ACCURACY ANALYSIS OF SENTINEL 2A AND LANDSAT 8 OLI+ SATELLITE DATASETS OVER KANO STATE (NIGERIA) USING VEGETATION SPECTRAL INDICES

This study explores the capabilities of Sentinel-2 over Landsat-8 Operational Land Imager (OLI) imageries for vegetation monitoring in the vegetated region of Minjibir LGA in Kano State. Accurate vegetation mapping is essential for monitoring crop and sustainable agricultural practice. Vegetation indices, comprising the Normalized Difference Vegetation Index (NDVI), Green Chlorophyll Index (GCI), Leaf Area Index (LAI) and Moisture Stress Index (MSI) were determined for each year. The findings showed an increase in Sentinel 2A value of the vegetation indices with respect to Landsat 8 throughout the time of the study (2015-2019). The best average performance over the supervised classification was obtained using Sentinel-2A bands, which are dependent on the training sample and resolution. While the spectral consistency of the data was inferred by cross-calibration analysis using regression analysis. The spatial consistency was assessed by descriptive statistical analysis of examined variables. Regarding the spatial consistency, the mean and standard deviation values of all variables were steady for all seasons excluding for the mean value of the LAI and MSI. Based on this finding, it is recommended that Sentinel-2A data could be used as a complementary data source with Landsat 8 OLI in vegetation assessment. * Corresponding author


General Instructions
Land-use mapping is a vital topic in the study of surface ecophysics, together with vegetation, soil, buildings, water, and other surface elements. Amongst them, vegetation is the most subtle to identifying surface climate change (Yuanhuizi et. al 2019). Additionally, vegetation is most closely associated with global and regional food security, planting intensity, crop yields, and other sustainable development goals. Future global climate change has increased the likelihood of severe, pervasive, and irreversible costs for human civilization and agriculture (IPCC, 2014). In the same perspective, the rapid and accurate mapping of vegetation has gradually become a crucial means for monitoring and evaluating agricultural development, and disaster monitoring/ management. Presently, remote sensing is an applied approach for vegetation appraisal by using vegetation variables, which varies vigorously in time and space (Yuanhuizi et.al 2019). The foremost role of environmental remote sensing to land resource management is its prospect to map vegetation resources and to observe changes that arise over prolonged epochs. Numerous satellite missions are launched with the most objective of observing changes within the vegetative cover over the world surface.
Most of such remote satellite missions are concerned with retrieving explicit vegetation parameters like the Normalized Vegetation Index (NDVI), the Fractional Vegetation Cover (FVC), Leaf Area Index (LAI), Green Chlorophyll Index (GCL), Soil Adjusted Vegetative Index (SAVI), Enhanced Vegetative Index (EVI), etc. (Melaas et al., 2013). Vegetation monitoring remains a fundamental focus within the science and practices of the remote sensing technology. However, the choice or accuracy of remote sensors used in vegetative monitoring remains of great importance. This study explores the Spatiotemporal changes between two remote sensing missions (Landsat 8 Operational Land Imager (OLI) and Sentinel 2A) and their spectral relationship in the region covering Minjibir Local Government Area (LGA) in the Kano State of Nigeria. This is with the objective of analysing the proficiencies of Sentinel-2 data over Landsat-8 OLI data for vegetation planning and monitoring. Four (4) Vegetation indices namely the NDVI, GCI, LAI, and MSI were determined for the period of 2015-2019.

Study Area
The study area is Minjibir LGA in Kano State. Minjibir lies between geographic latitude 12 o 00' 0'' and 12 o 20' 0'', and longitude 8 o 30' 00'' and 8 o 40'00''. The vegetation of Kano State is the semi-arid savannah. The Sudan Savannah is sandwiched by the Sahel Savannah in the north and the Guinea Savannah in the south. The savannah has been described as the zone that provides opportunity for optimal human attainment. This is because it is rich in faunal and floral resources, it is suitable for both cereal agriculture and livestock rearing, and the environment is relatively easy for movement of natural resources and manufactured goods. The figure 1 shows the study area.

