CoChat

Quality control and correction method for air temperature data from a citizen science weather station network in Leuven, Belgium

Open access
Beele, Eva, Reyniers, Maarten, Aerts, Raf, Somers, Ben · 2022 · Earth System Science Data Discussions

Preprint — published as 10.5194/essd-14-4681-2022 · Paper page

The DOI matches its Crossref record; 47 of the 47 references it cites were found in a registry or an index; 1 of the 69 sources its claims cite were checked for support.

Abstract

The dataset is accompanied by a newly developed station specific temperature quality control (QC) and correction procedure.

Overview

CoChat Summary

TL;DR

The paper presents the Leuven.cool citizen-science network of around 100 low-cost WH2600 weather stations and a three-level station-specific quality control and correction procedure for air temperature. The method removes implausible measurements and corrects inter-station and intra-station biases, reducing the mean temperature difference versus official RMI stations from 0.15 ± 0.56 K to 0.00 ± 0.28 K.

Key findings

  • The QC method reduced the mean temperature difference ± standard deviation between four low-cost reference stations and official RMI stations from 0.15 ± 0.56 K to 0.00 ± 0.28 K.
  • A random forest model performed best for intra-station bias prediction, with R² = 0.741 and RMSE = 0.279, compared with boosted regression trees (R² = 0.658, RMSE = 0.319) and simpler radiation-based models (e.g. radiation60: R² = 0.321, RMSE = 0.450).
  • For the Leuven.cool network, only 0.5 %–1 % of observations per year were flagged as outliers during QC level 1; QC level 2 gave a mean station offset of 0.010 °C, with station LC-102 highest at 0.349 °C and LC-074 lowest at −0.220 °C.
  • QC level 3 reduced summer mean temperatures and standard deviations by up to −0.40 °C and −0.36 °C, respectively, with little change in winter.

Methods

  • The study uses the Leuven.cool network of 106 low-cost Fine Offset WH2600 stations across Leuven, Belgium, with data from July 2019 to December 2021 aggregated to 10 min observations; four reference stations (LC-R01, LC-R02, LC-R04, LC-R05) were installed next to official RMI AWS in Uccle, Diepenbeek and Humain.
  • The QC procedure has three levels: L1 outlier detection (range, temporal step/persistence, and spatial Z-score tests), L2 inter-station fixed bias correction using uniform-temperature episodes, and L3 intra-station bias correction using a random forest trained on 60 % of LC-R data and validated on 40 %.
  • Evaluation compares LC-R station temperatures with AWS temperatures after each QC level; the method is then applied to the Leuven.cool LC-X network.

Limitations

  • The paper states that transferring the method to other networks or regions requires recalibration of QC parameters; QC L1.1 needs climate thresholds, QC L1.3 and L2 require a dense station network and are less suitable for single or few stations, and QC L3 needs additional data such as radiation and wind speed.
  • The paper notes that the reference-station spatial outlier test is not ideal because each reference station has only one neighbour, except LC-R01 and LC-R02 in Uccle during a 4-month overlap; it also notes possible bias because the RF model was trained on 60 % of the LC-R dataset and applied to the complete LC-R dataset, with test-only results giving 0.00 ± 0.28 °C versus 0.00 ± 0.22 °C for the complete dataset.
  • Evident from the text: only air temperature undergoes quality control and correction, while other variables receive only a qualitative assessment; the network is unevenly distributed across LCZ classes and some LCZ classes are missing due to practical installation constraints.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.5194/essd-2022-113

