University of Twente Student Theses
Mapping cropland cover in Mozambique with expert knowledge and remote sensing
Lemgo, Lawrence Mawutor (2024) Mapping cropland cover in Mozambique with expert knowledge and remote sensing.
|
PDF
5MB |
| Abstract: | Creating accurate and up-to-date land cover maps are essential for managing natural resources, strategic land use planning, and monitoring changes that occur in our environment. This study employed remote sensing to evaluate the effectiveness of Sentinel-2 and PlanetScope satellite imagery in distinguishing different maize cropping systems in Gaza province, Mozambique. Using a Random Forest (RF) classifier, the research investigated the relative importance of spectral indices, textural, and topographic features derived from high spectral resolution (Sentinel-2) and high spatial resolution (PlanetScope) data. Key research questions addressed included identifying the most effective features for distinguishing maize cropping systems and determining whether the two datasets had a statistically significant difference in classification accuracy. The study area encompassed diverse landscapes characterized by smallholder farms and mixed cropping systems. Sentinel-2 and PlanetScope imagery from 2023 were processed using the Google Earth Engine (GEE) platform followed by the extraction of spectral, textural, and topographic features. The results indicated that elevation, red edge bands (Sentinel-2) and visible bands (PlanetScope) were among the most significant features for classification. Feature importance analysis using Mean Decrease Gini (MDG), SHapley Additive exPlanations (SHAP), and Permutation Importance methods consistently highlighted these features. The RF model showed high accuracy for classes like trees, shrubs, and water bodies but recorded low performance with grass and mixed fields primarily due to spectral overlaps and imbalanced training samples. Comparative analysis using McNemar’s test revealed no statistically significant difference in classification accuracy between Sentinel-2 and PlanetScope for discriminating maize cropping systems. Despite the high spatial resolution of PlanetScope its classification performance was similar to Sentinel-2. This emphasizes the importance of spectral resolution. This research contributes to the understanding of the strengths and limitations of high spectral and spatial resolution datasets in agricultural monitoring. It provides insights into cost-effective methods for accurate land cover mapping, essential for informed agricultural policymaking and resource management in heterogeneous landscapes like Gaza Province. The study also emphasizes the need for balanced datasets and high-precision data collection to improve model performance and reliability. |
| 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/101951 |
| 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