Discharge estimation and forecasting by MODIS and altimetry data in Niger-Benue River

Remote Sensing of Environment 195 (2017) 96–106

Contents lists available at ScienceDirect

Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse

Discharge estimation and forecasting by MODIS and altimetry data in Niger-Benue River Angelica Tarpanelli a,⁎, Giriraj Amarnath b, Luca Brocca a, Christian Massari a, Tommaso Moramarco a a b

Research Institute for Geo-hydrological Protection, National Research Council, Via Madonna Alta 126, 06128 Perugia, Italy International Water Management Institute (IWMI), 127 Sunil Mawatha, Pelawatte, Battaramulla, Colombo 10120, Sri Lanka

a r t i c l e

i n f o

Article history: Received 16 June 2016 Received in revised form 23 March 2017 Accepted 12 April 2017 Available online xxxx Keywords: Flood forecasting Modis Altimetry Remote sensing Discharge Niger-Benue river

a b s t r a c t Flooding is one of the most devastating natural hazards in the world and its forecast is essential in flood risk reduction and disaster response decision. The lack of adequate monitoring networks, especially in developing countries prevents near real-time flood prediction that could help to reduce the loss of lives and economic damages. In the last few years, increasing availability of multi-satellite sensors induced to develop new techniques for retrieving river discharge and especially in supporting discharge nowcasting and forecasting activities. Recently, the potential of radar altimetry to estimate water levels and discharge in ungauged river sites with good accuracy has been demonstrated. However, the considerable benefit derived from this technique is attenuated by the low revisit time of the satellite (10 or 35 days, depending on the satellite mission) causing delays on the predicting operations. For this reason, sensors with a higher temporal resolution such as the MODerate resolution Imaging Spectroradiometer (MODIS), working in visible/Infra-Red bands, can support flood forecasting. In this study, we performed the forecast of river discharge by using MODIS and we compared it with the radar altimetry and in-situ data along the Niger-Benue River in Nigeria to develop an operational flood forecasting scheme that could help in rapid emergency response and decision making processes. In the first step, four MODIS products (daily and, 8-day from the TERRA and AQUA satellites) at two gauged sites were used for discharge estimation. Secondly, the capability of remote sensing sensors to forecast discharge a few days (~4 days) in advance at a downstream section using MODIS is analyzed and also compared with the one obtained by the use of radar altimetry by ENVISAT and Jason-2. The results confirmed the capability of the MODIS data to estimate river discharge with performance indices N 0.97 and 0.95 in terms of coefficient of correlation and Nash Sutcliffe efficiency. In particular, RMSE does not exceed 1300 m3/s and the fractional RMSE ranges between 0.15 and 0.23. For the forecasting exercise, both altimetry and MODIS provide satisfactory results with positive coefficient of persistence considering 4 days of lead time (N0.34). Although altimetry was found to be more accurate in the forecasting of river discharge (RMSE ~350 m3/s), the much higher temporal resolution of MODIS guarantees a continuity that is more suitable to address operational activities. © 2017 Elsevier Inc. All rights reserved.

1. Introduction Flooding is one of the most recurring, widespread, and disastrous natural hazard of the world and its destructive impact may be enormous (Emergency Events Database – EM-DAT, 2016). The effect of particularly intense rainfall events is aggravated by the socio-economic condition of some countries that often are insufficient to cope with this natural disaster (Ayala, 2002). Moreover, the lack of vulnerability assessment and of preparedness of emergency response does not facilitate actions ⁎ Corresponding author. E-mail addresses: [email protected] (A. Tarpanelli), [email protected] (G. Amarnath), [email protected] (L. Brocca), [email protected] (T. Moramarco).

http://dx.doi.org/10.1016/j.rse.2017.04.015 0034-4257/© 2017 Elsevier Inc. All rights reserved.

for evacuation of people, and the results are inevitably devastating (Amangabara and Obenade, 2015). A recent flood disaster in 2012 in Nigeria claimed more than 363 lives and adversely affected more than 7 million people with an estimated damages loss of 500 million USD. This example highlights the importance to monitor floods particularly in data scarce regions, where an alternative source of monitoring as the remote sensing technology is required (Aich et al., 2014; Tarhule, 2005; Emergency Events Database – EM-DAT, 2016). Further, with the growing population along the river banks and flood plains, anthropogenic activities such as dumping of wastes in water channel and the built up areas, lead to increased runoff (Ali, 2005). In this context, the construction and improvement of drainage networks, the collaborative efforts of government and stakeholders to support the forward planning, engineering and other professional

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106

agencies become extremely important in reducing the risk of flooding in Nigeria or elsewhere (Hula and Udoh, 2015). Moreover, accurate stage and/or discharge predictions with an appropriate forecast lead-time supported by a hydro-meteorological network operating on-line can mitigate the negative effects of a flood. The inadequate monitoring networks operating in Nigeria and the lack of timely access to information along the transboundary rivers hinders the estimation of the discharge and make near real-time flood forecasting difficult, leaving populations exposed to flooding. In recent years, satellite data, especially altimetry, have demonstrated their potential for discharge estimation through the use of approaches based on empirical formulas (Negrel et al., 2011), rating curves (Kouraev et al., 2004; Frappart et al., 2015; Tourian et al., 2013), rating curves plus flow routing method (Leon et al., 2006; Birkinshaw et al., 2014; Getirana, 2010; Tarpanelli et al., 2013a), hydraulic models (Domeneghetti et al., 2014, 2015; Yan et al., 2014; Neal et al., 2012), and assimilation techniques (Michailovsky et al., 2013). Some first attempts were done to support river discharge nowcasting and forecasting (Biancamaria et al., 2011; Hirpa et al., 2013; Pauwels et al., 2001; Hossain et al., 2014), but only one example is available in Nigeria (Pandey and Amarnath, 2015). The interesting preliminary study by Pandey and Amarnath (2015) used a “forecast rating curve” to relate the upstream water level derived by altimetry (ENVISAT, Jason-2 and Saral/Altika) to downstream observed river discharge in order to predict the discharge at the downstream section of Makurdi along the Benue River in Nigeria 5 days in advance. The approach is simple and practical to be used for expeditiously estimating the river flows. The main drawback is the low temporal resolution of the radar altimeter observation (10 or 35 days depending on the satellite mission), that delays the predicting operations. To overcome the gap due to the poor temporal sampling, data from NASA's Moderate Resolution Imaging Spectroradiometer (MODIS), freely available on a daily basis makes it possible to estimate discharge over large areas for flood forecasting applications. MODIS was already employed for estimating the variation of discharge both for gauged and ungauged river sites by Tarpanelli et al. (2013b, 2015) and for assessing the downstream discharge starting from the knowledge of the upstream river widths (Smith and Pavelsky, 2008; Gleason and Smith, 2014). Unlike MODIS, the radar altimetry provides a direct estimation of the water level through radar measurements, discharge estimation from MODIS is based on the difference between the reflectance measured for wet and dry pixels in the Near Infra-Red (NIR) band. The reflectance of water surface in the NIR band is lower than the reflectance on dry areas and their ratio, increasing with water extent, is found to give good estimation of discharge. The key concept was introduced by Brakenridge et al. (2007) that used microwave passive sensor AMSR-E for mapping flood events at global scale based on the brightness temperature at 36 GHz (http://floodobservatory.colorado.edu/). In the present study, the potential to forecast river discharge in Niger-Benue by using remote sensing data is tested and, based on the temporal resolution of the data, it is expected that information coming from MODIS could provide a more comprehensive understanding of the evolution of the flood event such as to provide a more reliable prediction with respect to radar altimetry data which does not guarantee a frequent sampling. On this basis, the purpose of this paper is twofold: 1) to test the procedure of Tarpanelli et al. (2013b) to estimate river discharge by using MODIS images in large rivers such as the NigerBenue which has a hydrological regime and climatic conditions different from the ones used to develop the approach; 2) to forecast the discharge in a downstream section based on the information obtained from MODIS images and radar altimetry in an upstream section with a lead time of four days. This study represents a proof-of-concept about the possibility to estimate and forecast the discharge in poorly gauged river basins by using globally and freely available data.

