Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

Selection No 1172405_

Technical Proposal
Section D

D - Description of Approach, Methodology and Work Plan

In this section, we describe the technical approach, methodology and work plan of the
technical components of the assignment, “Development of Flood Forecasting for the Ganges
and the Brahmaputra Basins using satellite based precipitation, ensemble weather forecasts,
and remotely-sensed river widths and height”. In addition, we describe the structure and
composition of our team assembled to meet this assignment’s objectives.

Technical  Approach  and  Methodology  

In this sub section we explain our understanding of the objectives of the assignment and
methodology for carrying out the activities and obtaining the expected outputs within the
overarching objective of this assignment: to support the assessment of strategic improvements
to regional flood forecasting capacity. We reproduce each Objective given in the RFP below,
and give our response directly beneath each one:

Objective (i) Implement long-lead time, public-access flood forecasting systems for the
Ganges and Brahmaputra basins spanning India, Bangladesh and Nepal utilizing new
satellite precipitation estimates and ensemble weather forecasts from multiple centers.
The Climate Forecasting Applications for Bangladesh (CFAB) river flow forecasting system
(Hopson & Webster, 2010), currently operational only for Bangladesh, will be extended into
India, Nepal, and Bangladesh, implemented and re-calibrated basin-wide for all the tributaries
and corresponding sub-basins of Ganges and Brahmaputra. The existing forecast modeling
will then be improved as follows:
(a) by optimally-combining additional satellite precipitation estimates (new products NOAA
Hydroestimator and JAXA/EORC GSMaP products will be added to the existing utilization of
NASA TRMM and NOAA CMORPH products);
(b) ensemble weather forecasts from multiple centers will be used (CMA, CMC, JMA, KMA,
MeteoFrance, NCEP, and UKmet in addition to the current ECMWF);
(c) utilizing a higher resolution DEM;
(d) routing model improvements;
(e) new pre- and post-processing statistical tools (“quantile regression based) used in
ensemble streamflow prediction;
(f) improvements to snow modeling (SNOW-17, operational US model);
(g) and finally, the system will also be extended out an additional 5 days (from the current 10
days), creating an enhanced lead-time (1-15 day) probabilistic river flow forecasting scheme.
Selection No. 1172405
Technical Proposal
Section D
Response: The CFAB
hydrological forecast model is a
hydrologic multi-modeling
system whose soil moisture
states are initialized by NASA
and NOAA precipitation
products (e.g., TRMM 3B42,
Huffman et al., 2005, 2007;
CMORPH, Joyce et al., 2004),
and whose states and fluxes are
forecasted forward using
ensemble weather forecast data
products and conditionally post-
processed to produce calibrated
probabilistic forecasts of river
discharge for key river reach
locations. Already operational
over the Brahmaputra and
Ganges river basins but only
providing operational flow
forecasts at points within
Bangladesh (the Ganges at
Hardinge Bridge and the
Brahmaputra at Bahadurabad, in particular) , this system will be extended and calibrated for
subcatchments upstream in India to show the potential benefits for upstream users as well. To
reach this goal, this Objective calls out very specific approaches that can be used to further
improve the skill and accuracy of the CFAB model, especially as it will be used to provide
flood forecasts at much smaller catchment scales than previously used (e.g. improvements
such as higher DEM resolution, more quickly updated satellite precipitation estimates such as
the NOAA Hydroestimator (Scofield & Kuligowski, 2003), and use of an expanded set of
ensemble weather forecasts, provided freely by the Thorpex-Tigge archive
(http://tigge.ecmwf.int, as opposed to sole reliance on the forecasts from ECMWF which were
used previsously as part of the CFAB project for Bangladesh). If NCAR receives this
consultancy, we will build each of this sublist of improvements directly into the CFAB model
that was used for Bangladesh. The expected outputs from this will be web displays that
provide 3-hr updated imagery of subcatchment flood risk, and clickable pop up time-series
plots showing past and forecasted river flows for each subcatchment.

As an example of the subcatchments to be forecasted for, previous work by the consultant


Team Leader has delineated the Ganges and Brahmaputra catchments and subbasins using
Hydro1k, shown below. (However, for this consultancy, we would generate more refined
delineations using higher resolution SRTM data.)
Selection No. 1172405
Technical Proposal
Section D

Figure 2: Ganges (upper panel) and Brahmaputra (lower panel) sub-catchments individually
to be forecasted, both for extreme rainfall events and main-stem flood levels.

