In a possible global survey, the key variables are areas of forest land, other wooded land and other land as well as their changes over time. Optional variables are tree stem volume and tree biomass, together with changes. Earlier remote sensing studies support the assumption that a breakdown into rough species groups should be possible by means of remote sensing aided analysis. The parameters which are possible to estimate using remote sensing aided method depend on the intensity of the field sampling. The estimation of the volume of growing stock and biomass of trees presumes a moderate field sampling intensity and measurements of tree components. There are many studies concerning the accuracy of the above mentioned variables with field measurements and remote sensing data. Rather dense field sampling grid has been applied in most studies. There is a lack of studies with very sparse field sampling intensity. The accuracy of estimates with sparse field sampling could be one topic in a possible global pilot study.
A couple of survey cases are given in the following. They are only tentative. The final error analysis needs more study and pilot studies. A method to estimate sampling error through simulation is outlined in Appendix 1. The FRA 2000 land cover map of FAO, produced by EROS Data Center has been applied in that study. Other global forest cover maps may also be available for the possible global survey. These will be very useful data sources in planning sampling designs in a similar way.
In order to get an idea of the level of the field measurement costs and to be able to compare those with remote sensing aided surveys, a rough calculation is first given in which only field data are used. The area represented by one field plot varies here from 30 000 ha in areas with no changes to 10 000 ha with a relative annul change of 0.8 %. The total costs with these assumptions exceed USD 100 mill., and USD 10 mill. if only one tenth of the plots are applied (Table 5). The costs of images corresponding 10 % Landsat ETM+ cover together with 100 field plots per image scene are given in Table 6. Comparison of the costs of the Table 5 and Table 6 shows that field data dominate the cost.
Table 5: An example of the number and costs of field plots in a global survey utilising field data only
Region |
Land area |
Forest area |
Net change 1990-2000 |
Net change 1990-2000 |
Area represented by one plot |
Number of field plots |
Estimated costs |
mill. ha |
mill. ha |
mill.ha/year |
% / year |
US $ | |||
Africa |
2978 |
650 |
-5,3 |
-0,815 |
13692 |
69221 |
30457393 |
Asia |
3085 |
548 |
-0,4 |
-0,073 |
28540 |
30010 |
13204558 |
Europe |
2260 |
1039 |
0,9 |
0,087 |
28268 |
44751 |
19690412 |
North and Central America |
2137 |
549 |
-0,6 |
-0,109 |
27814 |
27421 |
12065345 |
Oceania |
849 |
198 |
-0,4 |
-0,202 |
25960 |
10898 |
4794989,9 |
South America |
1755 |
886 |
-3,7 |
-0,418 |
21648 |
49035 |
21575347 |
Total |
13064 |
3869 |
-9,4 |
-0,243 |
231336 |
101788045 |
Table 6: Estimated costs with 100 field plots per Landsat ETM image and 440 US $ per plot
Region |
Field data |
Image data |
Total costs |
Africa |
4253940 |
58008 |
4311949 |
Asia |
4406785 |
60093 |
4466878 |
Europe |
3228309 |
44022 |
3272332 |
North and Central America |
3052609 |
41626 |
3094236 |
Oceania |
1212759 |
16538 |
1229296 |
South America |
2506939 |
34186 |
2541125 |
Total |
18661342 |
254473 |
18915815 |
For demonstrating the behaviour of the standards errors computed with the methods given in Appendix 1, we conducted the following experiment. The area of Europe and CIS was covered with two optional Landsat ETM+ densities 1) with 150 scenes and 2) with 300 scenes. The first option is nearly the density used in tropics for FRA 1990 and FRA 2000. We assumed that a digital image analysis of one Landsat TM scene requires 5 working days. A cost of USD 600 was assumed for one Landsat scene and a cost of USD 400 for one working day. The field data costs were not included in the first two options (Table 7). In the options 3 and 4 (Table 8), the costs of 100 field plots per one Landsat scene are included, USD 44 000 per scene. The number of Ikonos scenes was computed in such a way that total costs, including the image cost and two working days for one Ikonos scene, were same as with Landsat options. Two different prices were assumed for Ikonos, 1) USD 2700 and 2) USD1000 in anticipation of a potentially more competitive market in the future.