Cites

47 of 47 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 The WH2600 wireless digital weather station outdoor unit (a) at Mathieu de Layensplein in Leuven (LC-105) and (b) next to the official AWS equipment in Humain (LC-R05) (photograph credit: Maarten Reyniers).
Figure 1 The WH2600 wireless digital weather station outdoor unit (a) at Mathieu de Layensplein in Leuven (LC-105) and (b) next to the official AWS equipment in Humain (LC-R05) (photograph credit: Maarten Reyniers).CC-BY · From the PDF
Figure 2 The Leuven.cool network (LC-X) with (a) LCZ classification and (b) BBK (Bodembedekkingskaart; land use map) classification. (c) Belgium delineated by the three official regions (Flanders, Wallonia and Brussels) with the location of Leuven and the three RMI stations (AWSs) used in this study. The background map is from Esri (ESRI World Topographic Map, 2022).
Figure 2 The Leuven.cool network (LC-X) with (a) LCZ classification and (b) BBK (Bodembedekkingskaart; land use map) classification. (c) Belgium delineated by the three official regions (Flanders, Wallonia and Brussels) with the location of Leuven and the three RMI stations (AWSs) used in this study. The background map is from Esri (ESRI World Topographic Map, 2022).CC-BY · From the PDF
Figure 3 Histograms of the temperature bias between the low-cost reference stations (LC-R stations) and the official RMI stations (AWSs) for day and night (a) , with daytime defined by radiation > 0 Wm 2 (b) and night-time defined by radiation D 0 Wm 2 (c) . Mean biases and their standard deviations are given above the graphs. Note that the ranges on the y axis differ for the different subplots. The temperature bias was calculated for all measurements between the installation date of each LC-R and December 2021.
Figure 3 Histograms of the temperature bias between the low-cost reference stations (LC-R stations) and the official RMI stations (AWSs) for day and night (a) , with daytime defined by radiation > 0 Wm 2 (b) and night-time defined by radiation D 0 Wm 2 (c) . Mean biases and their standard deviations are given above the graphs. Note that the ranges on the y axis differ for the different subplots. The temperature bias was calculated for all measurements between the installation date of each LC-R and December 2021.CC-BY · From the PDF
Figure 4 Temperature bias ( C) as a function of hour of the day and month of the year for all LC-R stations (a) ; temperature bias ( C) as a function of radiation and wind speed for all LC-R stations (b) ; temperature bias ( C) as a function of radiation and wind speed for all LC-R stations, where the values written in each cell signify the sample size (c) ; and temperature bias ( C) as a function of radiation and wind speed for all LC-R stations, where cells with a sample size lower than 10 are not shown in the graph (d) . Note that the ranges on the y axis differ for the different subplots. Background colours, ranging from blue ( 1.7 C) to red (1.7 C), represent the temperature bias. The temperature bias was calculated for all measurements between the installation date of each LC-R station and December 2021.
Figure 4 Temperature bias ( C) as a function of hour of the day and month of the year for all LC-R stations (a) ; temperature bias ( C) as a function of radiation and wind speed for all LC-R stations (b) ; temperature bias ( C) as a function of radiation and wind speed for all LC-R stations, where the values written in each cell signify the sample size (c) ; and temperature bias ( C) as a function of radiation and wind speed for all LC-R stations, where cells with a sample size lower than 10 are not shown in the graph (d) . Note that the ranges on the y axis differ for the different subplots. Background colours, ranging from blue ( 1.7 C) to red (1.7 C), represent the temperature bias. The temperature bias was calculated for all measurements between the installation date of each LC-R station and December 2021.CC-BY · From the PDF
Table 7. The Pearson correlation matrix of temperature bias with other meteorological variables measured by the low-cost station.
Table 7. The Pearson correlation matrix of temperature bias with other meteorological variables measured by the low-cost station.CC-BY · From the PDF
Figure 6 The corrected T bias after validation with the RF model for daytime (a) and night-time (b) . Mean biases and their standard deviations are given above the graphs. Note that the ranges on the y axis differ for the different subplots.
Figure 6 The corrected T bias after validation with the RF model for daytime (a) and night-time (b) . Mean biases and their standard deviations are given above the graphs. Note that the ranges on the y axis differ for the different subplots.CC-BY · From the PDF
Image 7
CC-BY · From the PDF
Image 8
CC-BY · From the PDF
Figure 9 Histograms of temperature difference ( 1T D T LC-R T AWS) for each reference station, (a) LC-R01, (b) LC-R02, (c) LC-R04 (d) and LC-R05, at QC level 0 and QC level 1. The mean differences and their standard deviations are given above the graphs. Note that the ranges on the y axis differ for the different subplots. The temperature difference was calculated for all measurements between the installation date of each LC-R and December 2021.
Figure 9 Histograms of temperature difference ( 1T D T LC-R T AWS) for each reference station, (a) LC-R01, (b) LC-R02, (c) LC-R04 (d) and LC-R05, at QC level 0 and QC level 1. The mean differences and their standard deviations are given above the graphs. Note that the ranges on the y axis differ for the different subplots. The temperature difference was calculated for all measurements between the installation date of each LC-R and December 2021.CC-BY · From the PDF
Figure 10 Temperature difference ( 1T D T LC-R T AWS) as a function of hour of the day and month of the year for each reference station, (a) LC-R01, (b) LC-R02, (c) LC-R04 and (d) LC-R05, at QC level 1. Background colours, ranging from blue ( 1.7 C) to red (1.7 C), represent the 1T . The temperature difference was calculated for all measurements between the installation date of each LC-R and December 2021.
