Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Mapping desertification: towards an approach for mapping regional land degradation in drylands
Land degradation in drylands (desertification) is an issue that potentially impacts nearly half of the world's human population living on over a third of the Earth's land surface. Despite global concern of the impact of desertification on people and the environment, there is no universal method to a...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Department of Biological Sciences
2020
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613208336924672 |
|---|---|
| access_status_str | Open Access |
| author | Bell, Wesley |
| author2 | Hoffman, Michael Timm |
| author_browse | Bell, Wesley Hoffman, Michael Timm |
| author_facet | Hoffman, Michael Timm Bell, Wesley |
| author_sort | Bell, Wesley |
| collection | Thesis |
| description | Land degradation in drylands (desertification) is an issue that potentially impacts nearly half of the world's human population living on over a third of the Earth's land surface. Despite global concern of the impact of desertification on people and the environment, there is no universal method to assess and map desertification. Methods to assess desertification at the local to regional scale that can fit into a broader global desertification narrative are more appropriate. The overall objective of this thesis is to assess regional desertification using field and Earth observation data for the Namaqualand Hardeveld bioregion of South Africa. Field data on the condition of the land from 277 plots was analysed using Latent Class Analysis (LCA) and found to cluster into three separate states. The first state (S1) was comprised primarily of degraded plots. The third state (S3), on the other hand, was comprised primarily of non-degraded plots, while the plots in state two (S2) generally fell between those which were assigned to S1 and S3. Through the LCA, each plot was assigned a probability of belonging to each state, and the most important variables in distinguishing the three states (perennial plant cover and bare ground cover) were identified. A total of 16 remote sensing variables were determined for the project area. Five vegetation indices (NDVI, EVI, SAVI, OSAVI, MSAVI), as well as spectral mixture analysis (SMA) cover estimates for perennial vegetation, bare ground and bare rock were calculated using both Landsat 8 and Sentinel-2A data. These variables were used in a series of Partial Least Squares regression (PLSr) models to predict either the probability of a plot belonging to one of the three latent states, or the field estimated perennial plant and bare ground cover. The best performing PLSr model had ten remote sensing variables predicting the field estimates of cover (R2Ycum = 0.592; Q2cum = 0.554). Both Sentinel-2A and Landsat 8 SMA cover estimates were better at predicting field cover than any of the vegetation indices. Estimates of bare ground and perennial plant cover were projected over the project area using the PLSr model and ground truthed using data from 61 independent field test plots. There was a significant correlation between the PLSr estimates and the field estimates for both perennial plant cover and bare ground cover for the test plots with the best correlation found to be between the PLSr estimate of bare ground and field estimated bare ground cover (r = 0.827, p < 0.001, CI [0.727, 0.893]). The trendline slope and percentile range of a time series of the Landsat SMA bare ground estimate were used to create raster images. These images, along with images for the PLSr bare ground and perennial plant cover estimates, were converted into images representing membership values between zero and one for the habitat condition archetype. These three images were then combined to produce one raster representing the overall membership of the project area to the habitat condition archetype. The importance of five potential drivers of land degradation (elevation, slope aspect, slope steepness, rainfall trend, and land tenure) in predicting PLSrestimated perennial plant and bare ground cover were evaluated using a random forest model. All drivers were found to be important predictors of cover and were included in the construction of the final, multi-band archetype image. If habitat condition classes are designated according to the mean archetype membership value ± one / two standard deviations, then 17% of the project area could be considered moderately degraded, with just over 3% severely degraded. This novel method of assessing and mapping desertification leads to improved accuracy in predicting habitat condition in the context of potential drivers of change. The utility of SMA over traditional vegetation indices is supported for this particular environment. This methodology can be improved with better endmember designation as well as improved spatial data on the potential drivers of change in drylands. The archetype approach ensures less subjectivity in map production, and the retention of pertinent information in map products. The approach developed in this thesis will allow for more accurate desertification reporting for UNCCD member states and will ultimately improve efforts to combat desertification globally. |
| format | Thesis |
| id | oai:open.uct.ac.za:11427/32201 |
| institution | University of Cape Town (South Africa) |
| language | eng |
| last_indexed | 2026-06-10T12:32:29.432Z |
| license_str | Not specified — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from UCTD — University of Cape Town Open Access Repository |
| publishDate | 2020 |
| publishDateRange | 2020 |
| publishDateSort | 2020 |
| publisher | Department of Biological Sciences |
| publisherStr | Department of Biological Sciences |
| record_format | dspace |
| source_str | UCTD — University of Cape Town Open Access Repository |