Table 7: Two Landsat ETM+ cover options with a processing time of 5 working days per image and the number of Ikonos scenes with two price options corresponding to the same total costs as Landsat options. Ikonos is assumed to require 2 working days per scene. The cost of 400 USD has been used for one working day.
Europe & CIS |
150 Landsat scenes a 600 USD |
300 Landsat scenes a 600 USD |
No of Landsat |
150 |
300 |
costs |
390000 |
780000 |
No of Ikonos scenes, a 2700 USD |
111 |
223 |
No of Ikonos scenes, a 1000 USD |
217 |
433 |
Table 8: Two Landsat ETM+ cover options with a processing time of 5 working days per image and the number of Ikonos scenes with two price options corresponding to the same total costs as Landsat options. Ikonos is assumed to require 2 working days per scene. The cost of 400 USD has been used for one working day. With Landsat ETM+, the costs of 100 field plots with unit costs of USD 440 have been included.
Europe & CIS |
150 Landsat scenes a 600 USD |
300 Landsat scenes a 600 USD |
No of Landsat |
150 |
300 |
costs |
6990000 |
13980000 |
No of Ikonos scenes, a 2700 USD |
1997 |
3994 |
No of Ikonos scenes, a 1000 USD |
3883 |
7767 |
The number of Ikonos samples varies under these assumptions from n=111 to n=7767. Note that field data costs are not included in the Ikonos options. We have assumed that forest area and change in forest area can be interpreted from Ikonos or similar images without field data. The validity of this hypothesis must be tested in future analysis.
We use simulations to evaluate the feasibility of multiple-resolution remote sensing for FRA 2010 objectives. The simulation has the following components (see Appendix 1).
1. Construct a realistic hypothetical population that fully covers a very large region. The selected region was Europe and the Commonwealth of Independent States (CIS). The combined land area1 is 2.7-billon hectares, of which 38% is forestland and another 4% is other wooded land.
2. Simulate changes in forest cover in this hypothetical population over a 10-year time interval.
3. Simulate estimates of forest area from satellite data with multiple resolutions (1-km, 30-m and 1-m) for the entire hypothetical population at the beginning (2000) and end of the 10-year time interval (2010).
4. Compare alternative sampling designs to statistically estimate the area of forest cover and changes in forest cover during the 10-year interval.
5. Predict the statistical precision and cost of these alternative designs for possible use by FRA 2010 in the future.
The standard errors and coefficients of variation for the estimates obtained with Landsat image analysis are given in Table 9. The corresponding errors for estimates obtained with Ikonos image interpretation with two different Ikonos image price assumptions are presented in Table 10 when field measurement costs are excluded from Landsat options, and in Table 11 when field measurement costs are included in Landsat image options. A relatively small change rate makes the coefficient of variation high for the changes of forest area in all alternatives. The error level is near acceptable in all cases.
A somewhat surprising result is that the sampling errors with much smaller Ikonos images are about at the same level as, or even smaller than, those with about same number of much larger Landsat images. This appears to be caused by methods used to reduce variability with sub-strata (Appendix 1). This technique appears more successful with smaller sampling units, at least with the hypothetical population constructed for this study.
A fact which can affect the competitiveness of high resolution and very high-resolution images is the availability of the very high-resolution images. It is much more uncertain than that of high-resolution images. The needed double coverage of images for change analysis makes the situation more complicated with high-resolution images (see conclusions and discussion).
Our simplified simulation model shows that high resolution and very high-resolution image based forest survey can meet the accuracy requirements of a possible independent remote sensing aided forest survey with a moderate cost. The final conclusions would need further analysis and more resources than what was available in this study. The procedure described here could be repeated in the other regions of the globe as well, with some other remote sensing based material, and with other variables than which were used here.
