Figure 5
(a) Schematic (upper) and experimental (lower) setup of field measurement. (b) Nitrate concentration of an agricultural field is measured continuously with four low-cost nitrate ISEs over a period of 22 days. Part of the voltage readings are adapted from Ref 8,33. Different faults are observed- slow drift over time (S1), increasing sensitivity to temperature with time (S2), sudden jump in output (S4). If we do not correct drifts, three sensors become unreliable after 12–13 days and MLE fails to estimate correct concentration and credibility. (c) MLE assigns high credibility to S2 although it is not reliable at later days. Again, S3 is assigned low credibility at certain times although it is reliable. (d) Drift correction corrects the drift ahead of time, so that drift-corrected estimated signal follows the ground truth more accurately. (e) After drift correction, MLE correctly assigns lower credibility to S2 and S4.
