Detecting, Extracting, and Monitoring Surface Water From Space Using Optical Sensors: A Review

A robust Multi-Band Water Index (MBWI) for automated extraction of surface water from Landsat 8 OLI imagery

Satellites allow large-scale surveys to be conducted in short time periods with repeat surveys possible at intervals of <24 h. Very-high-resolution satellite imagery has been successfully used to detect and count a number of wildlife species in open, homogeneous landscapes and seascapes where target animals have a strong contrast with their environment. However, no research to date has detected animals in complex heterogeneous environments or detected elephants from space using very-high-resolution satellite imagery and deep learning. In this study, we apply a Convolution Neural Network (CNN) model to automatically detect and count African elephants in a woodland savanna ecosystem in South Africa. We use WorldView-3 and 4 satellite data –the highest resolution satellite imagery commercially available. We train and test the model on 11 images from 2014 to 2019. We compare the performance accuracy of the CNN against human accuracy. Additionally, we apply the model on a coarser resolution satellite image (GeoEye-1) captured in Kenya, without any additional training data, to test if the algorithm can generalize to an elephant population outside of the training area. Our results show that the CNN performs with high accuracy, comparable to human detection capabilities. The detection accuracy (i.e., F2 score) of the CNN models was 0.78 in heterogeneous areas and 0.73 in homogenous areas. This compares with the detection accuracy of the human labels with an averaged F2 score 0.77 in heterogeneous areas and 0.80 in homogenous areas. The CNN model can generalize to detect elephants in a different geographical location and from a lower resolution satellite. Our study demonstrates the feasibility of applying state-of-the-art satellite remote sensing and deep learning technologies for detecting and counting African elephants in heterogeneous landscapes. The study showcases the feasibility of using high resolution satellite imagery as a promising new wildlife surveying technique. Through creation of a customized training dataset and application of a Convolutional Neural Network, we have automated the detection of elephants in satellite imagery with accuracy as high as human detection capabilities. The success of the model to detect elephants outside of the training data site demonstrates the generalizability of the technique.

Using very‐high‐resolution satellite imagery and deep learning to detect and count African elephants in heterogeneous landscapes

Satellites allow large-scale surveys to be conducted in short time periods with repeat surveys possible In this study we apply a Convolution Neural Network (CNN) model to automatically detect and count African elephants in a woodland savanna ecosystem in South Africa. We use WorldView-3 and 4 satellite data – the highest resolution satellite imagery commercially available. We train and test the model on eleven images from 2014-2019. We compare the performance accuracy of the CNN against human accuracy. Additionally, we apply the model on a coarser resolution satellite image (GeoEye-1) captured in Kenya to test if the algorithm can generalise to an elephant population outside of the training area. Our results show the CNN performs with high accuracy, comparable to human detection capabilities. The detection accuracy (i.e., F2 score) of the CNN models was 0.78 in heterogeneous areas and 0.73 in homogenous areas. This compares with the detection accuracy of the human labels with an averaged F2 score 0.77 in heterogeneous areas and 0.80 in homogenous areas. The CNN model can generalise to detect elephants in a different geographical location and from a lower resolution satellite. Our study demonstrates the feasibility of applying state-of-the-art satellite remote sensing and deep learning technologies for detecting and counting African elephants in heterogeneous landscapes. The study showcases the feasibility of using high resolution satellite imagery as a promising new wildlife surveying technique. Through creation of a customised training dataset and application of a Convolutional Neural Network, we have automated the detection of elephants in satellite imagery with as high accuracy as human detection capabilities. The success of the model to detect elephants outside of the training data site demonstrates the generalisability of the technique.

https://biorxiv.org/content/10.1101/2020.09.09.289231v1.full.pdf

Using very high-resolution satellite imagery and deep learning to detect and count African elephants in heterogeneous landscapes

Wetland hydroperiod consists of different compo- nents, including frequency, duration and depth. A significant proportion of the seasonally flood-pulsed Okavango Delta is inundated for part of each year. Variation in hydroperiod, driven by the interaction of climate and ecological factors, results in a mosaic of vegetation communities. These commu- nities are highly dynamic over temporal and spatial scales. This study aimed to identify quantitative relationships be- tween hydroperiod components and floodplain vegetation, to better understand the potential effects of hydrological change. A stratified random sample of 30 floodplains was surveyed for species composition and abundance. Hydroperiod compo- nents for samplequadrats were estimated fromremotesensing and field measurement. Ordination demonstrated strong cor- relations between species composition and flood frequency, duration, years-since-last-flood and depth. Eight statistically discrete vegetation communities were derived from cluster analysis. Generalized linear models of species distributions along hydrological gradients were used to derive frequency and duration optima. Means of these parameters differed significantly between communities. Increasing hydroperiod corresponded with a progression from grasslands through sedge-dominated to aquatic communities. Species in infre- quently flooded areas indicated disturbance and a succession trend towards open woodland. In the sedgelands, average depth was the strongest correlate, while in grasslands and aquatic communities, this was duration.

