See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/282040360 Application of the Nordtest method for "real-time" uncertainty estimation of on-line ﬁeld measurement Article in Environmental Monitoring and Assessment · September 2015 DOI: 10.1007/s10661-015-4856-0 CITATIONS READS 0 223 6 authors, including: Teemu Näykki Bertil Magnusson Finnish Environment Institute Research Institutes of Sweden & Trollboken AB 15 PUBLICATIONS 207 CITATIONS 60 PUBLICATIONS 1,618 CITATIONS SEE PROFILE Tero Väisänen Finnish Environment Institute 19 PUBLICATIONS 387 CITATIONS SEE PROFILE Some of the authors of this publication are also working on these related projects: Quality Control View project P2-Measurement uncertainty in medical laboratory View project All content following this page was uploaded by Bertil Magnusson on 25 September 2015. The user has requested enhancement of the downloaded file. SEE PROFILE Environ Monit Assess (2015) 187:630 DOI 10.1007/s10661-015-4856-0 Application of the Nordtest method for Breal-time^ uncertainty estimation of on-line field measurement Teemu Näykki & Atte Virtanen & Lari Kaukonen & Bertil Magnusson & Tero Väisänen & Ivo Leito Received: 26 January 2015 / Accepted: 9 September 2015 # Springer International Publishing Switzerland 2015 Abstract Field sensor measurements are becoming more common for environmental monitoring. Solutions for enhancing reliability, i.e. knowledge of the measurement uncertainty of field measurements, are urgently needed. Real-time estimations of measurement uncertainty for field measurement have not previously been published, and in this paper, a novel approach to the automated turbidity measuring system with an application for Breal-time^ uncertainty estimation is outlined based on the Nordtest handbook’s Electronic supplementary material The online version of this article (doi:10.1007/s10661-015-4856-0) contains supplementary material, which is available to authorized users. T. Näykki (*) : L. Kaukonen Environmental Measurement and Testing Laboratory, Finnish Environment Institute, Hakuninmaantie 6, 00430 Helsinki, Finland e-mail: [email protected] A. Virtanen Data and Information Centre, Finnish Environment Institute, Mechelininkatu 34a, 00251 Helsinki, Finland B. Magnusson Chemistry, Materials and Surfaces, SP Technical Research Institute of Sweden, 50115 Borås, Sweden measurement uncertainty estimation principles. The term real-time is written in quotation marks, since the calculation of the uncertainty is carried out using a set of past measurement results. There are two main requirements for the estimation of real-time measurement uncertainty of online field measurement described in this paper: (1) setting up an automated measuring system that can be (preferably remotely) controlled which measures the samples (water to be investigated as well as synthetic control samples) the way the user has programmed it and stores the results in a database, (2) setting up automated data processing (software) where the measurement uncertainty is calculated from the data produced by the automated measuring system. When control samples with a known value or concentration are measured regularly, any instrumental drift can be detected. An additional benefit is that small drift can be taken into account (in real-time) as a bias value in the measurement uncertainty calculation, and if the drift is high, the measurement results of the control samples can be used for real-time recalibration of the measuring device. The procedure described in this paper is not restricted to turbidity measurements, but it will enable measurement uncertainty estimation for any kind of automated measuring system that performs sequential measurements of routine samples and control samples/reference materials in a similar way as described in this paper. T. Väisänen Environmental Measurement and Testing Laboratory, Finnish Environment Institute, Paavo Havaksen tie 3, 90570 Oulu, Finland I. Leito Institute of Chemistry, University of Tartu, Ravila 14a, 50411 Tartu, Estonia Keywords Measurement uncertainty . Field measurement . Nordtest approach . Water quality monitoring . Quality control . Turbidity 630 Page 2 of 12 Introduction The state of the water environment is usually monitored by carrying out the measurements of physical or chemical parameters of water. Information of measurement uncertainty is essential when evaluating the real value of the measurement result. Nowadays, several chemical laboratories estimate uncertainty for their measurement results according to the guides and standards available (Eurachem/CITAC 2012; Eurolab 2007; International Organization for Standardization 2012; Joint Committee for Guides in Metrology 2008; Magnusson et al. 2011). At the same time, there is a global trend in which the measurements are carried out more often in situ by field measurement devices or sensors (Arola 2012; Azzaro 2013; Bourgeois et al. 2001; Clinch and Worsfold 1987; Glasgow et al. 2004; Tschmelak et al. 2005; Valkama et al. 2007; Wang et al. 2004). These may function automatically or they are operated manually. The need to employ field measurements is apparent, since