Chapter 14 . Time-Series Prediction of Chlorophyll a in Lakes
281
••• • ••••••• _ booIstrap aggregate modal RMS OH'"
o _
nodes , 2 tiddon nocl .. , 5 hicIdon nodes
:
t
2: t.115 I o.a Q ! 2: 1.15 1 0.6 0 j 2 1.15 I 0.1 0
i
Kasumlgaura • same day modal
o - . nodes ! 2 hlddan nodes ! 5 hIddon nodes
2 C
2 U, Q.60 ! 2 U, 0..60 ! 2 1.1, 0.&0
Myponga • same day modal
21.110..6° 1 2 lS1 aso j, ," 0.60
Soyang . same d~y model
~ _ dlOI_ 01 RMS enora • • oudler
~
0 hldd Kasumlgaura • 30 day modal
o hldd d
2: 1.&, o.ao I2"" 0-40 j 2 ... , 0"'0
M~nga. 30 ~y model
o hlddon nodes 12 hlddon nocl .. i 5 hlddon nodes
!! f
2:
U 1 0 .1 0 I 2' U I 0.1 o! . z 1,1 I 0..6 0
Soyang • 30 day modal
Error at which Training Stopped
Figure 14.8. The effect of training time and the number of hidden layer nodes on
test set RMS error. a Kasumigaura same day model. b Kasumigaura 30 day ahead
model. c Myponga same day model. d Myponga 30 day ahead model. e Soyang
same day model. f Soyang 30 day ahead model.