---
id: po.wb.dca.decline-analysis
title: Decline Curve Analysis
summary: "How much oil one well will recover, from its monthly rate history through four decline fits to an EUR, a booked model and its sensitivity."
category: Production
requires:
  - po-addin
  - excel-365
tags:
  - decline-curve-analysis
  - Arps
  - modified-hyperbolic
  - Duong
  - ple
  - stretched-exponential
  - data-cleaning
  - outlier
  - weighted-fit
  - model-selection
  - history-match
  - forecast
  - terminal-decline
  - economic-limit
  - EUR
  - reserves
  - sensitivity
  - tornado
---

## Description

How much oil will this well recover, and how far can that number move? You paste one well's monthly oil rates; the workbook screens out the months that disagree with their neighbours, fits Arps, Duong, the power-law exponential (PLE) and the stretched exponential (SEDM) to the cleaned history, and forecasts each to an economic limit. You read off each model's EUR and life, the model to book, and how much each decline parameter moves the EUR. Field units; time in months, rates in STB/d.

**How to use it:** Enter monthly volumes over days on production on the History sheet, from month 1 down with no gaps (up to 36 months). Check every flagged month against the field record (allocation, downtime, tests) before trusting its cleaned rate. On the Forecast sheet, take the terminal decline from older analogue wells and the economic limit from the operating cost over the netback, and pick the model to book. On the Sensitivity sheet, take the low and high values from the fit's spread or from analogues. The Summary gathers the answers and the checks.

**Method:** The screen flags a month whose rate lies more than a set number of standard deviations from its rolling window, and interpolates over it. Each model is fitted by Nelder-Mead to minimize the weighted log error sum w (ln q − ln q_fit)² over the typed months, with month t of a history ending at T weighed (t/T)^p; the History sheet's two Arps fits are unweighted. Arps q = qi/(1 + b Di t)^(1/b); Duong q = q1 t^−m exp(a/(1 − m) (t^(1−m) − 1)) + q∞; PLE q = qi exp(−D∞ t − Di t^n); SEDM q = qi exp(−(t/τ)^n). The secant effective decline is 1 − (1 + b Di)^(−1/b) with Di per year; the modified hyperbolic switches to an exponential at the terminal decline at t = (Di/Dlim − 1)/(b Di). Each model's EUR and life are its cumulative and time to the economic limit, the cumulative in STB/d·month times 30.4375 days a month. The tornado moves one input at a time; the low and high estimates move all four at once.

**Example:** The decline family's reference well (24 months from 895 to 199 STB/d) with three bad months typed over it: month 7 at 1,150 STB/d, 14 at 120 and 19 at 480. A 7-month window at 3 standard deviations flags exactly those three, 12.5 % of the months, and cleans them to 492, 309.5 and 239.5 STB/d. Arps fitted to the raw rates gives qi 1,060.4 STB/d, Di 0.15102/month and b 0.843; fitted to the cleaned rates, 1,013.1, 0.15034 and 0.882, with an RMSE of 2.8 % of the mean rate. Cleaning moves the EUR from 663.3 to 680.7 MSTB, +2.6 %, inside the 5 % allowed. With linear late-time weights the four fits score 0.007677 (Arps: qi 1,024.4, Di 1.858/yr, b 0.900), 0.008257 (Duong: q1 822.7, a 1.423, m 1.327), 0.008781 (SEDM: qi 1,556.1, τ 4.80 months, n 0.448) and 0.008795 (PLE: qi 1,508.0, Di 0.4725, n 0.458, with D∞ at its lower bound of zero). Arps is best by 7.6 %, and every fit matches the history within 5.0 % of the mean rate. The Arps fit declines 66.4 %/yr (secant) in its first year, to 343.7 STB/d. With the terminal decline at 8 %/yr it switches after 12.7 years at 32.6 STB/d. To 15 STB/d the EURs are 664.5 MSTB modified hyperbolic (22.0 years, the booked model), 694.1 Arps (26.2), 709.5 Duong (27.6), 527.5 SEDM (12.3) and 522.1 PLE (12.0). They lie 36 % apart, more than the 25 % allowed, so book a range. On the Arps fit, b moves the EUR most: 53 % from 0.7 to 1.0, more than the 40 % one input may carry. The low and high estimates are 336.5 and 1,206.4 MSTB, with the base case between them.

## Validity

Decline models are empirical and hold at constant operating conditions; a workover, choke change or interference starts a new decline. A rolling Z-score can't tell a reporting error from a real event, and a run of bad months longer than half the window hides itself. The weighted fits favour the late months; on 24 months of a tight well's transient flow they agree within 5 % on the history and still differ by a third in EUR, which is the EUR spread the Forecast sheet reports. On this history the PLE fit takes D∞ to its lower bound of zero, so its forecast is close to SEDM's; a fixed D∞ needs a hand fit. Arps with b above 1 is transient and its EUR rests on the limit alone: book the modified hyperbolic. The sensitivity is Arps only and moves one input at a time, ignoring that a fit's qi, Di and b move together; the low and high estimates are deterministic scenarios, not P90/P10. Alternatives: rate-transient analysis or flowing material balance where pressures are measured, type wells from offsets, a Monte Carlo on the fit's covariance, AIC or BIC to penalize Duong's fourth parameter.

