3 T T T T T T T 
- Without yields ts 
  
  
  
T T T \ T T T 5 T T 7 
Without yields k\ With yields z / 
  
  
© = N OO A O00 O 
T 
| 
F 
ett 
/ 
zi NN / Logistic 
Expected Value for Normal Distribution 
& 
-L 
Estimate of Ln(Yield) 
T 
ELLE 
ß 
T 
Impact on Ln (Yield) or on "LN" 
T 
s 
| 
  
  
  
  
  
  
  
1 2 1 L if 1 1 -3 l 1 
6 b 15. 16. 5 0 5 10 15 C 0 5 10 15 
. Count Count x Slope in orchard (%) 
  
  
  
m 
Estimate 
Figure 11. — Regression model results based on 29 sites with positive 
yield data and extrapolation to sites with “0” yields": 
a: Z-scores of mango yields. 
b: Probability to expect a certain mango yield. 
c: Impact by slope on the Linear Part (LN) of the logistic 
model and on the Ln(yield) estimates of the regression 
model. 
7. Multiple linear regression to predict Ln(Yield+1) 
Both presented models include terrain, texture and water holding capacity co- 
variables. Testing of their interactions proved useful just as the term ‘canopy cover 
x use of a tractor for weeding’ (based on Figure 7%). Use of a motor sprayer 
occurred only when pruning was done and the two co-variables were re-combined 
into 2 new ones. The 10 variables included in the provisional model explained 
89% (Adjusted-R?) of the total variability of yields (Table 3). 
  
  
  
Table 3. Linear multiple regression results of Ln(Yield+1) of mango 
Adjusted multiple R^: 0.893 
Cases are weighted by (96 of mando trees/orchard x orchard size). 
Effect Coefficient | P(2 Tail) | R^when entered 
Constant -1.109 0.330 
If spraying by motor sprayer AND pruning done 1.139 0.000 49 
Year effect (1=good, O=normal, -1=bad) 1.165 0.000 66 
If sprayed with Azodrin 1.322 0.000 73 
If not in hills AND if poor water holding capacity -1.845 0.000 78 
If weeded by tractor MULTIPLIED BY canopy cover 0.008 0.004 82 
If ability to apply supplementary irrigation water 0.777 0.001 85 
If on footslopes -0.398 0.076 87 
pH of the top-soil 0.354 0.004 89 
If poor water holding capacity 0.870 0.013 91.5 
If pruning done AND not sprayed by motor sprayer) 0.523 0.033 92 
  
  
  
  
  
  
The one-sample t test of model residuals showed that the mean of -0.40 is not 
significantly different from zero (P = 1.5%). The Kolmogorov-Smirnov One Sample 
(2-tail) Test using the Normal (-0.40,1.05) distribution suggested a probability of 
  
334 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.