Hydrologic landscape regionalisation using deductive classification and random forests.

Area of Science:

• Ecohydrology
• Geospatial analysis
• Environmental science

Background:

• Landscape classification and hydrological regionalization are crucial for aquatic resource management and research.
• Existing methods often rely on predefined spatial units (e.g., catchments), leading to loss of internal variability.
• There is a need for classification methods that retain fine-scale hydrological variability across diverse spatial scales.

Purpose of the Study:

• To develop and present a novel methodology for classifying hydrologic landscapes using spatial environmental variables.
• To overcome limitations of previous approaches by avoiding reliance on arbitrary or predefined spatial units.
• To create a flexible and accurate regionalization framework applicable from local catchments to continental scales.

Main Methods:

• Employed non-parametric statistics and hybrid image classification (random forests) for landscape classification.
• Utilized statistical clustering to identify an optimal number of distinct hydrologic groups (23 identified).
• Extended the classification across ~228,000 km² of south-eastern Australia without predefined spatial units.

Main Results:

• Achieved highly accurate hydrological regionalization at 30-m and 2.5-km resolutions.
• A 10-km resolution classification was less accurate but suitable for continental-scale applications.
• A case study confirmed the method's ability to preserve intra- and inter-catchment hydrological variability.

Conclusions:

• The developed method simplifies existing ecohydrological classification frameworks.
• It effectively retains small-scale hydrological variability, crucial for understanding broader trends.
• The approach shows promise for predicting streamflow in ungauged catchments, advancing ecohydrology research.