Objective (ii) Combine CFAB ensemble discharge forecasts with DFO remotely-sensed
river discharge estimates to produce optimal river discharge estimates at select locations
along the river course. The Dartmouth Flood Observatory (DFO), in partnership with the
Joint Research Council, is providing multi-site estimates of river width for both the Ganges
and Brahmaputra (see JRC-Ispra, http://www.gdacs.org/floodmerge/ and DFO,
Selection No. 1172405
Technical Proposal
Section D
http://www.dartmouth.edu/~floods/). These estimates are publically-available at
http://old.gdacs.org/flooddetection/. Actual measurements are of upstream relative changes in
emissivity due to changes in river width (microwave imagery visible through cloud cover).
This signal is then transformed into discharge (see Hirpa et al. 2013) and effectively
“advected” to downstream forecasting locations. This river discharge system is independent
of the CFAB system discussed in Objective (i) above, the results of which will be optimally
combined with the CFAB system, with weighing coefficients optimized by forecast lead-time.
Skill comparisons between the CFAB and DFO-based system will be made, in addition to the
benefits in their optimal combination.

Response: One of the drawbacks of the previous CFAB system, as designed for Bangladesh,
was the lack of data availability of near-real-time river stage measurements upstream in the
Ganges and Brahmaputra catchments. As such, the CFAB modeling structure treated the
upstream river catchments as being “ungauged”. One approach to instead provide a “work
around” on this data limitation is to explore remotely-sensed information that do not rely on in
situ measurements being taken and being timely provided. The DFO RiverWatch data provide
one such source of such data, especially since the imagery this system relies on (microwave)
can penetrate cloud cover (which should be potentially plentiful during times of flood risk).
Thus, Figs. 3
and 4 indicate
what has been
entirely missing
but is now
feasible for
South Asia. It
provides a
subscene of the
present DFO
River Watch 2
map view
(automated,
updated daily)
of current river
flow severity
status of
monitored sites
(Fig. 3), and
also a sample
time-series of the existing present and historic status output at one site (Fig. 4).

This Objective, then calls for building upon previous work that has explored “advecting”
these signals downstream to provide flood forecast information (Hirpa et al 2013, for which
key personnel on this consultancy were co-authors on), and calls for combining flood
forecasts derived by this approach with the skillfully-proven CFAB approach (whose skill is
derived independently through the time-lags in the transport of observed satellite precipitation
estimates through watersheds, lead-times provided by weather forecasts, and time-series
Selection No. 1172405
Technical Proposal
Section D
forecasting methods). To reach the goals of
this objective, we will utilize a “new and
improved” (improvments made via the
steps of Objective (i)) CFAB system (which
our personnel have extensive knowledge
of), with the similar technologies used as
discussed in the Hirpa 2013 study. In
addition, the Hirpa study will need to be
“operationalized” (the study was done in
“hindcast mode”), and the CFAB and Hirpa
results will also be combined using a
“quantile regression” approach that has
been already developed as part of leveraged
technologies stemming from NCAR’s Figure 5 Daily time series of observed river discharge (solid)
ATEC project (see Section B.1.6). Note that and model nowcast (dash) based on the satellite-based river
discharge estimates for Ganges River at Hardinge Bridge
the benefits of utilizing quantile regression ground station in Bangladesh. Satellite-derived information
for this task are that optimal combinations at locations with distances ranging from 63 KM to 1828KM
upstream Hardinge Bridge station were used. (From Hirpa et
can be generated for each quantile (10%,
al., 2013.)
20%, etc.) of a forecasted probability
distribution, such that a best estimate and “error bounds” on the possible uncertain range is
also immediately provided. We would expect to report on how much skill can additionally be
provided through their optimal combination of these two approaches, as part of this
consultancy.

Objective (iii) Transform forecast flood discharges into inundation extent maps, using
analysis of past microwave and optical sensor imagery of actual inundation extent. A
unique enhancement to river flow forecasting will be to transform modelled forecast flood
discharges into (likely) inundation extent maps using historic flood imagery. This provides
accurate geo-location of flooding important for disaster relief efforts in complex terrain,
where numerical inundation modeling would normally fail. This will be accomplished using
analogue approaches to select pairings of archived, remotely sensed maps of inundation extent
matched with river discharges similar to model predictions.

Response: Providing flood forecasts of discharge magnitude are certainly vital to effectively
mitigate flood hazards for vulnerable citizens. If extensive flood mapping surveys have been
carried out, then the spatial extent of oncoming flood waters can be a priori determined.
However, for most regions of the world, such hydraulic modeling and spatial mapping have
not been carried out in sufficient detail.

To get around this limitiation, historical context could be used: if the discharge magnitudes
can be placed into the context of historical flows, a sense of the flood return period can be
estimated (10yr flood, 25yr flood, etc.), and citizens with intimiate knowledge of their region
could have context for the spatial extent the flood waters will occupy. However, preferable to
this latter approach would be to relate forecasted flood discharges to archived imager over the
region of interest that occurred during periods of similar river flows, thus giving citizens and
relief workers more specific a priori knowledge of the extent oncoming flood waters could
Selection No. 1172405
Technical Proposal
Section D
occupy, without reliance on detailed surveys. Note, however, one of the limitations of this
approach is the reliance on the assumption of geomorphological consistency – past and future
river channel characteristics (e.g. in crosssection, conveyance, and river channel location)
need to remain similar for this approach to be accurate. Thus, this method (or any of the three
methods discussed, actually) will be less accurate on rivers that are highly geomorpholocially
active.

