Histogram-based gradient boosted regression tree model of mean ages of shallow well samples in the Great Lakes Basin, USA

Dates

Publication Date: 2024-03-20
Start Date: 1890
End Date: 2020

Citation

Kauffman, L.J., Green, C.T., Ransom, K.M., and Ha, W.S., 2024, Histogram-based gradient boosted regression tree model of mean ages of shallow well samples in the Great Lakes Basin, USA: U.S. Geological Survey data release, https://doi.org/10.5066/P9LFX0XP.

Summary

Green and others (2021) developed a gradient boosted regression tree model to predict the mean groundwater age, or travel time, for shallow wells across a portion of the Great Lakes basin in the United States. Their study applied machine learning methods to predict ages in wells using well construction, well chemistry, and landscape characteristics. For a dataset of age tracers in 961 water samples, the mean travel time from the land surface to the sample location (center of saturated open interval) was estimated for each sample using parametric functions. The mean travel times were then modeled using a gradient boosting machine algorithm with cross validation tuning of model hyperparameters. The model contained in this data release [...]

Summary

Green and others (2021) developed a gradient boosted regression tree model to predict the mean groundwater age, or travel time, for shallow wells across a portion of the Great Lakes basin in the United States. Their study applied machine learning methods to predict ages in wells using well construction, well chemistry, and landscape characteristics. For a dataset of age tracers in 961 water samples, the mean travel time from the land surface to the sample location (center of saturated open interval) was estimated for each sample using parametric functions. The mean travel times were then modeled using a gradient boosting machine algorithm with cross validation tuning of model hyperparameters. The model contained in this data release was converted from the original model in the R language developed by Green and others (2021) to a python-based histogram-based gradient boosting regression tree (HGBRT) model (Pedregosa and others, 2011). Conversion to python facilitate the model's use as a support model for a groundwater nitrate decision support tool by Juckem and others (2024). The hyperparameters of the HGBRT model were adjusted using a Bayesian optimization algorithm (Head and others, 2021), with a goal of producing similar results as the original model by Green and others (2021). A total of 72 predictor variables were used for model development, including basic well characteristics, soil properties, aquifer properties, hydrologic position on the landscape, recharge and evapotranspiration rates, water quality constituents, and land use. Model results indicate that the mean of the natural logarithm of mean groundwater age for the wells used to train and test the model is 3.39 ln(years) with a root mean square error (RMSE) of 0.76 ln(years) for the holdout data to the HGBRT model. The RMSE for the holdout data (0.76) is similar to the RMSE from the original model for holdout data (0.84) reported by Green and others (2021). When the simulated values from the HGBRT model are back transformed from log space, the mean groundwater age is 55.9 years with an RMSE of 35.4 years for the testing data (Green and others, 2021 do not report matching results; simulated ages are reported for predicted ages at 14,335 non-sampled wells). The increased relative RMSE for back transformed ages reflects increasing error as values increase in the untransformed values.
Aside from the overall HGBRT methods contained as part of a python script, this data release includes a self-contained model directory for recreating the HGBRT model published in this data release. Three directories are available within this data release that define: 1. python attributes and input predictor variables, 2. model input and 3. the output model. The output directory also includes a model object (age_ml_model.joblib) for the HGBRT model used to predict the natural logarithm of the mean groundwater age. This model object is used directly by the groundwater nitrate decision support tool by Juckem and others (2024).

Contacts

Point of Contact :: Paul F Juckem
Originator :: Leon J Kauffman, Christopher T Green, Katherine M Ransom, Wonsook S Ha
Metadata Contact :: Paul F Juckem
Publisher :: U.S. Geological Survey
Distributor :: U.S. Geological Survey - ScienceBase
USGS Mission Area :: Water Resources
SDC Data Owner :: Upper Midwest Water Science Center

Attached Files

sorting

Metadata First

Recent First

Alphabetical

Click on title to download individual files attached to this item.

HGBR_groundwater_age.xml Original FGDC Metadata	View	127.92 KB	application/fgdc+xml
DataDictionary_AgeML.xlsx		25.79 KB	application/vnd.openxmlformats-officedocument.spreadsheetml.sheet
python-attributes.zip		24.3 MB	application/zip
ndst_pyml_opt.yml		486 Bytes	text/plain
ndst_pyml_opt.txt		25.76 KB	text/plain
model.zip		292.37 KB	application/zip
output.zip		983.06 KB	application/zip
readme.txt		13.93 KB	text/plain
modelgeoref.txt		786 Bytes	text/plain

Related External Resources

Type: Related Primary Publication

Juckem et al., 2024, Design and calibration of a Nitrate Decision Support Tool for groundwater wells in Wisconsin, USA. Environmental Modeling and Software	https://doi.org/10.1016/j.envsoft.2024.105999

Type: Source Code

Schachter, et al., 2024, A Groundwater Nitrate Decision Support Tool (GW-NDST) for the State of Wisconsin, version 1.1.0. U.S. Geological Survey Software Release	https://doi.org/10.5066/P13ETB4Q

Purpose

A machine learning model (histogram-based boosted regression tree) was developed to predict the mean groundwater age of wells (in years) for the State of Wisconsin, in support of the Groundwater Nitrate Decision Support Tool (GW-NDST) framework.

Histogram-based gradient boosted regression tree model of mean ages of shallow well samples in the Great Lakes Basin, USA

Dates

Citation

Summary

Summary

Contacts

Attached Files

sorting

Metadata First

Recent First

Alphabetical

Purpose

Map

Spatial Services

Communities

Associated Items

Tags

Provenance

Additional Information

Identifiers

Item Actions

View Item as ...

Save Item as ...

View Item...

Dates

Citation

Summary

Summary

Contacts

Attached Files sorting Metadata First Recent First Alphabetical

Related External Resources

Purpose

Map

Spatial Services

Communities

Associated Items

Tags

Provenance

Additional Information

Identifiers

Item Actions

View Item as ...

Save Item as ...

View Item...

Attached Files

sorting

Metadata First

Recent First

Alphabetical