Figure 10 Temperature difference ( 1T D T LC-R T AWS) as a function of hour of the day and month of the year for each reference station, (a) LC-R01, (b) LC-R02, (c) LC-R04 and (d) LC-R05, at QC level 1. Background colours, ranging from blue ( 1.7 C) to red (1.7 C), represent the 1T . The temperature difference was calculated for all measurements between the installation date of each LC-R and December 2021.CC-BY · From the PDF
Figure 11 Temperature difference ( 1T D T LC-R T AWS) as a function of radiation and wind speed for each reference station, (a) LCR01, (b) LC-R02, (c) LC-R04 and (d) LC-R05, at QC level 1. Cells with a sample size lower than 10 are not shown in the graph. Note that the ranges on the y axis differ for the different subplots. Background colours, ranging from blue ( 1.7 C) to red (1.7 C), represent the 1T . The temperature difference was calculated for all measurements between the installation date of each LC-R and December 2021.
Figure 11 Temperature difference ( 1T D T LC-R T AWS) as a function of radiation and wind speed for each reference station, (a) LCR01, (b) LC-R02, (c) LC-R04 and (d) LC-R05, at QC level 1. Cells with a sample size lower than 10 are not shown in the graph. Note that the ranges on the y axis differ for the different subplots. Background colours, ranging from blue ( 1.7 C) to red (1.7 C), represent the 1T . The temperature difference was calculated for all measurements between the installation date of each LC-R and December 2021.CC-BY · From the PDF
Figure 12 Offsets during the selected episodes (dots), showing the median offset (diamond) and its error bar (mean standard deviation) for each reference station (LC-R). For reference, the zero line is plotted in red.
Figure 12 Offsets during the selected episodes (dots), showing the median offset (diamond) and its error bar (mean standard deviation) for each reference station (LC-R). For reference, the zero line is plotted in red.CC-BY · From the PDF
Figure 14 Same as Fig. 10 but for QC level 2.
Figure 14 Same as Fig. 10 but for QC level 2.CC-BY · From the PDF
Figure 16 Same as Fig. 10 but for QC level 3.
Figure 16 Same as Fig. 10 but for QC level 3.CC-BY · From the PDF
Figure 17 Same as Fig. 11 but for QC level 3.
Figure 17 Same as Fig. 11 but for QC level 3.CC-BY · From the PDF
Figure 18 Offsets during the selected episodes (grey dots), the median offset (black diamond) and its error bar (mean standard deviation) for each LC-X station active during at least one episode. For reference, the zero line is plotted in red.
Figure 18 Offsets during the selected episodes (grey dots), the median offset (black diamond) and its error bar (mean standard deviation) for each LC-X station active during at least one episode. For reference, the zero line is plotted in red.CC-BY · From the PDF
Figure 19 (a) Prediction of the temperature bias ( C) as a function of hour of the day and month of the year for all LC-X stations and (b) prediction of the temperature bias ( C) as a function of radiation and wind speed for all LC-X stations, although cells with a sample size lower than 10 are not shown in the latter graph. Background colours, ranging from blue ( 1.7 C) to red (1.7 C), represent the temperature bias. The prediction of the temperature bias was calculated for all LC-X measurements from July 2019 to December 2021.
Figure 19 (a) Prediction of the temperature bias ( C) as a function of hour of the day and month of the year for all LC-X stations and (b) prediction of the temperature bias ( C) as a function of radiation and wind speed for all LC-X stations, although cells with a sample size lower than 10 are not shown in the latter graph. Background colours, ranging from blue ( 1.7 C) to red (1.7 C), represent the temperature bias. The prediction of the temperature bias was calculated for all LC-X measurements from July 2019 to December 2021.CC-BY · From the PDF
Image 18
CC-BY · From the PDF
Figure B1. Decision tree used to delineate training areas for urban LCZ classes.
Figure B1. Decision tree used to delineate training areas for urban LCZ classes.CC-BY · From the PDF
Figure B2. Training areas used for the creation of the LCZ map. Delineation is based on land use, building height, building density and green ratio data. The background map is from Esri (ESRI World Topographic Map, 2022).
Figure B2. Training areas used for the creation of the LCZ map. Delineation is based on land use, building height, building density and green ratio data. The background map is from Esri (ESRI World Topographic Map, 2022).CC-BY · From the PDF
Figure C1.
Figure C1.CC-BY · From the PDF
Figure C1.
Figure C1.CC-BY · From the PDF
Figure C1. Scatterplots of LC-R versus AWS dew point temperature, humidity, radiation and wind speed for each reference station: LCR01, LC-R02, LC-R04 and LC-R05. The identity line is shown in black. The colour scale indicates the density of observations, with yellow indicating the highest and purple indicating the lowest. The scatterplots include all measurements between the installation date of each LC-R and December 2021. For each variable, the same ranges are used on the x and y axes.
Figure C1. Scatterplots of LC-R versus AWS dew point temperature, humidity, radiation and wind speed for each reference station: LCR01, LC-R02, LC-R04 and LC-R05. The identity line is shown in black. The colour scale indicates the density of observations, with yellow indicating the highest and purple indicating the lowest. The scatterplots include all measurements between the installation date of each LC-R and December 2021. For each variable, the same ranges are used on the x and y axes.CC-BY · From the PDF
Figure D1. Same as Fig. 9 but for QC level 3. Comparison between 1T for the test dataset (a-d) and 1T for the complete dataset (e-h) .
Figure D1. Same as Fig. 9 but for QC level 3. Comparison between 1T for the test dataset (a-d) and 1T for the complete dataset (e-h) .CC-BY · From the PDF
Figure D2. Same as Fig. 10 but for QC level 3. Comparison between 1T for the test dataset (a-d) and 1T for the complete dataset (e-h) .
Figure D2. Same as Fig. 10 but for QC level 3. Comparison between 1T for the test dataset (a-d) and 1T for the complete dataset (e-h) .CC-BY · From the PDF
Figure D3. Same as Fig. 11 but for QC level 3. Comparison between 1T for the test dataset (a-d) and 1T for the complete dataset (e-h) .
Figure D3. Same as Fig. 11 but for QC level 3. Comparison between 1T for the test dataset (a-d) and 1T for the complete dataset (e-h) .CC-BY · From the PDF