To meet this Objective, we will


utilize the DFO archive of
catalogued imagery. NASA’s
orbital technology has been used
at DFO extensively, since the
launch of the twin MODIS
sensors in early 2000 and 2002,
to map flooding in South Asia.
Unlike other remote sensing-
based organizations active in
flood response, DFO maintains a
large and growing archive of such
Figure 6. The DFO record of flooding in this portion of the Ganges Basin is map data, in digital (GIS) format,
shown as light blue (2000 to 20111), light red was flooding in the 10 days prior to and for use in making
map update date, and red was current flooding. The numbers indicate River
Watch discharge measurement sites. This is a small subscene of the complete comprehensive regional displays
Surface Water Record display for this region. indicating the history of
inundation as well as on-going
flooding (Fig. 6). The archival flood information is exceptionally valuable, providing as it
does a view of flood hazard.

This large archive of such mapped inundation resident at DFO will allow production of this
innovative flood
prediction product. As
illustrated in Fig. 7, past
inundation extent can be
matched to the
corresponding remote
sensing-derived
discharge values (the
same approach can be
used for any ground
station sites for which
data output is available
publicly). Linkage to the
appropriate inundation
map can be provided at
Figure 7. Left: MODIS imaging and mapping of 2003 flooding along the Ganges River
between river measurement sites 200 and 201. At site 200 (uncalibrated) peak discharge the individual site
was 8500 m3/sec. Right: mapping of 2004 flooding. The uncalibrated peak discharge here displays: when a
is only ~3500 m3/sec.
particular discharge and
Selection No. 1172405
Technical Proposal
Section D
flood threshold is predicted, the user can call up the inundation that resulted, historically, from
the same values. Similarly, mapping inundation maps to the ensemble of river forecasts
produced by the CFAB model could then produce a range of possible inundation extent
scenarios.

Objective (iv) Validate Radar Altimetry Satellite Data for Operational Flood Forecast.
Synthetic aperture radar (satellite-based microwave (sic) sensor) altimetry data will be
collected and synthesized over river measurement sites to monitor changes in river water
heights to test their degree of accuracy for flood forecasting applications. These
measurements have the potential of being converted into river flow measurements to
supplement in-situ discharge data and are used to assess river heights at the regional scale
across all sub-basins. In addition, this approach will also test monitoring of reservoir levels, to
remotely-assess water release schedules of management agencies.

Response: Satellite radar altimeters have a certain capacity for monitoring the variations in
surface water level over the world’s largest lakes, reservoirs, wetlands and river channels. As
such the altimeters have played a role in basic research for the last two decades, and they are
also currently contributing to applied programs via the delivery of near real time products to
meet operational objectives. Just as with the DFO RiverWatch measurements discussed
above, satellite altimetry measurements can also be employed to measure upstream river
conditions (where river stage gauges are not available) whose information can then be used
(due to advective travel-time delays) to forecast the onset of flood waves downstream. The
benefits of satellite altimetry are its greater accuracy as compared to the RiverWatch data
(very roughly a factor of 10), but its drawbacks are its lower sampling frequency (roughly
every 10-days as compared to daily for RiverWatch at a fixed location). However, the
sampling frequency for a given river (as opposed to fixed location) is much higher, and in
particular, we believe combining both the RiverWatch and altimetry approaches together can
have real power to monitor upstream conditions of rivers with no real-time river stage
reporting. An important anticipated finding of this objective will be to test how far upstream
(as river widths diminish) the altimetry approach can retain some level of accuracy to benefit
flood forecasting, and over what types of terrain and at what sampling frequency and spatial
intervals.

For this Objective, Sub Consultant Dr. Charon Birkett at the Earth System Science
Interdisciplinary Center (ESSIC) of the University of Maryland will be employed. Dr. Birkett
is the Principal Investigator of the G-REALM program, the Global Reservoir and Lake
Monitor, is a NASA/USDA funded program that ingests raw altimetric data sets and delivers
water-level variations products for the world’s largest lakes and reservoirs. The products are
utilized by USDA/FAS regional analysts to determine i) short-term agricultural drought
conditions, and ii) long-term hydrological drought status. They also help assess irrigation
potential in basins for which in situ data is often sporadic or has restricted access. The
technique of deriving lake water-level variations is similar for rivers, floodplains and wetland
regions, and the G-REALM system could equally be used to output products for identified
river channel crossings.
Selection No. 1172405
Technical Proposal
Section D
The G-REALM water level products (in graphical and text format) are created using a suite of
altimeter data sets. Current operational products are based on IGDR data from the
NASA/CNES Jason-2/OSTM mission, and in the near future, from the ISRO/CNES SARAL
mission. The two figures below provide preliminary analysis from the use of this system to
meet this particular Objective of this consultancy.

