the Testing RICT predictions of expected values using an
Transcription
the Testing RICT predictions of expected values using an
Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data A Report to the Environment Agency J. Davy-Bowker R.T. Clarke November 2015 Research Contractor This document was produced by the Freshwater Biological Association: † † John Davy-Bowker and Ralph T. Clarke † The Freshwater Biological Association, River Laboratory, East Stoke, Wareham, Dorset, BH20 6BB, United Kingdom. Project Funders This project was funded by the Environment Agency. Disclaimer Whilst this document is considered to represent the best available scientific information and expert opinion available at the stage of completion of the report, it does not necessarily represent the final or policy positions of the project funders or contractors. Dissemination status Unrestricted Environment Agency Project Manager Environment Agency’s project manager for this contract was: John Murray-Bligh (EA) FBA Project Manager FBA’s project manager for this contract was: John Davy-Bowker FBA Project Code S/0025/R The Freshwater Biological Association The Freshwater Biological Association The Ferry Landing Far Sawrey, Ambleside Cumbria, LA22 0LP, United Kingdom The Freshwater Biological Association River Laboratory East Stoke, Wareham Dorset, BH20 6BB, United Kingdom Web site: www.fba.org.uk Email: [email protected] Registered Charity No. 214440 Company Limited by Guarantee No. 263162, England UKPRN No. 10018314 Registered Office: The Ferry Landing, Far Sawrey, Ambleside, Cumbria, LA22 0LP, United Kingdom EXECUTIVE SUMMARY Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Project funders: Environment Agency Background to research The environment agencies in the UK (the Environment Agency; Scottish Environment Protection Agency; Natural Resources Wales and the Northern Ireland Environment Agency) use the River Invertebrate Classification Tool (RICT) to classify the ecological quality of rivers for Water Framework Directive compliance monitoring. The current system uses RIVPACS observed (O) to expected (E) ratios (EQIs) of the two macroinvertebrate indices WHPT NTAXA and WHPT ASPT. Since the first launch of the RICT software a variety of research and development projects on RIVPACS have been undertaken to further develop and enhance the models. Whilst not all of these upgrades have been applied to the operational version of the RICT software, many have and this has necessitated a series of software upgrades by SEPA appointed programmers and SEPA in-house IT specialists. It is now a fitting time to test the implementation of the various upgrades that have been made by comparison to an independently coded version of the RIVPACS prediction model. It is also appropriate to do this using new test data with a wide geographical spread of test sites and range of environmental qualities. The derivation of new test data and the RICT testing exercise are reported here. Objectives of research To derive new RIVPACS/RICT test data for current (and future) RIVPACS/RICT testing purposes. To test the current RICT software to see if its predictions of single season (Spring) Expected values of the raw and reference quality-adjusted abundance-weighted WHPT indices for the test sites match those of an independently constructed version of the same RIVPACS IV prediction model and adjustment algorithms. Key findings and recommendations Our independent calculation of each of the algorithm steps involved in derving RIVPACS IV GB model predictions of the raw and adjusted Expected values of the WHPT indices suggest that these are correctly coded and calculated within the RICT software (i.e. RICT as of 13 March 2015). This is based on predictions of spring sample Expected values for 12 test sites. The RICT software correctly predicts the abundance-weighted WHPT indices based on Taxonomic Level 2 data, but using the Composite families version of WHPT rather than the Distinct families version. This should be made clear in the RICT software and manual, and may need be rectified within RICT in due course. The RICT code to calculate the values of the numerous derived environmental variables used in RICT predictions (notably Latitude, Longitude and mean and range of air Temperature) agrees with our independent RIVPACS calculations and checks The RICT code to calculate probability of end-group agrees with our independent tests. The RICT code using the probabilities of end-group and end-group biotic index means to calculate (unadjusted) Expected biotic index values agreed with our independent calculations (when checked using the two abundance-weighted WHPT indices for spring samples). The minor adjustments to WFD Reference state of the raw Expected values of WHPT NTAXA and WHPT ASPT for the 12 test sites appears to have been coded correctly in RICT, based on our independent code, but the RICT code could be tested further using artificial more-widely varying adjustment factors. Key words: River Invertebrate Prediction and Classification System, RIVPACS, River Invertebrate Classification Tool, RICT, Testing, Test Data. Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Table of Contents 1. BACKGROUND 2. DERIVING NEW RIVPACS/RICT TEST DATA 2.1 Geographical coverage throughout Great Britain 2.2 Range of Environmental Predictor Variables 2.3 Coverage of RIVPACS IV Biological End Groups 2.4 Reference and Degraded Samples 2.5 Environmental and Biological Test Data Files 3. RUNNING THE TEST DATA IN RICT 4. COMPARING RICT PREDICTIONS OUTPUT WITH INDEPENDENT CODE VALUES FOR THE TEST DATA 4.1 RIVPACS derived Environmental variables for the predictions 4.2 Discriminant functions coefficients 4.3 End-group means for WHPT biotic indices 4.4 Mahalanobis distances and Probabilities of end-group for test sites 4.5 (Unadjusted) Expected values of the (WHPT) biotic indices 4.6 Adjustment of Expeced values for the quality of reference sites involved 4.7 Summary of testing of RICT predictions of Expected values of WHPT indices 5. REFERENCES 6. APPENDIX 1: FULL LIST OF THE RIVPACS/RICT BIOLOGICAL TEST DATASET 1 2 4 4 4 4 8 8 11 12 12 13 14 14 14 15 16 17 18 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data 1. BACKGROUND Since the original research projects that led to the development of the current RIVPACS IV models in RICT and the subsequent launch of the first version of the RICT software (Clarke et al., 2006; Davy-Bowker et al., 2007a, 2007b; 2008), a variety of further research and development work has been carried out on the RIVPACS IV models and the RIVPACS bioassessment methodologies. These are summarised below: SNIFFER - project WFD100 (Davy-Bowker et al., 2010) Supplied new data files to support species-level taxonomic outputs from RICT. Allocated numerical abundance values to the RIVPACS database. Calculated a new range of species-level biotic indices in the RIVPACS database and supplied the files necessary for RICT to predict expected values. Produced a list of new predictive variables to offset the loss of predictive power associated with the future removal of variables affected by stress. SNIFFER - project WFD119 (Clarke et al., 2011) Derived new alternative predictive variables that are not affected by stressors with particular emphasis on hydrological/acidification metric predictors. Constructed several new RIVPACS models using stressor independent variables. Reviewed the performance of WFD reporting indices notably AWIC (species), LIFE (species), PSI and WHPT. Environment Agency - Deep Rivers (Jones et al., 2012; Davy-Bowker et al., 2014) Reviewed the results of previous deep-water methods comparison studies and made recommendations on the preferred deep water sampling method(s) and the threshold between methods for sampling wadeable and deep rivers. Examined the potential discontinuities in RIVPACS predictive models that