Heckman methodology and Cross-sectional data

Hi,
I am working on a cross sectional data that is of 3 years. (different observations in each year). I have data on 2005, 2008 & 2011.
I am basically tying to generate engel curves with the help of expenditures and incomes of household while controlling for various controls. Controls are same for each engel curve.
After estimating expenditures I came to know that there are many zero observation in an expenditure. i.e. travel expenditure, tobacco expenditure. I am applying Heckman for that to control for the zero expenditure.

I want to know why should I first run a dprobit regression and after that apply heckman command while controlling foe selectivity bias? what is the significance of the dprobit command?
I know the number of zero expendutres and they are coming to be above 10% of the data set. so why run the dprobit command first?

Also one important question is that can I run a twostep heckman methodology in a Cross-sectional setting? I fail to find any relevant articles on this approach.

Thankyou in advance
Amna Naveed