Figure 8: Google Earth imagery showing the location of the satellite radar altimeter “Jason-
2/OSTM” ground track crossings over the Ganges River. These crossings have a temporal
resolution or repeat period of 10-days. For example, satellite passes 003, 079, and 155 etc.
ascend over the river from south-west to north-east. Descending passes 192, 014, 090 etc.
descend over the river from the Himalayas. Time series of water height variations can be
constructed at each channel-crossing site.
Selection No. 1172405
Technical Proposal
Section D
Figure 9 (below): Preliminary altimetric time series depicting water level variations for
specific ascending pass channel crossings on the Ganges River. Data source: Jason-2/OSTM
satellite mission (2008 to present day), GDR data.
Selection No. 1172405
Technical Proposal
Section D
Work  Plan  

In this sub section we discuss the main activities of the assignment, including their content
and duration, and dates of the milestones to be reached. A list of the final documents, reports,
and trainings to be delivered is also provided. With reference to the key tasks to be undertaken
please refer to Figure 1 above (the flow chart of CFAB system forecast production). With
reference to the timing and duration of these tasks, please also see Annex 3 – Work Schedule.
As with the previous subsection, we reproduce each Task as given in the RFP below, and give
our response to each directly beneath each one, providing anticipated task begin date relative
to the beginning of the consultancy period of performance (PoP) and duration (in weeks).
These tasks are:

Task 1: GIS Basin Delineation


• Delineate sub-basins and river networks using SRTM (30m horizontal resolution)
data. The delineated sub-basins using the USGS Hydro1k (1km resolution stream
networks and flow connectivity) will be used, but use of the STRM data will be
employed to repeat the process at higher resolution.
Response: Our team has technical experience delineating watersheds across the globe,
including for the Ganges and Brahmaputra (see Figure 2 above). Given the foundational
importance of this task, we anticipate completing this task, including ancillary outputs
necessary for improved river routing, within the first week of this consultancy:
Duration/start date: 1week beginning week 1 of PoP

• Calculate grid-weighting per sub-basin for gridded rainfall observation, estimates, and
forecast products. This process will generate (fractional) weights that will efficiently
combine precipitation values from neighboring grids to produce sub-basin average
rainfall, based on the grid-box spatial overlap with the basin.
Response: At the finest subcathment scale, basins will be treated as single units with routing
routines linking and accumulating the subcatchment flows downstream. As such, subbasin
(spatially-)averaged rainfall needs to be calculated from gridded rainfall products
(observations or forecasts). To accomplish this, weights of each grid (of the precipitation
products) as applied to each subcatchment need to be calculated. Our team has developed
scripts that will be modified for generating subbasin areal-based weighting for each grid point
and for each subbasin (with a unique set of weights for each gridded products’ grid-spacing,
aka for each unique Thorpex-TIGGE weather center grid spacing weather forecasts and
satellite precipitation estimates grid spacing). Given our team’s technical experience
delineating watersheds and its foundational importance, and we anticipate completing this
task in one week, starting the 2nd week of the PoP:
Duration/start date: 1week beginning week 2 of PoP
Reported Milestones: we would provide the World Bank figures showing the newly
delineated subcatchment for which we will forecast for.
Selection No. 1172405
Technical Proposal
Section D

Task 2: Rain Gage and Stream Gage Data Collection and Processing
• Collect archived data – for use in model calibration and analogue-technique selection
Response: The data outlined in Task 2 are essential for data assimilation and verification of
the consultancy forecast outputs. Our team will update our archives of raingauge (gridded
0.5X0.5 degree data from NOAA and the World Meteorological Organization (WMO) global
telecommunication system (GTS)), along with river stage information for the downstream
boundary condition locations of the Ganges (Hardinge Bridge in Bangladesh) and the
Brahmaputra (Bahadurabad in Bangladesh), including any observed river discharge data for
these gaging locations collected from the Bangladesh Flood Forecasting and Warning Center
(FFWC), along with any other data provided by the World Bank.
Duration/start date: 2 weeks beginning week 1 of PoP

• Derive rating curves from observations – available river discharge and stage
measurements will be used to generate rating curves at select sites.
Response: where available, we will derive our own rating curves, using our own developed
algorithms, for each location that we have both stage and observed discharge records.
Duration/start date: 2 weeks beginning week 2 of PoP

• Automate real-time download of available rain and river stage gage sites.
Response: we have download scripted templates which will be modified for each of the
products listed above (including any new possible data provided by the World Bank). In
addition, we will automate the capture of stage measurements available upstream in India
provided at the website www.india-water.gov.in/ffs (Figure 2 below).
Duration/start date: 1 week beginning week 4 of PoP.
Reported Milestones: provide the World Bank access to view the updated stage measurements
over the web.
Selection No. 1172405
Technical Proposal
Section D

Figure 2: imagery of stage measurements available for automated download through the website
www.india-water.gov.in/ffs

Task 3: Satellite Precipitation Estimates


• Collect archived data from NOAA, NASA and JAXA/EORC for calibration purposes
Response: To calibrate hydrologic models, the best estimate of historical precipitation over
the watersheds need to be used. In raingauge data sparse regions, these data are often provided
by satellite precipitation estimates. Previous work on the CFAB project by the Team Leader
has archived NOAA CMORPH and NASA TRMM precipitation estimates. However, these
archives will need to be updated, along with new archives for the NOAA HydroEstimator and
JAXA products (new products to be tested) would be generated.
Duration/start date: 2 weeks beginning week 3 of PoP.