97

The manuscript is structured as follows. In Section 2, an overview of the examined study area is given. In Section 3, the satellite and in-situ data are described, whereas procedure is detailed in Section 4. In Section 5 the results are presented and discussed. In the last Section, perspectives and conclusions are drawn. 2. Study area Nigeria, a sub-Saharan West African country, is the most populous (170 million) in Africa continent located between latitude 4°9′N to 13°46′N and longitude 3°45′E to 16°54′E. The total geographical area is 923,770 km2 bordered in south by Atlantic Ocean while sharing land borders with Benin Republic, Chad, Niger and Cameroon. Niger River is the third longest transboundary river in Africa flowing 4100 km and draining 2.2 million km2. Almost 63% of Nigeria's total geographical area is drained by Niger River system. It is divided into four sections namely, upper Niger (in the north-west); inner delta; middle Niger and the lower Niger (in the south) river system. The transboundary Niger River system and its main tributary Benue are partially regulated by networks of dams such as Markala dam, Bamako dam and Sélingué Dam in Mali, Kainji dam and Jebba dam in Nigeria; and Lagdo dam in Benue River located near Garoua Cameroon. The percentage of total population below poverty line increased by 15% from 2004 to 2010. Of the 36 administrative divisions in Nigeria, ten states are estimated to possess over 70% poverty rate signifying extremely low indicators of human well-being (UNDP, United Nations Development Programme, 2015). Nigeria's Human Development Index rank for 2014 is 152 out of 188 countries positioning it in low human development category with value of 0.51 (UNDP, United Nations Development Programme, 2015). Owing to the recent 2012 catastrophic floods in Nigeria, forecast requirement of Nigeria Hydrological Services Agency, NIHSA, is to estimate water level in the upstream rivers from Cameroon to minimize the risk of flooding. During the interactive workshop with officials, a flood forecasts lead time of 5–6 days and translating these forecasts into valuable early warning information is considered critical to reduce the flood impacts on the major cities/towns (i.e. Yelwa, Jebba, Makurdi, Lokoja, Onitsha, Lau, Numan and Ibi) along the Niger River and agricultural losses. 3. In-situ and satellite dataset 3.1. In-situ dataset The analysis is focused on the confluence between Niger and Benue Rivers where the gauged stations of Lokoja (Niger) and Makurdi (Benue) are selected (Fig. 1). Table 1 summarizes the main hydraulic characteristics of the two sites in terms of maximum, minimum and mean discharge values. The period of available data is also indicated. The period from July to September is critical for floods and heavy rainfall, and in 2012 Nigeria has suffered a terrible flood, that pushed flows to overtop the banks, and submerged hundreds of thousands of acres of farmland, with 363 casualties, 5851 injured and 3,871,530 displaced from their homes (GFDRR, 2014). For both the gauged stations, the data of daily water level, h, and discharge Q are available from 1970 to 2012 covering different locations along Niger–Benue River. Data were obtained from the NIHSA in cooperation with the Federal Ministry of Agriculture and the Rural Development (FMARD), Nigeria. However, the majority of recorded time series are not continuous and for the current study only Makurdi and Lokoja gauged sites are used to test our approach. 3.2. MODIS datasets MODIS is a multispectral sensor on-board the TERRA and AQUA satellites acquiring image data of Earth's surface simultaneously at

98

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106

Fig. 1. Map of study area (a and b), with the location of the Niger basin and the false color MODIS image in NIR band of Niger-Benue Rivers with the location of the two gauged stations Lokoja and Makurdi and the ENVISAT and Jason-2 altimetry tracks (c). The Virtual Station, VS where the satellite track 530 of ENVISAT altimeter overpasses the river is shown.

multiple wavelengths. It provides high radiometric sensitivity in 36 discrete spectral bands ranging in wavelength from 0.4 μm (visible) to 14.4 μm (thermal infrared) and acquires at three different spatial resolutions: 250, 500 and 1000 m. Differently from the previous analyses (Tarpanelli et al., 2013b, 2015) in which the level-1 product was analyzed, in the present study level-2 and level-3 products are tested. MOD09GQ and MYD09GQ (TERRA and AQUA at daily resolution, respectively) and MOD09Q1 and MYD09Q1 (TERRA and AQUA at 8-day resolution) are made available by the USGS (http://e4ftl01.cr.usgs.gov/ ). We selected the Earth surface reflectance of band 2 (Near-Infrared, 841–876 nm) at 250 m. In the case of daily products, band 1 (Red 620–670 nm) is also selected for removing the clouds as proposed by Tarpanelli et al. (2013b). The 8-day product is less affected by clouds because level-3 MOD09Q1 (MYD09Q1) images contain the best possible observation during 8-day period of MOD09GQ (MYD09GQ) images (Vermote and Kotchenova, 2008). The purpose for the use of different MODIS products is twofold. First, the analysis carried out with products characterized by a different temporal resolution can provide some insights about whether or not lower temporal resolution products perform equally well to higher temporal resolution products thus allowing to significantly decrease the amount of data needed to obtain comparable results. To this end, it is worthy to note that the 8-day product could be feasible for large basins where the variability of the discharge is compatible with weekly temporal resolution, whereas it might be not appropriate for smaller basins characterized by a significant dynamic at shorter time scales (e.g. daily or sub-daily). Second, with level-2 and level-3 the steps of pre-processing

Table 1 Characteristics of the gauged stations: period of the analysis, maximum (Qmax), minimum (Qmin) and mean (Qm) discharge. River

Gauged station

Niger Lokoja Benue Makurdi

Period (years)

Qmax (m3 ∙s−1)

01/01/2001–12/31/2012 31,692.00 09/02/2006–22/06/2012 16,387.50

Qmin (m3 ∙s−1)

Qm (m3 ∙s−1)

820.64 228.88

6,372.10 3,373.98

of the images needed in the case of level-1 are not required offering significant advantages for the application of the procedure at operational level by non-experts in the field of satellite data processing. The analysis was focused on eight years of satellite data from 2006 to 2013 for a total of 2916 and 368 images for daily and 8-day products, respectively. The total number in the case of daily images decreased after the cloud filtering, with a final number of images equal to 1203 (TERRA) and 1446 (AQUA) at the Lokoja station and to 1006 (TERRA) and 1266 (AQUA) at the Makurdi station. The forecasting exercise has an important parameter to take into account, that is the latency of the data that could reduce the warning time. The latency of the L2 MODIS product is about 1–3 h (O′Neal, 2005), perfectly compatible with the forecasting in Nigeria. For L3 product, the latency is higher because they have to take into account the mean of 8 images, and for this reason is expected to be not optimal for a daily forecasting. 3.3. Altimetry datasets Satellite altimetry uses the two-way travelling time of a pulse emitted by an on-board antenna and its subsequent reflection from the earth's surface to calculate the distance between the satellite and the Earth's surface or water body, called range (Fu and Cazenave, 2001). The satellite altitude with reference to an ellipsoid is also accurately known from orbitographic modelling (Baup et al., 2014). By applying appropriate correction factors for the propagation delays of electromagnetic waves upon its the interactions with the atmosphere and geophysical entities, the height of the reflecting water surface with reference to an ellipsoid or a geoid can be estimated (Pandey et al., 2014; Baup et al., 2014). h ¼ H−R−C ionosphere −C drytroposphere −C wettroposphere −C solidEarthtide −C pole tide ð1Þ where h is the height of reflecting water surface (water body or river section), H is the altitude of satellite with reference to an ellipsoid, R is

