Replac static crop mask with a dynamic one

[ellaampy]

Global open-source dataset of monthly irrigated and rainfed cropped areas (MIRCA-OS) for the 21st century

[michielkallenberg]

As requested by a reviewer of our paper:

Static crop masks: Crop masks play a crucial role in processing yield predictors, as they help eliminate irrelevant pixels and reduce noise in the data. However, this study applies a static crop mask across multiple years, which is problematic. This approach does not align with the standards typically expected in benchmarking studies.

We acknowledge that dynamic crop masks could improve the accuracy of yield predictions by better reflecting spatial-temporal variability, as we also mentioned in the limitations section of the manuscript. However, the availability of consistent, high-quality, dynamic crop masks at the global scale remains limited — a constraint shared by many yield prediction and benchmarking efforts (Kebede et al., 2025). To the best of our knowledge, several established benchmark datasets
similarly rely on static or infrequently updated crop masks, as dynamic alternatives are often unavailable for large parts of the world (Yeh et al., 2021).

That said, CY-Bench is designed as a living resource that will evolve as the research community develops or identifies improved datasets including higher-quality crop masks, thematic predictors, and expert-curated features that can enhance subnational yield prediction.

Looking ahead, two options exist for incorporating dynamic crop masks:

1. regional, high-resolution products — such as the USDA Cropland Data Layer for the U.S. (Boryan et al., 2011) which would improve local accuracy but would also break the harmonization of the dataset, as these region-specific masks would not be applicable globally. This presents a challenge, as one of our goals is to enable fair intercomparisons between models across diverse regions.
2. the very recently released MIRCA-OS crop dataset (Kebede et al., 2025), which provides global crop masks at 5-year intervals. While this resource offers an exciting opportunity, its relatively coarse spatial resolution and recency mean we will first evaluate its suitability before integrating it into CY-Bench.