Key takeaways
- ERA5 reports about 50% more rainfall on average than Veriground's ground stations across the same cocoa-growing locations (4.15mm vs. 2.74mm/day, 189,278 paired observations, Jan 2020-Feb 2026).
- The gap is widest in Ecuador, where ERA5 estimates roughly five times more rainfall than Veriground during the rainy season.
- The two sources disagree on whether it even rained: Veriground recorded zero rainfall on 70% of days, while ERA5 reported zero rainfall on just 6% of days.
- Across 36 separate statistical comparisons by country, season, and El Niño/La Niña conditions, ERA5 and Veriground showed no reliable relationship (R² under 11% in every case) - so one dataset can't be used to correct the other.
A new Veriground white paper comparing six years of ground-station rainfall data against ERA5, the widely used satellite-and-model reanalysis dataset, finds the two sources disagree sharply across cocoa-growing regions in Côte d'Ivoire, Ghana, Nigeria, and Ecuador. Across 189,278 paired daily observations from January 2020 to February 2026, ERA5 reported about 50% more rainfall on average than Veriground's ground stations (4.15mm vs. 2.74mm per day) - a gap that widens to roughly five times higher in Ecuador during the rainy season.
The two sources also disagree on how often it rains at all. Veriground's ground stations recorded zero rainfall on nearly 70% of days, while ERA5 reported zero rainfall on just 6% of days - implying measurable rain on almost every day of the week. The most extreme single discrepancy: on April 10, 2023, ERA5 estimated 561mm of rainfall at an Ecuador location where the matching Veriground station recorded just 88.6mm that same day - and the ERA5 figure is larger than any rainfall Veriground has ever recorded, at any station, across the full six-year study.
Veriground measures rainfall directly with ground-based rain gauges reporting within hours; ERA5 estimates it by combining satellite observations, historical records, and computer models at roughly 30km resolution - well suited to global climate research, but not built to capture hyperlocal conditions on working farms. To see whether the two datasets could be combined to sharpen rainfall estimates, researchers ran 36 separate regression comparisons across different countries, seasons, and El Niño/La Niña conditions. Every comparison came back weak, with R² values staying under 11% throughout - meaning one dataset cannot reliably correct or improve the other.
For commodity desks and agribusinesses tracking cocoa crop conditions, the takeaway is straightforward: reanalysis products like ERA5 are not a substitute for ground-truth measurement in these regions, particularly for judging whether - and how much - it actually rained on a given day.