Improving Accuracy of Quantitative Precipitation Estimates at The Climate Corporation

The ability to gather, store, and use data on a large scale has led to improved efficiency and the creation of new industries. In increasing the digitization of agriculture, experiential knowledge is now joined by many new datasets that are becoming important tools to the modern farmer. The Climate Corporation (Climate) is developing tools utilizing cutting edge data science to sustainably increase farmer’s yields. These tools provide greater control and insight into common practices, such as fertility treatments and seed placement. An important component of this is field-level precipitation amounts. Climate has recently developed an in-house Quantitative Precipitation Estimation (QPE) data set for customer and internal consumption to improve the product’s precipitation accuracy. Customers receive rainfall reports on individual fields designed to help plan daily activities. For many internal models, precipitation estimates are an essential piece of information to drive various computations. Having high-resolution, field-level data can help improve decision making and better planning and optimized application of field treatments, can lead to cost savings across the farming operations

A specific focus on the US “corn belt,” an area characterized by low station density and sub-optimal radar coverage, drove development of Climate’s QPE. The farming community actively monitors weather and is highly engaged with products that they view as useful. Growers and numerical simulations require precipitation data available at all times, and dealing with any potential disruption in the QPE pipeline is a potential challenge. It is important, therefore, to have a robust system in place. While performance of the data set in comparison to other radar-based precipitation data sets using more typical scientific metrics (MAE, bias) has been important, these are not the only metrics we used to determine accuracy. Another challenge is communicating product uncertainty. Climate’s QPE is a 1 km gridded data set that can sufficiently capture synoptic- and mid meso-scale variability (down to ~ 2 - 5 km). However, natural small mesoscale variability of precipitation is more difficult to discern. Maintained rain gauge networks provide a “ground truth” and sense of statistical variability. These data provide single point comparisons and though they have some inherent error, the information is useful because it is an independent point source of data that cannot be inferred from using the more numerous, but coarse measurements of the WSR-88D network. Additional rain gauges with unknown data quality can be found in individual fields as well, and may pose challenges in comparison of these data to field-level precipitation estimates. It is key to communicate how QPE estimations are expected to compare to rain gauge measurements, including the underlying reasons for the departure.

This presentation will highlight lessons learned and challenges associated with development of a multisensor precipitation dataset tailored for digital agriculture. We will present an overview of the Climate’s QPE pipeline and discuss the effect of a data disruptions. The challenges of communicating inconsistencies and uncertainties between gridded and rain gauge point locations will also be presented.