I have a presence/absence data for a species occurence and I did a binomial GAM using sea surface temperature (sst) as a predictor variable.
My df object is a data.frame with the presence/ansence data for the species (0/1), the date and the coordinates of each point (WGS84) and the correspondent sst extracted from satellite observations.

gam<-mgcv::gam(sfr_presence ~ s(sst), family=binomial, data=df, method="REML")

Family: binomial 
Link function: logit 

Formula:
sfr_presence ~ s(sst)

Parametric coefficients:
            Estimate Std. Error z value Pr(>|z|)    
(Intercept) -3.17801    0.05595   -56.8   <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Approximate significance of smooth terms:
         edf Ref.df Chi.sq p-value    
s(sst) 5.853  6.862    553  <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

R-sq.(adj) =  0.0385   Deviance explained = 9.54%
-REML = 5886.8  Scale est. = 1         n = 25297

plot(gam,trans=plogis)

sst fitted function with plogis transformation

I’m trying to predict the probability of occurence of this species for a new dataset, but the values that result from the predict function are very low compared with the scale of the fitted function of sst from the gam.
The new dataset includes sst data from the same source and for the same area considered for model fitting.

pred<-mgcv::predict.gam(gam,se.fit=T,newdata=df_test, type="response",backtransform=T)

summary(pred$fit)

Min.    1st Qu.  Median    Mean 3rd Qu.    Max.    NA's 
0.02190 0.09596 0.11609 0.10250 0.11829 0.12178      21

summary(df_test$sst)

Min.   1st Qu.  Median   Mean   3rd Qu.  Max.    NA's 
19.60   21.24   21.95   21.99   22.72   24.43      21 

If the average of sst in the new dataset is around 22ºC the average prediction shouldn´t be around 0.7?