ArcGIS REST Services Directory Login
JSON

Layer: NZ fish predictions Crow REC2 2014 (ID: 0)

Name: NZ fish predictions Crow REC2 2014

Display Field: nzsegment

Type: Feature Layer

Geometry Type: esriGeometryPolyline

Description: <DIV STYLE="text-align:Left;"><DIV><DIV><P><SPAN STYLE="font-weight:bold;">Predicting distributions of New Zealand freshwater fishes</SPAN></P><P><SPAN STYLE="font-style:italic;">Executive summary</SPAN></P><P><SPAN><SPAN>The New Zealand Freshwater Fish Database (NZFFD) and the River Environment Classification (REC1) were used to generate spatial predictions of freshwater fish probability of capture across New Zealand. These predicted data have assisted with various management decisions throughout New Zealand, but a recent updated version of the REC (REC2) required these predictions to be recalculated. NIWA was contracted by the Department of Conservation to update the previous probability of capture models calculated by Leathwick et al. (2008) using REC2. Additionally, the present study checked and corrected all assignments of NZFFD records to REC1 segment identifiers and assigned NZFFD records to an REC2 segment identifier. </SPAN></SPAN></P><P><SPAN><SPAN>Previous attempts to match each NZFFD record with the correct NZReach identifier strongly suggested that the process could not be fully automated. Subsequently, we adopted an incremental approach, using existing data sources to automate as many quality-checks as possible, but also to highlight records where there was any evidence of ambiguity in the assignment of the REC1 and REC2 identifiers to NZFFD cards. Any ambiguous assignments were then manually checked and reassigned as appropriate. We then used Regularized Random Forest (RF) models to relate fish presence to 86 predictor variables that described environmental conditions, spatial arrangement and hydrological variation. These predictors included all of the key variables previously identified by Leathwick et al. (2008). To reduce the influence of sampling method on the results we completed separate analyses for each fishing method for each species, where sufficient data were available. For range-restricted non-diadromous fishes we predicted the probability of capture for all of New Zealand and the probability of capture only within their known range of occurrence. </SPAN></SPAN></P><P><SPAN><SPAN>Area Under Curve (AUC) was used as an indicator of model performance that accounted for correctly predicting an observation by chance. AUC values for all models indicated very high predictive performance. Across all models in the present study, AUC values averaged 0.91 suggesting that the models correctly predicted the observed values in 91% of cases. Diadromous fish models using electric-fishing data predominantly used predictors relating to altitude, temperature and distance from the ocean to predict fish capture. Non-diadromous fish models predominantly used predictors relating to spatial location, average slope of the segments downstream, upstream January temperature and distance to ocean. </SPAN></SPAN></P><P><SPAN><SPAN>Species models were likely to contain sampling bias in the predictions given that some aspects of the sampling methodology could not be included in the predictions. Separate models were generated for longfins and brown trout with different fishing methods, but we suggest utilising the electric-fishing models primarily for these species. This is simply because there is a higher number of observations utilised for electric-fishing models. If the end-user is interested in alternative predictions using the other sampling methods, then these should be considered in combination with the electric-fishing models. Compared to the previous REC1 analysis of Leathwick et al. (2008), the present study displayed very similar predictive performance. Overall the present study displayed slightly higher AUC values for species with high levels of prevalence (i.e. longfins and shortfins), but slightly lower AUC values for species with lower prevalence (i.e. non-diadromous species). This may be because of differences between the statistical methods or the datasets used. Although the differences between the AUC values between the present study and Leathwick et al. (2008) are small, combining the results of the two analyses would maximise the predictive accuracy of fish capture for New Zealand.</SPAN></SPAN></P><P><SPAN STYLE="font-style:italic;">Crow, S.K; Booker D.J.; Sykes, J.R.E.; Unwin, M.J.; Shankar, U. 2014: </SPAN><SPAN STYLE="font-weight:bold;">Predicting distributions of New Zealand freshwater fishes. NIWA Client Report CHC2014-145. 100 p</SPAN><SPAN>.</SPAN></P><P><SPAN STYLE="font-style:italic;">Leathwick, J.R., Julian, K., Elith, J., Rowe, D.L. 2008: </SPAN><SPAN STYLE="font-weight:bold;">Predicting the distributions of freshwater fish species for all New Zealand's rivers and streams. NIWA Client Report HAM2008-005. 56 p. </SPAN></P><P><SPAN /></P></DIV></DIV></DIV>

Service Item Id: 5e9f78ef6bfe45de961ceceb57e28ee1

Copyright Text: Data created by NIWA under contract to Department of Conservation. This feature complied by Nicholas Dunn and Julian Sykes - Department of Conservation and NIWA respectively.

Default Visibility: true

MaxRecordCount: 2000

Supported Query Formats: JSON, geoJSON, PBF

Min Scale: 0

Max Scale: 0

Supports Advanced Queries: true

Supports Statistics: true

Has Labels: false

Can Modify Layer: true

Can Scale Symbols: false

Use Standardized Queries: true

Supports Datum Transformation: true

Extent:
Drawing Info: Advanced Query Capabilities:
HasZ: false

HasM: false

Has Attachments: false

HTML Popup Type: esriServerHTMLPopupTypeAsHTMLText

Type ID Field: null

Fields:
Supported Operations:   Query   Query Attachments   Query Analytic   Generate Renderer   Return Updates

  Iteminfo   Thumbnail   Metadata