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106

the altimeter range, Cionosphere is the correction for delayed propagation through the ionosphere, Cdry troposphere and Cwet troposphere are corrections for delayed propagation in the troposphere from pressure and humidity variations, respectively, and Csolid Earth tide and Cpolar tide are corrections that account for crustal vertical motions from the solid and polar tides, respectively. Water level time series over river sections intersected by the satellite track crosses can be extracted from repeated along-track nadir measurements (Fu and Cazenave, 2001). In this study, nadir radar altimeter data from RA-2 or Advanced Radar Altimeter onboard the ENVIronmental SATellite, ENVISAT (Wehr and Attema, 2001), and Poseidon-3 onboard Jason-2 (Desjonquères et al., 2010) were used to derive water levels for multiple locations along Niger River. ENVISAT changed its orbit in October 2010 and concluded its mission in April 2012, whereas Jason-2 started in 2008 and it is ongoing on a new orbit after the change in October 2016. At all points where a satellite track intersects a water body, or “virtual gauge stations”, VS, water level time series was extracted. For flood forecasting, a reach of about 170 km along the River Benue starting from the virtual station VS 530, where altimetry track 530 (ENVISAT) is available, to the Makurdi gauged station was considered, as shown in Fig. 1. To have a better temporal resolution more altimetry tracks upstream the VS 530 are combined. Specifically, tracks 315, 773 and 988 from ENVISAT and track 96 from Jason-2 were used. Data were downloaded from CTOH (http://ctoh.legos.obs-mip.fr/) and the range is obtained by using of Ice-1 re-tracking algorithms, which generally provide the best results compared to other existing re-tracker algorithms (Ocean, Ice-2 and Ocean-ice) for inland water bodies (Frappart et al., 2006; Santos da Silva et al., 2010). The processing of the ENVISAT and Jason-2 data for deriving the time series of water levels at the virtual gauge location was carried out through the Virtual ALtimetry Station (VALS, 2010) software (Santos da Silva et al., 2010). The three data processing steps in VALS consist of the selection of data points, data filtering through the removal of outliers which might represent reflection from mixture landform patterns based on user discretion and the calculation of water level using statistical measures (Frappart et al., 2006; Amarnath et al., 2016). Water stage time series between 2002 and 2013 (complete years) were derived at the designated virtual stations using VALS. In this study, the water level data were referenced to the EGM2008 geoid model (Pavlis et al., 2012). 4. Methodology Considering the two objectives of the analysis, we first describe the procedure used for discharge estimation using MODIS in gauged river sites of Lokoja and Makurdi as in Tarpanelli et al. (2013b). Then, we provide the description of the methods for implementing the flood forecasting operations by using MODIS, and altimetry for comparison. Note that the first step is mandatory for successful implementation of the second step. Both the steps are carried out considering the entire period available because the sample size is too small and discontinue for splitting the dataset in two sub-datasets, one for the calibration and one for the validation. 4.1. River discharge estimation by using MODIS images The approach to estimate discharge is based on the procedure developed for the AMSR-E product by Brakenridge et al. (2007) and De Groeve (2010) then extended to MODIS data by Tarpanelli et al. (2013b). The main concept is to take advantage of the different pixel values of surface reflectance between water and land from the NIR band. Indeed, a water pixel, M (for “measurements”), has lower reflectance value than a land pixel, C for (“calibration”). During a flood event, the water pixel becomes wetter triggering a further decrease in its reflectance value. For the land pixel (not affected by water), the reflectance is characterized by a lower variability and is assumed constant.

99

Thus, measuring the reflectance variation in time of pixel M with respect to pixel C provides an indirect measure of the variation of river discharge (Tarpanelli et al., 2013b). In particular, since both C and M are affected in the same way by disturbances determined from the atmosphere, the ratio between the reflectance values of the C and M pixels should theoretically enable a minimization of the atmospheric effects. For a detailed description of the procedure, the reader is referred to Tarpanelli et al. (2013b). A brief description of the main steps involved in the procedure is given below. The pixel M and C are selected within a box centered on the single gauged sites. Hence, from every MODIS images, the box is extracted obtaining a long time series. For the daily products (level-2), the pixels affected by cloud cover are first identified by using a fixed threshold on reflectance values of the band 1, R1 (R1 N 0.2) and removed. Then a visual inspection on band 2 excludes from the analysis the images in which the river is not clearly visible. Differently from Tarpanelli et al. (2013b), in which the optimization procedure to estimate the best location for M and C was carried out by analyzing the correlation among the observed discharge and all the possible combinations of the pixels C and M pixels within the selected box, here the location of the C pixel is selected a priori by considering the pixels characterized by lower temporal variability (i.e. low coefficient of variation). For every pixel in the box, the ones with a coefficient of variation lower than the 5th percentile (with respect to the values computed over all the pixels in the box) are chosen and averaged to obtain the C time series (De Groeve and Riva, 2009). The optimization is then only carried out for the location of the M pixel for all the products. In this way, the procedure suffers less from non-identifiability issues with respect to results obtained in Tarpanelli et al. (2013b). It is worth to note that the obtained C/M is a time series because a time series of MODIS images is considered. Due to the high variability of the surface reflectance values, the C/M time series appear quite noisy. To overcome the issue, an exponential smoothing filter is applied to the time series before its use (Albergel et al., 2008; Wagner et al., 1999) obtaining the transformed ratios C/M*. The parameter which drives the amount of smoothing is chosen equal to 16 days which is the revisit time of the TERRA and AQUA satellites. All the C/M* ratios are compared in terms of Pearson coefficient of correlation with the observed discharge time series. Since the discharges of Niger-Benue river are affected by strong seasonality, the comparison between C/M* and discharge is carried out considering the time series anomalies without the effects of the seasonal cycle. The seasonal cycle is computed by averaging the discharge values picked up in the same day for all the years of the analysis. For instance, the discharge value of the seasonal cycle relative to the 1st January is calculated by taking the mean of the discharge values of 1st January of all the analyzed years. The anomalies time series for C/M* and discharge are obtained by subtracting the respective seasonal cycle, repeated for all the analyzed years, from the time series. The C/M* anomalies time series providing the maximum correlation with the observed discharge anomalies is used to identify the position of the M pixel. Once M is established for all the selected products, a linear model (for each product) is fitted to describe the functional relationship between the anomaly of C/M* time series and the anomaly of observed river discharge. The algorithm allows to estimate the discharge anomaly once C/M* is known. The total discharge can be simply obtained by adding the observed seasonal cycle to the estimated discharge anomalies. Based on the rationale behind the approach, the discharge variation is represented by the amount of water variation inside a cell (M). Since the Niger and the Benue rivers are large about 800–900 m, the number of cells representing the river is about 3–4 (remember that the size of MODIS cells is ~250 m). In order to not select a pixel completely saturated (fully occupied by water), for which the water change is difficult to measure, the analysis is carried out also by aggregating more pixels. Specifically, all the images are resampled through the calculation of the averaged values among 2 by 2 (4 pixels) and 3 by 3 (9 pixels) pixels

100

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106

obtaining new images. This step enables to shorten the computation time for the optimization procedure of M, because of the lower number of pixels composing the image.