Data Types and Sources
Landsat 8 OLI and Sentinel 2A for three different years (2015, 2017 and 2019) were obtained. These images were used to generate the land use and land cover information, and vegetation indices (NDVI, GCI, LAI and, MSI) within the study area.
Atmospherically corrected surface reflectances of multispectral bands of Landsat 8 OLI were freely downloaded from the website of the United State Geological Survey (USGS) (see, http://earthexplorer.usgs.gov/ ). The Sentinels Scientific Data Hub website (see, https://scibub.copernicus. eu.dhus/) provided free download of the Sentinel 2A Levels-1C products. The downloaded images from Sentinel data hub images were atmospherically corrected by means of the European Space Agency's (ESA) Sen2Cor atmospheric correction toolbox that is an inherent procedure within the SentiNel Application Platform (SNAP) tool version 6.0 to supply the Level-2A (L2A) products.
The Table 1

Image Pre-processing
The portion of interest (Minjibir LGA) was subsetted from each of the larger scenes using ArcGIS software. Geometric and radiometric corrections were performed on them for the purpose of ortho-rectification. The Sentinel 2A images were obtained at spatial resolution of 10 m. Therefore, the Sentinel-2A images were up scaled to the same spatial resolution as the Landsat-8 (i.e. 30 m) to match the performances of the datasets of two satellites within the same spatial-scale. The two datasets were geo-referenced or geo-coded that is registered to a geographic frame of reference (UTM Zone 32).
During layer stacking, all bands of the sensor data excluding the thermal band were considered for Layers stacking. The nature of these different bands had to be considered to make a decision as to which three-band combination would be most helpful for classification and visual interpretation, thus the false colour composite was employed in this study.

Image Classification and Accuracy Assessment
A supervised classification was performed on false colour composites (bands 5, 4 and 3 for Landsat 8 OLI and bands 8, 4 and 3 for Sentinel 2A) into the following land use and land cover classes; Light vegetation Built-up area, Dense vegetation, Water body and Bare surfaces(see Table 2). The classification was done according to Anderson et al (1976). Information collected during the field surveys were combined with the digital satellite image, which was derived from SAS-planet software and used to assess the accuracy of the classification. Bare surfaces Lands devoid of vegetation, exposed soil 5 Waterbody Land dominated by rivers, lake and dams An accuracy assessment was done by determining a confusion matrix. This finds the relationships between the mapped class label and that observed on the ground or reference data for a sample of cases at specific locations. The overall accuracy can be determined by dividing the number of correctly classified pixels by the total number of reference pixels. Overall accuracy is considered as the most suitable method for calculating accuracy assessment. The Kappa coefficient of agreement was used to improve the overall accuracy.

Estimating Vegetative Indices
Landsat 8 OLI of 2015, 2017 and 2019 and Sentinel 2AMSI of 2015, 2017 and 2019 were used for estimating vegetative indices and subsequent analysis. Digital number was converted to spectral radiance for the Landsat 8OLI and then into reflectance. Different vegetative types have different spectral characteristics. Based on the understanding of the satellite spectral data, we obtained different spectral information regarding vegetation in the study area. Four (4) vegetative indices derived at all the satellite epochs as described in the succeeding subsection

