Spot5Take5: CNES-ESA simulate Sentinel-2 4

The farmland on the foothills of Mt. Kenya is characterized by a mix of small-scale subsistence and commercial large-scale farms. Due to the small-scale plots and the high cloud-coverage caused by two rain seasons as well as the daily weather pattern typical close to the equator it was so far not possible to successfully monitor cropland or even differentiate it by crop type. The 5-day data acquisition during the SPOT5Take5 campaign lasting about 5 months offered a great opportunity to observe cropland phenology more in detail. Based on this seasonal 5-day satellite data time series we first pre-processed and stacked the data. Then we performed supervised classification testing and evaluating a number of classification algorithms in order to identify cropland type as well as crop types.

Transfers of organic carbon from terrestrial ecosystems to surface water and oceans are key mechanism to global biogeochemical cycles. In particular, Arctic rivers drain organic rich watersheds, which contain half of the worldwide organic carbon stock in soils (Tarnocai et al., 2009). In the Arctic region, most of these transfers occur during the short freshet period (typically a few weeks). Despite its importance, this period is still under studied, mainly because the access to the arctic rivers is challenging in spring (Raymond et al., 2007).

Previous studies demonstrated that organic carbon in rivers can be remotely sensed by optical multispectral remote sensing, namely with Landsat imagery (Griffin et al., 2011). The colored fraction of dissolved organic carbon (DOC) which absorbs UV and visible light (Chin et al., 1994) is designed as chromophoric dissolved organic matter (cDOM). It is highly correlated to DOC in large arctic rivers and streams, allowing for remote sensing to monitor DOC concentrations from satellite imagery (Griffin et al., 2011).

The influence of cDOM absorption on remote sensing signal is the largest in the blue part of the spectrum but a combination of the green (520-600nm) and red (630-690nm) bands has recently been successfully used to monitor cDOM in surface waters of lakes (Kutser, 2012). Nevertheless, published studies on this topic are based on a sparse temporal sampling, sometimes with a time lag of several days between the date of sampling and the image acquisition (Griffin et al., 2011), which implies severe limitation on the validity of established models (Brezonik et al., 2015).

Therefore, high temporal and spatial resolutions brought by the SPOT5-Take 5 experiments (5 days, 20 m) are highly relevant for the study of DOC fluxes in a large Arctic river. The high temporal resolution allows correctly assessing this highly dynamic process, especially the spring freshet event (a few weeks in May). The high spatial resolution allows assessing the spatial variability within the stream and quantifying DOC transfer during the ice break period when the access to the river is almost impossible.

In this study, we estimate the temporal and spatial variability of DOC concentrations in the Arctic River Yenisei during 5 months from 9^th April to 28^th August 2015 with a particular focus in the freshet period (in May). First, thank to an intensive field campaign, the SPOT5-Take5 data have been calibrated using direct correspondence (same day) with in-situ measurements of DOC and DOM spectral absorbance (cDOM). Secondly, field samples were regressed against spectral data from green and red bands (bands 1 and 2) to develop empirical based algorithms.

Owing the strong correlation between DOC and cDOM, the annual variability has been assessed and resulting maps have been produced. Therefore, we observed a large spatial and temporal variability of DOC concentrations during the SPOT5-Take5 period. As expected, resulting maps highlight a strong intensification of the process during the freshet period (in May) with a gradual decrease until August.

Thus, such variability was previously never observed at this spatial and temporal scale, opening the way to several extensive analyses. In the context of rapid climate change, this study demonstrates that high resolutions brought by Sentinel-2 or Venus missions will be powerful and reliable tools to improve our understanding of carbon cycle in major Arctic rivers.

Grasslands are widely distributed across Europe and cover approximately 8 percent of Europe. The term 'grasslands' includes meadows, steppes and grasslands managed (grazing, cutting, burning) with variable management intensity, which largely overlap with agro-ecosystems. Grasslands are often defined as areas covered by grass-dominated vegetation with little or no tree cover. The location and geographic distribution of grasslands is an important factor for decision makers, because there is a wide range of legal and financial instruments affecting grasslands (eg. European Unions Common Agricultural Policy (CAP) or grassland habitats for nature conservation).

In Take 5 mission SPOT5 images are used as a simulator of the time series that Sentinel-2A and Sentinel-2B will provide. The SPOT5 (Take5) experiment started on April 8th, 2015 and finished on 8th September, 2015. Data are processed by THEIA land data center and are distributed with a free and open policy via the joint ESA-CNES webportal spot-take5.org.

Our study site is located near the Lake Balaton in Hungary. Twenty-two satellite images acquired during the experiment for our study area and 12 from them had cloud cover less than 5 percent. Grassland monitoring is a challenging task, because multiseasonal images required for accurate classification, and in most cases cloud cover or haze and pour revisit time of satellite does not allow it.

Normalized Difference Vegetation (NDVI) time series were used to select the key phenological dates for grassland mapping, because NDVI is one of the most commonly used indicator of vegetation phenology. Phenological profiles of land cover classes and vegetation type were derived from reference pixels extracted from SPOT5 images. We calculated separability using Jeffries Matusita Distance for signature extracted from NDVI images to determine the key phenological dates and minimum number of images required for grassland classification. Minimum and average values of Jeffries Matusita Disatance were the key parameter to determine the optimal dates for accurate classification.

