134 
them 
computer. Final computer output consists of figures on the value, tion 
cost, and labour for each crop and for each type of livestock for to i 
each farm. impr 
ivel 
When compared to ground data which were collected in 1972 
for each farm the results are as given in Table 3. The method 
was also applied using 1964 imagery and ground truth collected resu 
in 1972. do s 
obta 
TABLE 3 
SUMMARY OF RESULTS: DETERMINATION 
OF FARM INCOME FROM API 
All 1971 
1971 less l 2 
All 1964 
1964 less 3 
All correctly 
identified by type^ 
Number of 
Observations 
Mean-*- 
Error 
92 
-.2 
91 
-.9 
81 
3.1 
78 
-1.8 
134 
-.5 
Standard 
Deviation 
Min. 
Max 
15.6 
-35 
+65 
14.1 
-35 
+ 47 
16.4 
-27 
+73 
11.6 
-27 
+27 
12.2 
-35 
+48 
1 All mean, standard deviations, minimum and maximum 
values are in per cent. 
2 For 1971 and 1964 the calculations were made without 
the most inaccurate values to demonstrate their effect 
on variation. 
3 In all of the above all farms are included except those 
with very low productivity (for example only pasture), 
including farms incorrectly identified. This category 
includes only those properly identified by farm type. 
DISCUSSION AND CONCLUSION. 
The preliminary results are very encouraging. Using 
Table III for the 1971 data one may state that 95% of the time 
the estimated value will be within 30% of the actual production 
value for all farm types whether or not they are correctly identi 
fied, and this is obtained with the assumption of average yields, 
average capacities, etc. The production values range from under 
$100.00 to over %900.00. The mean accuracy values speak for