might arise from the methods used to collect reference samples. Identified existing RIVPACS sites that have been inappropriately sampled (given their depth) and examined the distribution of deep water sites in the current model. Evaluated the suitability of the metrics EQR ASPT and NTAXA for deep rivers and examined the potential need for additional environmental variables for deep rivers. Produced clear guidelines for sampling deep rivers for inclusion in future Environment Agency sampling manuals. Undertook an ergonomic assessment of airlift sampling and provided a specification for a ‘standard’ airlift sampling device. SEPA - abundance weighted indices project (Clarke & Davy-Bowker, 2014) Develop algorithms and uncertainty parameter estimates for the incorporation of abundance-weighted WHPT, LIFE and PSI into RICT. Estimated sampling uncertainty components in the abundance-weighted WHPT, LIFE and PSI indices. Provided estimates of sampling variance for WHPT NTAXA, WHPT ASPT, LIFE and PSI, together with detailed algorithms for incorporation into confidence of class simulations. Analyzed a dataset of 427 samples to determine the biases (i.e. differences) between the observed (pre-audit) sample value and the audit-corrected sample value of each index. Provided algorithms to simulate the estimated sample processing biases in the abundance-weighted WHPT indices. Provided a detailed algorithms section to enable the RICT software programmers to encode these new methods and uncertainty parameter estimates for the abundanceweighted WHPT indices into the next version of RICT. 2 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Scottish Executive - abstraction and sediment project (Davy-Bowker et al., In prep) Developed algorithms and uncertainty parameter estimates for the incorporation of abundance-weighted classification indices WHPT, LIFE and PSI into RICT. Developed algorithms and parameter estimates for the incorporation of sample biases of abundance-weighted indices LIFE and PSI, into RICT. Produced the basic statistical procedures needed classify by LIFE and PSI in RICT. Derived an initial set of statistically based WFD class boundaries for LIFE and PSI taking into account of the range of pressures assessed by these metrics. In each of these upgrades a contract report was produced that contained a mixture of either new data, new code and/or new algorithms to permit the implementation of the work in RICT. Whilst not all of these upgrades have been applied to the operational version of the RICT software, this being dependent on the priorities of the UK Agencies, many have. These upgrades to RICT have been carried out by SEPA appointed programmers and SEPA inhouse IT specialists: It is now a fitting time to test the implementation of the various upgrades that have been made to RICT by comparison to an independent stand-alone version of RIVPACS. An independent version of RIVPACS can be constructed by virtue of the fact that all of the necessary data, code and algorithms have always been published (see above). Ralph Clarke (author) has extensive expertise in building RIVPACS models and wrote the software for RIVPACS III+, the version immediately preceding the RIVPACS IV models currently within RICT. Ralph also developed the the current RIVPACS IV statistical predictive models and algorithms that were supplied to the SEPA-employed programmers in their development of the original RICT software back in 2008. Given the importance of the testing process it was considered appropriate to base these tests on a new Great Britain-wide set of test data. Whilst various pieces of test data have existed both now and in the past (e.g. the 3-site test data supplied with earlier pre-RICT RIVPACS versions, and the current test data downloadable from the RICT website), these have limited coverage of stream types and limited geographical spread . The opportunity was therefore taken to devise a new set of test data with enhanced geographical spread, better coverage of the range of environmental predictor variables, better coverage of RIVPACS TWINSPAN biological end groups, and samples both in reference condition and degraded status. This research therefore had two objectives: To derive new RIVPACS/RICT test data for current (and future) RIVPACS/RICT testing purposes. To test the current (Spring 2015) RICT software to see if its predictions match those of an independently constructed version of the same RIVPACS IV model. 3 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data 2. DERIVING NEW RIVPACS/RICT TEST DATA A new set of test data was required with enhanced geographical spread, better coverage of the range of environmental predictor variables, better coverage of RIVPACS IV TWINSPAN biological end groups, and samples both in reference condition and degraded status. Sites were drawn from the RIVPACS reference site database (Davy-Bowker et al., 2007a), the dataset underpinning the development of all UK RIVPACS models. Candidate RIVPACS reference sites were manually assessed for suitability using the following overall considerations 1 2 3 Good geographical coverage (of Great Britain) Good spread of environmental predictor variables compared to the full range of those variables across the whole GB reference site dataset. Good coverage across the RIVPACS IV biological TWINSPAN end groups After examining various combinations of candidate sites, a group of twelve RIVPACS reference sites were identified (Table 1) that had good geographical coverage, had a good range of environmental variables, and were dispersed across all of the major TWINSPAN biological end groups. The three considerations of geographical coverage, environmental predictor variable range and coverage of biological end groups are discussed below. 2.1 Geographical coverage throughout Great Britain The geographical coverage of the twelve RIVPACS reference sites is shown in Figure 1. The sites are distributed across England (7 sites), Scotland (4 sites) and Wales (1 site) with the number of sites in each country in approximate proportion to their relative land areas. The sites cover a good north-south gradient (Shetland to the south west of England) and a good east-west gradient (Islay to Suffolk). The sites are also quite well dispersed in relation to each other, with only two sites (near Sheffield and Derby) in close proximity. 2.2 Range of Environmental Predictor Variables The range of the eight user-supplied environmental variables used in RIVPACS predictions for the twelve RIVPACS reference sites is shown in Figure 2. The open circles in Figure 2 represent the values of the environmental predictor variables for the twelve test sites. Underlying these are frequency distributions of the same variables across 685 RIVPACS reference sites. Figure 2 therefore shows the range of each variable that is likely to be encountered in British streams and rivers and the representativeness of the test data across those environmental gradients. For most of the variables the coverage is good, perhaps only weakening towards sites that are at high altitudes, have high water depth, or very high width. Overall the coverage was considered adequate. 