Automatic Generation of Water Masks from RapidEye Images

Abstract

Water is a very important natural resource and it supports all life forms on earth. It is used by humans in various ways including drinking, agriculture and for scientific research. The aim of this research was to develop a routine to automatically extract water masks from RapidEye images, which could be used for further investigation such as water quality monitoring and change detection. A Python-based algorithm was therefore developed for this particular purpose. The developed routine combines three spectral indices namely Simple Ratios (SRs), Normalized Green Index (NGI) and Normalized Difference Water Index (NDWI). The two SRs are calculated between the NIR and green band, and between the NIR and red band. The NGI is calculated by rationing the green band to the sum of all bands in each image. The NDWI is calculated by differencing the green to the NIR and dividing by the sum of the green and NIR bands. The routine generates five intermediate water masks, which are spatially intersected to create a single intermediate water mask. In order to remove very small waterbodies and any remaining gaps in the intermediate water mask, morphological opening and closing were performed to generate the final water mask. This proposed algorithm was used to extract water masks from some RapidEye images. It yielded an Overall Accuracy of 95% and a mean Kappa Statistic of 0.889 using the confusion matrix approach.

Automated SAR Image Thresholds for Water Mask Production in Alberta’s Boreal Region

“…This is an active area of research, and therefore, multiple methods for surface water detection using remote sensing exist. A brief summary of state-of-the-art surface water masks is provided, where the developed methods were applied to either optical [23][24][25] or SAR data [26,27].…”

Predicting terrestrial dispersal corridors of the invasive African clawed frog Xenopus laevis in Portugal

“…Based on remote sensing variables using a threshold-based water detection method, the larger still and flowing freshwater bodies within the study area were detected (Klemenjak et al 2012;Tetteh and Schönert 2015). To improve the algorithms' capacity to detect water, the precise locations of all verified water bodies were identified manually in Google Earth and used to train a bioclim model using the DISMO package for R (Hijmans, Phillips et al 2013;R Core Team 2017).…”

Small reservoirs for a sustainable water resources management

“…Both Sentinel-2 platforms are equipped with a multispectral sensor that acquires 13 spectral bands with 12 bit radiometric resolution and different geometric resolutions at 10 m (blue, green, red and NIR), 20 m (six bands including SWIR) and 60 m, as reported in Table 1. Remote sensed data are used for lake identification through water index (WI) evaluation by means of the NDWI (Normalized Difference Water Index) and the MNDWI (Modified Normalized Difference Water Index; Xiucheng et al, 2017;Gordana and Ugur, 2017;Yun et al, 2016;Dominici et al, 2019;Avisse et al, 2017;Gideon and Maurice, 2015;Mishra and Prasad, 2015).…”

“…The water body identification using remote sense data was widely investigated in several geographical contexts. Different satellite platform products were used to compute the NDWI and MNDWI indices (Di Francesco and Giannone, 2019): Landsat images produced an accuracy of extracted water bodies greater than 88 % (Mishra and Prasad, 2015), while for RapidEye imagery the overall accuracy of 95 % was reached (Gideon and Maurice, 2015). Finally, Sentinel-2 multispectral images were tested in Venice coastland (Italy) obtaining an overall accuracy greater than 94 % (Yun et al, 2016) and in the urban area of Beijing and Yantai (China) obtaining an overall accuracy greater than 97 % (Xiucheng et al, 2017).…”