The procedure for discharge forecasting through radar altimetry needs the knowledge of the rating curve at the downstream section that relates the discharge, Qd, at the water stage, hd, by:

4.2. Discharge forecasting at a downstream section

Q d ðt Þ ¼ a  hd ðt Þb

Discharge forecasting along a river can be performed by exploiting the wave travel time between an upstream and a downstream cross section. In the absence of any ground measurements for the upstream cross section, satellite measurements can be considered. In this study, we focus on MODIS data and for comparison purposes we also used altimetry observations. Therefore, the upstream section will correspond to a Virtual Station, VS, where the radar altimetry satellite track overpasses the river; MODIS data are instead available for any cross section. Among all the available VSs along the selected river, the upstream section at a distance that allows an adequate lead time in flood early warning should be selected. An important assumption has to be made for this approach that the discharge contribution of the intermediate basin is proportional to the contribution at the upstream section. Therefore, the observation acquired some days before at an upstream section is informative of the downstream discharge. The procedure for discharge forecasting by MODIS and radar altimetry is explained in the following. The description is introduced by the evaluation of the wave travel time used in both the procedures.

where a and b are coefficients known at the downstream site. In order to use this rating curve a relationship between the water level at the upstream, u, and downstream section, d, is derived by taking into account the wave travel time according to:

4.2.1. Evaluation of the wave travel time For the estimation of the wave travel time or time lag, TL, water levels at the upstream section derived by altimetry data are compared with those observed at downstream section – shifted by a different number of days – in terms of correlation coefficient, starting from the date of the satellite pass and gradually increasing one by one up to 10 days. The time step that provides the maximum coefficient of correlation is used as the wave travel time. The same analysis was carried out also using the discharge time series extracted by MODIS obtaining the same TL. 4.2.2. Discharge forecasting procedure by MODIS images Assuming the upstream section as the VS location and once the wave travel time between the upstream section and Makurdi is estimated, the forecasting by MODIS is made by adopting nearly the same procedure implemented for gauged station, but for VS. The only remarkable difference in this case is that the benchmark time series of discharge used for the selection of the M pixel and for the linear fitting is the discharge time series measured at the station located downstream. In particular, after the selection of C and M pixel and filtering, the linear model is fitted between the anomalies of C/M* at time t and the observed discharge anomalies downstream at t + TL and it allows to obtain the forecast of the discharge anomalies in the future with a warning time of TL. For obtaining the total discharge, the mean seasonal cycle is simply added (as before). 4.2.3. Discharge forecasting procedure by radar altimetry data In order to obtain a densified time series at VS 530, the bias between the different missions has to be removed and the travel time between the VSs and the reference VS 530 has to be considered. The biases are computed based on the difference in the mean and the standard deviation between water level time series from ENVISAT and Jason-2 missions (see Fig. 1) and the water level observed at Makurdi (Tourian et al., 2016). Considering the same flow velocity calculated between VS 530 and Makurdi (the assumption is plausible because the river slope, the roughness and the width are quite constant), the time lag is computed multiplying the flow velocity times the distance between each VS and the reference VS 530. The resulting altimetry time series is a fictitious water level time series located at the VS 530 with an improved temporal resolution and coverage with respect to the original altimetry tracks derived only by ENVISAT, VS530.

ð2Þ

hd ðt þ TLÞ ¼ F ðhu ðt ÞÞ

ð3Þ

where the water stage, hu, is given by the differences between the water level retrieved by altimetry Hu and the bottom of the cross section, h0. Therefore, forecasted discharge becomes: b

Q d ðt þ TLÞ ¼ a  F ðH u ðt Þ−h0 Þ

ð4Þ

F is a function operator and in this case, it is selected as a linear regression between the water stage upstream and those downstream. 4.3. Performance indices The estimated discharge obtained by the use of remote sensing data are validated through the use of the following performance indices: Pearson coefficient of correlation, r, the Nash-Sutcliffe efficiency, NS (Nash and Sutcliffe, 1970), and the root mean square error, RMSE, and the fractional RMSE, fRMSE, defined as fRMSE ¼

RMSE σ obs

ð5Þ

where σobs is the standard deviation of the observed time series. It varies in the range from 0 (perfect estimates) to infinity with values N1 indicating low performances (noise) (Draper et al., 2013). The reliability of the forecasted discharge is evaluated by adopting the coefficient of persistence, PC, (Kitanidis and Bras, 1980), which compares the prediction of the forecast model Qti , sim with the observation at t-TL, Qt−TL i ,obs , by assuming that the forecast coincides with the most recent observed value, through the following equation:
2 n ∑i¼1 Q ti;obs −Q ti;sim PC ¼ 1−
2 n ∑i¼1 Q ti;obs −Q t−TL i;obs

ð6Þ

PC values lower than 0 denote bad performances while values greater than zero indicate good to perfect (PC = 1) forecast. Another performance index is the percentage of the RMSE, pRMSE, expressed as follows:

pRMSE ¼

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u  t 2 u n Q i;sim −Q t−TL i;obs u∑ t−TL t i¼1 Q i;obs n

ð7Þ

where Qti ,sim is the discharge forecasted at time t of the i-th element of the time series, whereas Qt−TL i ,obs is the observed discharge at the present time, t-TL. Additionally, three categorical metrics are evaluated for a specific threshold representative of flood event for Makurdi station: Probability of Detection (POD), False Alarm Ratio (FAR) and Threat Score (TS). POD refers to the fraction of all qualifying events correctly predicted and FAR is the fraction of predicted events that are actually non-events. TS is the integrated measure of the overall categorical metrics and it is defined as the number of events successfully estimated over the total of hit, missed and false events. The optimal values are represented by POD = 1, FAR = 0 and TS = 1 (Brocca et al., 2014).

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106

5. Results and discussion Results of the procedure for the estimation of discharge by MODIS at gauged stations of Lokoja and Makurdi are presented for different products (TERRA and AQUA), temporal resolution (daily and 8-day) and spatial aggregations (1, 4 and 9 pixels) of the images. Once demonstrated that MODIS is able to provide reliable estimates of discharge, the forecasting at Makurdi station is assessed starting from the upstream VS in which MODIS and altimetry data are both available. 5.1. River discharge estimation at Lokoja and Makurdi For all the collected MODIS images and the stations (Lokoja and Makurdi), a box of about 15 km by 15 km (60 by 60 pixels) centered on the investigated gauging site is selected as shown in Fig. 2a (Lokoja) and 2b (Makurdi) in which the median value of reflectance of band 2 (daily image) at each pixel is represented. The water course is easily visible by the pixels with values lower than 0.4. The large river width (N 1 km) facilitates the identification of the water course and allows to discard quite easily the images with clouds in the daily analysis. The cloud-free images finally selected for the analysis represent about the 45% and 39% of the total images available for Lokoja and Makurdi, respectively (see 3.2). The locations of the C pixels (characterized by low temporal variability) are in different areas if we analyze the daily and 8-day products, whereas similar patterns are found among the products derived by