pressures to streamline the production of environmental information have increased, mainly for economic reasons. Field measurement devices are able to produce time-resolved measurement results much more efficiently than laboratory equipment and personnel. Other benefits for applying field measurement devices are the lower price compared to laboratory instruments and the fact that sample stability during transportation from field to the laboratory does not have to be considered. However, some deficiencies remain in applying sensor measurements that need to be taken into account. It may be difficult to monitor the appropriate quality of measurement results due to uncontrolled sensor degradation e.g. by applying similar quality control procedures than in chemical laboratories (Hovind et al. 2011). In addition, many of the field sensor operators do not have any kind of knowledge about the measurement uncertainty of their sensor measurement results (Leivuori et al. 2013; Näykki et al. 2013). The uncertainty can be estimated for the sensor in the field using quality control and validation data using the Nordtest handbook (Magnusson et al. 2011), for example. However, what is needed is an estimate of the uncertainty of the online results produced by the field measurement, taking into account any degradation of any performance of the sensor. According to the Nordtest measurement uncertainty handbook (Magnusson et al. 2011), the combined Environ Monit Assess (2015) 187:630 measurement uncertainty is broken down into two main components—the within-laboratory reproducibility uRw (also called intermediate precision) and the uncertainty due to possible method and laboratory bias ub. Both of these can be conveniently estimated from internal quality control data, i.e. results of routine sample replicates and reference materials (Hovind et al. 2011). SYKE has published a practical, user-friendly, opensource, freeware measurement uncertainty estimation software package for laboratories (MUkit website; Näykki et al. 2012), which is based on the Nordtest handbook (Magnusson et al. 2011). Presently, similar routine and daily basis quality control procedures as applied in laboratories (Hovind et al. 2011) are not used for online measurements, and the measurement results are reported without any information about measurement uncertainty. In this paper, we describe a procedure where quality assurance data is applied for real-time measurement uncertainty estimation to an online field measurement. This will promote the comparability of the measurement results since the actual uncertainty at that time is reported together with the results. A case study was selected for a parameter where short-time inhomogeneity is present—turbidity measurement. It is of course easier to monitor a parameter which is homogeneously distributed in the water. In water with high turbidity, e.g. high concentrations of clay, also the repeatability has to be taken into account in the uncertainty estimate. Turbidity is an essential parameter for describing the quality of water. Several thousands of turbidity measurements are carried out annually in Finnish lake, pond, river and brackish waters. Water turbidity is caused by particles and suspended solids causing changes in water quality, reduced light penetration, diminished recreational values and aesthetics, as well as direct and indirect impacts on fish, invertebrates and aquatic plants (Kerr 1995; Appleby and Scarratt 1989; European Inland and Fisheries Advisory Commission 1964). A large number of the testing laboratories worldwide follow the international standard method ISO 7027 in their turbidity measurements (International Organization for Standardization 1999). Turbidity is measured nephelometrically using an instrument calibrated with formazine standard solutions. The turbidity of the tested water sample is expressed in formazine nephelometric units (FNU). The measurement of turbidity is based on the scattering of light. The procedure described in this paper is not restricted to turbidity measurements; instead, it will enable Environ Monit Assess (2015) 187:630 measurement uncertainty estimation for any kind of automated measuring system that performs sequential measurements of routine samples and control samples/ reference materials similarly as described in this paper. Automation of measurement uncertainty estimation in field measurements There are two main requirements for the estimation of real-time measurement uncertainty of an online field measurement: (1) setting up an automated measuring system that can be (preferably remotely) controlled, which measures the samples (water to be investigated as well as synthetic control samples) in the way the user has programmed it and stores the results in a database, and (2) setting up automated data processing (software), where the measurement uncertainty is calculated from the data produced by the automated measuring system. Figure 1 describes the general procedure for measurement uncertainty estimation along with how the calculated uncertainty value is combined with the measurement result before transferring these into water quality