## References

- Arps, J.J. (1945). Analysis of Decline Curves. *Trans. AIME* 160, 228-247.
- Fetkovich, M.J. (1980). Decline Curve Analysis Using Type Curves. *JPT* 32(6), 1065-1077.
- Robertson, S. (1988). Generalized Hyperbolic Equation. SPE-18731-MS.
- Ilk, D., Rushing, J.A., Perego, A.D. and Blasingame, T.A. (2008). Exponential vs. Hyperbolic Decline in Tight Gas Sands. SPE-116731-MS.
- Valko, P.P. and Lee, W.J. (2010). A Better Way to Forecast Production from Unconventional Gas Wells. SPE-134231-MS.
- Duong, A.N. (2011). Rate-Decline Analysis for Fracture-Dominated Shale Reservoirs. *SPE Res Eval & Eng* 14(3), 377-387.
- SPEE (2016). *Monograph 4: Estimating Ultimate Recovery of Developed Wells in Low-Permeability Reservoirs*.

## Inputs

| Input | Unit | Default | Name |
|---|---|---|---|
| Monthly oil rate, months 1 to 36 | STB/d | reference well (24 months) with 3 bad months | `HIS_Q` |
| Rolling window, months (at least 3) | months | 7 | `HIS_Window` |
| Standard deviations beyond which a month is an outlier | — | 3 | `HIS_NSigma` |
| Largest share of months flagged | % | 15 | `HIS_FlagMax` |
| Largest EUR change cleaning may make | % | 5 | `HIS_ShiftMax` |
| Late-time weight exponent (0 weighs every month alike) | — | 1 | `FIT_WPow` |
| Fewest months a fit should rest on | months | 12 | `FIT_MinMonths` |
| Largest history mismatch allowed (RMSE over mean rate) | % | 10 | `FIT_FitMax` |
| Economic-limit oil rate | STB/d | 15 | `FCS_QEcon` |
| Terminal decline, effective | 1/yr | 0.08 | `FCS_Dlim` |
| Lowest and highest usual terminal decline, effective | 1/yr | 0.05, 0.10 | `FCS_DlimLo`, `FCS_DlimHi` |
| Largest EUR spread to book without a range | % | 25 | `FCS_SpreadMax` |
| Grid step, months per row | months | 7 | `FCS_Step` |
| Booked model (from a list) | — | Modified hyperbolic | `FCS_Book` |
| Initial rate, low and high | STB/d | 920, 1,120 | `SNS_qiLo`, `SNS_qiHi` |
| Initial decline, low and high | 1/yr | 1.6, 2.2 | `SNS_DiLo`, `SNS_DiHi` |
| Hyperbolic exponent, low and high | — | 0.7, 1.0 | `SNS_bLo`, `SNS_bHi` |
| Economic limit, low and high | STB/d | 10, 25 | `SNS_QEconLo`, `SNS_QEconHi` |
| Largest EUR swing one input may carry | % | 40 | `SNS_SwingMax` |

## Sheets

- `Guide` — what the workbook answers and the order to work through its sheets.
- `History` — the open table of months, a rolling Z-score that flags the months that disagree with their neighbours, the cleaned rates, and Arps fitted to the raw and the cleaned rates with each fit's EUR.
- `Fits` — Arps, Duong, PLE and SEDM fitted to the cleaned months in log rate with late-time weights, each fit's weighted log error and RMSE, and the best fit by the lowest error.
- `Forecast` — the Arps fit's secant and tangent effective declines and its switch to a terminal decline (the modified hyperbolic); the five models' 10-year cumulative, EUR and life to one economic limit with their spread; the model to book, picked from a list; a 35-year forecast grid with rate and cumulative charts.
- `Sensitivity` — the Arps EUR with a low and a high qi, Di, b and limit, a tornado, the low, best and high estimates, and sweeps of b, Di and the limit.
- `Summary` — the answers and every check under one verdict.
- `Test Cases` — the worked example and other cases, with expected and computed results.

Seven charts: raw and cleaned rates with both Arps fits; the four weighted fits on the cleaned history; five models' rates to the limit; their cumulatives; the EUR tornado; the EUR against b; the EUR against the limit.

## Functions used

`PO.DCA.Data.OutlierFlag`, `PO.DCA.Data.Clean`, `PO.DCA.Arps.Fit`, `.Arps.WFit`, `.Duong.WFit`, `.PLE.WFit`, `.SEDM.WFit`, `PO.Stats.RMSE`, `PO.Stats.WLogSSE`, `PO.DCA.Arps.Rate`, `.Prod`, `.EUR`, `.Time`, the same four for `ModHyp`, `Duong`, `PLE` and `SEDM`, `PO.DCA.Convert.NominalToEffective`, `.EffectiveToNominal`, `PO.DCA.Arps.SecantEff` (LAMBDA) and `PO.DCA.ModHyp.Tswitch` (LAMBDA).
