Your scattered data points must be fit to a surface. There are a number of ways to get a surface from scattered data, so you'll have to make some decisions.
First, convert lat and lon to some proper projected x and y coordinates. This is necessary because spacing of latitude lines does not equal spacing of longitude lines. Fitting to a surface will require some even spacing in x and y. If you have the Mapping Toolbox, convert to map x,y coordinates with the projfwd function. With your scattered data in x,y coordinates, you might want to try gridfit to generate a surface. The two big decisions you'll have to make with gridfit are
1. spatial resolution, which affects the time it takes to solve gridfit, and
2. smoothness parameter. Tuning the smoothness parameter may take some trial and error. It's essentially fitting a pliable plate to your scattered data points. The smoothing parameter defines how stiff that plate is.
After gridfit you can plot with surf, pcolor, imagesc, mesh, etc.