register. In this study, an automated turbidity measuring system was constructed, and it functioned by taking water samples via tubing into the instrument’s flow-through cell. Using valves, it was possible to automatically introduce natural water samples, as well as turbidity control samples to the turbidity detector. A challenge for the real-time uncertainty estimation is the autonomous analysis of reference samples/control solutions in the field. This can be done with automated spectrophotometers or by the turbidity measuring system described below, for example. For probes, the automation is not straightforward as they are usually deployed in situ and they do not have a flow-through cell. However, even for the probes, constructions there can be found, where the sensor is placed in a flow cell type of container, and the water sample or synthetic control sample is introduced via tubing to the cell for the measurement. Set up for automated turbidity measurement In order to apply the Nordtest method for Breal-time^ estimation of measurement uncertainty, the instruments need to analyse both (1) routine samples and (2) reference materials/control sample solutions automatically in the field. A cabin next to a river was employed for Page 3 of 12 630 setting up the automated measuring system. The distance from the river to the cabin was ca. 8 m, and the pumping height was ca. 5 m. The pump P1 (Biltema, 259755 water pressure pumps) controlled by the valve V1 (Danfoss, AVSI valve frames with Danfoss 018 F7397 magnetic coils DC 24 V) delivered water as scheduled by the operator (i.e. time interval of the sampling and the number of replicates measured) from the river to the ISO 7027 (International Organization for Standardization 1999) compliant flow-through sensor model 7998 016 (ABB, Switzerland) and the wall-mount mounted analyser model 4690 500 (ABB). In addition to the above-mentioned components, the measuring system in the cabin consists of pumps (P2-P3) and valves (V2-V5), tubings, containers for the control samples and wastes, an ABB automation module, a Raspberry Pi® computer with wireless communication, a Codesys® web-virtualisation interface for controlling the ABB automation module and a PhpMySQLAdmin® webinterface for controlling measurement data (Fig. 2). The technical details of the automated turbidity measuring system will be published later. The possible biofouling of the tubing used in the pumping system for delivering samples to the measuring system is one possible source of contamination and will depend on the sample types measured as well as the seasonal variations such as water temperature. The representativeness of the sample water should be regularly (e.g. once in a month) monitored with manual sampling associated with laboratory analysis. In addition, possible contamination or alteration of the control sample during the storage in the cabin may be a factor to be taken into account. Relevant stability studies should be conducted to ensure that fresh control samples are available. Instability of the control sample is included as an uncertainty component when estimating the uncertainty of the reference value for the control sample. Principle for real-time measurement uncertainty estimation A software application (AutoMUkit) was programmed for carrying out the real-time measurement uncertainty estimation. The software will be published later as freeware including the open-source program code, and the details for downloading it can be found on the Envical SYKE website after publication (MUkit website). In this paper, we describe the most important features of this 630 Page 4 of 12 Environ Monit Assess (2015) 187:630 Fig. 1 The general procedure for measurement uncertainty estimation and combination of the uncertainty value from the measurement result software. The application is programmed using Java® and consists of two main parts: (1) a measurement uncertainty solving class library and (2) a console application using the measurement uncertainty solving Fig. 2 Automated turbidity measurement system, which sequentially measures river samples and two control samples (CTRL A and CTRL B) and reports the data to a database class library, which saves uncertainties to Extensible Markup Language (XML) files or to a Java Database Connectivity (JDBC) compatible SQL database (Structured Query Language). Environ Monit Assess (2015) 187:630 Ladder of errors—components contributing to uncertainty Similarly to that described in (Hovind et al. 2011), the sources of errors affecting the possible deviation from a reference value for an analytical result may be described by the ladders presented in Fig. 3. Step 1 The method bias—a systematic effect owing to the method used Step 2 The instrument bias—a systematic effect (for an individual instrument or sensor) Step 3 The day-to-day variation—a combination of random and systematic effects owing to time effects, for example Step 4 The repeatability—a random effect occurring between replicate determinations performed within a short period of time; the sample inhomogeneity is part of the repeatability Each of these steps on the ladder adds its