Table 2:

CKMRnn performance with varying levels of spatial sampling bias, as depicted in Figure 4a. Here, the population sizes of both the tested simulations and the simulations used to train the model were constant in time (i.e., no population trend). Relative absolute error is calculated as |true − predicted| /true and relative error as (true − predicted) /true. Separate results are shown for simulations with true N above and below 12,000.

CKMRnn performance with sampling bias
Spatial sampling bias	Mean relative absolute error (MRAE)	MRAE N < 12000	MRAE N ≥ 12000	Mean relative error (MRE)	MRE N < 12000	MRE N ≥ 12000
uniform	0.063	0.064	0.060	0.011	0.034	−0.039
medium	0.051	0.052	0.049	0.001	0.014	−0.024
high	0.050	0.053	0.046	0.013	0.025	−0.011