/pdf/disaggregating-hydroperiod-components-of-the-seasonal-flood-50wm3adfxa.pdf

Disaggregating Hydroperiod: Components of the Seasonal Flood Pulse as Drivers of Plant Species Distribution in Floodplains of a Tropical Wetland

Keeping it simple: Monitoring flood extent in large data-poor wetlands using MODIS SWIR data

Characterising hydroperiod and vegetation for flood-pulsed wetlands is a critical first step towards understanding their ecology. In large, data-poor wetlands such as Botswana’s Okavango Delta, quantifying hydrology and ecology presents great logistic and financial challenges, yet relationships between hydrology and floodplain ecology are essential inputs to management. This paper describes an approach to improving ecological understanding by seeking relationships between archival remote sensing data and floodplain vegetation data. We produced a high spatial resolution (30 × 30 m) time series of annual flood frequency from Landsat 5TM imagery for the period 1989–2009. A second, lower spatial resolution (250 × 250 m) series of monthly flood extent was developed from a band 1 (0.62–0.67 μm) threshold of MODIS (MOD09Q1) imagery for the period 2000–2012. Vegetation composition and abundance was sampled in 30 floodplain sites, using a modified Braun-Blanquet approach. Interpreted flood extent from MODIS was 92 % accurate compared to the Landsat interpretation, and 89 % accurate when assessed against field data. Three major classes of floodplain vegetation were identified from ordination and cluster analysis: Occasionally flooded savanna, Seasonally flooded grassland, and Seasonally flooded sedgeland. Relationships identified between hydroperiod and vegetation communities were tested against five validation sites, in four of which indicator species occurrence was predicted with ≥60 % accuracy. The methods used are simple, objective, repeatable and inexpensive. Relating floodplain vegetation to hydrological history provides a means of predicting shifts in species composition and abundance for given changes in hydrology.

Remote Sensing-derived hydroperiod as a predictor of floodplain vegetation composition

The frequency and duration of inundation in river systems are important for a variety of applications, such as water resource management, floodplain mapping and habitat restoration. The Okavango Delta, the fifth largest Ramsar site in the world, has experienced a series of large floods since 2007, following the much lower flooding of the 1990s and early 2000s. This study aimed to establish inundation frequency and duration maps of the Okavango Delta for 2001 to 2012 using MOD09Q1 and MOD11A1. An earlier methods paper testing MODIS for flood mapping in the delta showed 99.4% overall accuracy for inundation mapping of the delta, with a kappa coefficient of 80%. A method based on a time-varying threshold derived from bimodal single-band histograms was used to classify the images into inundated and non-inundated areas. The method relies on the difference in reflectance between waterbodies and dry or vegetated soil, with the former having low reflectance and the latter displaying higher reflectance values. The...

Mapping inundation extent, frequency and duration in the Okavango Delta from 2001 to 2012

Knowledge of wetland vegetation spectral reflectance signatures can assist in spectral classification of remotely sensed images for monitoring of wetland hydroperiod. This study aimed at assessing the differences between wetland vegetation communities of varying species composition and density in terms of spectral reflectance. The investigation was carried out in floodplains at Nxaraga and Seronga in the Okavango Delta, Botswana. Spectral measurements were conducted during rising and receding flood stages. In each study area, 2 transects were located in homogeneous macrophyte stands, and in each transect 5–10 1 m2 quadrats were randomly set. In each quadrat, water depth and cover percentages (green leaved and senescent) of macrophyte vegetation were recorded, and a full reflectance spectrum between 325 and 1,075 nm wavelengths captured using a handheld spectroradiometer. Multiple regression analysis with stepwise selection of significant variables based on AIC was performed to determine the influence of various stand characteristics on spectral reflectance and vegetation indices (NDVI and EVI). The results showed that stand characteristics explained a large proportion of variance in spectral vegetation indices with r2 = 0.67 for NDVI and r2 = 0.75 for EVI. For NDVI, the significant explanatory variables were percentage green cover, senescent cover and water depth, while for EVI it was the total vegetation cover. Analysis of regression residuals for the various vegetation community classes showed significant differences between the classes in their NDVI, but no differences in EVI, with the exception of the Panicum repens class. Reflectance in visible wavelengths did not vary significantly between sites and seasons, but the NIR reflectance differed across sites and seasons. The results indicate that the simple vegetation indices in the Okavango Delta respond strongly to stand characteristics and thus can be used to derive these from remote sensing imagery. The scope for determination of vegetation communities from these indices is limited, however.

K. Thito

Papers

Keeping it simple: Monitoring flood extent in large data-poor wetlands using MODIS SWIR data

Remote Sensing-derived hydroperiod as a predictor of floodplain vegetation composition

Mapping inundation extent, frequency and duration in the Okavango Delta from 2001 to 2012

Spectral reflectance of floodplain vegetation communities of the Okavango Delta.