101

the two satellites (TERRA and AQUA) and for different spatial aggregation (1, 4 and 9 pixels). This is probably due to the temporal resampling of the 8-day product that produces different images even if not easily verifiable from a direct comparison. On the contrary, the optimization procedure indicates a more robust and consistent location of the M pixel that appears in the same area for all the images. Fig. 3 illustrates the comparison between the observed and estimated time series of the anomalies and total discharge for the two stations (Lokoja and Makurdi) and for the four MODIS products. A positive variation of the anomaly represents river discharge higher than the seasonal cycle, typically occurring during flood events. More in general, positive or negative anomalies can be produced by the anticipation or the delay of the discharge with respect to the seasonal cycle. In terms of anomalies, the ratios C/M* (the estimated discharge) follow quite well the behavior of the observed time series for all the MODIS products at both stations. In addition, we can also observe that for Lokoja during flood events of 2009–2010 and 2012 there is the underestimation of the anomalies whereas for Makurdi station only in 2009 and for TERRA product an underestimation is found. Correlation coefficients are higher than 0.55 for Lokoja and 0.69 for Makurdi while fRMSEs are smaller than 0.83 and NSs fall in the range 0.31–0.59 (see Table 2). Better performances are obtained for the AQUA satellite and, globally, for the single pixel configuration even if some exceptions are represented by the configuration at 4 pixels. Between the two temporal resolutions, the analyses show very similar performances with comparable values: for Lokoja 8-day product which outperforms the daily one even if the difference is

Fig. 2. Median value of reflectance in band 2 (a, b, c), coefficient of variation map (d, e, f) of the MODIS images time series calculated by the daily AQUA satellite product and the optical satellite image corresponding to the selected box taken from Google Earth™ (g, h, i) for Lokoja (a, d, g), Makurdi (b, e, h) and Virtual Station (c, f, i) boxes. In the upper figures also the location of M pixel is shown (white square). In Figure i) the yellow line is the ENVISAT 530 altimetry track. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

102

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106

Fig. 3. Estimated and observed discharge anomalies (upper two rows) and total discharge (lower two rows) for Lokoja and Makurdi stations by using MODIS images from the 4 products (daily and 8-day from Terra and Aqua).

negligible, while for Makurdi the results are inverted and the differences are larger. Moving to the total estimated discharges (and not the anomalies) an effective way to evaluate the proposed approach is to compare it with

the performance obtained by using the seasonal cycle of discharges that is used for calculating the discharge anomalies. Indeed, due to the high seasonality of the discharge, the errors obtained when comparing the seasonal cycle with the measured discharge time series is relatively

Table 2 Results for anomaly and total discharge in terms of coefficient of correlation (r), fractional root mean square error (fRMSE) and Nash-Sutcliffe efficiency (NS). For the total discharge also the RMSE is shown. Results refer to the analysis considering the original images (1 pixel) and images aggregated at 2 by 2 pixels (4 pixels) and 3 by 3 pixels (9 pixels). Two MODIS products (daily and 8-day) for each satellite (AQUA, A, and TERRA, T) are analyzed. MODIS product

Discharge anomalies

Total discharges

Lakoya

Makurdi

Lokoya

fRMSE

NS

r

fRMSE

NS

r 0.95

RMSE 1717

fRMSE 0.30

NS 0.90

r 0.97

RMSE 679

fRMSE 0.17

NS 0.94

0.72 0.69

0.69 0.72

0.52 0.48

0.77 0.72

0.64 0.69

0.59 0.52

0.99 0.98

854 1100

0.16 0.20

0.97 0.96

0.98 0.98

699 804

0.18 0.19

0.97 0.96

0.63 0.65 0.73 0.67

0.77 0.76 0.69 0.74

0.4 0.42 0.53 0.44

0.77 0.7 0.77 0.73

0.64 0.71 0.64 0.68

0.59 0.49 0.59 0.53

0.99 0.98 0.99 0.98

780 1160 858 1135

0.17 0.21 0.16 0.20

0.97 0.96 0.97 0.96

0.99 0.98 0.98 0.98

499 832 641 799

0.15 0.20 0.18 0.19

0.98 0.96 0.97 0.96

0.6 0.57 0.71 0.67

0.8 0.82 0.7 0.74

0.36 0.32 0.5 0.45

0.73 0.7 0.76 0.69

0.68 0.71 0.65 0.72

0.54 0.49 0.57 0.48

0.99 0.97 0.99 0.98

823 1254 876 1134

0.17 0.22 0.17 0.20

0.97 0.95 0.97 0.96

0.99 0.98 0.98 0.98

522 833 699 842

0.16 0.20 0.18 0.20

0.97 0.96 0.97 0.96

0.61 0.55

0.79 0.83

0.37 0.31

0.7 0.7

0.71 0.72

0.49 0.48

0.99 0.97

809 1270

0.17 0.23

0.97 0.95

0.99 0.98

562 837

0.17 0.20

0.97 0.96

Seasonal cycle 1 Pixel

4 Pixels

9 Pixels

daily (A) 8-day (A) daily (T) 8-day (T) daily (A) 8-day (A) daily (T) 8-day (T) daily (A) 8-day (A) daily (T) 8-day (T)

Makurdi

r

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106

low as shown by the high values of the performances displayed in Table 2. Therefore, to be meaningful, the estimated discharge obtained through MODIS must produce better scores than the seasonal cycle. Results in Table 2 indicate overall performances higher than those of the seasonal cycle. For Lokoja station, the correlation and the NS obtained by MODIS are higher than 0.97 and 0.95 with respect to the seasonal cycle that provides values of 0.95 and 0.90. Also the fRMSE is lower than 0.23 whereas for the seasonal cycle it reaches 0.32. For Makurdi station the results are similar. The RMSE does not exceed the 1300 m3/s for Lokoja and 850 m3/s for Makurdi. Fig. 3 shows a comparison between the total observed and estimated discharge for the two stations: the estimated discharges are in agreement with the observed ones and the flood events are properly reproduced. An exception regards the flood event in 2012, which is well reproduced at the Makurdi station with the daily products (TERRA and AQUA), whereas at Lokoja station the underestimation characterizing the anomalies are reflected in the total discharge. Regarding the temporal resolution, the 8-day products show good results and the differences with the outcomes of the daily product are very small. The analyses with the aggregated MODIS images (4 and 9 pixels) show results similar to the ones obtained with a single pixel and the performances are decreasing with increasing the aggregation cells. Overall, only a slightly deterioration is observed, confirming that a similar analysis can shorten the computational burden without loss of quality. Finally, it has to be noted that at both stations the availability of the satellite data can compensate the lack of in-situ data. Indeed, estimated discharges are obtained for the year 2013 and in the case of Makurdi station also in 2010, period in which ground data are missing. This represents a huge advantage for flood monitoring and disaster response activities and application in water resource management because it gets over the actual availability of the measurements.

