Browsing by Author "Mashimbye, Zama Eric"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Assessing the influence of DEM source on derived streamline and catchment boundary accuracy
    (South African Water Research Commission, 2019) Mashimbye, Zama Eric; De Clercq, Willem Petrus; Van Niekerk, Adriaan
    Accurate DEM-derived streamlines and catchment boundaries are essential for hydrological modelling. Due to the popularity of hydrological parameters derived mainly from free DEMs, it is essential to investigate the accuracy of these parameters. This study compared the spatial accuracy of streamlines and catchment boundaries derived from available digital elevation models in South Africa. Two versions of Stellenbosch University DEMs (SUDEM5 and DEMSA2), the second version of the 30 m advanced spaceborne thermal emission and reflection radiometer global digital elevation model (ASTER GDEM2), the 30 and 90 m shuttle radar topography mission (SRTM30 and SRTM90 DEM), and the 90 m Water Research Commission DEM (WRC DEM) were considered. As a reference, a 1 m GEOEYE DEM was generated from GeoEye stereo images. Catchment boundaries and streamlines were extracted from the DEMs using the Arc Hydro module. A reference catchment boundary was generated from the GEOEYE DEM and verified during field visits. Reference streamlines were digitised at a scale of 1:10 000 from the 1 m orthorectified GeoEye images. Visual inspection, as well as quantitative measures such as correctness index, mean absolute error, root mean squares error and figure of merit index were used to validate the results. The study affirmed that high resolution (<30 m) DEMs produce more accurate parameters and that DEM source and resampling techniques also play a role. However, if high resolution DEMs are not available, the 30 m SRTM DEM is recommended as its vertical accuracy was relatively high and the quality of the streamlines and catchment boundary was good. In addition, it was found that the novel Euclidean distance-based MAE and RMSE proposed in this study to compare reference and DEM-extracted raster datasets of different resolutions is a more reliable indicator of geometrical accuracy than the correctness and figure of merit indices.
  • Thumbnail Image
    Item
    Climate-based regionalization and inclusion of spectral indices for enhancing transboundary land-use/cover classification using deep learning and machine learning
    (MDPI, 2021) Kavhu, Blessing; Mashimbye, Zama Eric; Luvuno, Linda
    Accurate land use and cover data are essential for effective land-use planning, hydrological modeling, and policy development. Since the Okavango Delta is a transboundary Ramsar site, managing natural resources within the Okavango Basin is undoubtedly a complex issue. It is often difficult to accurately map land use and cover using remote sensing in heterogeneous landscapes. This study investigates the combined value of climate-based regionalization and integration of spectral bands with spectral indices to enhance the accuracy of multi-temporal land use/cover classification using deep learning and machine learning approaches. Two experiments were set up, the first entailing the integration of spectral bands with spectral indices and the second involving the combined integration of spectral indices and climate-based regionalization based on Koppen–Geiger climate zones. Landsat 5 TM and Landsat 8 OLI images, machine learning classifiers (random forest and extreme gradient boosting), and deep learning (neural network and deep neural network) classifiers were used in this study. Supervised classification using a total of 5140 samples was conducted for the years 1996, 2004, 2013, and 2020. Average overall accuracy and Kappa coefficients were used to validate the results. The study found that the integration of spectral bands with indices improves the accuracy of land use/cover classification using machine learning and deep learning. Post-feature selection combinations yield higher accuracies in comparison to combinations of bands and indices. A combined integration of spectral indices with bands and climate-based regionalization did not significantly improve the accuracy of land use/cover classification consistently for all the classifiers (p < 0.05). However, post-feature selection combinations and climate-based regionalization significantly improved the accuracy for all classifiers investigated in this study. Findings of this study will improve the reliability of land use/cover monitoring in complex heterogeneous TDBs.
  • Thumbnail Image
    Item
    Remote sensing of salt-affected soils
    (Stellenbosch : Stellenbosch University, 2013-03) Mashimbye, Zama Eric; De Clercq, W. P.; Van Niekerk, Adriaan; Stellenbosch University. Faculty of AgriSciences. Dept. of Soil Science.
