NEAR REAL TIME FLOOD MONITORING TOOL

DITAG - Politecnico di Torino, Corso Duca degli Abruzzi 24, 10135 Torino –franca.disabato@polito.it KEY WORDS: Remote Sensing, Flood, Monitoring, MODIS ABSTRACT: Risk reduction and disaster respond experts working on flood related issues could greatly benefit from the availability of information such as the real extent of past flood events and the monitoring of flood extent during ongoing events. To respond to the afore mentioned needs, ITHACA is developing a tool devoted to the automated detection of flooded areas with a worldwide coverage, basically trough the identification of water bodies on the ground and the comparison to a reference water extent based on the analysis of a 10 year remotely sensed data archive. The main input data is satellite imagery, specifically the MODIS/Terra Surface Reflectance Daily products (MOD09GQK and MOD09GQ), characterized by a spatial resolution of 250 meters. These products allow to have a daily estimate of the surface spectral reflectance investigating two bands of the e.m. spectrum, the red one (620-670 nm) and the IR one (841-876 nm). The main advantage of this product is that it contains the surface spectral reflectance as it would be measured at ground level in the absence of atmospheric scattering or absorption. The main disadvantage is that, being MODIS a passive sensor, the cloud coverage can be a relevant obstacle to the analysis. Therefore a cloud, cloud-shadow, and hill-shade masking procedure has to be used. Furthermore a multi-temporal approach has to be adopted in order to reduce the presence of no-data cells due to the cloud coverage. The temporal compositing algorithm is based on daily images representing water bodies on the ground, generated by the use of band ratios and threshold techniques. For each cell of the image, we calculated the number of days during which the relative area on the ground is classified as flooded. Defining a number of days threshold is therefore possible to state if the area is covered by water, as well as to provide a reliability parameter based on the cloud cover persistence. The comparison of the classified water bodies with average values related to the same period will finally allow to state with high probability if the area is flooded or normally covered by water. Preliminary results related to specific case studies will be shown. 1. INTRODUCTION ITHACA is a non-profit association, envisioned as a centre of applied research developing IT products and services in support of emergency management, focused especially on early-warning and damage assessment phases. Specifically for early-warning, ITHACA has developed and implemented an alert system based on precipitation analysis and related to historical flooded areas detection. The purpose of the tool, operational with a nearly global coverage and in near real-time, is to monitor actual precipitation rates, derived from a satellite platform, and to compare those with rainfall thresholds corresponding with historical flood events occurred on the field, detecting in that way extreme events. In the field of early-impact and damage assessment, main objective of ITHACA activities is to provide georeferenced information (mainly in form of map products) about affected areas and population. Main final user of those elaboration is United Nations World Food Programme (WFP). Since the beginning of their activity (January 2007), ITHACA has been activated in occasion of more than 50 events, most of them related with flood caused by intense precipitations, cyclones, hurricanes, etc. More than 200 rapid mapping products were delivered. Timely availability of geographic information about water bodies extension and, possibly, depth at an appropriate scale is fundamental for all response phases of the risk cycle. Both historical, to correctly identify the reference conditions or to produce scenarios based on historical events, and near real-time data, for the actual and precise identification of water covered areas, are thus essential. For that reason, ITHACA decided to activate a research activity finalized to the automation of the classification of water bodies from remote sensing images, with a dual objective: • to process an huge volume of historical data, in order to obtain dynamic reference data, related to different climatic periods/seasons; • to develop a near real-time monitoring system for flood events. 2. CLASSIFICATION OF WATER BODIES 2.1 Literature review Actually, water bodies identification on a scene acquired by optical sensors installed on satellite platforms is based on simple but effective histogram threshold techniques; those techniques exploit the behaviour of water in the infrared bands, where those surfaces have high absorption rates. Similar approach is valid also for processing radar images, due to the fact that, in case of calm water hit by an incident microwave beam, the specular response dominates the returned signal. As highlighted in previous articles, those methods are simple to apply but several disadvantages are also evident. Shadows, due to the local morphology or to the presence of clouds, are classified as water bodies and threshold values cannot be defined uniquely but adapted to the conditions at the moment of the acquisition. For those reason, an extensive review of water bodies classification techniques presented in literature was conducted, in order to identify those capable to solve or minimize the above mentioned problem. The analyzed techniques are mainly based on indexes derived from differential band ratios, to make threshold values independent from image acquisition parameters. In particular, according to literature the use of the Normalized Differential Water Index (NDWI) is commonly used to identify and classify flooded areas. Unfortunately, a unique definition of this index was not found, probably due to its adaptation to the different characteristics of spectral sensors