2.3 Coverage of RIVPACS IV Biological End Groups The construction of a RIVPACS model involves two main steps. Firstly a classification step where the taxonomic data from all the reference samples is split successively into biological end groups (typically using TWINSPAN), and secondly a Multiple Discriminant Analysis step where equations are built that can discriminate the end groups from each other using the physical predictor variables. Given that reference samples have been sought from all major stream types that exist, the biological end groups therefore cover the majority of the range of biological communities that are likely to occur. Designing the new test data to similarly cover a wide range of different biological stream types was clearly desirable and assessing this by virtue of their RIVPACS IV biological end groups was the simplest way to do this. 4 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Table 1. The twelve RIVPACS reference sites identified for use in a new test dataset and their associated environmental variables. Site ID River 3101 Derwent 9581 Site Langdale End Distance Discharge from Grid Ref Easting Northing Latitude Longitude Altitude Slope Category Source Alkalinity Mean Mean Mean Width Depth Substratum SE942910 494200 491000 54.31 -0.55 60 2.7 2 10 101.4 6.4 21.7 -5.63776 SK220646 422000 364600 53.18 -1.67 142 5 4 6 194.6 6 10.7 -3.2125 8805 Lathkill Alport Coombevalley Stream Kilkhampton SS246116 224600 111600 50.88 -4.49 100 40 1 1.7 50 1.4 7.8 -5.65594 2007 Blithe SK048259 404800 325900 52.83 -1.93 97 1.8 3 27 164 10 8.7 -2.8 2307 Colne Newton Fordstreet Bridge TL921272 592100 227200 51.91 0.79 15 1.3 3 28 217.7 6.6 14.4 -3.37 7145 Ed SU074105 407400 110500 50.89 -1.89 38 5.5 1 1.8 178 1.8 14.2 2.99505 6111 SEPA_ N06 SEPA_ W05 Pains Moor Hilgay Bridge Ouse/Cam TL604970 560400 297000 Shetland: Burn of Laxdale North Voxter HU437290 443700 1129000 Islay: Duich/Torra Torra Bridge NR344552 134400 655200 52.55 0.37 0 0.2 7 69 237.3 40 200 7.4375 60.044 -1.2154 5 11.5 1 6.6 17.5 3.6 12 -6.53 55.7173 -6.22961 46 11.4 3 8.5 3 4.53 20 -6.26 3785 Green Burn Dalmary NS515955 251500 695500 56.13 -4.39 30 45.7 1 4 9.5 3 17.5 -6.1375 NE01 Lossie Cloddach NJ203584 320300 858400 57.61 -3.33 44 8.3 4 27 23 15.1 25 -6.4 WE03 Afon Caseg Braichmelyn SH630663 263000 366300 53.18 -4.05 160 66.7 2 6.4 9.8 12 28 -6.985 5 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Figure 1. Geographical distribution of the 12 RIVPACS test sites. 6 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Figure 2. Environmental predictor variables of the 12 test sites (open circles) shown on top of frequency distributions of the same variables across 685 RIVPACS reference sites. 7 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Table 2 shows the RIVPACS IV biological end groups of the twelve reference samples. There were 43 end groups in RIVPACS IV and the test samples have been evenly spread across groups 1, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40 and 43. 2.4 Reference and Degraded Samples It was considered important that the test data set contained samples not only in reference condition, but also examples of sites in degraded status. For example, it would not be possible to use the test data in the future to examine the behaviour of RICT band limits if the samples did not contain data that spanned a range of biological qualities. Rather than adding another set of completely different samples of variously degrees of environmental degradation to the twelve reference sites identified above, it was thought useful to replicate the 12 reference samples, but this time with artificially simulated biological quality as this would give greater control over the gradient of biological degradation in the test data. Creating this second version of the twelve reference samples with simulated degradation was achieved by making percentage reductions in the reference values of the biotic indices. The simulated samples are therefore not ‘real’ in the sense that the indices are not derived from actual taxonomic level data, and as such no actual taxonomic data could necessarily create the particular set of observed biotic indices’ values that were made. However, given that this test data is being developed to test the correct coding of the classification steps in RICT, this is not a problem and the percentage reduction in indices approach does give a better and more even range of environmental qualities to achieve this. Table 2 summarises the final 24-site RIVPACS/RICT test data. In the first column are new test data set samples codes. These have been renumbered from their original RIVPACS site codes into a new structure where for example, in the case of sample TST-01-R, ‘TST’ indicates that this is a sample belonging to the RIVPACS/RICT test dataset, ‘01’ is a sequential number between 01 and 12, and the suffix R means ‘reference quality’ (the alternative ‘D’ indicating degraded quality). Table 2 also shows the two groups of reference and degraded samples, which GB country they are within, their TWINSPAN end group, and their original RIVPACS Site Code, River Name and Sample Name for cross referencing back to the RIVPACS database. 2.5 Environmental and Biological Test Data Files The final step in developing the new test data was to build RICT data input files for both the environmental predictor variables and the biological indices WHPT NTAXA and WHPT ASPT so that these data could be used for testing RICT. Note: the test dataset is also available as RICT formatted Excel ® files and is downloadable from the FBA website www.fba.org.uk (on the same webpage as this report). RICT currently performs WFD status classifications based on the indices WHPT NTAXA and WHPT ASPT and typically on a combination of spring and autumn samples. These were the combinations that this contract specification therefore requested for testing and comparison with an independently derived RIVPACS model. The environmental and biological data for these commonly used indices and seasons are given in Table 3 and Table 4. Given that RICT previously classified sites based on BMWP NTAXA and BMWP ASPT, and also that the LIFE and PSI indices may soon be added to RICT for similar purposes, it was considered useful to provide a more comprehensive list of biological index test data for indices other than just the two WHPT indices, and for all three seasons These are provided in Appendix 1. When combined with the Table 3 environmental data the Appendix 1 biological data will permit the future testing of RICT in terms of a much wider range of index-by-season combinations with a reduced likelihood of having to create another new set of test data. 8 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Table 2. The 12 sites were used to make 24 test samples in both reference state, and simulated degraded status with new site codes ‘TST-01R’ to ‘TST-12-R’ (where R denotes reference condition), and ‘TST-01-D’ to ‘TST-12-D’ (where D denotes simulated degraded) Test Data Site Number TST-01-R TST-02-R TST-03-R TST-04-R TST-05-R TST-06-R TST-07-R TST-08-R TST-09-R TST-10-R TST-11-R TST-12-R TST-01-D TST-02-D TST-03-D TST-04-D TST-05-D TST-06-D TST-07-D TST-08-D TST-09-D TST-10-D TST-11-D TST-12-D Reference/simulated degraded Reference Reference Reference Reference Reference Reference Reference Reference Reference Reference Reference Reference Degraded Degraded Degraded Degraded Degraded Degraded Degraded Degraded Degraded Degraded Degraded Degraded England/ Scotland/ Wales* Eng. Eng. Eng. Eng. Eng. Eng. Eng. Scot. Scot. Scot. Scot. Wales Eng. Eng. Eng. Eng. Eng. Eng. Eng. Scot. Scot. Scot. Scot. Wales TWINSPAN End Group (1-43) 20 24 28 32 36 40 43 1 4 8 12 16 20 24 28 32 36 40 43 1 4 8 12 16 Site Code 3101 9581 8805 2007 2307 7145 6111 SEPA_N06 SEPA_W05 3785 NE01 WE03 3101 9581 8805 2007 2307 7145 6111 SEPA_N06 SEPA_W05 3785 NE01 WE03 River Name Derwent Lathkill Coombevalley Stream Blithe Colne Ed Ouse/Cam Shetland: Burn of Laxdale Islay: Duich/Torra Green Burn Lossie Afon Caseg Derwent Lathkill Coombevalley Stream Blithe Colne Ed Ouse/Cam Shetland: Burn of Laxdale Islay: Duich/Torra Green Burn Lossie Afon Caseg *the number of samples was distributed proportionally across England, Scotland and Wales by land area. 9 Site Name Langdale End Alport Kilkhampton Newton Fordstreet Bridge Pains Moor Hilgay Bridge North Voxter Torra Bridge Dalmary Cloddach Braichmelyn Langdale End Alport Kilkhampton Newton Fordstreet Bridge Pains Moor Hilgay Bridge North Voxter Torra Bridge Dalmary Cloddach