5.2. Flood forecasting at Makurdi station The wave travel time calibrated through the comparison between the water levels derived by satellite altimetry and observed at Makurdi is 4 days. The achieved coefficient of correlation is very high, equal to 0.993. The box considered for the analysis with MODIS images is extracted at the VS 530 for facilitating the comparison between the satellite performances. We note that the MODIS box can be selected along the whole river provided that the section is upstream with respect to the location where the forecasting of discharge is necessary, and that no significant tributaries are present. Therefore, a further upstream section can be considered with MODIS thus increasing the lead time, with the only assumption that the inter-basin contribution is proportional. Along the analyzed river reach there are no significant concentrated tributaries and the contribution of the intermediate basin (16%) is negligible with respect to the basin area at Makurdi station. Moreover, in order to corroborate the assumption that the discharge between the Makurdi and the virtual stations at the upstream are not conditioned by substantial variations due to a different rainfall regime, an evaluation on the rainfall from Global Precipitation Climatology Project, GPCP (Huffman et al., 2001) was carried out. The cumulative monthly rainfall estimated within the basin upstream Makurdi are comparable with the ones obtained upstream the virtual stations 530, 988 and 296 (not shown for brevity). If rainfall is uniformly distributed across the basin, the assumption of proportional contribution can be considered accepted. In Fig. 2c the median value (or 50th percentile) of the reflectance is shown for each pixels of the analyzed box. Once the location of M is identified by the optimization with the discharge anomalies at Makurdi station 4 days ahead, the filtered C/M* ratio is calculated. The same analysis is carried out for the aggregated images at 4 and 9 pixels.

103

The results of the forecasting are reported in Table 3 and shown in Fig. 4 where the predicted discharges by altimetry and MODIS (TERRA and AQUA daily products) are plotted along with the observed gauge at Makurdi station. Despite the low temporal resolution, the altimetry outperforms MODIS in terms of fRMSE, NS and RMSE. An integrated use of multiple satellite altimetry sensors could represent a valid approach to improve the forecasting (Tourian et al., 2016). Therefore, we investigated also the merging of the MODIS products from AQUA and TERRA. Specifically, the results obtained with the two sensors are merged in a unique time series in order to fill the date gap by one of the two satellites. In the case the information is available from both the satellites, the value of reflectance is averaged. The results are shown in Table 3 (AQUA & TERRA). For MODIS sensors a good agreement is detected between observed and estimated discharges, both for the single and integrated use. All the products provide a coefficient of persistence N 0, assuring a good level of prediction with respect to the actual data available from ground observation. 8-day AQUA product provides the higher performance. If the differences between the temporal resolution (daily and 8-day) are quite evident, for the spatial resolution (in the case of 4 and 9 pixels, not shown for brevity) there is no dominant behavior and the errors are consistent. In 2006 and 2007 a slightly overestimation is observed, whereas good prediction for the 2012 event is obtained as represented in the second-last column of Table 3 and in the zoom of Fig. 4b and c. In this case, the 8-day products are preferred in the rising limb, because they are temporally continuous along all the period. Indeed, in the daily product many images are discarded due to the cloud coverage during the month of August and the data are missing for a long period. However, in terms of peak value, daily products outperform the 8-day products due to their larger temporal resolution that does not allow a proper sampling of the flood event especially at high flow conditions. Same conclusions can be drawn for the altimetry as shown in Fig. 4d. For the use of 8-day products, it should be better forecast the discharge at least 8 days in advance in order to be consistent with the latency of the product. For daily products, as the latency is about 4–6 h, the forecast can be considered reliable and feasible. In Fig. 4b–d, the annual mean average value representing the seasonal values of the discharge is plotted demonstrating that it is not sufficient to forecast the event due to its big magnitude. For flood forecasting, we are also interested to evaluate the accuracy of the forecast during the occurrence of high flood events, i.e., during the monsoon season. Last column in Table 3 shows the results of the analysis for the period July–November, in which the Benue river experiences high discharges. In this case the PC value is high, confirming the potential and the benefit to apply the procedure during extreme events. With the purpose to use the procedure within an early warning system, the evaluation of the number of correct predictions or false alarms is carried out. Table 4 resumes the categorical metrics above a fixed threshold of 10, 000 m3/s that well represents the flood events for Makurdi station. The POD values range from 0.82 and 0.97 for the analysis of single pixel, and in all the cases they are higher than 0.79. The false alarms are relatively low with the FAR not exceeding 0.17. The TS-values are in a very high range (0.71–0.97). For comparison, the performance metrics are calculated for altimetry as well, but the performance are lower, with POD, FAR and TS equal to 0.79, 0.18 and 0.68, respectively. A further analysis is carried out to compare the performance of altimetry and MODIS in the day-by-day forecasting activities. Specifically, for each day it is computed the forecasted discharge at time t + TL (4 days ahead in this case) by using the extrapolation of the last two measurements available from the satellite at time t. Therefore, for ENVISAT data the extrapolation is done also for N 30 days (35-day revisit time) thus providing unreliable results not shown here for brevity. Indeed, the problem of the temporal resolution of the radar altimetry is evident with respect to MODIS and the performance considerably

104

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106

Table 3 Evaluation of errors on the forecasted discharge in terms of coefficient of correlation, r, Root Mean Square Error, RMSE, fractional RMSE, fRMSE, percentage RMSE, pRMSE, Nash-Sutcliffe efficiency, NS, and Persistence Coefficient, calculated on the entire time series, PC, on the 2012 flood, PC-2012 and during the monsoon season, PC-monsoon season. Also the number of the sample, N°, for which the errors are calculated is reported. The results refer to the analysis considering the two MODIS products daily and 8-day acquired by the satellite AQUA and TERRA (1 pixel) and ENVISAT/RA-2 altimetry. Product

r

RMSE (m3/s)

fRMSE

PRMSE

NS

PC

n°

PC (2012)

PC monsoon season

Altimetry Daily (Aqua) 8-day (Aqua) Daily (Aqua & Terra) Daily (Terra) 8-day (Terra) 8-day (Aqua & Terra)

0.965 0.991 0.988 0.988 0.993 0.987 0.988

351.02 543.03 622.90 622.69 417.57 631.51 612.12

0.09 0.14 0.16 0.15 0.12 0.16 0.16

0.20 0.47 0.69 0.26 0.35 0.74 0.62

0.99 0.98 0.98 0.98 0.99 0.97 0.98

0.59 0.53 0.83 0.35 0.79 0.82 0.84

242 854 237 910 657 237 236

0.55 0.55 0.1 0.68 −0.2 0.17 0.29

0.79 0.60 0.83 0.39 0.87 0.87 0.87

decreases achieving coefficient of correlation equal to 0.56 for altimetry and 0.93 (0.96) for daily (8-day) MODIS product. 6. Conclusions The analysis showed the potential of MODIS to estimate discharge and flood forecasting along the Niger Benue River in Nigeria. Specifically, four MODIS products are tested derived by two satellites (TERRA and AQUA) and with two temporal resolutions (daily and 8-day). An attempt of aggregated products is also analyzed through a resampling of 4 (2 by 2) and 9 (3 by 3) pixels. The discharge estimation at two river sections, Lokoja and Makurdi, demonstrated that MODIS is able to reproduce observations, with fRMSE b0.20 and NS higher than 0.96. Even if the differences among the four products are low, the daily products are more reliable to represent the discharge with respect to 8-day products. For the two satellites no conclusions can be drawn, as AQUA is better for Lokoja station and TERRA for Makurdi station. The resampling of the images is not so relevant and even if the pixels are aggregated, the performances are maintained.

For flood forecasting, the performances of the MODIS products are similar with overall good results (persistency coefficient N0.53). Differently from radar altimetry, the temporal resolution of MODIS provides a significant improvement but it is still limited in terms of model performance. Indeed, even if the high frequency of MODIS images fosters the evaluation of the discharge in forecasting phase, the imagery technique affects the accuracy. Moreover, the procedure can be applied all over the river reach and no restrictions exist for the choice of the box provided that the assumptions are met. The only limitation of MODIS data being an optical sensor which is affected by clouds can hinder during the forecasting phase. The current procedure of discharge estimates can be tested for other satellites data covering both the optical and radar sensors namely Landsat, Sentinel 1 and 2 which are freely available at a global scale and with a latency of a few hours (from 1 to 12). In contrast, radar altimetry provides improved accuracy with respect to MODIS procedures and it is found to be the most reliable tool to estimate river discharge in large rivers, but its success is constrained both spatially and temporally. Indeed, the forecasting is linked to the location of the satellite track and the number of observations is limited to the

Fig. 4. a) 4-day ahead forecasted discharges at Makurdi gauged station by using MODIS from AQUA and TERRA satellite (daily product) and radar altimetry from ENVISAT and Jason-2. Closer view of the 2012 flood event for daily (b) and 8-day (c) MODIS products and for altimetry (d).

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106 Table 4 Number of hit and missed events and false alarms along with the ccategorical metrics, POD (Probability of Detection), FAR (False Alarm Ratio) and TS (Threat Score) values, for observed discharge values above the threshold of 10, 000 m3/s. Product

Hit

Missed

False

POD

FAR

TS

Altimetry Daily (Aqua) 8-day (Aqua) Daily (Terra) 8-day (Terra) Daily (Aqua & Terra) 8-day (Aqua & Terra)

23 89 25 64 23 108 22

6 13 3 2 5 18 4

5 4 4 0 4 4 4

0.79 0.87 0.89 0.97 0.82 0.86 0.85

0.68 0.04 0.14 0.00 0.15 0.04 0.15

0.18 0.84 0.78 0.97 0.72 0.83 0.73

revisit time of the satellite. Further developments will be addressed to test the procedure taking advantage of the improved revisit time of recent satellite missions such as Sentinel-3 (or the future SWOT). Alternatively, new techniques can be applied for densifying the temporal series by altimetry considering all the satellite tracks available along the interested river reach. The availability of future altimetry missions should allow improvement in flood forecasting that can help in better early warning and forecasting for community resilience in reducing the risk of floods.

Acknowlegment This research is funded by the Federal Ministry of Agriculture and Rural Development (FMARD), Nigeria, (AGRIC/MOU/LU/116/2013) and the CGIAR Research Program on Water, Land and Ecosystems (WLE). The authors are grateful to Nigeria Hydrological Services Agency (NIHSA) and Nigerian Meteorological Agency (NIMET) for sharing insitu observations of water levels. Second author is thankful to Rajesh Pandey (IWMI) to support in altimetry data extraction.

References Aich, V., Koné, B., Hattermann, F.F., Müller, E.N., 2014. Floods in the Niger basin – analysis and attribution. Nat. Hazards Earth Syst. Sci. Discuss. 2, 5171–5212. Albergel, C., Rüdiger, C., Pellarin, T., Calvet, J.C., Fritz, N., Froissard, F., Suquia, D., Petitpa, A., Piguet, B., Martin, E., 2008. From near-surface to root-zone soil moisture using an exponential filter: an assessment of the method based on in-situ observations and model simulations. Hydrol. Earth Syst. Sci. 12, 1323–1337. Ali, P.I.O., 2005. Flood Damage Assessment in Makurdi Town. Unpulished M.Sc Thesis. Department of Geography, Benue State University, Makurdi, Nigeria. Amangabara, G.T., Obenade, M., 2015. Flood vulnerability assessment of Niger Delta states relative to 2012 flood disaster in Nigeria. Am. J. Environ. Prot. 3 (3), 76–83. Amarnath, G., Pandey, R.K., Smakhtin, V., Williams, T.O., 2016. Examining the utility of satellite altimetry for improved flood forecasting in the Niger-Benue river basin. J. Hydrol. Eng. (under review). Ayala, I.A., 2002. Geomorphology, natural hazards, vulnerability and prevention of natural disasters in developing countries. Geomorphology 107–124. Baup, F., Frappart, F., Maubant, J., 2014. Combining high-resolution satellite images and altimetry to estimate the volume of small lakes. Hydrol. Earth Syst. Sci. 18, 2007–2020. Biancamaria, S., Hossain, F., Lettenmaier, D.P., 2011. Forecasting transboundary river water elevations from space. Geophys. Res. Lett. 38, L11401. Birkinshaw, S.J., Moore, P., Kilsby, C.G., O'Donnell, G.M., Hardy, A.J., Berry, P.A.M., 2014. Daily discharge estimation at ungauged river sites using remote sensing. Hydrol. Process. 28 (3), 1043–1054. Brakenridge, G.R., Nghiem, S.V., Anderson, E., Mic, R., 2007. Orbital microwave measurement of river discharge and ice status. Water Resour. Res. 43, W04405. Brocca, L., Ciabatta, L., Massari, C., Moramarco, T., Hahn, S., Hasenauer, S., Kidd, R., Dorigo, W., Wagner, W., Levizzani, V., 2014. Soil as a natural rain gauge: estimating global rainfall from satellite soil moisture data. J. Geophys. Res. 119 (9):5128–5141. http:// dx.doi.org/10.1002/2014JD021489. De Groeve, T., Riva, P., 2009. Early flood detection and mapping for humanitarian response. In: Landgren, J., Nulden, U., Van de Walle, B. (Eds.), Proceedings of the 6th International ISCRAM Conference – Gothenburg, Sweden, May 2009. De Groeve, T., 2010. Flood monitoring and mapping using passive microwave remote sensing in Namibia. Geomat. Nat. Haz. Risk 1 (1), 19–35. Desjonquères, J.-D., Carayon, G., Steunou, N., Lambin, J., 2010. Poseidon-3 radar altimeter: new modes and in-flight performances. Mar. Geod. 33 (SI), 53–79. Domeneghetti, A., Tarpanelli, A., Brocca, L., Barbetta, S., Moramarco, T., Castellarin, A., Brath, A., 2014. The use of remote sensing-derived water surface data for hydraulic model calibration. Remote Sens. Environ. 149, 130–141.

105

Domeneghetti, A., Castellarin, A., Tarpanelli, A., Moramarco, T., 2015. Investigating the uncertainty of satellite altimetry products for hydrodynamic modelling. Hydrol. Process. 29 (23), 4908–4918. Draper, C., Reichle, R., De Jeu, R., Naeimi, V., Parinussa, R., Wagner, W., 2013. Estimating root mean square errors in remotely sensed soil moisture over continental scale domains. Remote Sens. Environ. 137, 288–298. Emergency Events Database – EM-DAT, 2016. http://cred.be/sites/default/files/2015_ DisastersInNumbers.pdf (Accessed on 13 January 2017). Frappart, F., Calmant, S., Cauhopé, M., Seyler, F., Cazenave, A., 2006. Preliminary results of ENVISAT RA-2 derived water levels validation over the Amazon basin. Remote Sens. Environ. 100 (2), 252–264. Frappart, F., Papa, F., Marieu, V., Malbéteau, Y., Jordy, F., Calmant, S., Durand, F., Bala, S., 2015. Preliminary assessment of SARAL/AltiKa observations over the Ganges-Brahmaputra and Irrawaddy Rivers. Mar. Geod. 38 (S1), 568–580. Fu, L.L., Cazenave, A. (Eds.), 2001. Satellite Altimetry and Earth Sciences. Academic Press, San Diego, CA, p. 463. GFDRR (2014). Nigeria post-disaster needs assessment: 2012 floods. https://www.gfdrr. org/sites/gfdrr/files/NIGERIA_PDNA_PRINT_05_29_2013_WEB.pdf (Accessed on 13 January 2017). Getirana, A.C.V., 2010. Integrating spatial altimetry data into the automatic calibration of hydrological models. J. Hydrol. 387, 244–255. Gleason, C.J., Smith, L.C., 2014. Toward global mapping of river discharge using satellite images and at-many-stations hydraulic geometry. Proc. Natl. Acad. Sci. 111 (13): 4788–4791. http://dx.doi.org/10.1073/pnas.1317606111. Hirpa, F.A., Hopson, T.M., De Groeve, T., Brakenridge, G.R., Gebremichael, M., Restrepo, P.J., 2013. Upstream satellite remote sensing for river discharge forecasting: application to major rivers in South Asia. Remote Sens. Environ. 131, 140–151. Hossain, F., Siddique-E-Akbor, A.H.M., Mazumdar, L.C., Shah Newaz, S.M., Biancamaria, S., Lee, H., Shum, C.K., 2014. Proof of concept of an operational altimeter-based forecasting system for transboundary flow. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 7 (2):587–601. http://dx.doi.org/10.1109/JSTARS.2013.2283402. Huffman, G.J., Adler, R.F., Morrissey, M.M., Bolvin, D.T., Curtis, S., Joyce, R., McGavock, B., Susskind, J., 2001. Global precipitation at one-degree daily resolution from multi-satellite observations. J. Hydrometeorol. 2, 36–50. Hula, M.A., Udoh, J., 2015. An assessment of the impact of flood events in Makurdi, Nigeria. Civ. Environ. Res. 7 (10), 53–60. Kitanidis, P.K., Bras, R., 1980. Real time forecasting with a conceptual hydrologic model. 2. Applications and results. Water Resour. Res. 16 (6), 1034–1044. Kouraev, A., Zakharova, E.A., Samain, O., Mognard-Campbell, N., Cazenave, A., 2004. Ob' river discharge from Topex/Poseidon satellite altimetry. Remote Sens. Environ. 93, 238–245. Leon, J.G., Calmant, S., Seyler, F., Bonnet, M.P., Cauhope, M., Frappart, F., Filizola, N., Fraizy, P., 2006. Rating curves and estimation of average water depth at the upper Negro River based on satellite altimeter data and modeled discharges. J. Hydrol. 328, 481–496. Michailovsky, C.I., Milzow, C., Bauer-Gottwein, P., 2013. Assimilation of radar altimetry to a routing model of the Brahmaputra River. Water Resour. Res. 49 (8), 4807–4816. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models, part I: a discussion of principles. J. Hydrol. 10 (3), 282–290. Neal, J., Schumann, G., Bates, P., 2012. A subgrid channel model for simulating river hydraulics and floodplain inundation over large and data sparse areas. Water Resour. Res. 48 (11). Negrel, J., Kosuth, P., Bercher, N., 2011. Estimating river discharge from earth observation measurement of river surface hydraulic variables. Hydrol. Earth Syst. Sci. 15 (6), 2049–2058. O’Neal, J., 2005. The Near Real Time Processing Effort. Crosslink Newsletter, Winter 2005. Pandey, R.J., Crétaux, J.F., Bergé-Nguyen, M., Tiwari, V.M., Drolon, V., Papa, F., Calmant, S., 2014. Water level estimation by remote sensing for the 2008 flooding of the Kosi River. Int. J. Remote Sens. 35 (2), 424–440. Pandey, R., Amarnath, G., 2015. The potential of satellite radar altimetry in flood forecasting: concept and implementation for the Niger-Benue river basin. Proc. IAHS. 370, pp. 223–227. Pauwels, V.R., Hoeben, R., Verhoest, N.E., De Troch, F.P., 2001. The importance of the spatial patterns of remotely sensed soil moisture in the improvement of discharge predictions for small-scale basins through data assimilation. J. Hydrol. 251 (1), 88–102. Pavlis, N.K., Holmes, S.A., Kenyon, S.C., Factor, J.K., 2012. The development and evaluation of the Earth Gravitational Model 2008 (EGM2008). J. Geophys. Res. Solid Earth 117. Santos da Silva, J., Calmant, S., Seyler, F., Rotunno Filho, O.C., Cochonneau, G., Mansur, W.J., 2010. Water levels in the Amazon basin derived from the ERS 2 and ENVISAT radar altimetry missions. Remote Sens. Environ. 2010 (114), 2160–2181. Smith, L.C., Pavelsky, T.M., 2008. Estimation of river discharge, propagation speed, and hydraulic geometry from space: Lena River, Siberia. Water Resour. Res. 44 (3). Tarhule, A., 2005. Damaging rainfall and flooding: the other Sahel hazards. Clim. Chang. 72 (3), 355–377. Tarpanelli, A., Barbetta, S., Brocca, L., Moramarco, T., 2013a. River discharge estimation by using altimetry data and simplified flood routing modeling. Remote Sens. 5 (9), 4145–4162. Tarpanelli, A., Brocca, L., Lacava, T., Melone, F., Moramarco, T., Faruolo, M., Pergola, N., Tramutoli, V., 2013b. Toward the estimation of river discharge variations using MODIS data in ungauged basins. Remote Sens. Environ. 136, 47–55. Tarpanelli, A., Brocca, L., Barbetta, S., Faruolo, M., Lacava, T., Moramarco, T., 2015. Coupling MODIS and radar altimetry data for discharge estimation in poorly Gauged River basins. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 8 (1), 141–148. Tourian, M.J., Sneeuw, N., Bardossy, A., 2013. A quantile function approach to discharge estimation from satellite altimetry (ENVISAT). Water Resour. Res. 49 (7), 4174–4186.

106

A. Tarpanelli et al. / Remote Sensing of Environment 195 (2017) 96–106

Tourian, M.J., Tarpanelli, A., Elmi, O., Qin, T., Brocca, L., Moramarco, T., Sneeuw, N., 2016. Spatiotemporal densification of river water level time series by multimission satellite altimetry. Water Resour. Res. 52. UNDP, United Nations Development Programme, 2015. Human Development Report 2015: Work for Human Development. New York. http://hdr.undp.org/sites/default/ files/2015_human_development_report_1.pdf. VALS Tool Virtual ALtimetry Station. VALS Version 0.6.2, 2010. Available online: http:// www.ore-hybam.org/index.php/eng/Software/VALS (accessed on 13 January 2017). Wagner, W., Lemoine, G., Rott, H., 1999. A method for estimating soil moisture from ERS scatterometer and soil data. Remote Sens. Environ. 70, 191–207.

Wehr, T., Attema, E., 2001. Geophysical validation of ENVISAT data products. Adv. Space Res. 28, 83–91. Yan, K., Tarpanelli, A., Balint, G., Moramarco, T., Di Baldassarre, G., 2014. Exploring the potential of radar altimetry and SRTM topography to support flood propagation modeling: the Danube case study. J. Hydrol. Eng. 20 (2). Vermote, E.F., Kotchenova, S.Y., March 2008. MOD09 (Surface Reflectance) User's Guide, Version 1.1.