Question: do groundwater levels in long-record US wells stay lower than precipitation predicts after multi-year droughts (non-recovery or a pumping ratchet)? Approach: we used USGS field measurements of depth to water (parameter 72019) for 5,993 wells with records from before 1976 to after 2022, water-year median levels, NOAA climate-division precipitation and GHCN_CAMS temperature. Droughts are 3 or more consecutive water years below the 1951-2025 division mean with a mean deficit of 10 % or more. A reference model (precipitation lags 0-3, temperature, linear or spline trend) was fitted on normal years only. The metric is the mean standardised residual R in the 5 post-drought years. A time-shift placebo (58 common shifts) gave the null distribution. Results: for 2,579 droughts bracketed by fitted years in 1,576 wells, the mean R was -0.08 SD (linear trend, placebo 95 % range -0.21 to +0.36, p = 0.12) and -0.17 SD (spline trend, -0.30 to +0.41, p = 0.12). Across 13 variants the mean R was -0.28 to +0.19 SD. The residual was positive during droughts, lowest in post-drought years 3-4 (-0.26 to -0.38 SD, at or below the placebo 2.5th percentile) and near 0 by year 6. Thus the response was delayed, not persistent. With equal weight per division the mean R was -0.18 SD (p = 0.034, linear) and -0.12 SD (p = 0.085, spline), approximately 0.3 ft. Injected deficits of 0.25 SD in all droughts were detected in 20 of 20 draws. Wells with long-term declines and High Plains wells did not have a larger deficit; no group passed a Benjamini-Hochberg correction. Limitations: division-scale precipitation, no pumping data, model-dependent sign, an uneven well network, and some design choices made after a preliminary run on two thirds of the wells.
Filters: anzsrc-for-2020: 370703 ×
Hurwitz et al. (2008) reported that annual mean eruption intervals of Old Faithful and four other Upper Geyser Basin geysers decrease when Madison River discharge increases (Old Faithful r = -0.98, 1967-2006), and that intervals increase seasonally after the snowmelt peak with a lag for each geyser. We did an out-of-sample test for 2007-2025 with the public GeyserTimes archive (1.54 million records) and USGS daily discharge. With GeyserTimes data for 1997-2006 we reproduced the annual correlation for Old Faithful (r = -0.99, 8 years). For 2007-2025 it was -0.26 (19 years; one-sided p = 0.27 against AR(1) surrogate discharge), and the correlation of year-to-year changes was +0.03. The 1997-2006 regression forecast the 2007-2025 July-September mean interval worse than the mean of the last three in-sample years (RMSE 4.5 vs 3.4 min). Daisy and Aurum changed sign; Castle was negative (-0.54) but had the opposite sign in 1997-2006. The in-sample correlation came mostly from the 1997-2001 loss of short Old Faithful eruptions (46% to 2%). In 2007-2025 the short fraction did not decrease in dry years; it reached 17-19% in the driest years, 2024-2025. The seasonal cycle recurred at the published lags (mean within-year r: Old Faithful +0.27, 16 of 20 years positive; Daisy +0.36, 17 of 19; Aurum +0.39, 10 of 13). But discharge of other years correlated equally well (Old Faithful 0.32 vs 0.28 for the same year; Daisy 0.44 vs 0.40), and the size of the Old Faithful seasonal increase did not scale with the snowmelt pulse after 2006 (r = +0.13, p = 0.30). Thus the data show a stable calendar cycle, but not control by the snowmelt of each year. Limitations: at most 19 out-of-sample years per geyser, changes in observation methods (loggers, visual, webcam), fewer recorded Old Faithful eruptions after 2019, river discharge as the only recharge index, no earthquake model, and many tested variants. A weak annual relation (about r = -0.3) is not excluded.