University of Twente Student Theses
Mapping deforestation in the Gaza province (Mozambique) with random forest machine learning algorithm on historical landsat satellite imagery
Boateng, Prince (2024) Mapping deforestation in the Gaza province (Mozambique) with random forest machine learning algorithm on historical landsat satellite imagery.
|
PDF
2MB |
| Abstract: | This study provides analysis of deforestation dynamics and forest regeneration in Gaza Province, Mozambique, spanning the 30-year period from 1993 to 2023. Utilizing a Random Forest machine learning algorithm applied to historical Landsat satellite imagery, the research integrates a diverse range of spectral bands, spectral indices, topographic features, tasselled cap and texture measures to map and quantify landcover changes. The analysis reveals a substantial 37% decrease in forest cover equating to a total loss of 7,288.8 km² with an average annual loss rate of 242.96 km². In contrast, forest regeneration efforts resulted in the recovery of only 2,456 km² of forest averaging 81.9 km² per year. This significant disparity between forest cover loss and regeneration underscores the inadequacy of current reforestation and conservation initiatives in the region. The temporal breakdown of the data highlights critical periods where forest cover loss intensified. Between 1993 and 2004 forest regeneration accounted for 1,085 km² representing 10% of the forest cover while the loss during the same period amounted to 2,323.6 km² of forest cover. In the subsequent periods, forest regeneration rates diminished considerably with 655 km² (6.9%) regenerated from 1998 to 2008, 282 km² (3.3%) from 2004 to 2013, and a mere 152 km² (2%) from 2013 to 2023. Forest cover loss fluctuated but remained substantial peaking at 2,280.2 km² (29%) between 2008 and 2018 marking the most significant reduction in forested land during the study period. Geographically, the study identifies Mabalane district as experiencing the highest levels of forest regeneration particularly in the earlier periods compared to Chókwè and Bilene districts. Despite these localized efforts the overall trend points to a persistent and severe reduction in forest cover across the province. This trend is further exacerbated by the expansion of cropland and built-up areas driven by agricultural development and urbanization which have significantly altered the landscape.The Random Forest algorithm employed in this study proved highly effective in classifying landcover types, achieving overall classification accuracies between 80% and 86% across the different years analysed. The model was particularly accurate in distinguishing forest and cropland categories with user and producer accuracies consistently exceeding 80%. However, challenges arose in accurately classifying built-up and other land categories reflecting the complexity of these landcover types. The study's use of spectral indices like NDVI as well as topographic and texture measures contributed to the robustness of the model's performance. These findings underscore the critical need for more effective and scalable conservation strategies in Gaza Province. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Subject: | 43 environmental science |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/103421 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page