Braichmelyn Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data PEBBLES_GRAVEL SAND SILT_CLAY 2 4 1 3 3 1 7 1 3 1 4 2 2 4 1 3 3 1 7 1 3 1 4 2 BOULDER_COBBLES 2.7 5 40 1.8 1.3 5.5 0.2 11.5 11.4 45.7 8.3 66.7 2.7 5 40 1.8 1.3 5.5 0.2 11.5 11.4 45.7 8.3 66.7 ALKALINITY DISCHARGE 60 142 100 97 15 38 1 5 46 30 44 160 60 142 100 97 15 38 1 5 46 30 44 160 MEAN_DEPTH SLOPE 91000 64600 11600 25900 27200 10500 97000 29000 55200 95500 58400 66300 91000 64600 11600 25900 27200 10500 97000 29000 55200 95500 58400 66300 MEAN_WIDTH ALTITUDE 94200 22000 24600 04800 92100 07400 60400 43700 34400 51500 20300 63000 94200 22000 24600 04800 92100 07400 60400 43700 34400 51500 20300 63000 DIST_FROM_SOURCE Northing SE SK SS SK TL SU TL HU NR NS NJ SH SE SK SS SK TL SU TL HU NR NS NJ SH VELOCITY Easting SITE TST-01-R TST-02-R TST-03-R TST-04-R TST-05-R TST-06-R TST-07-R TST-08-R TST-09-R TST-10-R TST-11-R TST-12-R TST-01-D TST-02-D TST-03-D TST-04-D TST-05-D TST-06-D TST-07-D TST-08-D TST-09-D TST-10-D TST-11-D TST-12-D NGR Table 3. RIVPACS/RICT environmental test data. 10 6 1.7 27 28 1.8 69 6.6 8.5 4 27 6.4 10 6 1.7 27 28 1.8 69 6.6 8.5 4 27 6.4 6.4 6 1.4 10 6.6 1.8 40 3.6 4.53 3 15.1 12 6.4 6 1.4 10 6.6 1.8 40 3.6 4.53 3 15.1 12 21.7 10.7 7.8 8.7 14.4 14.2 200 12 20 17.5 25 28 21.7 10.7 7.8 8.7 14.4 14.2 200 12 20 17.5 25 28 101.4 194.6 50 164 217.7 178 237.3 17.5 3 9.5 23 9.8 101.4 194.6 50 164 217.7 178 237.3 17.5 3 9.5 23 9.8 58 30 65 25 12 0 1 73 67 70 70 83 58 30 65 25 12 0 1 73 67 70 70 83 36 47 30 53 80 38 2 27 33 25 30 17 36 47 30 53 80 38 2 27 33 25 30 17 3 20 3 15 8 13 3 0 0 5 0 0 3 20 3 15 8 13 3 0 0 5 0 0 1 3 3 7 0 50 94 0 0 0 0 0 1 3 3 7 0 50 94 0 0 0 0 0 Site TST-01-R TST-02-R TST-03-R TST-04-R TST-05-R TST-06-R TST-07-R TST-08-R TST-09-R TST-10-R TST-11-R TST-12-R TST-01-D TST-02-D TST-03-D TST-04-D TST-05-D TST-06-D TST-07-D TST-08-D TST-09-D TST-10-D TST-11-D TST-12-D WHPT NTAXA Abund 24 25 28 31 24 31 35 11 18 15 22 27 23 20 18 16 8 6 33 9 12 8 8 5 WHPT ASPT Abund 6.512 6.476 6.739 6.587 4.908 5.503 4.017 5.436 7.767 7.033 7.636 7.604 5.783 5.650 5.778 5.125 3.625 2.833 3.636 4.667 6.417 5.250 5.250 4.200 Season_ID 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 WHPT NTAXA Abund WHPT ASPT Abund Site TST-01-R TST-02-R TST-03-R TST-04-R TST-05-R TST-06-R TST-07-R TST-08-R TST-09-R TST-10-R TST-11-R TST-12-R TST-01-D TST-02-D TST-03-D TST-04-D TST-05-D TST-06-D TST-07-D TST-08-D TST-09-D TST-10-D TST-11-D TST-12-D Season_ID Table 4. RIVPACS/RICT Spring (left) and Autumn (right) WHPT ASPT and WHPT NTAXA biological test data (the same data are coloured blue and green respectively in Appendix 1). 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 6.348 5.892 6.132 6.648 4.881 5.452 3.959 5.845 8.018 7.146 7.617 7.048 5.615 5.150 5.222 5.125 3.455 2.667 3.500 5.000 6.818 5.286 5.750 3.286 27 25 28 31 31 29 27 11 17 13 24 33 26 20 18 16 11 6 26 9 11 7 8 7 10 These environmental and biological RIVPACS/RICT test data files are available to download as RICT formatted Excel ® files from the FBA website www.fba.org.uk and are stored on the same webpage as this report. This is the simplest way to access and use the test dataset with RICT. Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data 3. RUNNING THE TEST DATA IN RICT The first step in making a comparison between the predictions generated by RICT, and those generated by an independent version of the same RIVPACS IV model, was to obtain the RICT classification results. An initial exploratory RICT run was performed first to ensure that the test data were correctly formatted for RICT (e.g. columns correctly labelled and in the correct sequence and site names in the environmental file and biological files matched each other). This was also an opportunity to perform a visual check of the grid references by comparing the map in Figure 1 with the map produced inside RICT. RICT currently has the ability to produce both single season results (runs where prediction and classification results are derived from just one season of biological data) or combined season runs (runs where two separate single season runs are integrated within the software into an overall combined seasons classification). In the second case, RICT automatically produces single season ‘child’ runs before subsequently combining them into the overall classification. It was useful to take the opportunity to verify that the results of separate season runs were in agreement with those of the child components of combined season runs (as far as the basic Face Value Bias Uncorrected EQI values at least – testing further than this was beyond the scope of this project). Testing confirmed that the Face Value Bias Uncorrected EQI values produced in single season runs and the EQIs produced in combined season child runs were in mutual agreement. This simplified the subsequent comparison of RICT with the independently generated RIVPACS model by making it only necessary to compare single season runs. The newly developed RIVPACS/RICT test data were therefore put through RICT as two separate season classification runs, one for spring and one for autumn using: Environmental data and Spring WHPT NTAXA & ASPT biological data Environmental data and Autumn WHPT NTAXA & ASPT biological data The output files from these two runs were downloaded from the RICT software and saved. The results of these two RICT classification runs were then extracted into a more convenient spreadsheet format ready for comparison with the independently derived RIVPACS IV prediction model. 11 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data 4. COMPARING RICT PREDICTIONS OUTPUT WITH INDEPENDENT CODE VALUES FOR THE TEST DATA Because of a limited budget and hence time for this testing work, it was agreed with John Murray-Bligh (Environment Agency) that the testing would concentrate on making predictions of the expected values of the abundance-weighted WHPT NTAXA and WHPT ASPT indices for spring samples, both as raw un-adjusted expected values and as adjusted expected vaues (i.e. adjusted for the quality of the RIVPACS reference sites actively involved in the prediction for any specific test site). All of the statistical algorithms used to provide the RIVPACS IV model predictions of expected values of biotic indices in the original (and current) RICT software were developed by Ralph Clarke and supplied in complete detail in the SNIFFER project WFD72C final report (Davy-Bowker et al 2008) upon which the RICT software was orginallly developed. Back in 2007-08, Ralph Clarke originally developed and assessed the RIVPACS IV models involved in RICT using a mixture of statistical software and his own MINITAB statistical software macro codes. For this RICT-testing project, Ralph adapted parts of this code to make independent calculations of the predictions for the 12 sites in the new Test Data described in sections 2 and 3 above. 