own uncertainty. A new term Binstrument reproducibility^ is introduced in this paper for field measurements to replace the term Bwithin-laboratory reproducibility^, which is applied for laboratory measurements. Instrument reproducibility is largely similar to within-laboratory reproducibility, covering the variation resulting from shortterm repeatability, and also day-to-day random Fig. 3 The ladder of errors for a measurement procedure used in field measurements Page 5 of 12 630 variation. Instrument reproducibility can be assumed to be lower than within-laboratory reproducibility, since the latter also includes the variation from analysts, which may change in a laboratory between different days. General requirements For the measurement uncertainty estimation, all components contributing to uncertainty need to be taken into account. An outline according to Nordtest TR 537 (Magnusson et al. 2011), where these components are divided into random and systematic errors, is described also in detail in ISO 11352 (International Organization for Standardization 2012) (Fig. 4). Regarding the turbidity case study, these are obtained through the analysis of replicate samples of the water to be investigated in addition to synthetic turbidity control samples. 1. Measurement results of routine samples Replicate measurements of the routine samples (e.g. river water) are carried out, and their measurement results reflect the random variation (repeatability) resulting from the sample inhomogeneity, for example, due to particles in the routine samples or the restrictions of the device’s measurement capabilities (ur,range in Fig. 4). 2. Measurement results of control samples The control sample solution is a synthetic solution with known concentration of the analyte. In this study, a formazine turbidity standard (HACH, Loveland, CO, USA) with a certified value of 4000 Formazine Nephelometric Unit (FNU) was used for the preparation of diluted control samples. Typically, it is advisable to use more than one control sample having different analyte values or concentrations. This is mandatory since the measurement uncertainty should also be estimated both for low concentrations near the limit of quantification and for high concentrations at the optimal range of detection. The measurement results of diluted control samples are used for calculating the uncertainty component reflecting the instrument reproducibility (termed as within-laboratory reproducibility in uRw,stand in Fig. 4, (Hovind et al. 2011; International Organization for Standardization 2012; Magnusson et al. 2011) and bias values. 630 Page 6 of 12 Environ Monit Assess (2015) 187:630 Fig. 4 Flow chart describing the estimation of measurement uncertainty using quality control and validation data. Scheme is reproduced from Fig. 1 in ISO 11352 (International Organization for Standardization 2012) courtesy of ISO, Geneva In this way, part of the repeatability component will be included twice, but usually, it is small in comparison to the between-days variation (Magnusson et al. 2011). Uncertainty of the certified value (ucref in Fig. 4) is obtained from the certificate of the solution, taking into account the uncertainty of the dilution steps during the preparation of the control samples. The automated measurement system is required to produce a CSV (comma-separated values) file in the following form: … A,2014-05-14 15:00:00,3.523 A,2014-05-14 15:00:05,3.545 A,2014-05-14 15:00:10,3.530 D,2014-05-14 15:00:15,0.000 A,2014-05-14 16:00:00,3.427 A,2014-05-14 16:00:05,3.455 A,2014-05-14 16:00:10,3.441 D,2014-05-14 16:00:15,0.000 B,2014-05-14 16:30:00,3.987 B,2014-05-14 16:30:05,3.976 B,2014-05-14 16:30:10,3.969 … where the comma separated columns are 1. Type of sample A = Routine sample B = Control sample for low concentration range C = Control sample for high concentration range (type of sample not shown in the example data) D = A marker sequence showing that a replicate series of routine sample measurements has finished Environ Monit Assess (2015) 187:630 2. Date in the form YYYY-MM-dd HH/mm/ss (Y= year, M=month, d=day, H=hour, m=minute, s= second) 3. Measurement result with a dot as the decimal separator. A row defines a measurement result. The rows are separated with a new line. The order of the columns and the format of the date string can also be modified from the settings file of the application. The types for the control samples (B and C in the CSV presented above) are used as identifiers for the control samples. There can be one or more control samples. Alternatively, an SQL database having JDBC drivers can also be used. SQLite™ and MySQL™ databases have been tested to work. The SQL database structure is presented in Fig. 5. Page 7 of 12 630 6. Timespan for the selection of control samples results (used as BCRM^ according to Nordtest (Magnusson et al. 2011)) used for the uncertainty calculation (bias calculation) 7. Timespan for the selection of control samples results (used as BControl sample^ according to Nordtest (Magnusson et al. 2011)) used for the uncertainty calculation (uRw, stand calculation) 8. Separate parameters (six different) for the minimum number and minimum timespan of (a) Routine (b) Control (c) Breference^ sample results required for the uncertainty calculation to be successful. 