Skip to main content
. 2018 Mar 31;218(4):1430–1449. doi: 10.1111/nph.15123

Figure 2.

Figure 2

Neural network (NN)‐based predicted vs observed gross primary productivity (GPP). (a) Predicted values are based on the NN model estimating actual light use efficiency, NN act, using all input variables (temperature, vapour pressure deficit (VPD), photosynthetically active radiation (PAR), soil moisture) and ‘all days’ data. (b) Predicted values are based on the NN model estimating potential light use efficiency, NN pot, trained at data from days above the soil moisture threshold (‘moist days’), using temperature, VPD and PAR as input and evaluated only on ‘moist days’ data. (c) Same as (b) but evaluated on ‘dry days’ data. (d) Predicted values based on NN pot vs predicted values based on NN act, evaluated only on ‘moist days’ data. NSE, Nash‐Sutcliffe model efficiency; RMSE, root‐mean‐square error.