Best practices for field boundaries

Ensure optimal Monitor results with the following considerations for field boundaries.

Before submitting your fields to Monitor API, consider the following best practices to ensure optimal results:

1. Valid geoJSON geometry object. Monitor takes in field boundaries in geoJSON format, and can accept valid polygon and multi-polygon objects. Below are a few examples of boundaries that may result in unreliable results.
   
2. Small fields. Monitor API interprets satellite imagery to predict field practices. On smaller fields there is less available imagery to interpret, which can make results less reliable. While Monitor can technically support fields of any size, you may want to consider only submitting fields larger than one acre to optimize results (or otherwise review results from smaller fields to ensure the field data supports your use case).
3. Multi-polygons representing areas under different management practices. While multi-polygons are a valid geometry shape for Monitor, we suggest checking to ensure that any multi-polygons submitted are representing the same management practices. 
   Pro tip: It can be helpful to check that the polygons in multi-polygon fields are close in proximity, as polygons that are far apart from each other typically indicate different management areas or a mistake in the boundary.
   
   1. For example, if one area of a field is growing corn one season and the other portion is growing soybeans: those areas should be submitted to Monitor as separate field boundaries.
   2. A valid example of a multi-polygon for Monitor is a field that is divided by a stream all the way through, where the stream is excluded from the geometry.
4. Boundaries should represent a majority of agriculture land. Monitor is trained to detect agricultural practices. When a boundary is submitted with non-ag land, the results will be unreliable.
   1. A small amount of non-ag land may be ok (for example if a farm building is in the middle of the field but represents less than 5% of the total area).
   2. Larger areas of non-ag land will cause confusion in the models. For example, if 30% of a field boundary includes a border hedge row or natural lands it can cause the models to confuse the actual crop type grown on a field.