4.1 RIVPACS derived Environmental variables for the predictions RIVPACS predictions of the probability of RIVPACS TWINSPAN end-groups membership for any river site are based on previously-derived multivariate discriminant analyses (MDA) of the RIVPACS Reference sites. These MDA discriminant functions are applied to a specific set of environmental variables, most of which are derivatives of the original environmental variables supplied by the User as input to the RICT software. Specifically the derived variables are: Latitiude and Longitude - derived from site National Grid reference Air Temperature Mean and Range - derived from site National Grid reference Mean substratum composition (in phi units) - derived from percentage cover of each of boulders/cobbles, pebbles/gravel, sand and silt/clay and also the logarithm (to base 10) of: stream distance from source (DFS) stream width and depth altitude and slope at site alkalinity (log alkalinity and alkalinity are both used in the predictive equatons) The precise order of all of the MDA environmental variables and their values for the 12 test data sites are given in Table 5. Within the RICT software the NGR of a site is used to derive its Latitude and Longitude and estimates of the Mean and Range of Air Temperature of the site. These derived variables are based on a complex set of trigonometric and geographic interpolation equations and background temperature map data, which were all supplied to the RICT programmers within the WFD72C project and final report. To check these have been coded correctly within the RICT software, we compared the values of these variables output from RICT (RICT output file ‘PEV.XML’) with those previously derived by us (prior to RICT) using the RIVPACS III+ (RPBATCH) software. There were no differences in values (to 2 decimal places) between the RICT and former RPBATCH derived values for any of these variables or for Mean 12 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data substratum composition. The log (to base 10) transformation of the appropriate environmental variables was also checked and found to be correct for the 12 test sites. SITE LATITUDE LONGITUDE MEAN AIR TEMP AIR TEMP RANGE DISCHARGE CATEGORY ALKALINITY MEAN SUBSTRATUM LOG ALTITUDE LOG DISTANCE FROM SOURCE LOG WIDTH LOG DEPTH LOG ALKALINITY LOG SLOPE Table 5. RIVPACS/RICT MDA environmental predictor variables’ values for the test data (derived variables displayed to three decimal places); variables are in the correct order for input to RICT TST-01-R 54.306 -0.552 9.486 11.953 2 101.4 -5.638 1.778 1.000 0.806 1.336 2.006 0.431 TST-02-R 53.178 -1.671 9.750 12.562 4 194.6 -3.213 2.152 0.778 0.778 1.029 2.289 0.699 TST-03-R 50.877 -4.493 10.761 9.848 1 50.0 -5.656 2.000 0.230 0.146 0.892 1.699 1.602 TST-04-R 52.830 -1.929 9.590 12.650 3 164.0 -2.800 1.987 1.431 1.000 0.940 2.215 0.255 TST-05-R 51.910 0.793 10.098 13.670 3 217.7 -3.370 1.176 1.447 0.820 1.158 2.338 0.114 TST-06-R 50.893 -1.895 10.558 12.496 1 178.0 2.995 1.580 0.255 0.255 1.152 2.250 TST-07-R 52.547 0.366 9.645 13.584 7 237.3 7.438 0.000 1.839 1.602 2.301 2.375 0.740 0.699 TST-08-R 60.044 -1.215 7.530 8.880 1 17.5 -6.535 0.699 0.820 0.556 1.079 1.243 1.061 TST-09-R 55.717 -6.230 9.427 9.091 3 3.0 -6.265 1.663 0.929 0.656 1.301 0.477 1.057 TST-10-R 56.129 -4.390 8.757 12.242 1 9.5 -6.138 1.477 0.602 0.477 1.243 0.978 1.660 TST-11-R 57.609 -3.334 8.567 11.206 4 23.0 -6.400 1.643 1.431 1.179 1.398 1.362 0.919 TST-12-R 53.176 -4.050 10.224 10.298 2 9.8 -6.985 2.204 0.806 1.079 1.447 0.991 1.824 Summary: the RICT code to calculate the values of the numerous derived environmental variables used in RICT predictions agrees with our independent RIVPACS calculations and checks. 4.2 Discriminant functions coefficients In the independent check of the RICT software it was assumed that the coefficients of the 13 MDA discriminant functions used in the RIVPACS predictions were those supplied from the WFD72C project as file: 'DFCOEFF_GB685_sent101207.DAT' The 13 MDA discriminant functions mean values for the reference sites in each of the 43 TWINSPAN end-groups were extracted from the RICT software and supplied by David Colvill (SEPA). These were used in the independent RICT calculations testing. 13 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data 4.3 End-group means for WHPT biotic indices The RIVPACs IV model TWINSPAN end-group mean values of each of a wide range of biotic indices are available form the RIVPACS Reference sites main database available from the FBA (www.fba.org.uk) and other sources. On checking, we eventually determined that the form of abundance-weighted WHPT NTAXA and WHPT ASPT currently used in RICT are derived from the end-group means of the two WHPT indices based on Taxonomic Level 2 data, but using the BMWP Composite taxa version of the WHPT indices rather than the Distinct families version (both versions were supplied by us in 2008 for the original version of RICT). This is important as some Agency staff may have thought that this set of RICT predictions were for abundance-weighted WHPT NTAXA and WHPT ASPT based on the use of Distinct families taxonomic data and consequently would have incorrectly compare observed index values for Users’ sites based on Distinct family data values with expected values based on Composite values. WHPT NTAXA for distinct families can only be as great or greater than WHPT NTAXA for Composite families; when greater there would be some over-estimation of EQI NTAXA. RICT software and User documentation needs to: (i) make clear that the abundance-weighted WHPT indices are currently based on BMWP Composite families data (ii) be updated to include the ability to make predictions for the abundanceweighted WHPT indices based on Distinct families taxonomic data. 4.4 Mahalanobis distances and Probabilities of end-group for test sites RIVPACS (and thus RICT) predictions of the end-group probabilities of any site involve calculating the Mahalanobis distances (in MDA multivariate space) of the test site to each end-group. These Mahalanobis distances can be extracted from RICT output files. The distance for each of the 12 test sites to each of the 43 GB model end-groups agree with our independent calculations (all differences in distances were less than 0.0001). The RICT output of its calculations of the probability of belonging to each of the 43 endgroups for each of the 12 test sites agreed with our independent calculations (all differences in probabilities were less than 0.0001). For any fixed set of values of the environmental variables for a site, the probability of belonging to each end-group is fixed in that it does not depend on either the seasons and/or years samples to be assssed or on the biotic indices to be used in the assessment or on their observed values. Summary: the RICT code to calculate the probabilities of end-group for each test site agrees with our independent calculations. 4.5 (Unadjusted) Expected values of the (WHPT) biotic indices In this testing study, the requirement was check the accuracy of the RICT calculation of the raw (i.e. unadjusted) Expected values of the two currently used biotic indices, namely abundance-weighted WHPT NTAXA and WHPT ASPT. In section 4.3 of this report, we have already deduced that the current RICT software predictions were based on the Composite family form of these indices (rather than the Distinct family form). In our independent calculation checking we therefore used the same end-group means of the Composite family forms of the two abundance-weighted WHPT indices. 14 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Our checks were based on comparing predictions for spring samples only. The RICT output agreed with our independent calcuations of the (unadjusted) expected spring sample values of WHPT NTAXA (all differences were less than 0.007) and WHPT ASPT (all differences were less than 0.002). The same RICT code is almost certainly used for other season(s) samples, so the only remaining cause of error in this stage of the predictions is that the wrong season(s) end-group means for the required biotic indices are selected. Summary: the RICT code using the probabilities of end-group and end-group biotic index means to calculate (unadjusted) Expected biotic index values agreed with our independent calculations - when checked using the two abundance-weighted WHPT indices for spring samples. 