Bewertung der Erfassungswahrscheinlichkeit für globales Biodiversitäts-Monitoring: Ergebnisse von Sampling GRIDs aus unterschiedlichen klimatischen Regionen

An assessment of sampling detectability for global biodiversity monitoring: results from sampling GRIDs in different climatic regions

von Dirk Nemitz
Masterarbeit
Datum der mündl. Prüfung:2008-12-05
Erschienen:2015-03-25
Betreuer:Prof. Dr. Michael Mühlenberg
Gutachter:Falk Huettmann
Gutachter:Prof. Dr. Christoph Kleinn
Zum Verlinken/Zitieren: http://dx.doi.org/10.53846/goediss-4966

 

 

Dateien

Name:Nemitz_MINC_Thesis1.pdf
Size:3.86Mb
Format:PDF
Description:Dirk Nemitz MINC Thesis 2008

Lizenzbestimmungen:

Zusammenfassung

Englisch

This thesis provides important input for the development of a cost-effective global biodiversity assessment and monitoring system. The study is embedded in a larger project to evaluate possibilities of multiple-species surveys using biodiversity GRIDs. As a pilot study six GRIDs in diverse ecosystem settings are sampled. Sampling methods used for animal species are point transects for birds and trapping webs for arthropods; additionally a line transects add-on protocol is used at some study areas for amphibians, reptiles and butterflies. Within this framework the task is taken over to develop predictive models for sampled animal species with Random Forests. Additionally the data is analyzed to derive abundance estimates with multiple covariate DISTANCE sampling and occupancy estimates through the software PRESENCE.  A total of 5,007 observations from six study areas from all over the world are analyzed in detail. Total sampling time is about 12 weeks. High quality non-random predictive models with a ROC value > 0.5 are gained with Random Forests analysis for 116 described animal narratives. Half of these observations origin from point transect sampling, the other half from trapping web catches. The line transects add-on protocol results in another 3 predictive models. Abundance and occupancy estimates are derived from the data for 46 animal narratives, 23 of those for point transect data, 22 for trapping web data, and 1 for line transect data. Predictive modeling with Random Forests proves to be a very powerful tool. DISTANCE sampling estimates from this study show large confidence interval ranges, but are extremely cost-efficient to gather initial information for multiple species rapidly. PRESENCE estimates are partly unsatisfying because of a large portion of animal narratives with perfect occupancy estimates (Psi = 1.0). It is assumed that this is an effect of small sampling size which will not be problematic for larger amounts of data. This has to be kept in mind when comparing DISTANCE and PRESENCE results. Correlation between DISTANCE and PRESENCE detection probability estimates is negative, while correlation between DISTANCE abundance estimates and PRESENCE occupancy estimates is positive for all but one study area. It is recommended to repeat the comparison when data from more plots is available. On one hand the results, the cost-effectiveness of the study, and possibilities opened by this kind of multiple-species multi-method sampling are promising, on the other hand funding for this visionary approach was not available.
Keywords: Alaska; Costa Rica; Nicaragua; Papua New-Guinea; Russia; Biodiversity monitoring; Sampling GRID; DISTANCE; PRESENCE; Random Forests
 
Statistik