9. A limiting date (such as calibration date), before which routine, control and Breference^ sample values are not selected Parameters The AutoMUkit software application requires the operator to set parameters for the measurement uncertainty estimation. The operator needs to determine the following parameters: 1. Number of ranges used for measurement uncertainty estimation (there must be exactly one control/ reference sample for each value/concentration range) 2. Minimum and maximum values/concentrations for the ranges In the sample input file (presented in BGeneral requirements^), there are two types of control sample (BB^ and BC^). Therefore, our pilot application needs to have two value/concentration ranges. For each value/concentration range, the following parameters need to be set: 1. The period of time for which one measurement uncertainty value is valid for 2. Control sample reference value/concentration 3. Uncertainty of the reference value/concentration 4. Is absolute or relative uncertainty estimation required 5. Timespan for the selection of routine sample replicate measurement results used for the uncertainty calculation The software application has two modes: Bforced^ and Bautomatic^. Using the forced mode, the application evaluates measurement uncertainties within a specified period of time. Using the automatic mode, the application calculates measurement uncertainties starting from the end date of the last successful measurement uncertainty calculation for each value/concentration range. For the automatic mode, an additional parameter needs to be set. This defines the earliest date in the past, which is included in the evaluation in case no successful evaluations are found. Since the results of the past measurements are used for calculating uncertainty, the choices that the operator make in the settings will affect the measurement uncertainty estimation. If the time interval for collecting control sample measurement results is too long, the results may not fully reflect the current state of the measurement bias. The replicate measurement of the routine samples has to be carried out within a short time interval. If the time interval of the replicate measurements is too short, the measurement result does not reflect any possible random variation and the estimated uRw may not be representative enough. If the time interval is too long, the measurement results no longer represent the same test sample when the sample is changing very fast, for example in river water. Nordtest TR 537 (Magnusson et al. 2011) suggests having at least 50 results and a timespan of 1 year. This is true for laboratory measurements and quality control procedures 630 Environ Monit Assess (2015) 187:630 Page 8 of 12 Fig. 5 The SQL database structure from which the measurement results can be read for the measurement uncertainty calculation. The types of column values are specified by the Java types in brackets. ReplSeries is a database table for replicate series, with a column for the average time (repl_series_avg_time) of the replicates in the series. MeasResult is a table for the measurement values (column value) including columns for the time of the measurement (meas_time) and the average time of the replicate series (repl_series_avg_time) the measurement result belongs to. RefResult is a table for the reference/control sample measurement values having columns for the time of the measurement (meas_time), the measurement value (value) and an identifier for the concentration range the reference belongs to (conc_range_id). The identifiers for the data presented for the CSV would be BB^ and BC^ applied there. Field sensor measurements can be carried out 24 h per day, and the amount of data is much higher than in laboratory measurements. For real-time uncertainty estimation, the authors suggest that at least 30 measurement results should be collected over a time period of 10 days or more. In the river water experiments, successive measurements carried out in 5-s intervals were regarded as replicate measurements. Three replicates were recorded for one test sample. A new test sample was measured every 60 min. Control samples were measured every 12 h. Binstrument reproducibility^ uRw. Equations 1 and 2 are used for calculating the repeatability component. If measurement uncertainty is calculated as an absolute value, the repeatability component ur,range is X nr c −c i¼1 ðiÞmax ðiÞmin ð1Þ ur;range ¼ nr ⋅d Calculation of measurement uncertainty with AutoMUkit The queries of the application are performed, and the measurement uncertainty is calculated according to how the parameters were set. Figure 6 shows the raw measurement data consisting of one measurement result per row with information about the sample type, the measurement date and time. Three parallel turbidity measurements are carried out within a 5-s interval from the routine (river water) sample BA^. After 60 min, another three replicates are measured. Later in the sequence, measurement results can be seen