Table 2.

Forward variable selection process for the GWR models. The global component of the four fitted GWR is shown. Differences AIC values and Akaike weights (w_i) are shown for each model, in comparison with the best ranking model, both from the global regression (ΔAIC[GR]; w_i [GR]) and from the GWR model (ΔAIC[GWR]; w_i [GWR]). Global estimated coefficients are presented for all included variables, along with the p values (p; starred p values are significant at 0.05 level). Response variable: residuals from the linear model (detrended_residuals). The first step, considering an intercept-only model, is used as null model.

Variables	Estimate	p	ΔAIC [GR]	wi [GR]	ΔAIC [GWR]	wi [GWR]
Step 1: detrended_residuals ~ 1			35.41	1.60×10−8	162.46	5.28×10−36
Intercept	0.00
Step 2: detrended_residuals ~ pc.frag1			15.09	4.13×10−4	282.37	4.83×10−62
Intercept	0.00
pc.frag1	0.08	*<0.01
Step 3: detrended_residuals ~ pc.frag1 + c.fnp			7.14	0.02	255.74	2.92×10−56
Intercept	− 0.06
pc.frag1	0.08	*<0.01
c.fnp	0.01	*<0.01
Step 4: detrended_residuals ~ pc.frag1 + c.fnp + c.fpl			2.75	0.20	228.05	3.02×10−50
Intercept	− 0.10
pc.frag1	0.08	*<0.01
c.fnp	0.01	*<0.01
c.fpl	1.92	*0.01
Step 5: detrended_residuals ~ pc.frag1 + c.fnp + c.fpl + temp_jul			0.00	0.78	0.00	1.00
Intercept	− 0.62
pc.frag1	0.08	*<0.01
c.fnp	0.01	*<0.01
c.fpl	1.95	*0.01
temp_jul	0.10	*0.03