• Automate real-time download of these products and combine for operational


hydrologic model and in-stream flow initialization
Response: The developed CFAB codes of the Team Leader Hopson will be used as templates
for automatic download (for operational updating) of the NOAA, NASA, and JAXA satellite
Selection No. 1172405
Technical Proposal
Section D
precipitation products, which are essential as part of the objective of producing an automated
river forecast system.
Duration/start date: 1 week beginning week 5 of PoP

• Optimal combination of products, based on available rain gage values


Response: where reporting raingauge data are available, we will use them to calibrate the
satellite rainfall estimates for: 1) locations where no raingauge data is available, but still
required for hydrologic model calibration, and for 2) for quicker responding operational flood
forecasting, given that satellite data availability times are typically significantly quicker than
rain gauge reports. The calibration will be done for: 1) probability of precipitation, and for 2)
the rainfall amount, given a rainfall event occurs.
Duration/start date: 2 weeks beginning week 5 of PoP
Reported Milestones: We would provide the World Bank links to web-based near-real-time
plots of the combined rainfall products showing current rainfall conditions for each
subcatchment relative to its own historical climatology, providing a “return period” contextual
estimate of the severity of the rainfall event.

Task 4: Thorpex-TIGGE NWP Ensembles


• Collect archived data – It will require downloading from ECMWF data
Response: Download scripts will be developed by NCAR’s Computational Information
Systems Laboratory (CISL) will be developed to assist this project, downloading these data
from the ECMWF Thorpex-TIGGE portal (based at Reading, UK).
Duration/start date: 2 weeks beginning week 4 of PoP

• Automate real-time download


Response: Download scripts will be automated for 6-times-daily automated download of 6
weather forecast center ensemble data products (for a total of over 200 dynamic ensemble
members forecasting potential rainfall at 6-hour time-steps over the Ganges and Brahmaputra
catchments).
Duration/start date: 1 week beginning week 5 of PoP

• Pre-process and optimal combination of products, based on available rain gauge


values, utilizing a quantile-regression and analog based approach, generating
calibrated probability distribution function inputs into hydrologic model.
Response: Previous calibration algorithm templates developed by Team Leader Hopson as
part of the CFAB, Google, and ATEC projects (subsections B.1.1, B.1.2, and B.1.6 above)
will be leveraged to combine precipitation forecasts for this consultancy. Just as for the
satellite precipitation estimates, the ensemble precipitation forecast products will also need to
be calibrated relative to raingauge values where available, before ingesting them into
hydrologic models (otherwise creating the potential for biased river discharge forecasts).
Selection No. 1172405
Technical Proposal
Section D
Duration/start date: 2 weeks beginning week 5 of PoP
Reported Milestones: Similar to what is proposed for satellite-derived rainfall estimates (Task
3), we would provide the World Bank links to web-based near-real-time plots showing current
rainfall conditions for each subcatchment relative to its own historical climatology, providing
a “return period” contextual estimate of the severity of the forecasted rainfall event.

Task 5: Hydrologic Modeling and Multi-modeling


• Implement lumped and semi-distributed 2-layer models for each sub-basin and
gauging locations (maximum of three).
Response: The CFAB hydrologic forecast system used here consists of both “lumped” and
“semi-distributed” modeling components (see Figure 1 above). The lumped model treats
everything upstream of the forecast location as “lumped” (aka a single spatially-averaged
rainfall value). For this consultancy, each location along the Ganges and Brahmaputra rivers
where we have river stage measurements (see triangles in Figure 2 above) could be treated as
a distinct implementation of the “lumped” modeling approach (aka some hundreds of points),
using each reported stage value ingested as part of the algorithm. Although the consultancy
specifies a maximum of three staging locations be done, we will consider implementing the
approach at more than just three sites within the Ganges and Brahmaputra catchments,
depending on the computational demands of each process and the computational resource still
available.
Duration/start date: 3 weeks beginning week 7 of PoP

• River routing using


o Constant travel time in lower reaches.
o Muskinghum-Cunge in upper catchments of the Ganges and Brahmaputra.
Response: As part of the implementation of the semi-distributed modeling component of the
CFAB forecasting system, both a constant travel time (simplest) and Muskinghum-Cunge
(more computationally demanding) routing scheme would be implemented and the results
compared at different stage observation locations to see the added (if any) benefits of the
Muskinghume-Cunge over the constant travel time, and whether the former justifies the extra
computational burden.
Duration/start date: 4 weeks beginning week 10 of PoP
Reported Milestones: The results of this comparison study would be made available to World
Bank staff at the completion of this task.

• Automate model calibration.


Response: Both the CFAB lumped and semi-distributed hydrologic schemes have calibration
coefficients as part of their algorithms. We will implement automated calibration (CRON
computer background) processes to optimize these calibrations as more data comes in – this is
especially important, since at the beginning of this consultancy, very little upstream stage data
Selection No. 1172405
Technical Proposal
Section D
would be available from online stage data sites (see Figure 2), but will continue to be
generated over time, continually improving hydrologic calibrations.
Duration/start date: 1 week beginning week 10 of PoP

• Implement auto-regressive corrections to each model, before combining into optimal


multi-model
Response: An important error correction component to the CFAB forecasting system are auto-
regressive error corrections applied to the forecast time series. These autoregression
corrections rely on near-real-time stage (or discharge) measurements. This algorithm would
be applied to the river forecast location where stage readings are available (Figure 2).
Duration/start date: 2 weeks beginning week 13 of PoP

• Optimal combination of model outputs, combine models for each lead time and post-
process and combine using quantile-regression and analogue based-approach
generating calibrated final probability distribution function.
Response: One of the strengths of using more than one hydrologic model (in this case, the
lumped and semi-distributed approaches), is that they can optimally be combined and
weighted to essentially mitigate random errors of each, and improve overall foreasting skill.
Here we propose to utilize a quantile regression calibration system (similar as discussed in
Tasks 3 and 4), that generates an ensemble of forecasts, where each ensemble represents a
specific optimized quantile of the forecast output probability distribution function.
Duration/start date: 3 week beginning week 14 of PoP
Reported Milestones: As each modeling component (“lumped”, semi-distributed, etc.,
becomes available, we would provide web-based plots of forecasted discharge outputs as they
become available at the completion of each sub-task listed above, for near-real-time display.

Task 6: Incorporate DFO Measurements


• Increase the number of daily-updated River Watch river flow measurement sites to 90
locations while receiving input from regional specialists and collaborators in locating
these sites; and test accuracy to available local ground station data.
Response: Previous work by our team (Feyera et al. 2013) utilized approximately 26
RiverWatch sites along each of the two rivers of this consultancy (aka Ganges and
Brahmaputra). Skill could potentially be further improved beyond that shown in this past
work by the provision of more data, to further improve the skill of downstream advective
forecasting, and also as data that could be used in data assimilation algorithms to update
upstream hydrologic model states above the RiverWatch sites.
Duration/start date: 6 weeks beginning week 1 of PoP

• Implement new routing routine in CFAB model to overlap DFO measurement sites
with model grid points.
Selection No. 1172405
Technical Proposal
Section D
Response: The benefits of having a consistent set of hydrologic model routing sites and DFO
measurement sites are that this allows the DFO RiverWatch data to be available for directly
updating (via data assimilation) the river flow states in the CFAB semi-distributed model
routing routine.
Duration/start date: 3 weeks beginning week 13 of PoP

• Create real time data links, from DFO to NCAR, to enable their planned discharge
measurements assimilation task.
Response: Real time data links are essential for operational flood forecasting, and being
timely enough allow data assimilation into hydrologic model initial states before forecast
generation ensues.
Duration/start date: 2 weeks beginning week 15 of PoP

• Develop assimilation algorithm to update CFAB model flow estimates with River
Watch measurements.
Response: Given funding levels and time constraints of this consultancy, the data assimilation
algorithm would be based on simple “optimal interpolation” of instream hydrologic model
routing states, which potential future consultancies can build from to implement more
complex data assimilation schemes (as well as ingest satellite altimetry data as well).
Duration/start date: 3 weeks beginning week 15 of PoP

• Automate all data ingest and data assimilation processes.


Response: As above, crucial to near real-time operational forecast generation over the scale
required of this consultancy (aka roughly 1.5 million square kilometers), all tasking of the
forecast system implementation will need to be automated. Here we will build upon
experience and scripts developed from the original CFAB forecast system implementation.
Duration/start date: 2 weeks beginning week 16 of PoP

• Ensure DFO-only river forecast systems are produced periodically.


Ensure combined CFAB-DFO River forecast systems are also produced periodically
Response: A stand-along DFO/RiverWatch advection-based hydrologic forecast will be daily
produced (essentially, operationalizing the system shown in hindcasts from Feyera et al. 2013,
but using the expanded RiverWatch sites proposed by this consultancy), and optimal
combination of both the CFAB and DFO systems will be combined together, again, utilizing
similar quantile-regression algorithms as discussed in Tasks 3, 4, and 5 above.
Duration/start date: 5 weeks beginning week 18 of PoP
Reported Milestones: Web-based plots of these forecasts will be made available to the World
Bank for viewing at the completion of this task.

Task 7: Transforming forecasts into informative visualizations and Dissemination


Selection No. 1172405
Technical Proposal
Section D
• Couple the flood discharge estimates, on a selected site and reach basis, to the MODIS
observed inundation, to produce online libraries of measured or predicted discharge
and matching floodplain inundation maps.
Response: To provide the linkage between the DFO MODIS imagery and the CFAB ensemble
discharge forecasts, our first step will be to run the CFAB model over the historical period
that the archive corresponds to, associating a modeled discharge to an archived image of a
given reach (note that this process will have to be carried out on a reach-by-reach basis, to the
spatial extent of the image itself). In realtime mode, each CFAB forecast ensemble (for each
river reach) will search over its own historical model discharges, find the dates for which the
“nearest neighbors” occurred, and then pair the ensemble with the DFO imagery occurring on
the same date.
Duration/start date: 4 weeks beginning week 7 of PoP

• Devise satellite-based microwave sensor altimetry virtual data points to collect river
measurement sites to monitor changes in river water heights to compute river flow for
validations.
Response: The work will be undertaken by Dr. Charon Birkett. Due to finanancial and time
limiations, we view this task as a somewhat separate “stand-alone” project, laying the
foundation for potential further consultancies to assimilate these measurements into the CFAB
model for improved hydrologic prediction. As such the timeline for this will be set by the sub
contractor (in consultation with Hopson), with a final report due six months after the
beginning of the period of performance. More detailed tasking is provided below:
Ø Investigation of radar altimetry data sets over the Ganges/Brahmaputra Rivers and
over a selection of lakes/reservoirs in India. Data will be Jason-2/OSTM GDR
(primarily) and SARAL GDR (secondary), i.e. archival (not operational) data and the
number of river channel and lake/reservoir crossings to be investigated will be as time
(FTE) permits;
Ø Creation and validation of altimetric time series utilizing in situ gauge data sets;
Ø Summary of potential virtual stations and lake/reservoirs for operational monitoring;
Ø Deliverables will include i) (static) water-level variation products in both graphical
and ascii text format, ii) Google imagery displaying satellite ground track locations,
iii) a summary report which will include details of instrument performance, data set
limitations, and recommendations for further analysis and transition to operations.
Duration/start date: 4 weeks beginning week 7 of PoP

• Establish real-time data links between forecast model outputs and the River Watch site
displays, such that predicted discharge are provided as well as the present discharge
values.
Response: Real-time data links are essential if the CFAB ensemble forecasts are to be timely
linked to imagery, showing where potential inundation areas are forecasted to occur, and if
these areas are to have timely forewarning.
Duration/start date: 3 weeks beginning week 20 of PoP
Selection No. 1172405
Technical Proposal
Section D
•Create graphical displays providing: (i) public access to both predicted and present
river discharge status of a relatively dense network of measurement points, and (ii) to
estimated and forecasted inundation maps.
Response: Through the NCAR-DFO partnership, the results of the collaborative work will be
incorporated and published on the DFO web site, which is currently a well-used international
portal for the dissemination of both automated discharge measurements and flood inundation
mapping products. Through this website, as well as a mirrored NCAR site, we will
disseminate project results to World Bank staff and national and subnational level
practitioners (with the prior approval of World Bank staff), as well as via interactive
engagements such as trainings and workshops.

Forecast dissemination will also occur at different levels. Since the probabilistic (ensemble)
forecasts can be presented in different ways, during the dissemination and input workshops,
we will work with end-users and prospective end users to determine the most useful ways to
present “best estimate” and uncertainty flood forecast information. The Dartmouth Flood
Observatory (DFO) River Discharge Measurements web site is a model format on which we
can build.1 “River Watch 2” is an existing automated processor supported by NASA Earth
Science Research and Applications Programs. In support of our proposed work, the
University of Colorado is making available extensive space on River Watch 2 for South Asia-
focused displays to provide a prototype portal for both satellite-based present status
information and the model-based discharge prediction and flood warning information.
Duration/start date: 3 weeks beginning week 22 of PoP
Reported Milestones: The displays discussed above will be made available to the World Bank
for viewing at the completion of each sub-task.

Task 8: Training and Reporting


• Conduct workshops on current technologies and future developments and operational
management of full system;
Response: There are two workshops called for in this RFP: one 5 months after the beginning
of the period of performance (PoP), and one 6 months after PoP. Because of the complexity
and range of the technologies being employed in this consultancy, we propose the first
workshop be devoted to training strictly on the technologies themselves, with the 2nd
workshop devoted to presentations on the forecast outputs, displayed information and visuals,
technical findings on forecast skill, and steps forward for potential future consultancies. The
first workshop will rely on material Team Leader Hopson has presented at 3 previous week-
long workshops in India covering long-lead flood forecasting (Patna, Bihar, August 2009;
National Water Academy, Pune, June 2012; National Institute of Hydrology, Roorkee,
November 2013), but also include more extensive discussions on RiverWatch, DFO, and
satellite altimetry technologies.
Duration/start date: 2 weeks beginning week 20 (1st week prep, workshop given week 21) of
PoP (1st week prep, workshop given week 21); and 2 weeks beginning week 25 of PoP (1st
week prep, workshop given week 26)

1
See: http://floodobservatory.colorado.edu/CriticalAreas/DischargeAccess.html
Selection No. 1172405
Technical Proposal
Section D
Reported Milestones: This reporting milestone is self-evident.

