Kirimi, Fridah: Remote sensing based assessment of land cover and soil moisture in the Kilombero floodplain in Tanzania. - Bonn, 2018. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5n-51590
@phdthesis{handle:20.500.11811/7614,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5n-51590,
author = {{Fridah Kirimi}},
title = {Remote sensing based assessment of land cover and soil moisture in the Kilombero floodplain in Tanzania},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2018,
month = aug,

note = {Wetlands provide important ecological, biological, and social-economic services that are critical for human existence. The increasing demand for food, arable land shortage and changing climate conditions in East Africa have created a paradigm shift from upland cultivation to wetland use due to their year-round soil water availability. However, there is need to control and manage the activities within the wetlands to ensure sustainable use while negating any negative effects caused by these activities. This is implemented through the decisions made by the land managers within the wetlands. Providing the users of the wetlands with scientific knowledge acts as a support tool for policy-making geared towards the sustainable use of the wetlands. The overall research contains two main components: First, the need for timely land cover maps at a reasonable scale, and secondly, the assessment of soil moisture as a major contributor to agricultural production.
The objectives of the study were to generate land cover maps from multi-sensor optical datasets and to assess the performance of single-polarized Sentinel-1 Gray Level Co-occurrence Matrix (GLCM) texture and Principal Component Analysis (PCA) features by applying multiple classification algorithms in a floodplain in the Kilombero catchment. Furthermore, soil moisture spatial-temporal patterns over three hydrological zones was assessed, estimation of soil moisture from radar data and generation of soil moisture products from global products was investigated. The correlation of the merged products to Normalized Difference Vegetation Index (NDVI) measures was also investigated.
RapidEye, Sentinel-2 and Landsat images were used in determining the areal extents of four major land cover classes namely vegetated, bare, water and built up. The acquisition period of the images ranges from August 2013 to June 2015 for the RapidEye images, December 2015 to August 2016 for the Sentinel-2 images and 2013 to 2016 Landsat-8 images were included in the land cover time series dynamic study. However, the major challenge arising was cloud coverage and hence Sentinel-1 images were tested in the application of Synthetic Aperture Radar (SAR) in wetland mapping. Variograms were used in spatial-temporal assessment of soil moisture data collected from three hydrological zones, riparian, middle and fringe. A roughness parameter was derived from a semi-empirical model. Soil moisture was retrieved from TerraSAR-X and RadarSAT-2 with the retrieved roughness parameter as an input in a linear regression equation. Triple collocation was applied in error assessment of the global soil moisture products prior to development of a merged product. Cross-correlation was applied in relating NDVI to soil moisture.
Optical data (RapidEye, Landsat-8, and Sentinel-2) generated land cover maps used in assessing the land cover dynamics over time. The land cover ratios were related to depth to groundwater. As the depth to groundwater reduced in June the bare land coverage was 45-57% while that of vegetation was 34-47%. In December when the depth to groundwater was highest, bare land coverage was 62-69% while that of the vegetated area was 27-25%. This indicates that depth of groundwater and vegetation coverage responds to seasonality. During the dry season, 68-81% of the total vegetation class is within the riparian zone.
In the classification of the SAR images, the overall accuracies for the single polarized VV images ranged from 54-76%, 60-81% and 61-80% for Random Forest (RF), Neural Network (NN) and Support Vector Machine (SVM) respectively. GLCM features had overall accuracies of 64-86%, 65-88% and 65-86% for RF, NN, and SVM respectively. PCA derived images had similar overall accuracies of 68-92% for NN, RF, and SVM respectively. The PCA images had the highest overall accuracy for the entire time series indicating that reduction in the number of texture features to layers containing the maximum variance improves the accuracy.
The standard deviation of soil moisture was noted to increase with increasing soil moisture. Soil texture plays a key role in soil moisture retention. The riparian fields had a high water content explained by the high clay and organic matter content. A roughness parameter was derived and utilized in the retrieval of soil moisture from SAR resulting to R2 of 0.88- 0.92 between observed and simulated soil moisture values from co-polarized RadarSAT-2 HH and TerraSAR-X HH and VV.
Merged soil moisture product from FEWSNET Land Data Assimilation System_NOAH (FLDAS_NOAH), ECMWF Re-Analysis Interim (ERA-Interim) and Soil Moisture and Ocean Salinity (SMOS) and FLDAS_Variable Infiltration Capacity (VIC), ERA-Interim and SMOS had similar patterns attributed to FLDAS_NOAH and FLDAS_VIC forced by the same precipitation product (RFE). Cross-correlation of Moderate-resolution Imaging Spectrometer (MODIS) NDVI and the merged soil moisture products revealed a 2-month lag of NDVI. Hence, the relationship is useful in determining the Start of Season from soil moisture products.
In conclusion, the successful land cover mapping of the study area demonstrated the use of satellite imagery for wetland characterization. The vast coverage and frequent acquisitions of optical and microwave remotely sensed data additionally make the approaches transferable to other locations and allow for mapping at larger scales. Soil moisture assessment from point data revealed varied soil moisture patterns whereas global remotely sensed and modeled products rather provide complementary information about growing conditions, and hence a situational assessment tool of potential of physical availability dimension of food security. This study forms a baseline upon which additional monitoring and assessment of the Kilombero wetland ecosystem can be performed with the current results marked as a reference. Moreover, the study serves as a demonstration case of remote sensing based approaches for land cover and soil moisture mapping, whose results are useful to stakeholders to aid in the implementation of adapted production techniques for yield optimization while minimizing the unsustainable use of the natural resources.},

url = {https://hdl.handle.net/20.500.11811/7614}
}

Die folgenden Nutzungsbestimmungen sind mit dieser Ressource verbunden:

InCopyright