GitHub

Fit with Baseline Correction

Peak fitting in fit_peaks includes a polynomial baseline by default, but for strongly curved backgrounds you may need to correct the baseline first.

Built-In Baseline

fit_peaks fits peaks on top of a polynomial baseline. Control the polynomial order with baseline_order:

using OpticalSpectroscopy
using CurveFitModels

x = collect(range(1900, 2200, length=500))
y = lorentzian([0.4, 2060.0, 20.0, 0.0], x) .+ 0.02 .+ 0.004 .* randn(length(x))

# Linear baseline (default)
result = fit_peaks(x, y, (1900, 2200))

# Constant baseline (flat offset)
result = fit_peaks(x, y, (1900, 2200); baseline_order=0)

# Quadratic baseline
result = fit_peaks(x, y, (1900, 2200); baseline_order=2)

Access the fitted baseline:

bl = predict_baseline(result)       # baseline on fit region
bl_custom = predict_baseline(result, x)  # baseline on custom x

Pre-Correction for Difficult Baselines

When the background is too complex for a low-order polynomial (e.g., fluorescence in Raman, broad solvent absorption in FTIR), correct the baseline before fitting.

Using correct_baseline

bl = correct_baseline(x, y; method=:arpls, λ=1e6)
# Returns a NamedTuple: (x, y, baseline) — `bl.y` is the corrected signal.

result = fit_peaks(bl.x, bl.y, (1900, 2200))

Using find_peaks with Baseline

Peak detection can apply baseline correction internally:

peaks = find_peaks(x, y, baseline=:arpls)

This affects which peaks are detected but does not change the data passed to fit_peaks.

Combining Both

For the best results with challenging data:

# 1. Correct baseline for accurate peak detection
peaks = find_peaks(x, y, baseline=:arpls, min_prominence=0.05)

# 2. Fit on original data with polynomial baseline
result = fit_peaks(x, y, (1900, 2200); peaks=peaks)

This uses baseline-corrected data for detection (finding the right peaks) but fits the original data (avoiding artifacts from baseline subtraction).

See Also