Statistical predictions of groundwater levels and related spatial diagnostics for a hydrogeologic framework of the Chicot aquifer system, southwestern Louisiana, 1983–2023, from the mmlCHICOTla statistical research software

Results of predicted groundwater levels and related spatial diagnostics, produced by an applied prediction-modeling statistical-research workflow, are provided as multi-banded GeoTIFF rasters for two custom bundles (classifications) of local aquifer codes for many thousands of wells in the hydrogeologically complex Chicot aquifer system, southwestern Louisiana. The modeling workflow was centered on application of the covCHICOTla (data preprocessing) and mmlCHICOTla (statistical predictions of water levels and diagnostics) research software (Asquith and Killian, 2025a,b). The mmlCHICOTla software creates and operates three forms of statistical learning methods (cubist, random forest, and support vector machine) for prediction of water levels and related quantities associated with statistical uncertainties at the 1-kilometer (km) grid scale using the National Hydrogeologic Grid (Clark and others, 2018). The three learning forms are combined in software to make final predictions and prediction diagnostics. The workflow of mmlCHICOTla, centered on the multiple methods of machine learning (MML), was operated and stewarded as the May 6, 2025, model run as documented within mmlCHICOTla (refer also to software path mmlCHICOTla/model_preserve/README.md). Extensive and comprehensive documentation, including citations to authoritative literature, is made within the software home pages of covCHICOTla (https://code.usgs.gov/map/gw/levs/altaqfrs/chicot/covCHICOTla) and mmlCHICOTla (https://code.usgs.gov/map/gw/levs/altaqfrs/chicot/mmlCHICOTla). 

Local aquifer codes represent a coding scheme used by the U.S. Geological Survey (USGS) National Water Information System (NWIS). Asquith and Lindaman (2025) provide comprehensive citations, data sources, data filtering decisions, and other details leading to wells being classified into either CHCTLYR1 or CHCTLYR2 classes suitable for statistical production of typical well-depth geospatial products provided in this data release. The CHCTLYR1 (combination of USGS NWIS local aquifer codes 112CHCTU, 11202LC, 112ACFL, 112CHCT, and 112UPTC) represents comparatively "shallow" Chicot aquifer system conditions and effectively includes the entire horizontal expanse of the aquifer. The CHCTLYR2 (combination of USGS NWIS local aquifer codes 112CHCTL, 11205LC, and 11207LC) represents comparatively "deep" aquifer conditions and is applicable for only part of the Chicot aquifer system expanse.

Predictions of water levels, as water-level altitudes or as depths to water, are intended to provide first-order general estimate of mean groundwater levels on a monthly basis for January 1983 to December 2023 (492 months) for the two classifications (CHCTLYR1 and CHCTLYR2). The MML used 4,620 wells having 12,239 monthly water levels (CHCTLYR1) and 612 wells having 3,042 monthly water levels (CHCTLYR2). Gridded statistical predictions as 492-layer GeoTIFF files for computational themes that include (1) water-level altitude ("pol"), (2) depth to water ("nhgd2w"), (3) two forms of model errors ("modnorerr" and "modpdqerr"), (4) ratio of model error to total error ("modtorat"), (5) two forms of total error (combined statistical and model) assuming no land-surface variation within 1-km grid cell ("norerr" and "pdqerr"), and (6) two forms of total error including land-surface variation within 1-km grid cell ("norerrnhg" and "pdqerrnhg"). In the software and data nomenclature, the normal probability distribution is abbreviated ("nor") and symmetrical polynomial quantile-density probability is abbreviated ("pdq") as the structural form of the unknown but estimable error distribution. The 90-percent prediction limits are represented by lower ("lwr") and upper ("upr") limits for both error-model distribution forms ("norlwr", "norupr" and "pdqlwr", "pdqupr") assuming no land-surface variation within 1-km grid cell. Lastly, the 90-percent prediction limits are represented by lower ("lwr") and upper ("upr") limits for both error-model distribution forms ("norlwrnhg", "noruprnhg" and "pdqlwrnhg", "pdquprnhg") including land-surface variation within 1-km grid cell. A total, therefore, of 13 computational themes for each CHCTLYR1 and CHCTLYR2 were produced having 492-layers for each month (1983–2023). The gridded predictions are the result of statistical research and algorithm development (2024–25) for the Chicot Aquifer system and are intended as science products to aid derivative studies of the hydrogeologic framework and hydrologic history of the aquifer.