Innovative techniques such as acoustic Doppler current profiler (ADCP) allow for seamless mapping of fine-scale instream habitat conditions in shorter time. Such detailed information can be used to quantify the importance of habitat heterogeneity (or environmental diversity) which is believed to be important in ecology but difficult to quantify.

     Habitat suitability modelling is a way to illustrate how a target species response to a given habitat condition. While we observed tremendous development in modelling approaches using predictive machine, there still exist some issues in data quality affecting model performance and habitat information retrieved from the models. Given the aforementioned advanced techniques, it should now be possible to use field-observed large and precise instream habitat data.

     This study aims to demonstrate how ADCP data can be used for modelling microhabitats of freshwater fish in a small stream in Japan. Instream habitat survey was conducted using an ADCP equipped with RTK GPS in 75-m reach, based on which 2D maps of water depth and depth-averaged flow velocity were created. Fish were sampled at three points in a cross section arranged at 3-m intervals longitudinally, resulted in totaled 75 sampling points. At each sampling point, instream habitat conditions within 1.5 m radius were extracted using a GIS software. Of these data including 13 fish species, habitat models were constructed using machine learning-based approaches for its predictive capability. Results revealed the microhabitat uses by different species with different body sizes. Model accuracy can be discussed with reference to data quality.