for the synthetic control sample BB^ with a turbidity value in the low concentration range. As seen from Fig. 6, the turbidity standard solution is used both as a control sample and a reference material for the measurement. The results of routine sample replicates and control sample replicates are used for calculating the If measurement uncertainty is calculated as a relative value, the repeatability component ur,range% is 0 1 X nr B cðiÞmax −cðiÞmin C @100%⋅ c A i¼1 ðiÞmax þ cðiÞmin 2 ur;range% ¼ ð2Þ nr ⋅d wherecmax is the maximum and cmin the minimum concentration of a replicate series. nr is the number of replicate series. d denotes a factor for converting the mean difference to the standard deviation and depends on the amount of replicate series (International Organization for Standardization 2012). Equations 3 and 4 are used for calculating the instrument reproducibility component (u Rw,stand and uRw,stand%) from control sample replicates respectively: uRw;stand ¼ S Rw;stand C avg uRw;stand% ¼ 100%⋅ ð3Þ S Rw;stand C avg ð4Þ Environ Monit Assess (2015) 187:630 Page 9 of 12 630 Fig. 6 The general functioning of the real-time uncertainty estimation where sRw,stand is the standard deviation of control sample measurement results and cavg is the average of control sample measurement results. The uRw or uRw% is calculated as described in Eqs. 5 and 6: uRw ¼ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ur;range 2 þ uRw;stand2 ð5Þ of the formazine turbidity control sample measured every 12 h. sﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 sb ð7Þ ub ¼ b2 þ pﬃﬃﬃ þ u2cref n where b or uRw% qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ¼ ur;range% 2 þ uRw;stand%2 sb ð6Þ Equation 7 is used for calculating the uncertainty component for bias (ub) from the measurement results n the difference between the average of the sensor measurement results of the control sample (during a certain period of time) and the actual reference value of the control sample the standard deviation of the sensor measurement results of the control sample the number of sensor measurement results of the control sample uCref is the standard uncertainty of the reference value 630 Page 10 of 12 Saving the measurement uncertainty calculations The AutoMUkit software application creates evaluations, which determine the measurement uncertainty value for a concentration range and a period of time (the length of the period is defined in the settings). These evaluations are calculated for the whole timespan for which measurement uncertainty is calculated (explicitly defined if the forced mode is used and from the last successful measurement uncertainty evaluation until present time if the automatic mode is used). Currently, the application supports two methods for saving the measurement uncertainty evaluation results: XML files or SQL database. If measurement uncertainty evaluations are saved to XML files, the application saves a file for each evaluation, so that the file name contains the start and end date for the time the measurement uncertainty evaluation applies for. The file name also includes the value/ concentration range, which the evaluation belongs to and whether the evaluation has been successful or has failed. An example of the filename is B2 0 1 4 _ 0 4 _ 0 2 _ 0 0 _ 0 0 _ 0 0 _ Z + 0 3 _ 0 0 2014_04_02_12_00_00 Z+03_00-B-success.xml^. The actual XML file contains information about the evaluation, including the expanded measurement uncertainty. If the SQL database implementation is used, measurement uncertainty evaluation information is saved to a database with the structure presented in Fig. 7. In addition to Fig. 5, the database contains the following tables: – – – – MeasUncEval: a successful measurement uncertainty evaluation (a measurement uncertainty value for a concentration range and a period of time). MeasUncFail: a failed measurement uncertainty evaluation. The requirements for the input information set in the parameters (i.e. minimum period of time, minimum amount of results) are not fulfilled. MeasUncAppRun: information about every time the AutoMUkit application’s measurement uncertainty evaluation has run. One run has evaluations (MeasUncEval/MeasUncFail) for each value/ concentration range. ConcRange: settings for value/concentration ranges presented earlier (conc_range_id is the identifier for the Bcontrol sample^ and the concentration range). These can alternatively be loaded from a file. Environ Monit Assess (2015) 187:630 The measurement uncertainty can then be combined with the measurement results of the samples, by either querying the measurement uncertainty XML files or the SQL database. User interface for parameter adjusting The parameters described earlier can be adjusted and their influence on the measurement uncertainty can be assessed visually using the developed visualisation tool (see the Electronic Supplementary Material (ESM)). In the visualisation tool, the user can set the parameters for the measurement uncertainty calculation. After that, the operator can see how the changes to the parameters affect the calculated measurement uncertainty. The visualisation tool