Rabo Bank’s Carbon Bank program is accelerating sustainable food production with initiatives that reduce and remove carbon emissions from the atmosphere. In order to support their Carbon Farming initiative, Seqana partnered with Rabo Carbon Bank to run a 2 phase pilot project.
Phase 1 was a “blind test” where Seqana utilizes a blend of proprietary data, commercial data, and public data sources to model a snapshot of the soil organic carbon (SOC) stocks of the 9 areas of interest (AOIs) requested by RaboBank in USA. The results of this blind test, showcased per hectare SOC stocks of each of the 9 AOIs with lower bound predictions at both a confidence interval (CI) of 90% and 95%. This means that the predicted SOC stock values are the most conservative (lowest) estimates for the respective confidence intervals.
Phase 2 of this pilot project incorporated actual ground truth data to calibrate and test our models. While the final accuracy of the model is to be tested by RaboBank, we had some interesting findings: The blind test model for SOC stocks in t/ha could explain 12.6% of the variance in the calibration reference dataset provided by RaboBank. The average errors range from 6-9.7 t/ha. With the new calibration data we could improve the modelling accuracy to explain 57% of the variance in the SOC stocks and reduce the errors to a range from 3.3-6.6 t/ha. Bulk density and SOC% could be predicted even better with our models explaining 77% and 79% of the variance in the dataset respectively. Since the reference datasets used for calculating these accuracies were quite small these accuracy estimates should still be treated with caution and we are curious as to the outcome of the accuracy analysis conducted by the RaboBank on the remaining sampled data from the pilot study sites.