T51 The Science of Groundwater Recharge, Coal Mine Hydrology/Geochemistry, Stream Restoration and its Application to the Public Good: In Honor of Mary W. Stoertz

Sunday, 28 October 2007
1:30 PM - 5:30 PM

A NEW MODELING APPROACH EVOLVING FROM STOERTZ'S INSPIRATION: WATER BALANCE, PATTERN RECOGNITION AND INITIAL MODEL CONCEPTUALIZATION

Paper number: T51-12
Program: Technical Sessions
Time: 4:20 PM – 4:35 PM

Authors

  • Yu-Feng Lin Presenting Author
    Univ. of Illinois at Urbana-Champaign Illinois State Geological Survey, Champaign, IL
  • Randy Hunt
    US Geological Survey Upper Midwest Water Science Center, Middleton, WI
  • Albert Valocchi
    University of Illinois at Urbana-Champaign, Urbana, IL
  • Peter Bajcsy
    University of Illinois at Urbana-Champaign, Urbana, IL
  • Jihua Wang
    University of Illinois at Urbana-Champaign, Urbana, IL
  • Chulyun Kim
    University of Illinois at Urbana-Champaign, Urbana, IL

Abstract

Stoertz modified MODFLOW code to estimate groundwater recharge and discharge (R&D) for two and three dimensional groundwater models in steady state based on water balance in 1989. However, if the grid spacing is too small relative to the spacing between observations, the method will break down due to the sensitivity to the groundwater flow gradient. Lin and Anderson (2003) developed a R&D pattern recognition and rate estimation procedure using a sophisticated surface interpolation, image processing algorithms and parameter estimation to post-process the results from Stoertz's water balance code. This R&D pattern recognition and rate estimation procedure works well with smaller model grid spacing than previous studies, allowing for improvement in model accuracy and stability.

The pattern recognition and rate estimation procedure inspired by Stoertz's work has been integrated as GIS plug-in utilities which can be applied to other R&D estimation methods and areal image processing in other disciplines. Our group is currently advancing the pattern recognition approach by exploiting machine learning algorithms to visually and analytically discover R&D model relationships with prior knowledge represented in ancillary field information. This approach will enable hydrogeologists to recognize R&D pattern patterns with quantifiable reliability indices. These indices from pattern learning results will bridge the gap between the traditional subjective zonal pattern estimations and advanced stochastic and uncertainty analysis.

The foundation of model analysis is the conceptual model which is based on the subjective judgment of the analyst. Scientists have recognized the importance of a good conceptual model to the whole modeling exercise. The impact inspired by Stoertz's work has been carried further than R&D study in our group. To improve the initial conceptual model and better understand some of its limitations (even before the execution of numerical model simulations and calibrations), one of our future studies is to incorporate uncertainties due to natural unpredictability and measurement inaccuracy in order to understand spatial and temporal impacts and to improve conceptual models. This future research will develop algorithms and procedures which are novel in both computer and natural sciences.