de Bie, Kees 
  
  
  
  
Simultaneous use of different models is justified if (it is assumed that) each 
represents a different crop physiological mechanism and each defines its 
contribution to the final production independently. It is assumed that these 
mechanisms are ‘flower initiation’ and ‘fruit formation’ (from flowering to fruit 
maturity). This assumption tallies with the observation that several very lush and 
fully-grown orchards in Phrao failed to produce any fruits. The perfect continuum 
of yield data gathered (Figure 3^) hardly supports the assumption made. 
Therefore, the Ln (Yield+1) data were also subjected to multiple linear regression 
in spite of their non-normal behavior (section 7). 
All models proposed referred to weighed yield data. The weighing factor used was 
"orchard size x fraction of mango trees in the orchard". Weighing aimed to reduce 
the effect of "the total quantity involved" on sale proceeds and to reduce the effect 
of unequal mango tree densities. After the three models were established, all 
results were evaluated to identify the "best" approach to estimate the contributions 
of individual constraints. 
  
  
e 
e 
N 
T 
1 
68% RE 
6 confidence tet A 
-— 
  
  
  
  
  
  
  
  
  
  
  
  
= £ 
2 2 
= = 
2 2 
° 3 
à a 
s 1L. © / 
Ju EN eclipse 7 
2 ; = te 
- OF z w Of P zl 
50 50 : A U 
2 E zl 9 Ak a z 
© : © o 
$ | $ e 
S d Sot 
5 2 3 52 - ] 
o d. o 
o © 
E 3 | | di 3 I I Li a 
0 100 200 300 -1 0 1 2 3 4 5 6 
a. Yield b. Natural Log of Yield 
53 T T T T 25 TDI T T T 
3 ; a 
a 2. 6896 confidence 9 20 7 
2 eclipse ey eT ! T 
x y o9 1 
S 17 E any. à , 
E ik qn, 15} ' M 
Z ot. 00 4| E ! Mean 
S 8° 5 8 1 
o Eod 10r 1 À 
= -1- ee =} ! 
Su pe 1 
> 5 | 
Tr SF 
cU 
sr q : 7 
P A 
üi.3 l | | | | 0 i | 
0 1 2 3 4 5 6 0 1 2 3 4 5 6 
e. Natural Log of (Yield + 1) d. Natural Log of (Yield + 1) 
Figure 3. Transformation of mango yield data (‘000 Bath/ha) 
a: Z-Scores of original yield data 
b: Z-Scores of Ln(yield) with '0' yields omitted? 
c: Z-Scores of Ln(yield+1 yi 
d: plot of the distribution of Ln(yield+1) 
  
The probability that Ln(yield) is normally distributed is 62.3% (Kolmogorov-Smirnov 2.34, 1.64). 
9 
10 The probability that Ln(yield+1) is normally distributed is 3.6% (Kolmogorov-Smirnov 1.57, 1.69). 
  
326 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.