I don't know how you help someone improve their recall, but if you have them recall Information you already acquired in your previously collected data, you might obtain an idea as to whether some numbers are biased high or low.

If you are concerned that they may not want to answer yes/no questions honestly, you could look into the literature on "randomized response."

If you think larger farms may have different responses than smaller ones, you could do a simple random sample in each of two or more subpopulations based on farm size. I would not call that stratified random sampling because that would not indicate you want answers by group/category/subpopulation, and you probably would want that. In stratified random sampling, you want to reduce the sample size needed to obtain overall results, but you may have inadequate sampling in any given stratum to report for that stratum reliably, for itself.

...................

For the future, after determining best practices, if you instead just want to monitor total production from some point forward, you could consider a stratified random sample, by size. It may be best to make the stratum for the largest farms a certainty stratum (i.e., census the largest stratum) to more efficiently obtain estimated totals with good accuracy.

Better, for production totals, if an occasional census of the population is feasible, one might use a form of 'cutoff' sampling with a ratio model to infer to the entire population for intervening periods. Then you could consider the following notes I've made.

Subject: 

'Cutoff' sampling in Official Statistics

The following paper is with regard to the accurate use of 'cutoff' sampling for Official Statistics from establishment surveys.  It has been proven highly effective at the US Energy Information Administration (EIA), using ratio modeling.  (For example: retail electric sales volume by economic end-use sector.) 

https://www.researchgate.net/publication/374294842_Stratification_and_other_Issues_for_Best_US_Federal_Establishment_Survey_Quasi-Cutoff_Sampling

Also, especially for repeated establishment surveys, consider: 

"Application of Efficient Sampling with Prediction for Skewed Data," JSM 2022: 

https://www.researchgate.net/publication/362370770_Application_of_Efficient_Sampling_with_Prediction_for_Skewed_Data ,

which includes comments on

Guadarrama, Molina, and Tillé(2020), Survey Methodology, which found that regression modeling appeared to perform better than calibration in small domains, for cutoff sampling, in their excellent article found at https://www.researchgate.net/publication/342657185_Small_area_estimation_methods_under_cut-off_sampling.