Getting started¶

Install¶

The package is not on PyPI yet. Install it from GitHub:

pip install git+https://github.com/GeoPilot-AI/pumptest

It needs Python 3.11 or later, numpy and scipy.

Units: one consistent set¶

The solvers work in the units they are given, so lengths, times and rates must use one consistent set, and the results come back in it. Metres, days and cubic metres per day is the usual choice, and gives transmissivity in square metres per day. Convert field units first with pumptest.units:

from pumptest.units import convert, drawdown_from_levels

rate = float(convert(250, "lpm", "m3/d"))  # 250 litres per minute = 360 m3/d
minutes = [1, 2, 5, 10, 30, 60]
days = convert(minutes, "min", "d")
drawdown = drawdown_from_levels([12.31, 12.45, 12.70, 12.88, 13.20, 13.41], static_level=12.05)

convert knows metres, centimetres, millimetres, feet and inches; seconds, minutes, hours and days; and cubic metres, litres, US and imperial gallons and cubic feet per second, minute, hour or day (lpm, lps, m3/h, gpm, cfs and so on).

Describe the test¶

A test is a pumping well, a pumping schedule and one or more observations.

import numpy as np

from pumptest import Observation, PumpingTest, Schedule, Well

t = np.array([0.001, 0.002, 0.005, 0.01, 0.02, 0.05, 0.1, 0.2, 0.5])  # days
s = np.array([0.05, 0.11, 0.21, 0.29, 0.37, 0.48, 0.56, 0.64, 0.75])  # metres

test = PumpingTest(
    well=Well(radius=0.1),
    schedule=Schedule.constant(500.0),  # m3/d from time zero
    observations=[Observation("OW-1", distance=25.0, time=t, drawdown=s)],
    thickness=20.0,
)

distance=0 means the observation is in the pumping well itself.
Schedule.constant(500, stop=0.5) adds recovery from day 0.5; Schedule(((0, 300), (0.1, 600), (0.2, 900))) is a step test.
thickness is needed by the unconfined solution, and to report K = T/b.

Fit a model¶

from pumptest import Theis, fit

result = fit(test, Theis())
print(result.summary())
print(result.values["T"], result.standard_errors["T"])

result.values holds the estimates, result.standard_errors their one-sigma errors, result.derived quantities such as K, result.rmse the misfit in the drawdown unit, and result.residuals the misfit at each observation.

Hold a parameter at a known value with fixed, fit only some wells with observations, or give your own starting point with initial:

result = fit(test, Theis(), fixed={"S": 2e-4})

Compare models¶

from pumptest import HantushJacob, Neuman, compare

for r in compare(test, [Theis(), HantushJacob(), Neuman()]):
    print(f"{r.model:15} AICc {r.aicc:8.1f}   RMSE {r.rmse:.4f}")

compare ranks by AICc, which penalises every extra parameter: a leaky or unconfined model wins only when it fits enough better to earn its extra parameters.