Add instructions

README.md

@@ -12,6 +12,10 @@ configs:
 
 Synoptic-Bench is used to train Vision-Language Models to generate text descriptions of weather forecast data. It contains 14,159 total GFS forecasts issued four times per day and spanning from 2016-2025 with lead times ranging from 3 hours to 168 hours in 3 hour intervals. Precipitation rate, 2-meter temperature, 850 mb zonal wind velocity, 850 mb meridional wind velocity, and 500mb geopotential height is included in the forecast data. There are 1,367,041 total Area Forecast Discussions (AFDs) matched to GFS forecasts.  
 
+## Data Preprocessing:
+
+AFDs are highly complex and contain text that widely varies in the types of weather phenomena, lead times, and spatial scales that are discussed. It is therefore the responsibility of the user to filter the text and create images that properly align with the research question. A tutorial for preprocessing the data is included in preprocess_tutorial.ipynb. The example_images folder contains samples of images and filtered AFDs created from the dataset.
+
 ## Data Sources:
 
 This dataset is derived from publicly available U.S. government data: