The discharge prediction at a river site is fundamental for water resources management and flood risk prevention. Over the last decade, the possibility to obtain river discharge estimates from satellite sensors data has become of considerable interest. The Moderate Resolution Imaging Spectroradiometer (MODIS) can provide a proper tradeoff between temporal and spatial resolution useful for discharge estimation. It assures, in fact, at least a daily temporal resolution and a spatial resolution up to 250 m in the MODIS channel 1 and 2. In this study, the capability of MODIS data for discharge prediction is investigated. Specifically, the different spectral behavior of water and land in the Near Infrared (NIR) portion of the electromagnetic spectrum (MODIS channel 2) is exploited by computing the ratio of the MODIS channel 2 reflectance values between two pixels located within and outside the river. Values of such a ratio should increase with the presence of water and thus can be used for discharge estimation. The Upper Tiber River basin and the Po River basin are located in the Central and the North Italy, respectively. In particular, six gauged river sites are used to test the application of the approach: Ponte Nuovo and Monte Molino for the Tiber River, Piacenza, Cremona, Borgoforte and Pontelagoscuro for the Po River. Discharge and flow velocity data for all the gauged stations are used. MODIS is one of the sensors aboard the Earth Observing Satellites Terra and Aqua. For this study, MODIS channel 1 ( mm) and channel 2 ( mm) data were extracted from MODIS level 1b (MOD02QKM and MYD02 QKM) data sets for Aqua satellite. In particular, all the images acquired in the period over Italy were first processed obtaining surface reflectance values (R 1 and R 2 ) and then the sub- scene, centred on the two study areas, were extracted. A. Tarpanelli 1, L. Brocca 1, T. Lacava 2, M. Faruolo 2, F. Melone 1, T. Moramarco 1, N. Pergola 2,3 and V. Tramutoli 2,3 1 CNR-IRPI, Hydrological Section of the Research Institute for Geo-Hydrological Protection, Perugia, Italy; CNR-IMAA, Institute of Methodologies for Environmental Analysis, Tito Scalo (PZ), Italy; 3 Department of Engineering and Physics of the Environment, University of Basilicata, Potenza, Italy; Brakenridge et al. (2007) used the Advanced Microwave Scanning Radiometer (AMSR-E) for discharge estimation based on the ratio between the brightness temperature measured for a pixel unaffected by the river and a pixel centered over the river, respectively. This procedure has been implemented world-wide within the “Global Flood Detection System” ( verview.aspx) for more than monitoring areas (see Figure 3). verview.aspx Station 662 (Piacenza) Station 2671 (Pontelagoscuro) r=0.35 r=0.58 AQUA- MODIS Reflectance value of Channel 2 (10-Feb :10) Following the same approach, in this study, MODIS channel 2 data, acquired in the Near Infrared (NIR) portion of the electromagnetic spectrum with a spatial resolution of 250 m, are used. Specifically, the different spectral behavior of water and land at these wavelengths is exploited. In fact, compared to reflectance values of common surface land covers (like bare or vegetated soils), water shows a lower reflectance in the NIR region. So that, when water is present in a pixel or more water is added, its reflectance tends to decrease. Therefore, considering the surface reflectance of a water pixel, named M, located within the river with permanent presence of water, and the one of a land pixel, named C, located near the river in an area free of surface water even during high flood, then the ratio between C and M is a sensitive and consistent measure of surface water. Due to the high variability of the surface reflectance values, the ratio C/M changes rapidly and appears quite noisy. To reduce these, also related to atmospheric contribution, an exponentially smoothing filter (Wagner et al., 1999) is applied to the data stream. Such a filter uses a single tuning parameter (the characteristic time length, T) and continually calculates a smoothed value given by a weighted average of recent observations giving less weight to older data. Specifically, for each gauged river site, the following processing steps have been carried out: from each sub-scene of the MODIS images (acquired by the sensors aboard Aqua satellite), a box of 28x32 pixels centered on the investigated gauging station is selected; pixels affected by cloud cover and/or snow are identified by using a simple threshold on R 1 images (R 1 >0.15) and a visual detection and excluded from further data analysis; for each pixel of the box 28x32 the ratio C/M of R 2 data is calculated thus obtaining 895 time series of C/M ratios; the discharge, Q, and mean flow velocity, V, measured at each gauging