Posts

Validation measurements in Addis Ababa and Adama

Wegene Negese, an MSc student at Arba Minch University (Climate) and employee of the Ethiopian Meteorology Institute (EMI), has conducted validation measurements of the low-cost sensor system with SPS30 Sensirion at the EMI meteorological stations of Addis Ababa and Adama. For a period of four months, he collocated the sensor system with itself, and conducted gravimetry measurements. Gravimetry is the reference method for calibrating PM2.5 measurement instruments.

Sensors were installed at meteorology stations of Addis Ababa and Adama
Gravimetric analysis with filter measurements wasconducted to validate the sensor systems

He is currently working on his MSc thesis, but I can already present some preliminary results:

  • The coefficient of variation (CV; a measure of variation between two identical instruments) was 9.5% for two sensor systems in Addis Ababa (based on 12,677 10-minute averages), and 4.4% for two sensor systems in Adama (based on 4,135 10-minute averages). This indicates that the variation between two sensor systems is lower than 10%. 10% is set as a maximum allowed CV for measurement instruments by the NIOSH and the US EPA.
  • The sensor system systematically measures lower than gravimetry, but the correlation is strong. Linear regression of all data points of Addis Ababa and Adama combined (n=16) leads to a slope of 1.62 with an R2 of 0.99. The Pearson correlation is 0.97.

The data of Wegene confirms wat I found in earlier data with measurements in Arba Minch: the SPS30 Sensirion has low within-variation, and shows a stable bias both under ambient and indoor (high) concentration settings versus gravimetry measurements (see this publication). In other words: the SPS30 Sensirion appears to be a very good sensor under Ethiopian circumstances.

Arduino workshop sensor systems

Together with my colleague Afework Tademe (electronics) I organized a one-day Arduino workshop, in order to help colleagues get started with building their own low-cost sensor systems. Five colleagues got a crash-course in microprocessors, sensors and electronics. As part of the program, participants built themselves working systems of relative humidity and temperature sensors, LEDs and real-time clocks. Hopefully we will see a variety of locally developed low-cost sensor systems!

Arduino class
Microelectronics
Arduino build
previous arrow
next arrow
Background

Last June, I presented my work at the international symposium of my institute. That raised the question on whether the same principle (buying microelectronics and locally constructing sensor systems) could be used for the field of water- and soil studies. Over the summer I was able to bring to Ethiopia cheap soil moisture, water level and water temperature sensors. The one-day Arduino workshop on building sensor systems is meant to motivate colleagues to start building their own low-cost instruments.

Below you can see the slides used during the workshop.

 

Presentation at 2023 international symposium

I presented my work on low-cost research methods at the 21st International Symposium on Sustainable Water Resources Development (June 9-10, 2023). This symposium is organized by my institute (Arba Minch Water Technology Institute) and the Water Resources Research Center (WRRC).

Arba Minch University faced budget cuts, which reduces opportunities for local staff to conduct research. Reducing research costs expands research opportunities at this time, because many staff members are idle. Therefore, my focus on low cost research by using Do-It-Yourself (DIY) measurement setups and students as data collectors (student science) caught the attention of the research director of the WRRC. He invited me to present about these two cost-saving methods on the symposium. See here the slides of the presentation.

Apart from presenting, I could display different instruments on tables. I showed all components that go into the low-cost PM2.5 sensor system. Participants could try to register the highest CO2 concentration by blowing into an CO2 measurement instrument. Also, the recently launched awtiCode was on display, and participants could leave suggestions and questions for Python code.

Table presentation
LCS parts
CO2 competition
awtiCode
previous arrow
next arrow

DOI for Arba Minch University journals

Thanks to the Global Equitable Membership (GEM) program of Crossref, all journals published by Arba Minch University (AMU) now have DOI registration. AMU publishes the following journals:

Until recently, these journals did not have DOI registration. I came across the Crossref GEM program, and through this program Crossref has provided Arba Minch University with DOI registration capacities for free. Through DOI registration, articles published in these journals will be found more easily through platforms like Google Scholar, which can lead to more citations and higher scientific impact.

 

[UPDATED] Two of my articles are published in the EJWST, and thanks to Crossref now have permanent DOI links:

Launching awtiCode

Together with colleagues from the Arba Minch Water Technology Institute (AWTI), I have launched awtiCode: Python code by and for AWTI staff. With a core team, we are planning bi-weekly meetings, where progress and plans are discussed. We will inquire dataprocessing needs amongst colleagues, and (try to) code solutions. Code is hosted on GitHub repositories: /jddingemanse/awticode and /awticode.

Current members of the awtiCode core team are Awel Haji, Bahafta Gebresilassie, Beyene Senedu, Daniel Asele, Demiso Daba, Israel Gebresilasie, Mesele Markos and Nebiyu Waliyi. The launch of awtiCode is another step in promoting Python dataprocessing in my institute. I have given various trainings, such as a Basic Python training (see here and here), and a Training of Trainers. True programming skills, however, come from doing. The assumption is that awtiCode will result in coding experience for the core members.

Visitor offered Python training

My brother in law, Nimrod de Wit, came to Arba Minch to live and work with us for some months. He has quite some experience with Python. Based on this experience, our visitor offered a training on Basic Python for Environmental Data Processing to my colleagues of the Arba Minch Water Technology Institute. 31 persons attended one or more training sessions. 17 persons received certificates for attending all sessions and submitting an assignment in which they used Python for their own data.