    ENGLISH ABSTRACT: Concrete evidence of dryland salinity was observed in the Berg River catchment in the Western Cape Province of South Africa. Soil salinization is a global land degradation hazard that negatively affects the productivity of soils. Timely and accurate detection of soil salinity is crucial for soil salinity monitoring and mitigation. It would be restrictive in terms of costs to use traditional wet chemistry methods to detect and monitor soil salinity in the entire Berg River catchment. The goal of this study was to investigate less tedious, accurate and cost effective techniques for better monitoring. Firstly, hyperspectral remote sensing (HRS) techniques that can best predict electrical conductivity (EC) in the soil using individual bands, a unique normalized difference soil salinity index (NDSI), partial least squares regression (PLSR) and bagging PLSR were investigated. Spectral reflectance of dry soil samples was measured using an analytical spectral device FieldSpec spectrometer in a darkroom. Soil salinity predictive models were computed using a training dataset (n = 63). An independent validation dataset (n = 32) was used to validate the models. Also, field-based regression predictive models for EC, pH, soluble Ca, Mg, Na, Cl and SO4 were developed using soil samples (n = 23) collected in the Sandspruit catchment. These soil samples were not ground or sieved and the spectra were measured using the sun as a source of energy to emulate field conditions. Secondly, the value of NIR spectroscopy for the prediction of EC, pH, soluble Ca, Mg, Na, Cl, and SO4 was evaluated using 49 soil samples. Spectral reflectance of dry soil samples was measured using the Bruker multipurpose analyser spectrometer. “Leave one out” cross validation (LOOCV) was used to calibrate PLSR predictive models for EC, pH, soluble Ca, Mg, Na, Cl, and SO4. The models were validated using R2, root mean square error of cross validation (RMSECV), ratio of prediction to deviation (RPD) and the ratio of prediction to interquartile distance (RPIQ). Thirdly, owing to the suitability of land components to map soil properties, the value of digital elevation models (DEMs) to delineate accurate land components was investigated. Land components extracted from the second version of the 30-m advanced spaceborne thermal emission and reflection radiometer global DEM (ASTER GDEM2), the 90-m shuttle radar topography mission DEM (SRTM DEM), two versions of the 5-m Stellenbosch University DEMs (SUDEM L1 and L2) and a 5-m DEM (GEOEYE DEM) derived from GeoEye stereo-images were compared. Land components were delineated using the slope gradient and aspect derivatives of each DEM. The land components were visually inspected and quantitatively analysed using the slope gradient standard deviation measure and the mean slope gradient local variance ratio for accuracy. Fourthly, the spatial accuracy of hydrological parameters (streamlines and catchment boundaries) delineated from the 5-m resolution SUDEM (L1 and L2), the 30-m ASTER GDEM2 and the 90-m SRTM was evaluated. Reference catchment boundary and streamlines were generated from the 1.5-m GEOEYE DEM. Catchment boundaries and streamlines were extracted from the DEMs using the Arc Hydro module for ArcGIS. Visual inspection, correctness index, a new Euclidean distance index and figure of merit index were used to validate the results. Finally, the value of terrain attributes to model soil salinity based on the EC of the soil and groundwater was investigated. Soil salinity regression predictive models were developed using CurveExpert software. In addition, stepwise multiple linear regression soil salinity predictive models based on annual evapotranspiration, the aridity index and terrain attributes were developed using Statgraphics software. The models were validated using R2, standard error and correlation coefficients. The models were also independently validated using groundwater hydro-census data covering the Sandspruit catchment. This study found that good predictions of soil salinity based on bagging PLSR using first derivative reflectance (R2 = 0.85), PLSR using untransformed reflectance (R2 = 0.70), a unique NDSI (R2 = 0.65) and the untransformed individual band at 2257 nm (R2 = 0.60) predictive models were achieved. Furthermore, it was established that reliable predictions of EC, pH, soluble Ca, Mg, Na, Cl and SO4 in the field are possible using first derivative reflectance. The R2 for EC, pH, soluble Ca, Mg, Na, Cl and SO4 predictive models are 0.85, 0.50, 0.65, 0.84, 0.79, 0.81 and 0.58 respectively. Regarding NIR spectroscopy, validation R2 for all the PLSR predictive models ranged from 0.62 to 0.87. RPD values were greater than 1.5 for all the models and RMSECV ranged from 0.22 to 0.51. This study affirmed that NIR spectroscopy has the potential to be used as a quick, reliable and less expensive method for evaluating salt-affected soils. As regards hydrological parameters, the study concluded that valuable hydrological parameters can be derived from DEMs. A new Euclidean distance ratio was proved to be a reliable tool to compare raster data sets. Regarding land components, it was concluded that higher resolution DEMs are required for delineating meaningful land components. It seems probable that land components may improve salinity modelling using hydrological modelling and that they can be integrated with other data sets to map soil salinity more accurately at catchment level. In the case of terrain attributes, the study established that promising soil salinity predictions could be made based on slope, elevation, evapotranspiration and terrain wetness index (TWI). Stepwise multiple linear regressions soil salinity predictive model based on elevation, evapotranspiration and TWI yielded slightly more accurate prediction of soil salinity. Overall, the study showed that it is possible to enhance soil salinity monitoring using HRS, NIR spectroscopy, land components, hydrological parameters and terrain attributes.