Verification

Claims the paper makes

1 Valid · 0 Not valid · 42 Not checked · 114 Own results and interpretations

65 more claims

Create a free account to see all 157 claims and to verify the 6 references still open.

See all claims — free account
  • It may state two separate claims as one. 57
  • The quote only partly states this claim. 4
  • Probably not a substantial claim. 34

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 47 references

Research works printed but not linked: 3

  1. Ahrens, C. D.: Meteorology Today: An Introduction to Weather, Climate, and the Environment, 9th Edn., Brooks/Cole, 549 pp., 2009. · DOI
  2. Rizwan, A. M., Dennis, L. Y. C., and Lia, C.: A review on the generation, determination and mitigation of Urban Heat Island, J. Environ. Sci., 20, 120-128, https://doi.org/10.1016/S1001- 0742(08)60019-4, 2008. · DOI
  3. Stewart, I. D. and Oke, T. R.: Local climate zones for urban studies, B. Am. Meteorol. Soc., 93, 1879-1900, https://doi.org/10.1175/BAMS-D-11-00019. · DOI · Paper page

Cited by

See all 44

Metadata

Authors
Beele, E., Reyniers, M., Aerts, R., & Somers, B.
Affiliations
  • KU Leuven (BE)
  • Royal Meteorological Institute of Belgium (BE)
  • Sciensano (Belgium) (BE)
Venue
Earth System Science Data Discussions
Published
2022-05-02
Volume
14
Issue
10
Pages
4681-4717
ISSN
1866-3508 (ISSN-L)
OpenAlex
W4307361576
Cited by
44
References
46
Citations by year
2026: 16 · 2025: 13 · 2024: 7 · 2023: 4 · 2022: 4
FWCI
2.04
Type
journal article
Language
en
Open access
Locations
  • Earth system science data (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Lirias (KU Leuven) (submittedVersion, cc-by)
Funders
Fonds Wetenschappelijk Onderzoek (1SE0621N)
Topics
  • Physical Sciences › Environmental Science › Environmental Engineering › Urban Heat Island Mitigation
  • Physical Sciences › Computer Science › Computer Science Applications › Mobile Crowdsensing and Crowdsourcing
  • Physical Sciences › Environmental Science › Management, Monitoring, Policy and Law › Environmental Monitoring and Data Management
Keywords
air temperature, citizen science, data quality control, temperature bias correction, random forest, urban heat island, urban climate monitoring, low-cost weather station, Leuven, solar radiation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-11-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Quality control and correction method for air temperature data from a citizen science weather station network in Leuven, Belgium. https://app.cochat.ai/papers/10.5194/essd-2022-113
The paper
Beele, E., Reyniers, M., Aerts, R., & Somers, B. (2022). Quality control and correction method for air temperature data from a citizen science weather station network in Leuven, Belgium. Earth System Science Data Discussions. https://doi.org/10.5194/essd-2022-113

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.