uses the forced mode of AutoMUkit. The examples in the graphs are presented as Electronic Supplementary Material. The graphs include the presentation of measurement results and their uncertainties (k = 2) shown using error bars. The X chart is available for visual monitoring of the measurement results of control samples. An r% chart is also available, where each dot is calculated from the difference of routine sample replicate results as a percentage of their mean value (r% value (Hovind et al. 2011). The graph for the r% chart is very useful for monitoring the random variation as a function of the measured value or concentration, since the measurement uncertainty will normally vary by concentration when the instrumental methods are used. In the lower concentration ranges, the absolute measurement uncertainty is usually constant, while at higher concentrations, the relative uncertainty is constant (Eurachem/CITAC 2012; Magnusson et al. 2011; Magnusson and Koch 2012). Therefore, it is advisable to divide the measurable concentration range into parts. Typically, the decision for the limit concentration between the low and high ranges is based on the visual study of the behaviour of relative random variation (r% values). The operator of the sensor may set the parameters (concentration ranges) for the measurement uncertainty calculation according to the observation, where the random variation starts to increase or remains constant, respectively. r% values may also be viewed as a function of time, so the operator can observe possible changes in the random variation due to seasonal effects. Environ Monit Assess (2015) 187:630 Page 11 of 12 630 Fig. 7 Database structure of measurement uncertainty evaluation information Discussion and conclusions Field sensor measurements are becoming more common in environmental monitoring. Solutions for enhancing reliability, i.e. knowledge of the measurement uncertainty of field measurements, are urgently needed. Real-time estimations of measurement uncertainty for field measurement have not been published before, and in this paper, a novel approach for the automated turbidity measuring system with a computer software application for real-time uncertainty estimation based on the Nordtest handbook’s measurement uncertainty estimation principles was outlined. The software for measurement uncertainty estimation will be published later as freeware with source code included. The software will enable measurement uncertainty estimations for any kind of automated measuring systems that performs sequential measurements of routine samples and control samples/reference materials, as described in this paper. This automated turbidity measuring system has been developed as a pilot system for continuous uncertainty measurement system. Establishing such an automated measuring system may require a cabin and also heating during the winter time. The need for electricity is also required. These requirements may limit the deployment of the technology, and if the appropriate cabin is not readily available, establishing the infrastructure may not be cost-effective. The benefits of advanced utilisation of field measurements are enormous. The measurement results are produced at a very high frequency, and the use of the equipment does not require manual operation, except when maintenance is needed. The alteration of the sample during transportation to the laboratory and storage is not relevant in the field measurement. Sensor measurement results may occasionally drift for many reasons e.g. due to biofouling, but this can be monitored using the approach presented in this study. When control samples with a known value or concentration are measured regularly, information from changes in calibration can also be detected. An additional benefit is that any drift that occurs can be taken into account (in real-time) as a bias value in the measurement uncertainty calculation, or the measurement results of the control samples can be used for real-time recalibration of the measuring device. For the current version of AutoMUkit, the operator of the instrument and the software has to define the parameters, i.e. the frequency for the recalculation of the measurement uncertainty. As a future 630 Page 12 of 12 development area, the proportional-integral-derivative (PID) controller widely used in adjusting industrial processes could be used for controlling when the sensor degrading has reached a set limit and when the measurement uncertainty needs to be recalculated. The measurement results will then become much more useful when they are associated with the uncertainty information. They can be compared to the measurement results produced in a laboratory, and they can be used more easily for decisionmaking processes. Acknowledgments The Cleen Ltd MMEA programme and Maa-ja vesitekniikan tuki ry are acknowledged for their financial support. This work has also been partially supported by Graduate School BFunctional materials and technologies^ receiving funding from the European Social Fund under project 1.2.0401.09-0079 at the University of Tartu, Estonia. References Appleby, J. P., & Scarratt, D. J. (1989). Physical