CS34 River Dynamics
Date: Wednesday, June 12, 2002
Time: 4:15:00 PM
Location: View Royal
 
JeongKS, Pusan National University, Pusan, Republic Of Korea, pow5150@hananet.net
Kim, H, W, Sunchon National University, Sunchon, Republic Of Korea, hwkim@sunchon.ac.kr
Joo, G, J, Pusan National University, Pusan, Republic Of Korea, gjjoo@hyowon.pusan.ac.kr
 
Application of machine learing to zooplankton dynamics in a regulated river ecosystem: patterning and predicting cladoceran dynamics
image
In this study two types of artificial neural network algorithms were applied to cladoceran dynamics in a river-reservoir hybrid system. Self-Organizing Map(SOM) and Time-Delayed Recurrent Neural Network (TDRNN) were adapted to cladocerans to recognize time-series patterns and prediction. Weekly samples of limnological parameters, phytoplankton and zooplankton were obtained for four years (1994-97) from the lower Nakdong River in S. Korea. This river is a good example of river-reservoir ecosystem which exhibited a eutrophic status (average Chl. a for four years, 50.2 ug/L). River hydrology were highly controlled by multi-purpose dams and an estuarine barrage, and characterized by typical monsoon and typhoon rains during summer (June-August). Five species of cladocerans (Bosmina longirostris, Bosminopsis deitersi, Diaphanosoma brachyurum, Moina macrocopa, and M. micrura) were abundant in the lower part of the river (over 95% of total individuals). Six clusters were identified by means of SOM algorithm, which consisted of presence and absence of each species (quantization error, 0.217; topological error, 0.014). Mainly evaporation, discharge, water temperature, and chl. a were important variables on the changes of major cladoceran abundances. TDRNN was well trained (RMSE under 0.005), and sensitivity analysis showed clear patterns of species succession by those four variables. Machine learning techniques could suggest good information about the factors on cladoceran species changes in river-reservoir systems from time-series data. Regional river characteristics such as climate (concentrated rainfall during summer) as well as complex ecological interactions were shown to be important to cladoceran dyanmics. As the number of regulated river-reservoir system increased rapidly, adequate analyzing methodologies such as neural network enable to find good solution for ecosystem studies.