Flood Extent Mapping in the Caprivi Floodplain Using Sentinel-1 Time Series

Abstract

The deployment of the Sentinel-1 (S1) satellite con-1 stellation carrying a C-band Synthetic Aperture Radar (SAR) 2 enables regular and timely monitoring of floods from their 3 onset until returning to non-flooded (NF) conditions. The major 4 constraint on using SAR for near-real-time (NRT) flood mapping 5 has been the inability to rapidly process the obtained imagery 6 into reliable flood maps. This study evaluates the efficacy of 7 S1 time series for quantifying and characterising inundations 8 extents in vegetated environments. A novel algorithm based on 9 statistical time series modelling of flooded (F) and non-flooded 10 (NF) pixels is proposed for NRT flood monitoring. For each 11 new available S1 image, the probability of temporarily flooded 12 conditions is tested against that of NF conditions by means of 13 likelihood ratio tests. The likelihoods for the two conditions are 14 derived from early acquisitions in the time series. The algorithm 15 calibration consists of adjusting two likelihood ratio thresholds 16 to match the reference flooded area extent during a single flood 17 season.The proposed algorithm is applied to the Caprivi region 18 and the resulting maps were compared to cloud-free Landsat-8 19 (LS8) derived maps captured during two flood events. A good 20 spatial agreement, (85-87%), between LS8 and S1 flood maps was 21 observed during the flood peak in both 2017 and 2018 seasons.22 Significant discrepancies were noted during the flood expansion 23 and recession phases, mainly due to different sensitivities of the 24 data sources to the emerging vegetation. Overall, the analysis 25 shows that S1 can stand as an effective standalone or gap-filling 26 alternative to optical imagery during a flood event. 27 Index Terms-flood mapping, floating and emergent vegetation, 28 Sentinel-1, SAR, time series, land cover. 29 I. INTRODUCTION 30 Riparian areas, such as the Caprivi flood plain, are flooded 31 almost every year due to excess rainfall in the upper catch-32 ments [1]. Riverine flooding affects the ecological (e.g., 33 ecosystem productivity, species distribution and occurrence, 34 nutrients and sediment dynamics) and socio-economic systems 35

A novel change detection and threshold-based ensemble of scenarios pyramid for flood extent mapping using Sentinel-1 data

“…The ESP mean OA compares with the 85–87% found in Ref. [ 15 ] using change detection and thresholding based on combined optical and SAR data using a machine learning algorithm. Existing methods of FEM that include operationally challenging fieldwork [ 15 ], data intensive physical [ 67 ] and statistical-based models [ 42 ] and modern machine learning algorithms tend to have complex architectures and data challenges [ 66 ].…”

“…[ 15 ] using change detection and thresholding based on combined optical and SAR data using a machine learning algorithm. Existing methods of FEM that include operationally challenging fieldwork [ 15 ], data intensive physical [ 67 ] and statistical-based models [ 42 ] and modern machine learning algorithms tend to have complex architectures and data challenges [ 66 ]. To attain effective flood management objectives that include early warning, preparedness, evacuations and post flood recovery, detection and mapping of floods requires simple, fast, and accurate approaches [ 58 , 68 , 69 ].…”

“…Field surveys [ 12 ] and modelling [ 13 ] approaches have commonly been used for flood mapping. Typically, these approaches rely on field data and the often sparsely distributed weather stations, hence subjective, tedious, costly and prone to errors, especially in the developing countries with limited resources [ 14 , 15 ]. On the other hand, advances in remote sensing technology with freely available satellite images have opened up new opportunities for its adoption for flood mapping and monitoring over large spatial extents [ 15 , 16 ].…”

Impacts of seasonal flooding on geographical access to maternal healthcare in the Barotse Floodplain, Zambia

“…Optical sensors are sensitive to cloud cover which predominates during the wet season; extensive cloud cover above the Barotse Floodplain prevented floodwaters from being visible in MODIS imagery for the entire of March 2018, and others have reported difficulty in obtaining cloud-free imagery for the region [ 34 , 36 ]. Optical sensors also experience difficulty detecting vegetated waters, as do radar sensors albeit to a lesser extent [ 114 – 116 ]. Most readily-available flood extents rely on algorithms suitable for mapping open waters [ 115 , 117 ] leading to flood omission where inundated vegetation is present [ 113 ].…”

Mapping the 2021 October Flood Event in the Subsiding Taiyuan Basin by Multitemporal SAR Data

“…The ground range detected (GRD) and single look complex (SLC) data of VV polarization over the study area acquired by Sentinel-1 were collected for extracting flood change and time-series deformation, respectively. The accuracy of flood detection in VV polarization is higher than that in VH polarization [20], [21], so we selected the SAR images in VV polarization. The L2A products obtained by Sentinel-2 were processed to generate the modified normalized difference water index 3 > ID NUMBER < (MNDWI) [22] for validation.…”