Table 9: The standard errors and coefficients of variation when Landsat images are applied
150 Landsat scenes |
300 Landsat scenes | |
Standard error | ||
Forest area time 0 (km2) |
388 587 |
253 377 |
Other Wooded land area time 0 (km2) |
42 661 |
28 373 |
Change in forest area from time 0 to time 10 (km2) |
30 100 |
19 365 |
Change in other wooded land area from time 0 to time 10 (km2) |
7 912 |
5 254 |
Rate of change in forest area %/10-years |
0,38 % |
0,25 % |
Rate of change in other wooded land area %/10-years |
0,96 % |
0,65 % |
Coefficient of Variation | ||
Forest area time 0 (km2) |
3,85 % |
2,51 % |
Other Wooded land area time 0 (km2) |
3,90 % |
2,60 % |
Change in forest area from time 0 to time 10 (km2) |
122,70 % |
78,94 % |
Change in other wooded land area from time 0 to time 10 (km2) |
4,94 % |
3,28 % |
Rate of change in forest area %/10-years |
157,85 % |
104,40 % |
Rate of change in other wooded land area %/10-years |
6,58 % |
4,46 % |
Table 10: The standard errors and coefficients of variation when Ikonos images are employed. Field plot costs of Landsat images are excluded
111 Ikonos scenes |
217 Ikonos scenes |
223 Ikonos scenes |
443 Ikonos scenes | |
Standard error | ||||
Forest area time 0 (km2) |
310 644 |
221 359 |
214 476 |
152 643 |
Other wooded land area time 0 (km2) |
51 478 |
36 606 |
36 194 |
25 671 |
Change in forest area from time 0 to time 10 (km^2) |
36 469 |
26 533 |
25 436 |
18 248 |
Change in other wooded land area from time 0 to time 10 (km^2) |
10 817 |
7 791 |
7 629 |
5 441 |
Rate of change in forest area %/10-years |
0,34 % |
0,25 % |
0,24 % |
0,17 % |
Rate of change in other wooded land area %/10-years |
0,60 % |
0,43 % |
0,43 % |
0,30 % |
Coefficient of Variation | ||||
Forest area time 0 (km2) |
3,08 % |
2,19 % |
2,12 % |
1,51 % |
Woodland area time 0 (km2) |
4,71 % |
3,35 % |
3,31 % |
2,35 % |
Change in forest area from time 0 to time 10 (km2) |
148,66 % |
108,16 % |
103,69 % |
74,39 % |
Change in other wooded land area from time 0 to time 10 (km2) |
6,75 % |
4,86 % |
4,76 % |
3,40 % |
Rate of change in forest area %/10-years |
142,03 % |
103,58 % |
99,75 % |
71,43 % |
Rate of change in other wooded land area %/10-years |
4,10 % |
2,97 % |
2,90 % |
2,07 % |
Table 11: The standard errors and coefficients of variation when Ikonos images are employed. The field plot costs of Landsat analysis are included
1197 Ikonos scenes |
3883 Ikonos scenes |
3994 Ikonos scenes |
7767 Ikonos scenes | |
Standard error |
||||
Forest area time 0 (km2) |
71 972 |
51 381 |
50 695 |
36 194 |
Other wooded land area time 0 (km2) |
12 083 |
8 626 |
8 503 |
6 066 |
Change in forest area from time 0 to time 10 (km2) |
8 579 |
6 124 |
6 033 |
4 301 |
Change in other wooded land area from time 0 to time 10 (km2) |
2 549 |
1 820 |
1 795 |
1 279 |
Rate of change in forest area %/10-years |
0,08 % |
0,06 % |
0,06 % |
0,04 % |
Rate of change in other wooded land area %/10-years |
0,14 % |
0,10 % |
0,10 % |
0,07 % |
Coefficient of Variation |
||||
Forest area time 0 (km2) |
0,71 % |
0,51 % |
0,50 % |
0,36 % |
Other wooded land area time 0 (km2) |
1,11 % |
0,79 % |
0,78 % |
0,56 % |
Change in forest area from time 0 to time 10 (km2) |
34,97 % |
24,96 % |
24,59 % |
17,53 % |
Change in other wooded land area from time 0 to time 10 (km2) |
1,59 % |
1,14 % |
1,12 % |
0,80 % |
Rate of change in forest area %/10-years |
33,55 % |
23,94 % |
23,62 % |
16,82 % |
Rate of change in other wooded land area %/10-years |
0,97 % |
0,69 % |
0,68 % |
0,49 % |
1 Europe has a total land area of 565,930,000-ha; 31% is forestland and 7% is other wooded land (FAO 2000). CIS has a total area 2,213,036,000-ha; 40% is forestland and 4% is other wooded land. Europe covers 21% of the simulation population, while CIS covers 79%.