  • Thumbnail Image
    Item
    Remote sensing-based identification and mapping of salinised irrigated land between Upington and Keimoes along the lower Orange River, South Africa
    (Stellenbosch : University of Stellenbosch, 2005-04) Mashimbye, Zama Eric; Zietsman, H. L.; University of Stellenbosch. Faculty of Arts and Social Sciences. Dept. of Geography and Environmental Studies.
    Salinisation is a major environmental hazard that reduces agricultural yields and degrades arable land. Two main categories of salinisation are: primary and secondary soil salinisation. While primary soil salinisation is caused by natural processes, secondary soil salinisation is caused by human factors. Incorrect irrigation practices are the major contributor to secondary soil salinisation. Because of low costs and less time that is associated with the use of remote sensing techniques, remote sensing data is used in this study to identify and map salinised irrigated land between Upington and Keimoes, Northern Cape Province, in South Africa. The aim of this study is to evaluate the potential of digital aerial imagery in identifying salinised cultivated land. Two methods were used to realize this aim. The first method involved visually identifying salinised areas on NIR, and NDVI images and then digitizing them onscreen. In the second method, digital RGB mosaicked, stacked, and NDVI images were subjected to unsupervised image classification to identify salinised land. Soil samples randomly selected and analyzed for salinity were used to validate the results obtained from the analysis of aerial photographs. Both techniques had difficulties in identifying salinised land because of their inability to differentiate salt induced stress from other forms of stress. Visual image analysis was relatively successful in identifying salinised land than unsupervised image classification. Visual image analysis correctly identified about 55% of salinised land while only about 25% was identified by unsupervised classification. The two techniques predict that an average of about 10% of irrigated land is affected by salinisation in the study area. This study found that although visual analysis was time consuming and cannot differentiate salt induced stress from other forms; it is fairly possible to identify areas of crop stress using digital aerial imagery. Unsupervised classification was not successful in identifying areas of crop stress.
  • Thumbnail Image
    Item
    A synthesizing land-cover classification method based on Google Earth Engine : a case study in Nzhelele and Levhuvu catchments, South Africa
    (Springer, 2020-07-07) Zeng, Hongwei; Wu, Bingfang; Wang, Shuai; Musakwa, Walter; Tian, Fuyou; Mashimbye, Zama Eric; Poona, Nitesh; Syndey, Mavengahama
    This study designed an approach to derive land-cover in the South Africa with insufficient ground samples, and made a case demonstration in Nzhelele and Levhuvu catchments, South Africa. The method was developed based on an integration of Landsat 8, Sentinel-1, and Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM), and the Google Earth Engine (GEE) platform. Random forest classifier with 300 trees is employed as land-cover classification model. In order to overcome the defect of insufficient ground data, the stratified sampling method was used to generate the training and validation samples from the existing land-cover product. Likewise, in order to recognize different land-cover categories, the percentile and monthly median composites were employed to expand input metrics of random forest classifier. Results showed that the overall accuracy of the land-cover of Nzhelele and Levhuvu catchments, South Africa in 2017–2018 reached to 76.43%. Three important results can be drawn from our research. 1) The participation of Sentinel-1 data can slightly improve overall accuracy of land-cover while its contribution on land-cover classification varied with land types. 2) Under-fitting problem was observed in the training of non-dominant land-cover categories using the random sampling, the stratified sampling method is recommended to make sure the classification accuracy of non-dominant classes. 3) When related reflectance bands participated in the training process, individual Normalized Difference Vegetation index (NDVI), Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI), Normalized Difference Built-up Index (NDBI) have little effect on final land-cover classification result.