How is the error rate pointed by the arrow defined? How did you get it? thanks!
Readout error: do a bunch (1000's) of experiments preparing the qubit in either the 0 state or the 1 state and then immediately measuring the qubit state after each preparation. Two types of readout errors can occur: prepared 1 --> measured 0, and prepared 0 --> measured 1. The reported readout error is the average of the rate of those two errors (e.g. if the first occurred on 2% of trials and the second occurred on 1% of trials, then the total error rate would be 1.5% or 0.015.
Single-qubit U2 error rate: this is measured using randomized benchmarking (see https://arxiv.org/abs/1109.6887), in which sequences of gates (specifically Clifford gates) are applied with the goal of taking the qubit on a random walk among certain points on the Bloch sphere and returning it to the 0 state it started in. As the number of gates in the sequence is increased, the chance of returning to zero drops exponentially and eventually saturates near 50%. The gate error rate is extracted from the fit to this exponential decay.
CNOT error rate: two-qubit randomized benchmarking; same idea as for single-qubit errors, but now the gates are two-qubit Clifford gates.