| spelling | oai:open.uct.ac.za:11427/32201 Mapping desertification: towards an approach for mapping regional land degradation in drylands Bell, Wesley Hoffman, Michael Timm Visser, Vernon Biological Sciences Land degradation in drylands (desertification) is an issue that potentially impacts nearly half of the world's human population living on over a third of the Earth's land surface. Despite global concern of the impact of desertification on people and the environment, there is no universal method to assess and map desertification. Methods to assess desertification at the local to regional scale that can fit into a broader global desertification narrative are more appropriate. The overall objective of this thesis is to assess regional desertification using field and Earth observation data for the Namaqualand Hardeveld bioregion of South Africa. Field data on the condition of the land from 277 plots was analysed using Latent Class Analysis (LCA) and found to cluster into three separate states. The first state (S1) was comprised primarily of degraded plots. The third state (S3), on the other hand, was comprised primarily of non-degraded plots, while the plots in state two (S2) generally fell between those which were assigned to S1 and S3. Through the LCA, each plot was assigned a probability of belonging to each state, and the most important variables in distinguishing the three states (perennial plant cover and bare ground cover) were identified. A total of 16 remote sensing variables were determined for the project area. Five vegetation indices (NDVI, EVI, SAVI, OSAVI, MSAVI), as well as spectral mixture analysis (SMA) cover estimates for perennial vegetation, bare ground and bare rock were calculated using both Landsat 8 and Sentinel-2A data. These variables were used in a series of Partial Least Squares regression (PLSr) models to predict either the probability of a plot belonging to one of the three latent states, or the field estimated perennial plant and bare ground cover. The best performing PLSr model had ten remote sensing variables predicting the field estimates of cover (R2Ycum = 0.592; Q2cum = 0.554). Both Sentinel-2A and Landsat 8 SMA cover estimates were better at predicting field cover than any of the vegetation indices. Estimates of bare ground and perennial plant cover were projected over the project area using the PLSr model and ground truthed using data from 61 independent field test plots. There was a significant correlation between the PLSr estimates and the field estimates for both perennial plant cover and bare ground cover for the test plots with the best correlation found to be between the PLSr estimate of bare ground and field estimated bare ground cover (r = 0.827, p < 0.001, CI [0.727, 0.893]). The trendline slope and percentile range of a time series of the Landsat SMA bare ground estimate were used to create raster images. These images, along with images for the PLSr bare ground and perennial plant cover estimates, were converted into images representing membership values between zero and one for the habitat condition archetype. These three images were then combined to produce one raster representing the overall membership of the project area to the habitat condition archetype. The importance of five potential drivers of land degradation (elevation, slope aspect, slope steepness, rainfall trend, and land tenure) in predicting PLSrestimated perennial plant and bare ground cover were evaluated using a random forest model. All drivers were found to be important predictors of cover and were included in the construction of the final, multi-band archetype image. If habitat condition classes are designated according to the mean archetype membership value ± one / two standard deviations, then 17% of the project area could be considered moderately degraded, with just over 3% severely degraded. This novel method of assessing and mapping desertification leads to improved accuracy in predicting habitat condition in the context of potential drivers of change. The utility of SMA over traditional vegetation indices is supported for this particular environment. This methodology can be improved with better endmember designation as well as improved spatial data on the potential drivers of change in drylands. The archetype approach ensures less subjectivity in map production, and the retention of pertinent information in map products. The approach developed in this thesis will allow for more accurate desertification reporting for UNCCD member states and will ultimately improve efforts to combat desertification globally. 2020-09-09T16:28:47Z 2020-09-09T16:28:47Z 2020 2020-09-09T16:28:29Z Doctoral Thesis Doctoral PhD http://hdl.handle.net/11427/32201 eng application/pdf Department of Biological Sciences Faculty of Science |
| spellingShingle | Biological Sciences Bell, Wesley Mapping desertification: towards an approach for mapping regional land degradation in drylands |
| thesis_degree_str | Doctoral |
| title | Mapping desertification: towards an approach for mapping regional land degradation in drylands |
| title_full | Mapping desertification: towards an approach for mapping regional land degradation in drylands |
| title_fullStr | Mapping desertification: towards an approach for mapping regional land degradation in drylands |
| title_full_unstemmed | Mapping desertification: towards an approach for mapping regional land degradation in drylands |
| title_short | Mapping desertification: towards an approach for mapping regional land degradation in drylands |
| title_sort | mapping desertification towards an approach for mapping regional land degradation in drylands |
| topic | Biological Sciences |
| url | http://hdl.handle.net/11427/32201 |
| work_keys_str_mv | AT bellwesley mappingdesertificationtowardsanapproachformappingregionallanddegradationindrylands |