4.6 Adjustment of Expeced values for the quality of reference sites involved The RICT predicted Expected index values for any site is subsequently adjusted using the algorithms developed by Ralph Clarke and detailed in the Work Element 4.5 section of the WFD72C final report. These algorithms were converted into independent code by Ralph Clarke and used to provide an independent check of the RICT software code calculations of adjustment of raw Expected values to WFD Reference Expected values for the same 12 test sites. The RICT output of the adjusted Expected spring sample values agreed with our independent calcuations for WHPT NTAXA (all differences were less than 0.008) and WHPT ASPT (all differences were less than 0.002). However, because the adjustment factors (Q1,Q2, Q3 ,Q4, Q5) for both of these indices are are either 1.0 or quite close (Table 6), it is not necessarily a very sensitive check of the RICT code. At the next stage of the RICT checking project, it would be better if these true values of the adjustment factors were temporarily replaced with more extreme values (e.g. 1.0, 0.9, 0.8, 0.6, 0.4) in both RICT and our independent code and the adjusted Expected values recalculated and compared (the adjustment factors in RICT should of course be reset back to their true values after any such test). Further testing in a follow on project would also provide the opportunity for us to verify that RICT is using up-to date versions of the Adjustment Factors that reflect the latest work done on intercalibration by John Murray-Bligh. Table 6. Adjustment factors for reference site quality (Q1, Q2, Q3, Q4, Q5) for abundanceweighted WHPT NTAXA and WHPT ASPT Index WHPT NTAXA WHPT ASPT Q1 1 1 Q2 1 1 Q3 1 1 Q4 0.967 0.977 Q5 0.926 0.945 Summary: The minor adjustments to WFD Reference state of the raw Expected values of WHPT NTAXA and WHPT ASPT for the 12 test sites appears to have been coded correctly in RICT, based on our independent code, but the RICT code could be tested further using artificial more-widely varying adjustment factors. 15 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data 4.7 Summary of testing of RICT predictions of Expected values of WHPT indices Our independent calculation of each of the algorithm steps involved in derving RIVPACS IV GB model predictions of the raw and adjusted Expected values of the WHPT indices suggest that these are correctly coded and calculated within the RICT software (i.e. RICT as of 13 March 2015). This initial testing project has also highlighted the need for the RICT software and website to include clear software version numbering and a user-accessible log of updates and changes, along with the dates that these alterations were made. 16 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data 5. REFERENCES Clarke R.T. & Davy-Bowker J. (2006). Development of the scientific rationale and formulae for altering RIVPACS predicted indices for WFD reference condition. Scotland & Northern Ireland Forum for Environmental Research. Edinburgh, Scotland, UK. (SNIFFER project WFD72B). Clarke R.T. & Davy-Bowker J. (2014) River Invertebrate Classification Tool Science Development Project: Modifications for WHPT and other abundance-weighted indices. A report to the Scottish Environment Protection Agency. Clarke R. T., Davy-Bowker J., Dunbar M., Laize C., Scarlett P.M. & Murphy J.F. (2011) Enhancement of the River Invertebrate Classification Tool (RICT). Scotland & Northern Ireland Forum for Environmental Research. Edinburgh, Scotland, UK. (SNIFFER project WFD119). Davy-Bowker J., Arnott S., Close R., Dobson M., Dunbar M., Jofre G., Morton D., Murphy J., Wareham W., Smith S. & Gordon V. (2010) Further Development of River Invertebrate Classification Tool. Scotland & Northern Ireland Forum for Environmental Research. Edinburgh, Scotland, UK. (SNIFFER project WFD100). Davy-Bowker J., Clarke R., Corbin T., Vincent H., Pretty J., Hawczak A., Blackburn J., Murphy J. & Jones I. (2008). River Invertebrate Classification Tool. Scotland & Northern Ireland Forum for Environmental Research. Edinburgh, Scotland, UK. (SNIFFER project WFD72C). Davy-Bowker J., Clarke R., Furse M., Davies C., Corbin T., Murphy J. & Kneebone N. (2007a) RIVPACS Database Documentation. Scotland & Northern Ireland Forum for Environmental Research. Edinburgh, Scotland, UK. (SNIFFER project WFD46). Davy-Bowker J., Clarke R., Furse M., Davies C., Corbin T., Murphy J. & Kneebone N. (2007b) RIVPACS Pressure Data Analysis. Scotland & Northern Ireland Forum for Environmental Research. Edinburgh, Scotland, UK. (SNIFFER project WFD46). Davy-Bowker J., Clarke R.T., Jones J.I. & Murphy J.F. (In prep) River Invertebrate Classification Tool Science Development Project: Describing the impact of abstraction and fine sediment pressures on the biological communities in Scottish rivers. A report to the Scottish Government. Davy-Bowker J., Jones J.I. & Murphy J.F. (2014) Standardisation of RIVPACS for deep rivers: Phase I - deriving a standard approach to deep river sampling. Environment Agency, Bristol. Jones J.I. & Davy-Bowker J. (2012) Standardisation of RIVPACS for deep rivers: Phase I review of techniques for sampling benthic macro-invertebrates in deep rivers. Environment Agency, Bristol. 17 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data TL1 BMWP TL2 WHPT Score (nonAb,DistFam) TL2 WHPT NTAXA (nonAb,DistFam) TL2 WHPT ASPT (nonAb,DistFam) TL2 WHPT Score (nonAb,CompFam) TL2 WHPT NTAXA (nonAb,CompFam) TL2 WHPT ASPT (nonAb,CompFam) TL2 WHPT Score (AbW,DistFam) TL2 WHPT NTAXA (AbW,DistFam) TL2 WHPT ASPT (AbW,DistFam) TL2 WHPT Score (AbW,CompFam) TL2 WHPT NTAXA (AbW,CompFam) TL2 WHPT ASPT (AbW,CompFam) TST-01-R 1 132 22 6.000 157.7 24 6.571 156.0 24 6.500 156.3 24 6.512 154.6 24 6.442 TST-02-R 1 140 21 6.667 163.4 25 6.536 163.3 25 6.532 161.9 25 6.476 161.8 25 6.472 TST-03-R 1 164 26 6.308 194.2 28 6.936 192.8 28 6.886 188.7 28 6.739 187.2 28 6.686 TST-04-R 1 165 26 6.346 206.0 31 6.645 198.4 30 6.613 204.2 31 6.587 196.4 30 6.547 TST-05-R 1 107 22 4.864 119.5 24 4.979 116.1 23 5.048 117.8 24 4.908 114.6 23 4.983 TST-06-R 1 144 28 5.143 168.5 31 5.435 168.7 31 5.442 170.6 31 5.503 170.6 31 5.503 TST-07-R 1 146 31 4.710 152.7 35 4.363 147.0 32 4.594 140.6 35 4.017 135.6 32 4.238 TST-08-R 1 43 9 4.778 63.6 11 5.782 62.2 11 5.655 59.8 11 5.436 58.4 11 5.309 TST-09-R 1 119 17 7.000 137.3 18 7.628 135.8 18 7.544 139.8 18 7.767 138.3 18 7.683 TST-10-R 1 76 13 5.846 101.3 15 6.753 93.8 14 6.700 105.5 15 7.033 95.9 14 6.850 TST-11-R 1 134 20 6.700 160.4 22 7.291 152.6 21 7.267 168.0 22 7.636 160.1 21 7.624 TST-12-R 1 171 25 6.840 202.6 27 7.504 193.6 26 7.446 205.3 27 7.604 197.1 26 7.581 TST-01-D 1 112 21 5.333 134 23 5.826 133 23 5.783 133 23 5.783 131 23 5.696 TST-02-D 1 98 17 5.765 114 20 5.700 114 20 5.700 113 20 5.650 113 20 5.650 TST-03-D 1 90 17 5.294 107 18 5.944 106 18 5.889 104 18 5.778 103 18 5.722 TST-04-D 1 66 13 5.077 82 16 5.125 79 15 5.267 82 16 5.125 79 15 5.267 TST-05-D 1 27 8 3.375 30 8 3.750 29 8 3.625 29 8 3.625 29 8 3.625 TST-06-D 1 14 6 2.333 17 6 2.833 17 6 2.833 17 6 2.833 17 6 2.833 TST-07-D 1 124 29 4.276 130 33 3.939 125 30 4.167 120 33 3.636 115 30 3.833 TST-08-D 1 30 7 4.286 45 9 5.000 44 9 4.889 42 9 4.667 41 9 4.556 TST-09-D 1 65 11 5.909 76 12 6.333 75 12 6.250 77 12 6.417 76 12 6.333 TST-10-D 1 30 7 4.286 41 8 5.125 38 7 5.429 42 8 5.250 38 7 5.429 TST-11-D 1 34 7 4.857 40 8 5.000 38 7 5.429 42 8 5.250 40 7 5.714 TST-12-D 1 17 5 3.400 20 5 4.000 19 5 3.800 21 5 4.200 20 5 4.000 TST-01-R 2 154 24 6.417 176.2 27 6.526 168.6 26 6.485 179.9 27 6.663 172.7 26 6.642 TST-02-R 2 116 17 6.824 123.6 18 6.867 124.7 18 6.928 122.6 18 6.811 123.5 18 6.861 TST-03-R 2 141 24 5.875 174.1 28 6.218 166.7 27 6.174 167.7 28 5.989 160.3 27 5.937 TST-04-R 2 154 26 5.923 171.0 29 5.897 171.3 29 5.907 172.1 29 5.934 172.5 29 5.948 TST-05-R 2 100 20 5.000 115.0 22 5.227 115.7 22 5.259 117.7 22 5.350 118.4 22 5.382 TST-06-R 2 119 23 5.174 131.4 25 5.256 131.7 25 5.268 122.0 25 4.880 122.1 25 4.884 TST-07-R 2 142 29 4.897 147.1 34 4.326 140.0 31 4.516 132.4 34 3.894 126.0 31 4.065 TST-08-R 2 35 9 3.889 52.6 10 5.260 50.9 10 5.090 55.3 10 5.530 53.6 10 5.360 TST-09-R 2 120 18 6.667 138.1 19 7.268 136.4 19 7.179 132.7 19 6.984 131.0 19 6.895 TST-10-R 2 null null null null null null TST-11-R 2 122 19 6.421 147.0 21 7.000 139.2 20 6.960 150.9 21 7.186 142.1 20 7.105 TST-12-R 2 148 22 6.727 168.0 24 7.000 159.0 23 6.913 162.7 24 6.779 153.6 23 6.678 TST-01-D 2 131 23 5.696 150 26 5.769 143 25 5.720 153 26 5.885 147 25 5.880 TST-02-D 2 81 14 5.786 87 14 6.214 87 14 6.214 86 14 6.143 86 14 6.143 TST-03-D 2 78 16 4.875 96 18 5.333 92 18 5.111 92 18 5.111 88 18 4.889 TST-04-D 2 62 13 4.769 68 15 4.533 69 15 4.600 69 15 4.600 69 15 4.600 TST-05-D 2 25 7 3.571 29 8 3.625 29 8 3.625 29 8 3.625 30 8 3.750 TST-06-D 2 12 5 2.400 13 5 2.600 13 5 2.600 12 5 2.400 12 5 2.400 TST-07-D 2 121 28 4.321 125 32 3.906 119 29 4.103 113 32 3.531 107 29 3.690 TST-08-D 2 25 7 3.571 37 8 4.625 36 8 4.500 39 8 4.875 38 8 4.750 TST-09-D 2 66 12 5.500 76 12 6.333 75 12 6.250 73 12 6.083 72 12 6.000 TST-10-D 2 TST-11-D 2 31 7 4.429 37 7 5.286 35 7 5.000 38 7 5.429 36 7 5.143 TST-12-D 2 15 4 3.750 17 5 3.400 16 5 3.200 16 5 3.200 15 5 3.000 TL1 ASPT Site ID TL1 NTAXA Season Code 6. APPENDIX 1: Full List of the RIVPACS/RICT Biological Test Dataset Spring Reference Degraded Summer Reference null null null null null null null null null Degraded null null null null null null null 18 null null null null null null null null Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data TL2 WHPT ASPT (nonAb,DistFam) TL2 WHPT Score (nonAb,CompFam) TL2 WHPT NTAXA (nonAb,CompFam) TL2 WHPT ASPT (nonAb,CompFam) TL2 WHPT Score (AbW,DistFam) TL2 WHPT NTAXA (AbW,DistFam) TL2 WHPT ASPT (AbW,DistFam) TL2 WHPT Score (AbW,CompFam) 157 26 6.038 168.3 27 6.233 167.0 27 6.185 171.4 27 6.348 170.1 27 6.300 3 129 21 6.143 150.4 25 6.016 150.7 25 6.028 147.3 25 5.892 147.6 25 5.904 TST-03-R 3 149 25 5.960 176.7 28 6.311 175.3 28 6.261 171.7 28 6.132 170.1 28 6.075 TST-04-R 3 184 28 6.571 206.4 31 6.658 206.5 31 6.661 206.1 31 6.648 206.1 31 6.648 TST-05-R 3 123 25 4.920 151.1 31 4.874 144.1 29 4.969 151.3 31 4.881 143.8 29 4.959 TST-06-R 3 143 26 5.500 164.3 29 5.666 164.7 29 5.679 158.1 29 5.452 158.4 29 5.462 TST-07-R 3 123 26 4.731 116.5 27 4.315 120.0 27 4.444 106.9 27 3.959 111.8 27 4.141 TST-08-R 3 56 11 5.091 65.7 11 5.973 64.1 11 5.827 64.3 11 5.845 62.7 11 5.700 TST-09-R 3 110 16 6.875 134.4 17 7.906 134.2 17 7.894 136.3 17 8.018 136.1 17 8.006 TST-10-R 3 81 13 6.231 92.8 13 7.138 91.3 13 7.023 92.9 13 7.146 91.4 13 7.031 TST-11-R 3 151 22 6.864 181.5 24 7.562 173.7 23 7.552 182.8 24 7.617 174.9 23 7.604 TST-12-R 3 171 27 6.333 230.0 33 6.970 221.0 32 6.906 232.6 33 7.048 223.3 32 6.978 TST-01-D 3 133 25 5.320 143 26 5.500 142 26 5.462 146 26 5.615 145 26 5.577 TST-02-D 3 90 17 5.294 105 20 5.250 105 20 5.250 103 20 5.150 103 20 5.150 TST-03-D 3 82 16 5.125 97 18 5.389 96 18 5.333 94 18 5.222 94 18 5.222 TST-04-D 3 74 14 5.286 83 16 5.188 83 16 5.188 82 16 5.125 82 16 5.125 TST-05-D 3 31 9 3.444 38 11 3.455 36 10 3.600 38 11 3.455 36 10 3.600 TST-06-D 3 14 5 2.800 16 6 2.667 16 6 2.667 16 6 2.667 16 6 2.667 TST-07-D 3 105 25 4.200 99 26 3.808 102 26 3.923 91 26 3.500 95 26 3.654 TST-08-D 3 39 9 4.333 46 9 5.111 45 9 5.000 45 9 5.000 44 9 4.889 TST-09-D 3 61 10 6.100 74 11 6.727 74 11 6.727 75 11 6.818 75 11 6.818 TST-10-D 3 32 7 4.571 37 7 5.286 37 7 5.286 37 7 5.286 37 7 5.286 TST-11-D 3 38 8 4.750 45 8 5.625 43 8 5.375 46 8 5.750 44 8 5.500 TST-12-D 3 17 5 3.400 23 7 3.286 22 6 3.667 23 7 3.286 22 6 3.667 TL2 WHPT ASPT (AbW,CompFam) TL2 WHPT NTAXA (nonAb,DistFam) TL2 WHPT NTAXA (AbW,CompFam) TL2 WHPT Score (nonAb,DistFam) 3 TST-02-R TL1 ASPT TL1 BMWP TST-01-R Site ID TL1 NTAXA Season Code Appendix 2: continued... Autumn Reference Degraded Footnote: Reference biotic indices (Site ID suffixed ‘R’) were the biotic indices of the RIVPACS reference samples. Degraded biotic indices (Site ID suffixed ‘D’) were simulated. 19 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data TL4 AWIC(Sp) Murphy TL5 AWIC(Sp) Murphy TL4 SEPA % Acid Sensitive Taxa TL5 SEPA % Acid Sensitive Taxa TL1/2 LIFE(Fam) (CompFam) TL2 LIFE(Fam) (DistFam) TL3 LIFE(Fam) (DistFam) TL4 LIFE(Sp) TL5 LIFE(Sp) TL3 PSI(Fam) TL4 PSI(Sp) TL5 PSI(Sp) 1 4.636 5.875 5.875 7.812 7.812 11.982 11.982 7.100 7.100 7.100 8.000 8.000 53.659 69.565 72.093 TST-02-R 1 4.500 7.333 7.333 10.167 10.167 54.198 54.198 7.526 7.526 7.526 8.600 8.600 75.676 88.636 88.372 TST-03-R 1 4.360 6.316 6.316 8.789 8.789 54.099 53.051 7.625 7.625 7.625 8.419 8.667 66.667 69.118 72.131 TST-04-R 1 4.800 7.312 7.267 10.312 10.333 10.385 10.385 7.542 7.600 7.600 8.552 8.846 63.636 74.699 76.190 TST-05-R 1 5.100 7.200 7.200 10.100 10.100 2.844 2.844 6.550 6.524 6.524 7.200 7.455 29.545 43.860 45.455 TST-06-R 1 5.500 6.571 6.571 9.857 9.857 28.318 27.773 6.704 6.704 6.704 7.000 7.346 44.068 34.667 40.625 TST-07-R 1 5.519 9.000 9.000 14.000 14.000 2.065 2.064 5.759 5.656 5.656 5.667 5.761 13.699 6.140 6.140 TST-08-R 1 4.444 6.667 6.667 8.000 8.000 2.445 2.309 7.429 7.429 7.429 8.400 8.500 63.158 57.143 53.846 TST-09-R 1 3.882 5.875 5.875 7.312 7.312 5.234 5.234 8.200 8.200 8.200 8.667 8.667 82.353 94.286 94.286 TST-10-R 1 3.846 5.000 5.000 6.700 6.700 9.635 9.635 6.818 6.667 6.667 7.588 7.750 44.444 54.545 54.545 TST-11-R 1 4.300 7.000 7.000 9.333 9.333 46.679 46.679 8.222 8.263 8.263 9.087 9.091 76.364 90.000 88.679 TST-12-R 1 4.440 7.074 7.074 9.259 9.259 45.600 45.509 7.957 7.917 7.917 8.727 8.750 74.000 90.769 93.333 TST-01-D 1 3.941 4.994 4.994 6.640 6.640 10.185 10.185 6.035 6.035 6.035 6.800 6.800 45.610 59.130 61.279 TST-02-D 1 3.150 5.133 5.133 7.117 7.117 37.939 37.939 5.268 5.268 5.268 6.020 6.020 52.973 62.045 61.860 TST-03-D 1 2.398 3.474 3.474 4.834 4.834 29.754 29.178 4.194 4.194 4.194 4.630 4.767 36.667 38.015 39.672 TST-04-D 1 1.920 2.925 2.907 4.125 4.133 4.154 4.154 3.017 3.040 3.040 3.421 3.538 25.454 29.880 30.476 TST-05-D 1 1.275 1.800 1.800 2.525 2.525 0.711 0.711 1.638 1.631 1.631 1.800 1.864 7.386 10.965 11.364 TST-06-D 1 0.550 0.657 0.657 0.986 0.986 2.832 2.777 0.670 0.670 0.670 0.700 0.735 4.407 3.467 4.063 TST-07-D 1 4.691 7.650 7.650 11.900 11.900 1.755 1.754 4.895 4.808 4.808 4.817 4.897 11.644 5.219 5.219 TST-08-D 1 3.111 4.667 4.667 5.600 5.600 1.712 1.616 5.200 5.200 5.200 5.880 5.950 44.211 40.000 37.692 TST-09-D 1 2.135 3.231 3.231 4.022 4.022 2.879 2.879 4.510 4.510 4.510 4.767 4.767 45.294 51.857 51.857 TST-10-D 1 1.538 2.000 2.000 2.680 2.680 3.854 3.854 2.727 2.667 2.667 3.035 3.100 17.778 21.818 21.818 TST-11-D 1 1.075 1.750 1.750 2.333 2.333 11.670 11.670 2.056 2.066 2.066 2.272 2.273 19.091 22.500 22.170 TST-12-D 1 0.444 0.707 0.707 0.926 0.926 4.560 4.551 0.796 0.792 0.792 0.873 0.875 7.400 9.077 9.333 TST-01-R 2 4.833 6.769 6.769 9.077 9.077 49.341 49.232 7.409 7.348 7.348 8.333 8.333 58.000 73.077 69.231 TST-02-R 2 4.500 7.500 7.500 10.125 10.125 79.587 79.587 7.625 7.625 7.625 8.294 8.294 80.769 73.333 73.333 TST-03-R 2 4.792 6.692 6.692 9.231 9.231 51.297 51.240 7.409 7.348 7.348 8.125 8.333 53.968 60.000 65.217 TST-04-R 2 5.000 7.333 7.333 10.250 10.250 23.996 23.996 7.167 7.167 7.167 7.844 7.931 54.902 61.842 60.317 TST-05-R 2 5.105 6.800 6.800 10.000 10.000 7.073 7.073 7.278 7.278 7.278 7.955 8.471 58.333 63.333 65.957 TST-06-R 2 5.143 6.333 6.333 9.667 9.667 22.732 22.312 6.591 6.591 6.591 6.704 6.952 31.148 25.714 28.814 TST-07-R 2 5.560 9.000 9.000 13.500 13.500 4.528 4.601 5.643 5.516 5.516 5.415 5.439 18.182 6.034 6.034 TST-08-R 2 5.222 7.000 7.000 9.667 9.667 66.531 62.595 7.286 7.286 7.286 8.500 8.400 65.000 84.615 63.636 TST-09-R 2 4.444 6.455 6.455 8.455 8.455 0.818 0.818 7.750 7.750 7.750 8.400 8.429 81.818 87.500 87.097 TST-10-R 2 null null null null null null null null null null null null null null null TST-11-R 2 4.684 7.688 7.667 11.062 11.133 42.052 42.210 8.118 8.167 8.167 9.167 9.217 79.592 86.957 89.231 TST-12-R 2 4.636 7.765 7.765 10.176 10.176 25.000 25.000 7.750 7.762 7.762 8.217 8.217 74.359 91.892 91.429 TST-01-D 2 4.108 5.754 5.754 7.715 7.715 41.940 41.847 6.298 6.246 6.246 7.083 7.083 49.300 62.115 58.846 TST-02-D 2 3.150 5.250 5.250 7.088 7.088 55.711 55.711 5.338 