We created different scenarios that used various number of satellite images and compared the accuracy of classification results. Hungarian Land Parcel Identification Systems (LPIS) were used data as reference data for accuracy assessment. This image database provided by Take 5 mission could be proper base for grassland mapping, but we found in our previous research that images from March and October can be useful for grassland delineation in Hungary. In future research, Sentinel-2 satellite images can be added to this database.

A significant correlation exists between a high concentration of parasites and bacteria and some water quality parameters including turbidity. Suspended matter can carry viruses and bacteria pathogenic to humans and foster their development. Therefore, turbidity can be considered a vector of microbiological contaminants, which cause diarrheal diseases. Few studies have focused on water turbidity in rural Africa, while many cases of intestinal parasitic infections are due to the consumption of unsafe water from ponds, lakes and rivers. Diarrheal diseases are the second cause of infant mortality in sub-Saharan Africa. So, monitoring of this parameter represents a challenge for health improvement. Turbidity refers to the optical properties of water, so it may be monitored by remote sensing. Its spatial and temporal variability requires the use of high temporal and spatial resolution sensors like SPOT5 images of SPOT5 TAKE5 program, in anticipation of Sentinel-2.

The particularly high turbidity values found in tropical Africa, compared to many other regions of the world, challenge the use of remote sensing and questions the methods developed for less turbid water. In addition, high aerosol loadings (mineral dust and biomass burning) may be detrimental to turbidity retrieval.

The objectives of our study are to monitor turbidity by remote-sensing in West Africa, to document surfaces water according to their turbidity levels so as to provide a first step toward mapping of health risks related to turbidity.

Turbidity and Surface Suspended Sediment Concentration (SSSC) measurement were carried out from 16 April 2015 to 13 September 2015 every 5 days, simultaneously with the acquisitions of SPOT5 TAKE5 images near the outlet of the Bagre reservoir. Moreover, a boat trips were made during July-August 2015. Turbidity, SSSC and hyperspectral radiometry (performed using TriOS RAMSES radiometers operating in the 350-900 nm) measurements were collected in different areas of the lake to study their spatial and temporal variability and to establish the best retrieving methods. The various algorithms and indices proposed in the literature for inland waters were applied to ground reflectance and SPOT5 TAKE5 series (19 images totally or partially exploitable). The data and algorithms have been evaluated with field measurements.

The first results with Modis images show a very good relationship between SSSC and ground reflectance. The beginning of the rainy season displays increasing the NIR/R reflectance, indicating increasing turbidity. Turbidity increases until the middle of the rainy season (August). The results show both spatial and temporal variation of the reflectance during the rainy season. Reflectance usually decreases downstream. The upstream areas reach maximum reflectance in June, the median zones in July-August, and the downstream in September. Moreover, differences in reflectance between the upstream and downstream areas are important. Analysis of SPOT5 TAKE5 image should confirm these results and propose more specific information related to the high spatial and temporal resolution. The high turbidity of the Bagre Lake indicates clear health risks. The turbidity data will be analyzed in view of medical statistics from health centers on diarrheal diseases as well practices and uses of populations to document areas and periods causing the highest risks to the population health.

The methodology developed here for the Bagre Lake could be applied to Sentinel-2 data which will be the best resolution for monitor turbidity and health risks.

In Tunisia, oases are facing lots of problems including abandoned plots, poor water management, salinity and/or hydromorphy of soils. To deal with this, developing a system for monitoring oases appears essential. The time series of SPOT5 TAKE5 data are used to test the implementation of such monitoring system based on the analysis of temporal radiometric signatures, on part of the year from April to September.

Main objective is to highlight the spring-summer temporal signatures that characterize oases from a representative sample taken in the same region of southern Tunisia (Djerid: Nefta-Tozeur): can we develop a typology of these oases in accordance with their own characteristics (crop diversity, spatial organization of plantations and plots, diversity of species...)?

The processings applied to SPOT5 TAKE5 data are based on five types of data: 1) the observations made in the field; 2) thematic data available on the biophysical environment (soil, geomorphology, cropping practices…); 3) the digital surface model from SRTM; 4) national meteorological data; 5) bimonthly MODIS 250m data available since 2001 (mod13 Products: NDVI and EVI).

Research activities are organized in two steps:

1. Unsupervised clustering. This first step assesses the information content and discrimination capabilities of these data, without a priori from a representative sample of oases. For this, different methods of analysis were tested: a) the so-called Empirical Orthogonal Functions (EOF), to identify the different patterns in temporal data; b) unsupervised clustering of multivariate time series, adapting conventional methods (K-means and Hierachical Agglomerative Clustering) using Euclidean distance date to date and dissimilarity measure based on Dynamic Time Warping.

2. Discriminant analysis. This second step is based on the extraction of simple descriptive attributes of temporal dynamics identified in the first step and by means of a discriminant analysis of different types of oasis. This analysis has been done through factor analysis and supervised clustering methods like decision trees. It leads to a reduced set of discriminating variables associated with classification "rules".

This work prepares the use of upcoming sensors such as the sentinel-2 series, for monitoring during the year the current "state" of oases. Each new image in time series is used to update the temporal profiles of oases and determine their types in terms of changes in cultural practices, stage of evolution or health of plots.