effects of suspended solids on marine and estuarine fish and shellfish with special reference to ocean dumping: a literature review. Canadian Technical Report of Fisheries and Aquatic Sciences No. 1681. Arola, H. (Ed.). (2012). Jatkuvatoiminen sameusmittaus—Hyvät mittauskäytännöt ja aineistonkäsittely (Continuous turbidity measurement—best measurement practices and data processing), Environmental Administration Guidelines 2/2012 (In Finnish). Helsinki: Finnish Environment Institute. Azzaro, F. (2013). Automated nutrients analysis for buoys in seawater and intercalibration. International Journal of Environmental Monitor Analysis, 1(6), 315–322. Bourgeois, W., Burgess, J. E., & Stuetz, R. M. (2001). On-line monitoring of wastewater quality: a review. Journal of Chemical Technology and Biotechnology, 76(4), 337–348. Clinch, J. R., & Worsfold, P. J. (1987). An automated spectrophotometric field monitor for water quality parameters: determination of nitrate. Analitica Chimica Acta, 200, 523–531. Eurachem/CITAC Guide CG 4 (2012). Quantifying Uncertainty in Analytical Measurement, 3rd edition. Available via www. eurachem.org. Eurolab (2007). Measurement uncertainty revisited: alternative approaches to uncertainty evaluation. Available via: www. eurolab.org/documents/Technical_Report_Measurement_ Uncertainty_2007.pdf. European Inland and Fisheries Advisory Commission. (1964). Water quality criteria for European freshwater fish: report on finely divided solids and inland fisheries. Journal of Air and Water Pollution, 9(3), 151–168. Glasgow, H. B., Burkholder, J. M., Reed, R. E., Lewitus, A. J., & Kleinman, J. E. (2004). Real-time remote monitoring of water quality: a review of current applications, and advancements in sensor, telemetry, and computing technologies. Journal of Experimental Marine Biology and Ecology, 300(1–2), 409–448. View publication stats Environ Monit Assess (2015) 187:630 Hovind, H., Magnusson, B., Krysell, M., Lund, & U., Mäkinen, I. (2011). Nordtest Technical Report 569 - Internal Quality Control - Handbook for Chemical laboratories, 4th edition, Nordic Innovation, Oslo, Norway. Available via www.nordtest.info. International Organization for Standardization (1999). ISO 7027, Water quality — Determination of turbidity. International Organization for Standardization (2012). ISO 11352, Water quality — Estimation of measurement uncertainty based on validation and quality control data. Joint Committee for Guides in Metrology (2008). Evaluation of measurement data - Guide to the expression of uncertainty in measurement (GUM). JCGM 100:2008. Available via www. bipm.org. Kerr, S. J. (1995). Silt, turbidity and suspended sediments in the aquatic environment: an annotated bibliography and literature review. Ontario: Ontario Ministry of Natural Resources, Southern Region Science & Technology Transfer Unit Technical Report TR-008. 277 pp. Leivuori, M., Björklöf, K., Näykki, T., & Väisänen, R. (2013). Laboratorioiden välinen pätevyyskoe 5/2013, Kenttämittaukset –vesien happi, lämpötila, pH ja sähkönjohtavuus (In Finnish), Finnish Environment Institute, Helsinki, Finland. Available via http://www.syke. fi/download/noname/%7BFD00FB6D-21B3-4D8F-8C35424926B9E626%7D/92086. Magnusson, B., & Koch, M. (2012). Use of characteristic functions derived from proficiency testing data to evaluate measurement uncertainties. Accreditation and Quality Assurance, 17, 399–403. Magnusson, B., Näykki, T., Hovind, H., & Krysell, M. (2011). Nordtest Technical Report 537 - Handbook for the calculation of measurement uncertainty in environmental laboratories, 3rd edition, Nordic Innovation, Oslo, Norway. Available via www.nordtest.info. MUkit website: http://www.syke.fi/en-US/Services/Calibration_ services_and_contract_laboratory/MUkit__Measurement_ Uncertainty_Kit. Näykki, T., Virtanen, A., & Leito, I. (2012). Software support for the Nordtest method of measurement uncertainty evaluation. Accreditation and Quality Assurance, 17, 603–612. Näykki, T., Jalukse, L., Helm, I., & Leito, I. (2013). Dissolved oxygen concentration interlaboratory comparison: what can we learn? Water, 5(2), 420–442. Tschmelak, J., Proll, G., Riedt, J., Kaiser, J., Kraemmer, P., Bárzaga, L., Wilkinson, J. S., Hua, P., Hole, J. P., Nudd, R., Jackson, M., Abuknesha, R., Barceló, D., Rodriguez-Mozaz, S., López de Alda, M. J., Sacher, F., Stien, J., Slobodník, J., Oswald, P., Kozmenko, H., Korenková, E., Tóthová, L., Krascsenits, Z., & Gauglitz, G. (2005). Automated water analyser computer supported system (AWACSS) Part II: intelligent, remote-controlled, cost-effective, on-line, water-monitoring measurement system. Biosensors and Bioelectronics, 20, 1509–1519. Valkama, P., Lahti, K., & Särkelä, A. (2007). Automaattinen veden laadun seuranta Lepsämänjoella. (Automated water quality monitoring in the River Lepsämänjoki), Terra, 119(3–4), 195–206. Department of Geosciences and geography, University of Helsinki, Finland. Wang, Q., Li, Y., Obreza, T., & Munoz-Carpena, R. (2004). Monitoring stations for surface water quality. University of Florida, IFAS Extension, Fact Sheet SL 218. Available via http://ufdc.ufl.edu/IR00003153/00001.