wooden Stevenson screens, modern AWS radiation shields are often constructed from compact plastic. These shields are prone to trapping solar heat on calm, sunny days and, more critically, accumulating layers of dust. This dust accumulation lowers the shields' albedo (reflectivity) and physically obstructs natural ventilation pathways. While National Meteorological and Hydrological Services (NMHS) in developing regions often lack the resources for formal laboratory calibration, the specific quantitative impact of simple, low-cost physical cleaning on restoring data accuracy remains under-documented.

Methodology
The study was conducted at the Chisumbanje weather station in Zimbabwe, a site with high exposure to wind-blown dust due to local agricultural activities. The research employed a "before-and-after" experimental design over a 40-day period centered on a routine maintenance intervention executed on August 20, 2025.

Data Sources: Paired daily observations were collected from a co-located AWS (equipped with PT100 resistance and capacitive relative humidity sensors in a multi-plate plastic shield) and a Manual Weather Station (MWS) serving as the reference (using liquid-in-glass thermometers in a wooden Stevenson screen).
Intervention: The maintenance involved a thorough physical cleaning of the AWS multi-plate plastic shields with distilled water to remove dust layers and clearing particulate blockages from the relative humidity sensor's protective mesh cap.
Analysis Period: The dataset was split into a 20-day pre-maintenance period (July 31 – August 19) representing peak dust contamination, and a 20-day post-maintenance period (August 21 – September 9).
Statistical Metrics: Agreement between the AWS and MWS was evaluated for daily maximum temperature ( $T_{max}$ ) and relative humidity (RH) using Pearson's correlation coefficient ( $R$ ), Root Mean Square Error (RMSE), Mean Bias Error (MBE), and Percentage Bias (PBIAS).

Key Results
The study found that routine physical maintenance measurably improved data quality for both variables, though the magnitude of improvement varied:

Maximum Temperature ( $T_{max}$ ): The AWS already showed strong agreement with the MWS prior to maintenance ( $R = 0.992$ ). Post-maintenance, agreement improved further, with $R$ increasing to $0.999$ and RMSE decreasing from $0.863^\circ\text{C}$ to $0.305^\circ\text{C}$ . The Percentage Bias (PBIAS) narrowed significantly from $-2.85\%$ to $-0.83\%$ .
Relative Humidity (RH): This variable showed weaker initial agreement ( $R = 0.740$ ) and a larger systematic underestimation (PBIAS of $-20.13\%$ ). Following cleaning, the correlation improved to $0.876$, RMSE dropped from $5.015\%$ to $1.949\%$ , and PBIAS reduced to $-6.91\%$ .
Bias Direction: In both periods, the AWS consistently underestimated values relative to the manual station (negative MBE). However, the magnitude of this underestimation was substantially reduced after cleaning, particularly for RH. The study notes that while dust accumulation typically causes temperature readings to run higher due to reduced shield albedo, the AWS in this study consistently displayed negative biases, indicating an uncorrected sensor calibration drift superimposed by ventilation errors.

Key Contributions

Quantification of Dust Impact: The study provides empirical evidence that dust accumulation on AWS shields causes significant systematic bias, particularly for relative humidity, where the pre-maintenance error was roughly seven times larger than that of temperature.
Validation of Low-Cost Maintenance: It demonstrates that simple physical cleaning (distilled water rinsing and mesh clearing) is an effective, low-cost intervention for restoring sensor accuracy in resource-limited settings, recovering a significant portion of the lost data quality.
Differentiation of Error Sources: The results highlight that while cleaning restores ventilation and albedo, it does not fully correct underlying sensor calibration drift. The persistence of a negative bias in RH post-cleaning suggests that capacitive sensors retain microscopic contaminants within their protective mesh that field cleaning cannot fully remove, necessitating periodic laboratory recalibration.

Significance and Claims
The paper asserts that routine physical maintenance is a critical, often overlooked component of preventive maintenance for AWS networks in dust-prone regions. The findings emphasize that:

Operational Necessity: Regular servicing schedules are essential for networks supporting operational meteorology and early warning systems, as undetected sensor drift can propagate errors into downstream products like evapotranspiration models and heat indices.
Limitations of Field Cleaning: While field cleaning is highly beneficial, it is insufficient as a standalone solution for all sensor types. The study concludes that a combination of field cleaning and periodic laboratory recalibration is required to fully address sensor drift, particularly for relative humidity sensors.
Policy Implications: The authors recommend that NMHS enforce strict compliance with WMO guidelines regarding physical inspection and cleaning during routine checks. They further suggest integrating automated threshold flags in data management systems to trigger maintenance alerts when deviations exceed operational tolerances.

The study acknowledges its limitations, noting that the results are based on a single station and a specific 40-day window. Consequently, while the recovery magnitudes are promising, they may not directly transfer to the entire Zimbabwe AWS network without further multi-station, long-term validation.

Technical Summary: Improving Measurement Quality of Surface Meteorological Observations

More like this