Training sessions started with a small lecture
We used most time for practice in Jupyter Notebooks
Coffee and tea breaks were fundamental

I have offered a Python training to AWTI staff already multiple times (march 2021, june 2021, october 2021 and october 2022). Nimrod fully revised those materials. My trainings leaned heavily on Powerpoint slides and simultaneous small practice tasks in the Spyder IDE. Nimrod shifted to small (30 minute) introductions, after which participants could work by themselves through Juypter Notebooks. The training materials are on my GitHub repository.

I was glad to be the assistant rather than the trainer during a training for once. I am very glad for the revised training materials, which force the participants to practice much more. A training might help people to get started, but skills in something like programming are ultimately from a lot of hours of simply trying and using it.

Training participants could only receive a certificate after completion of a personal assignment.

Python Training of Trainers

I offered a Training of Trainers on Basic Python for Environmental Data Processing. Participants were selected from the four faculties of the Arba Minch Water Technology Institute (AWTI), as well as the Water Resources Research Centre (WRRC). The intention was to create in each of these faculties a Python team of 2-3 staff members that can provide courses and support within their respective faculty. Eleven participants received their certificate during a closing ceremony. During that ceremony, each participant presented their plan for what they would do with Python for AWTI. Scientific directors of both AWTI and the Arba Minch Institute of Technology (AMIT) were present as well.

Closing ceremony of the Python Training of Trainers. The empty tables were later on filled with delicious goat meat and injerra.

I have offered a Python training for AWTI staff already three times (in March 2021, June 2021 and October 2021). Approximately 100 colleagues have come to one or more training sessions. The number of staff really taking up Python, however, is low. Our newest idea was therefore to offer a focused Training of Trainers (TOT). Participants were asked to provide a motivation at the start of the training, and write a plan on what they will do with Python for AWTI after the training. Also, the training materials were extended a bit, and spread out over more sessions. The earlier Python trainings took five days – or ten mornings. This TOT took ten full days. The increased time was filled with a bit of new material, but especially more time for practice.

I have shared the training materials on an OSF repository.

Article on evaluation of low-cost sensors

Across various locations I have used the IQAir Airvisual Pro (IQAV), the UCB-PATS+ (PATS), and the Sensirion SPS30 as part of a locally assembled sensor system (SPSA). I have analyzed the data of different locations where those low-cost sensors (LCS) were collocated. At some of those locations, I also conducted gravimetry measurements with the UPAS. Based on this, I conducted an evaluation of the low-cost sensors PM2.5 data quality in three ways:

  • Within identical sensors (how do identical sensors compare to each other);
  • With each other (how do the measurements of one LCS compare to another LCS);
  • With the reference method (gravimetry is the golden standard for PM2.5 measurements).

Results of this comparison now have been published: “Evaluation of Three Low-Cost Particulate Matter (PM2.5) Sensors for Ambient and High Exposure Conditions in Arba Minch, Ethiopia“. All data and code for data processing and visualization is hosted on an OSF repository.

One of the collocation measurements was in a kitchen on main campus.

Data quality

My primary interest is the data quality of the SPSA. The IQAV and PATS cost about 300-500 euros. We construct the SPSA locally, and it costs about 60 euros. I found out that both under ambient and indoor (high and highly variable concentrations) circumstances, the SPSA data quality is equal to if not better than that of the IQAV and the PATS. More importantly, based on the data collected so far, the SPSA data quality compares well to international data quality requirements. The coefficient of variation (a measure of variation between identical sensors) ranged between 3 and 7% across low and high concentrations. This is lower than the required 10%. The accuracy versus gravimetric samples (a measure of how much it is ‘off’) was 16% under ambient and 15% under indoor circumstances. This is lower than the required 25%.

I will continue collecting data with the SPSA. For example, the gravimetry comparison under ambient conditions was only based on three samples. However, so far it is very promising that with a locally assembled sensor system we can surpass data quality of commercially available instruments and reach international standards.

Concentration levels

The analysis included data from several measurements, across low (<10 μg/m3) to high (>10,000 μg/m3) concentrations. Below, concentrations of all LCS across various locations are shown (Figures A1 and A2 of the article).

FigureA1
FigureA2
previous arrow
next arrow

Students find priority areas: publication

The student course on air quality in 2019 resulted in concentration data of various places on the campus of Arba Minch University. Together with three top students of this cohort (Muse Abayneh, Kirubel Getachew and Feyera Fekadu) I wrote an article concerning priority areas that could be distinguished from the student measurements. This article has now been published in the Clean Air Journal: Using student science to identify research priority areas for air pollution in a university environment: an Ethiopian case study.

Student measurements at various locations, relative to the respective guideline value. At multiple locations measurements are above the guideline value (>100%).

Python training for Biology

On request of Dr. Ashenafi Hailu, I constructed and offered a training in Basic Python for Biology Data Processing. 12 PhD students participated in the training. Daniel Asele assisted me during the training. The training took five full days, all within one week.

For this training, I followed the general outline of the training I gave earlier for environmental data: module 1 and 2 as the absolute basics (variables, conditions, loops, etcetera), and modules 3 and 4 for working with data (Pandas) and visualizing data (Matplotlib). I based data examples on the field of biology. The fifth module went into some applications more specific to biology: some statistics, (sequence) string manipulation, and working with the package BioPython. I have shared all training materials on an OSF repository.

This training was a (probably one-time-only) move out of my normal field. It cost me quite some time to get acquainted with terminology and techniques for biology. This spent time directly translated itself into more experience. However, in the future I prefer to stick to tools and knowledge used in my own field (air quality). Or, that of my direct colleagues (Arba Minch Water Technology Institute; environmental sciences).

Apart from that, this training was an experiment with offering it within one week (five days in a row). For a next training, I will shift back to one or two sessions per week. Five days in a row made it impossible for participants to practice with module topics before moving on to a next module.