Derivation of Moisture Stress Index(MSI)
This index is sensitive to the increase of leaf water content. It is used for analyzing vegetation coverage, predicting the productivity and modeling, analyzing the plant use conditions and studying the ecosystem physiology. The MSI is calculated as a ratio of the mid-infrared (MIR) and the near-infrared (NIR) (Hunt et al., 1989;Welikhe et al., 2017). The formula used to derive the MSI is shown in equation (1).
• Derivation of Green Chlorophyll Index (GCL) In remote sensing, the Green Chlorophyll Index is employed to estimate the content of leaf chlorophyll in countless species of plants. The chlorophyll content reveals the physiological condition of vegetation; it drops in strained plants and may therefore be used as a measurement of plant well-being. Enhanced estimation of chlorophyll amount with the GCI are often achieved by using satellite sensors that have broad NIR and green wavelengths (Gitelson et al., 2003). The formula used to derive the GCI is shown in equation (2).
The NDVI is the most important vegetation index in remote sensing. It is widely used for analysing land use changes, including vegetation and other factors. This index is suitable for areas with moderate and higher vegetation density since it is less susceptible to soil and the effects of atmosphere.
The NDVI is calculated from the visible red and near infrared bands. The rationale of the index is that healthy vegetation has a high reflectance in the near infrared (NIR) and a low reflectance in the red, thereby enhancing the interpretation of vegetation cover while suppressing subtle noise from other land cover types (Rouse et al., 1974). The formula used to derive NDVI is shown in equation (3).
Where r is reflectance and is defined as in equation (4), in the equation (4), L is spectral radiances at the sensor aperture, dr is the inverse square of earth-sun distance, sun E represents the mean solar exoatmospheric irradiances, θ is the solar zenith angle and, d is the distance from the earth to the sun.
• Derivation of Leaf Area Index(LAI) LAI can be determined directly by taking a statistically significant sample of foliage from a plant canopy, measuring the leaf area per sample plot and dividing it by the plot land surface area. Indirect methods measure canopy geometry or light extinction and relate it to LAI (Juutinen et al., 2017). This study adopted a relation between the NDVI and LAI as presented in equation (5) Total 431.14, 431.14 431.14, 431.14 431.14, 431.14 ** S2A = Sentinel2A, L8 = Landsat8 OLI

Image Classification Accuracy
Accuracy analysis was undertaken using the confusion matrix otherwise referred to as the error matrix presented in Tables 4. The confusion matrix involves different statistical measures such as producer's accuracy and user's accuracy for each of the classes, after which, the overall accuracy and kappa index for the classification were determined. The producer's accuracy was obtained by dividing the total number of pixels classified correctly in a category by the total number of pixels of that category as derived from the reference data. While the user accuracy, on the other hand, is expressed as the ratio of correctly classified pixels to the total number of pixels classified in that class. Hence, the result of the Landsat 8 (  Table 4: Classification Accuracy Assessment Report The overall accuracy is the number of correctly classified pixels (sum of the diagonal cells) divided by the total number of pixels checked. The overall accuracy of Sentinel 2A was found to be 98%, 98% and 98% for the three-epoch period with a kappa index of 70%, 78%, and 79% (Table 4). While the overall accuracy of Landsat 8 was found to be 88%, 93% and 91% for the three-epoch period with a kappa index of 72%, 76%, and 74%.In the accuracy assessment test results presented in Table  4, it is quite clear that Sentinel 2A has better user accuracy and producer accuracy for built-up, water, bare land and vegetation classes compared to Landsat 8.

Comparative Analysis of Landsat 8 OLI and Sentinel 2A Derived NDVI
The NDVI value as presented in Table 5 revealed that lower NDVI is associated with the developed settlements while high NDVI values are associated with the less developed natural surfaces. The results of NDVI maps are shown in appendix2.
In  Table 6.      In this study, the multi-temporal analysis was performed on two different satellite sensor (Sentinel 2Aand Landsat 8 OLI for Minjibir LGA of Kano State. Through the retrieved vegetation indices, it was observed that high vegetation indices are attributed with Sentinel 2Avalue with respect to Landsat 8 throughout the time of study. From the comparison, it was found that the supervised classification from both the two sources are corresponding. Correlation was also made between the NDVI, MSI, GCI and LAI of the Landsat 8 and Sentinel 2A data. The result of correlation between Landsat 8 and Sentinel 2A for the indices shows that the two are weakly correlated. However, the best average performance over the supervised classification was obtained using Sentinel-2 bands. The models based on Sentinel-2 data outperformed Landsat 8 models for all forest variables. Results were clearly shown in Sentinel 2A supervised classification due to the fine spatial resolution of 10 m. The study recommends the integration of datasets from satellite remoting sensing sensors for improved or optimized vegetation monitoring.