Using in situ Sensor Networks and Remotely Sensed Imagery to Improve Site-Specific Understanding of Maize Yield Variance

Crop simulation models, like DSSAT, allow for the exploration of crop performance under a variety of soil, climate, and management scenarios. While crop model results are imperfect and often represent idealized conditions, a main advantage of crop models is that they can be run at a scale not feasible for actual crop trials. Crop model simulations can also aid in evaluating management strategies under different soil and climate conditions.

Earth Observation (EO) data sets can improve crop modeling in several ways. Directly, EO data sets can be used as soil and climate inputs in crop models. Indirectly, EO data sets can be used to calibrate crop models, including calibrating cultivar coefficients, and providing realistic measurements of crop phenology, like biomass and leaf-area index (LAI). Satellite imagery has also been used together with crop model simulations in an empirical model to predict yield, the Scalable Crop Yield Mapper (SCYM).

This project focuses on how in situ pod sensors (Arable Marks) and remotely sensed data sets (Sentinel-2, Landsat) can complement crop model simulations used for yield variance decomposition. Pod sensors provide temporally dense measurements of crop phenology (e.g. vegetation indices, LAI). These pod sensor measurements can be integrated with spatially dense remotely sensed measurements of the same metrics.

The integration of pod sensor and satellite data, together with field measurements of LAI, and other imagery sources (e.g. UAV), allows for crop phenology to be measured at multiple scales. This multi-scale approach provides a richer base of crop phenology observations, which can be used to calibrate crop models to local soil and climate conditions. The locally tuned crop models can then be used to study how different management strategies affect yield variance.