5.338 5.338 5.806 5.806 56.538 51.333 51.333 TST-03-D 2 2.636 3.681 3.681 5.077 5.077 28.213 28.182 4.075 4.041 4.041 4.469 4.583 29.682 33.000 35.869 TST-04-D 2 2.000 2.933 2.933 4.100 4.100 9.598 9.598 2.867 2.867 2.867 3.138 3.172 21.961 24.737 24.127 TST-05-D 2 1.276 1.700 1.700 2.500 2.500 1.768 1.768 1.820 1.820 1.820 1.989 2.118 14.583 15.833 16.489 TST-06-D 2 0.514 0.633 0.633 0.967 0.967 2.273 2.231 0.659 0.659 0.659 0.670 0.695 3.115 2.571 2.881 TST-07-D 2 4.726 7.650 7.650 11.475 11.475 3.849 3.911 4.797 4.689 4.689 4.603 4.623 15.455 5.129 5.129 TST-08-D 2 3.655 4.900 4.900 6.767 6.767 46.572 43.817 5.100 5.100 5.100 5.950 5.880 45.500 59.231 44.545 TST-09-D 2 2.444 3.550 3.550 4.650 4.650 0.450 0.450 4.263 4.263 4.263 4.620 4.636 45.000 48.125 47.903 TST-10-D 2 null null null null null null null null null null null null null null null TST-11-D 2 1.171 1.922 1.917 2.766 2.783 10.513 10.553 2.030 2.042 2.042 2.292 2.304 19.898 21.739 22.308 TST-12-D 2 0.464 0.776 0.776 1.018 1.018 2.500 2.500 0.775 0.776 0.776 0.822 0.822 7.436 9.189 9.143 TL5 WFD AWIC(Sp) McFarland TL1 AWIC(Fam) TST-01-R Site ID TL4 WFD AWIC(Sp) McFarland Season Code Appendix 1: continued... Spring Reference Degraded Summer Reference Degraded 20 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data TL5 PSI(Sp) TL4 PSI(Sp) TL3 PSI(Fam) TL5 LIFE(Sp) TL4 LIFE(Sp) TL3 LIFE(Fam) (DistFam) TL2 LIFE(Fam) (DistFam) TL1/2 LIFE(Fam) (CompFam) TL5 SEPA % Acid Sensitive Taxa TL4 SEPA % Acid Sensitive Taxa TL5 WFD AWIC(Sp) McFarland TL4 WFD AWIC(Sp) McFarland TL5 AWIC(Sp) Murphy TL4 AWIC(Sp) Murphy TL1 AWIC(Fam) Site ID Season Code Appendix 1: continued... Autumn Reference TST-01-R 3 4.923 6.846 6.846 9.538 9.538 21.145 21.145 7.333 7.333 7.333 8.400 8.375 62.264 85.185 79.167 TST-02-R 3 5.000 7.500 7.500 10.875 10.875 82.877 82.877 6.895 6.895 6.895 8.227 8.600 49.020 56.364 67.391 TST-03-R 3 4.583 6.615 6.615 9.000 9.000 43.691 43.275 7.304 7.304 7.304 7.889 8.083 58.065 65.517 68.627 TST-04-R 3 4.962 7.200 7.143 10.067 10.000 20.841 20.841 7.769 7.769 7.769 8.059 8.367 66.071 70.130 75.862 TST-05-R 3 5.174 7.200 7.200 10.600 10.600 14.317 14.305 6.957 6.840 6.840 7.382 7.633 39.062 49.451 49.367 TST-06-R 3 5.120 6.889 6.889 10.111 10.111 34.361 34.134 6.680 6.680 6.680 7.222 7.565 33.871 33.803 37.705 TST-07-R 3 5.609 9.000 9.000 13.000 13.000 0.896 5.800 5.560 5.560 5.485 5.548 16.418 TST-08-R 3 4.727 5.800 5.800 7.000 7.000 14.511 14.286 7.778 7.778 7.778 8.714 8.833 68.182 77.778 68.421 TST-09-R 3 4.000 5.789 5.789 7.526 7.526 16.491 16.491 TST-10-R 3 3.750 5.857 5.857 7.571 7.571 TST-11-R 3 4.045 6.720 6.667 9.080 TST-12-R 3 4.481 6.826 6.826 TST-01-D 3 TST-02-D 3 TST-03-D 0.895 4.819 3.659 7.867 7.867 7.867 8.682 8.714 83.333 97.727 97.674 1.132 7.091 7.091 7.091 8.417 8.455 73.077 85.000 84.211 9.042 34.455 34.439 8.100 8.143 8.143 8.968 9.000 75.439 89.873 91.549 9.043 9.043 37.400 37.400 7.560 7.577 7.577 8.586 8.607 71.154 90.323 92.308 4.185 5.819 5.819 8.107 8.107 17.973 17.973 6.233 6.233 6.233 7.140 7.119 52.924 72.407 67.292 3.500 5.250 5.250 7.613 7.613 58.014 58.014 4.827 4.827 4.827 5.759 6.020 34.314 39.455 47.174 3 2.521 3.638 3.638 4.950 4.950 24.030 23.801 4.017 4.017 4.017 4.339 4.446 31.936 36.034 37.745 TST-04-D 3 1.985 2.880 2.857 4.027 4.000 8.336 8.336 3.108 3.108 3.108 3.224 3.347 26.428 28.052 30.345 TST-05-D 3 1.294 1.800 1.800 2.650 2.650 3.579 3.576 1.739 1.710 1.710 1.846 1.908 9.766 12.363 12.342 TST-06-D 3 0.512 0.689 0.689 1.011 1.011 3.436 3.413 0.668 0.668 0.668 0.722 0.757 3.387 3.380 3.771 TST-07-D 3 4.768 7.650 7.650 11.050 11.050 0.761 0.762 4.930 4.726 4.726 4.662 4.716 13.955 4.096 3.110 TST-08-D 3 3.309 4.060 4.060 4.900 4.900 10.158 10.000 5.445 5.445 5.445 6.100 6.183 47.727 54.445 47.895 TST-09-D 3 2.200 3.184 3.184 4.139 4.139 9.070 9.070 4.327 4.327 4.327 4.775 4.793 45.833 53.750 53.721 TST-10-D 3 1.500 2.343 2.343 3.028 3.028 0.472 0.453 2.836 2.836 2.836 3.367 3.382 29.231 34.000 33.684 TST-11-D 3 1.011 1.680 1.667 2.270 2.261 8.614 8.610 2.025 2.036 2.036 2.242 2.250 18.860 22.468 22.887 TST-12-D 3 0.448 0.683 0.683 0.904 0.904 3.740 3.740 0.756 0.758 0.758 0.859 0.861 1.181 Degraded 21 7.115 9.032 9.231 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data Season Code TL2 SPEAR(Fam) % TL4 SPEAR(Sp) % TL5 SPEAR(Sp) % TL4 CCI TL5 CCI Appendix 1: continued... TST-01-R 1 50.826 44.352 53.455 17.818 17.818 TST-02-R 1 55.179 56.564 59.371 16.800 16.800 TST-03-R 1 37.100 32.760 44.640 6.000 5.739 TST-04-R 1 47.228 42.921 51.305 12.115 11.591 TST-05-R 1 22.700 21.223 28.413 8.600 8.409 TST-06-R 1 28.756 33.532 38.908 13.323 12.542 TST-07-R 1 18.182 21.216 24.801 16.234 16.022 TST-08-R 1 28.478 41.777 54.572 1.200 1.200 TST-09-R 1 66.144 60.177 60.750 9.375 9.375 TST-10-R 1 37.404 23.368 27.622 5.471 5.250 TST-11-R 1 62.106 52.891 60.615 11.000 6.158 TST-12-R 1 61.869 63.781 64.359 11.000 11.000 TST-01-D 1 43.202 37.699 45.437 15.145 15.145 TST-02-D 1 38.625 39.595 41.560 11.760 11.760 TST-03-D 1 20.405 18.018 24.552 3.300 3.156 TST-04-D 1 18.891 17.168 20.522 4.846 4.636 TST-05-D 1 5.675 5.306 7.103 2.150 2.102 TST-06-D 1 2.876 3.353 3.891 1.332 1.254 TST-07-D 1 15.455 18.034 21.081 13.799 13.619 TST-08-D 1 19.935 29.244 38.200 0.840 0.840 TST-09-D 1 36.379 33.097 33.413 5.156 5.156 TST-10-D 1 14.962 9.347 11.049 2.188 2.100 TST-11-D 1 15.527 13.223 15.154 2.750 1.540 TST-12-D 1 6.187 6.378 6.436 1.100 1.100 TST-01-R 2 59.967 54.503 61.544 12.200 11.667 TST-02-R 2 44.442 41.212 40.980 17.733 17.733 TST-03-R 2 33.121 25.082 34.113 4.950 4.833 TST-04-R 2 39.479 39.002 45.536 11.129 11.071 TST-05-R 2 25.408 22.551 25.239 5.714 5.250 TST-06-R 2 15.538 15.729 20.289 4.957 4.579 TST-07-R 2 18.785 17.464 20.442 12.950 13.103 TST-08-R 2 29.634 23.498 33.328 1.000 1.000 TST-09-R 2 48.336 40.523 46.377 9.286 9.286 TST-10-R 2 null null null null null TST-11-R 2 58.874 56.473 58.750 5.842 5.842 TST-12-R 2 46.875 44.212 48.465 9.500 9.500 TST-01-D 2 50.972 46.328 52.312 10.370 9.917 TST-02-D 2 31.109 28.848 28.686 12.413 12.413 TST-03-D 2 18.217 13.795 18.762 2.723 2.658 TST-04-D 2 15.792 15.601 18.214 4.452 4.428 TST-05-D 2 6.352 5.638 6.310 1.429 1.313 TST-06-D 2 1.554 1.573 2.029 0.496 0.458 TST-07-D 2 15.967 14.844 17.376 11.008 11.138 TST-08-D 2 20.744 16.449 23.330 0.700 0.700 TST-09-D 2 26.585 22.288 25.507 5.107 5.107 TST-10-D 2 0.000 0.000 0.000 0.000 0.000 TST-11-D 2 14.719 14.118 14.688 1.461 1.461 TST-12-D 2 4.688 4.421 4.847 0.950 0.950 Site ID Spring Reference Degraded Summer Reference Degraded 22 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data TL4 CCI 47.239 10.435 9.773 45.915 39.175 43.239 20.000 20.632 TST-03-R 3 32.553 25.090 31.237 14.609 14.667 TST-04-R 3 40.109 34.659 43.678 10.806 10.000 TST-05-R 3 19.186 21.716 24.746 8.971 8.448 TST-06-R 3 27.263 28.993 29.615 9.800 9.773 TST-07-R 3 22.353 21.769 24.459 14.875 14.933 TST-08-R 3 46.580 43.844 52.606 1.000 1.000 TST-09-R 3 69.892 68.249 71.528 11.053 11.053 TST-10-R 3 54.503 50.644 52.618 14.000 14.000 TST-11-R 3 58.180 54.263 58.103 11.923 11.923 TST-12-R 3 59.933 57.354 58.728 11.800 11.250 TST-01-D 3 41.429 34.857 40.153 8.870 8.307 TST-02-D 3 32.141 27.423 30.267 14.000 14.442 TST-03-D 3 17.904 13.800 17.180 8.035 8.067 TST-04-D 3 16.044 13.864 17.471 4.322 4.000 TST-05-D 3 4.797 5.429 6.187 2.243 2.112 TST-06-D 3 2.726 2.899 2.962 0.980 0.977 TST-07-D 3 19.000 18.504 20.790 12.644 12.693 TST-08-D 3 32.606 30.691 36.824 0.700 0.700 TST-09-D 3 38.441 37.537 39.340 6.079 6.079 TST-10-D 3 21.801 20.258 21.047 5.600 5.600 TST-11-D 3 14.545 13.566 14.526 2.981 2.981 TST-12-D 3 5.873 1.180 1.125 TL5 CCI TL5 SPEAR(Sp) % 48.740 41.008 3 TL2 SPEAR(Fam) % 3 TST-02-R Season Code TST-01-R Site ID Autumn TL4 SPEAR(Sp) % Appendix 1: continued... Reference Degraded 5.993 5.735 23 Testing RICT predictions of expected values using an independent RIVPACS model and new RIVPACS test data 24