station and corresponding to the acquisition dates of the satellite sensor overpasses are selected; the smoothing exponential filter is applied to the C/M time series, thus obtaining C/M* time series, by calibrating the T-value against V observations; The agreement between C/M* and Q, V time series is finally assessed in terms of correlation coefficient, r, root mean square error, RMSE and Nash-Suitcliffe, NS. The first comparison is carried out between the C/M* time series and the flow velocity, V, considering as potential calibration pixel all the 895 pixels surrounding the gauging station (see Figure 1 and Figure 2). In particular, the pixel M was varied within the river while C in the remaining pixels outside the river. The best calibration pixel is selected by maximizing the linear correlation coefficient, r, between C/M* and V time series. For that, the optimal T-value of the exponential filter is found to be equal to 90 for the stations of Tiber River and equal to 10 for Po River. For each gauged station, the optimal position of the pixels M and C corresponding to the maximum value of the correlation coefficient is used for all the comparisons carried out in this study. As it can be seen in the Table, the unfiltered C/M ratio is found to be poorly correlated with both Q and V (and their logarithms), with r-values in the range , and it is clearly due to the noisy of the signal (i.e., atmospheric contribution, residual clouds or snow). On the other hand, if the exponential smoothing filter is applied, the r-values significantly increase (up to 0.8), confirming the potential of such an approach to reduce the impact of random signals on long data series. Generally the maximum values of correlation are found for velocity data. For that, considering the velocity data from all the stations for each basin, it is possible to identify a single empirical relationship between satellite data, in terms of C/M* e in situ measurements of velocity that allow us to estimate the discharge also for ungauged river sites based on MODIS data. An analysis for the evaluation of the capability of MODIS data to estimate river discharge is carried out. The study has shown that MODIS data can give good estimates of discharge time series for medium sized basins characterized by high temporal variability of discharge. Furthermore, the empirical relationships between satellite and in-situ time-series for the two basins were investigated in- depth to regionalize their parameters and, hence, to estimate the discharge also for ungauged river sites based on MODIS data. The capability of MODIS to estimate discharge can be efficiently employed for rating curve development at ungauged river sites by using simplified routing approaches, that will be object of future investigations. RIVER GAUGED STATIONS BASIN AREA (km 2 ) PERIOD (years) CORRELATION COEFFICIENTS VQ Regional relationsIn situ relationsRegional relationsIn situ relations C/MC/M* RMSENSRMSENSRMSENSRMSENS QVlog(Q)log(V)QVlog(Q)log(V) TIBERPONTENUOVO TIBERMONTEMOLINO POPIACENZA POCREMONA POBORGOFORTE POPONTELAGOSCURO PONTE NUOVO MONTE MOLINO PONTELAGOSCURO BORGOFORTE CREMONA PIACENZA REGIONAL ANALYSIS r=0.68 r=0.76 r=0.67 r=0.74 Brakenridge, G. R. Nghiem, S. V., Anderson, E. & Mic, R. Orbital microwave measurement of river discharge and ice status, Water Resources Research, 2007, 43, W Wagner, W., Lemoine, G. & Rott, H. A method for estimating soil moisture from ERS scatterometer and soil data, Remote Sensing of Environment, 1999, 70, pp The authors wish to thank R. Rosi for his technical assistance, Umbria Region for providing most of the analyzed data for Tiber River basin, ARPA (Agenzia Regionale Prevenzione e Ambiente ) Emilia Romagna and in particular, Eng. Federica Pellegrini, for providing the analyzed data for Po River basin. In the Table the performance in terms of RMSE and NS are reported by using both the regional and the at site relationships for velocity and discharge estimation (the discharge is estimated by the product of the velocity and the flow area). As expected, the performance of the regional relationship slightly decreases even though NS values are still quite high (NS= ) when discharge is considered. In principle, this relationship can be used for estimating discharge at ungauged sites as well. Figure3 Figure3 shows the comparison between the C/M data obtained by AMSR-E (available at Global Flood Detection System) and in situ observations of discharge for Piacenza and Pontelagoscuro sections. Correlation values (r=0.35 and r= 0.58) are quite lower than the ones obtained by MODIS (r=0.63 and r=0.73), confirming the potential of this data for improving discharge estimation over medium-sized ( km 2 ) basins. Figure 1 Figure 2 Figure 3 M C M C M C M C r=0.79 r=0.71 r=0.64