Academia.eduAcademia.edu

Figure 1 – uploaded by Daniel Botkin

See full PDFdownloadDownload figure
3) For an area of 409 km?, with 113 738 landscape elements, in the multi-use Superior National Forest (nonwilderness) near the Boundary Waters Canoe Area. A) For an area of 534 km2, with 148 406 landscape elements, in the protected wilderness of the Boundary Waters Canoe Area.

Table 1 3) For an area of 409 km?, with 113 738 landscape elements, in the multi-use Superior National Forest (nonwilderness) near the Boundary Waters Canoe Area. A) For an area of 534 km2, with 148 406 landscape elements, in the protected wilderness of the Boundary Waters Canoe Area.

Related Figures (10)

Fic. 1. Scatterplot showing correspondences between ecological states and reflected light intensity as measured in July 1983 by Landsat-4 MSS in wavelength band 4 (near-infrared, 0.8—-1.1 um) and band 1 (visible light near the green peak, 0.5- 0.6 um). The boxes define correspondences of spectral regions with ecological states. The units on each axis are digital counts (0-256) proportional to radiance as recorded by the Landsat MSS. Box boundaries were determined as described in the text (Methods: Decision regions . . .). The isopleths shown are regions of constant numbers of pixels. The interior isopleths enclose the higher abundances of pixels. For example, the innermost isopleths are >400 pixels. Fourteen ecological states were distinguishable in the July 1983 Landsat-4 MSS imagery. These states (clearings, lakes, and the 12 ecological states with lower case alphabetical designations) are represented in the scatterplot with lighter bordered boxes. In the June 1973 Landsat-1 MSS imagery, it was necessary to combine some of the 1983 states as shown by the boxes with darker borders. The six final states (clearings, regeneration, broadleaf, mixed, conifer, and lakes) were used to classify both the 1983 and 1973 Landsat images.
Fic. 1. Scatterplot showing correspondences between ecological states and reflected light intensity as measured in July 1983 by Landsat-4 MSS in wavelength band 4 (near-infrared, 0.8—-1.1 um) and band 1 (visible light near the green peak, 0.5- 0.6 um). The boxes define correspondences of spectral regions with ecological states. The units on each axis are digital counts (0-256) proportional to radiance as recorded by the Landsat MSS. Box boundaries were determined as described in the text (Methods: Decision regions . . .). The isopleths shown are regions of constant numbers of pixels. The interior isopleths enclose the higher abundances of pixels. For example, the innermost isopleths are >400 pixels. Fourteen ecological states were distinguishable in the July 1983 Landsat-4 MSS imagery. These states (clearings, lakes, and the 12 ecological states with lower case alphabetical designations) are represented in the scatterplot with lighter bordered boxes. In the June 1973 Landsat-1 MSS imagery, it was necessary to combine some of the 1983 states as shown by the boxes with darker borders. The six final states (clearings, regeneration, broadleaf, mixed, conifer, and lakes) were used to classify both the 1983 and 1973 Landsat images.
* Water and clouds are omitted; hence some columns do not sum to 100%. TABLE 2. Percent of land area by class.*
* Water and clouds are omitted; hence some columns do not sum to 100%. TABLE 2. Percent of land area by class.*
"ABLE 3. Markov projections of steady-state areal percentages occupied by ecological states. there are none (Type 2 errors), and an indication of no transitions when in fact there were (Type | errors). The error model of the Appendix quantifies the dependence of these errors in the classification-based transition probabilities on the magnitude of the classification likelihoods and the proportion of the scene classified as in the various ecological states. tions about errors generated by image processing. A promising approach would be image simulation. Image pairs representing changing landscapes could be sim- ulated with known transition probabilities, including spatial structures incorporating average patch sizes, proportions, etc., of real landscapes. These simulated image pairs could then be sampled at various spatial resolutions to examine the effect of mixed pixels on accuracy. The image pairs could also be misregistered to study the effects on apparent transition probabilities. Clearly, such a study should be done, but is beyond the scope of this investigation.
"ABLE 3. Markov projections of steady-state areal percentages occupied by ecological states. there are none (Type 2 errors), and an indication of no transitions when in fact there were (Type | errors). The error model of the Appendix quantifies the dependence of these errors in the classification-based transition probabilities on the magnitude of the classification likelihoods and the proportion of the scene classified as in the various ecological states. tions about errors generated by image processing. A promising approach would be image simulation. Image pairs representing changing landscapes could be sim- ulated with known transition probabilities, including spatial structures incorporating average patch sizes, proportions, etc., of real landscapes. These simulated image pairs could then be sampled at various spatial resolutions to examine the effect of mixed pixels on accuracy. The image pairs could also be misregistered to study the effects on apparent transition probabilities. Clearly, such a study should be done, but is beyond the scope of this investigation.
TABLE 4. Recurrence times [(T,)]. Recurrence time = the average number of years for a landscape element to return to a given state once it had left that state.
TABLE 4. Recurrence times [(T,)]. Recurrence time = the average number of years for a landscape element to return to a given state once it had left that state.
* Matrix elements C(a|d) are the estimated likelihoods (based on ground-observed patches) of classifying a ground-observed patch (pixel) into ecological state a if ground observations indicated that it was in ecological state b. That is C(a|b) = n(a|bV/N where n(a|b) is the number of patches that ground observation showed to be in ecological states of type b but that were classified as type a, and N is the total number of patches (pixels) in the ground-observed set. The diagonals of the matrix are the estimated likelihoods of correct classification. The off-diagonal elements are the estimated likelihoods of incorrect clas- sification. TABLE 5. Likelihoods of correct and incorrect classifications based on satellite remote sensing.*
* Matrix elements C(a|d) are the estimated likelihoods (based on ground-observed patches) of classifying a ground-observed patch (pixel) into ecological state a if ground observations indicated that it was in ecological state b. That is C(a|b) = n(a|bV/N where n(a|b) is the number of patches that ground observation showed to be in ecological states of type b but that were classified as type a, and N is the total number of patches (pixels) in the ground-observed set. The diagonals of the matrix are the estimated likelihoods of correct classification. The off-diagonal elements are the estimated likelihoods of incorrect clas- sification. TABLE 5. Likelihoods of correct and incorrect classifications based on satellite remote sensing.*
A) For an area of 534 km? with 148 406 landscape elements in the protected wilderness of the Boundary Waters Canoe Area. 1983 state B) For an area of 409 km? with 113 738 landscape elements in the multiuse Superior National forest nonwilderness near to the Boundary Waters Canoe Area.
A) For an area of 534 km? with 148 406 landscape elements in the protected wilderness of the Boundary Waters Canoe Area. 1983 state B) For an area of 409 km? with 113 738 landscape elements in the multiuse Superior National forest nonwilderness near to the Boundary Waters Canoe Area.
Related topics:
Remote SensingEcologyForest successionEcological SuccessionSensor CalibrationAtmospheric Effects
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved