DEVELOPMENT OF PATTERN RECOGNITION AND LEARNING UTILITIES FOR SHALLOW GROUNDWATER RECHARGE AND DISCHARGE ESTIMATION IN SEMI-HUMID REGIONS USING MULTIPLE DATA SOURCES

Groundwater recharge and discharge (R&D) rates and patterns are difficult to characterize, and currently no single method is capable of estimating R&D rates and patterns for all practical applications. Therefore, cross analyzing results from various estimation methods and related field information will likely be superior to using only a single estimation method.

We have developed a GIS plug-in utility, called the Pattern Recognition Utility (PRU), to help hydrogeologists estimate R&D in a more efficient way than conventional methods. The PRU uses numerical methods and image processing algorithms to estimate and visualize shallow R&D patterns and rates with GIS. The PRU includes (but is not limited to) a default R&D estimation code using a finite difference mass balance approach in 2D and steady state. This default R&D estimate code only requires data for water table, bedrock elevations and hydraulic conductivities. It can provide a fast initial estimate prior to planning labor intensive and time consuming field R&D measurements.

Another JAVA based software package, called Spatial Pattern 2 Learn (SP2L), was developed to cross analyze results from the PRU with ancillary field information, such as land coverage, soil type, topographic maps and previous estimates. The learning process of SP2L cross examines each initially recognized R&D pattern with the ancillary spatial dataset, and then calculates a quantifiable reliability index for each R&D map using a supervised machine learning technique called decision tree. The SP2L is capable of generating alternative R&D maps if the user decides to apply certain conditions recognized by the learning process. The reliability indices from SP2L will improve the traditionally subjective approach to initiating conceptual models by providing objectively quantifiable conceptual bases for further probabilistic and uncertainty analyses. Both the PRU and SP2L have been designed to be user-friendly and universal utilities for pattern recognition and learning to improve model predictions from sparse measurements by computer-assisted integration of spatially dense geospatial image data and machine learning of model dependencies.