• Deliver final report and prepare peer-reviewed papers for a joint publication;
Response: The final report will summarizing our findings with respect to the implementation
of the CFAB system for subcatchments within India, utilization of DFO RiverWatch for
forecast skill enhancements and inundation imagery displays, and the potential benefits of
utilizing satellite altimetrysatellite pre, and will be due 6 months after the beginning of the
PoP.
Duration/start date: 4 weeks beginning week 22 of PoP
Reported Milestones: This reporting milestone is self-evident.

• First draft of peer-reviewed journal article on research findings completed.


Response: In addition to the delivery of the final report, a peer-reviewed journal article is also
called for, with rough draft due 9 months after the PoP beginning. The overall PoP duration of
12 months allows additional time for paper revisions to be completed and payment of
publishing fees. Finally, results will also be disseminated via traditional, academic-conference
presentations (with prior consent of the World Bank).
Duration/start date: 13 weeks beginning week 27 of PoP
Reported Milestones: This reporting milestone is self-evident.

Deliverables and Schedule


The expected deliverables and their schedule are provided as below:

Deliverables Description Timing


(months
after
signing)
Inception Report Brief approach, methodology, detailed work plan, 1st
Description of models to be used, overall data month
compilation for Tasks 1 and 2.
Satellite estimates, NWP Technical document describing initial results from 4th
ensembles, and Multi-model setting up of hydrological model and calibration; month
compilation report outline of assumptions and challenges faced
(Task3-5).
Completion of flood Completion of the major tasks involved in 5th
forecasting for the Ganges producing sub-basin river forecasts. month
and Brahmaputra at sub-
basin levels
Training on technologies Also by the 5th month, a training will be conducted 5th
developed as part (in Delhi or at a smaller center) to provide month
hydrologists and users an overview of Long Lead
Flood Forecasting technologies being implemented
as part of this consultancy
Final Report and training -- Draft report on flood forecast development and 6th
sub-basins level flood forecast results as described month
Selection No. 1172405
Technical Proposal
Section D
in overall tasks, including results from setting up
of hydrological model and calibration; outline of
assumptions and challenges faced (Task3-5).
-- Training on the overall system will also be
provided in India
finalization of 1st draft of First draft of peer-reviewed journal article on 9th
peer-review paper research findings completed. month

Relation to Prior Research – leveraging previous resources


This assignment will leverage technologies developed under previous similar work at NCAR
and the DFO, some of which include: US National Science Foundation base funding (Hopson:
Advanced Study Program postdoc, 2006-8); USAID Grant USAID/OFDA AOT-A-00-00-
00262-00 (2000–03) (PI Peter Webster; Hopson) that provided Bangladesh with operational
forecasts of severe flooding at 1- to 10-day lead-times from 2003-2008; US Army Test and
Evaluation Command (ATEC) funding (Knievel and Hopson), which has developed effective
ensemble probability calibration tools; NASA funding (Brakenridge (PI)) through a research
feasibility project designed to define the pathway for sustainable implementation of a flood
mapping processor provided to the DFO; and the US Bureau of Reclamation funding, where
developed GIS tools for mapping numerical weather prediction outputs to river catchment
areal domains would be utilized.

Organization  and  Staffing  

In this sub section we describe the structure and composition of our team, listing the main
disciplines of the assignment, the key expert responsible, and proposed technical and support
staff. We refer to the Senior Personnel provided below, as well as the Objectives and Tasking
discussed above. Further details are referred to Section B – Consultants’s Experience, Annex
1 listing of Team composition, Task assignments, and Level of Effort, and Annex 2 which
provides a listing of the Curriculum Vitae (CV) of these key personnel.

Senior Personnel
NCAR
Hopson, Thomas. Team Leader / Prominent Researcher / Scientist: National Center for
Atmospheric Research
Hacker, Joshua. Prominent Researcher / Scientist: National Center for Atmospheric Research
Yates, David. Prominent Researcher / Scientist: National Center for Atmospheric Research
Wilhelmi, Olga. GIS program team leader / GIS specialist: National Center for Atmospheric
Research
Dumont, Arnaud. Analyst / Software engineer: National Center for Atmospheric Research
Knievel, Jason. Researcher / Scientist: National Center for Atmospheric Research
Sub Consultant - Dartmouth Flood Observatory
Brakenridge, G. Robert. Collaborator / Consultant / Prominent Researcher / Director
Dartmouth Flood Observatory: Dartmouth Flood Observatory, University of Colorado
Selection No. 1172405
Technical Proposal
Section D
Sub Consultant – University of Maryland
Birkett, Charon. Collaborator / Consultant / Prominent Researcher / Scientist: University of
Maryland
Sub Consultant
Islam, A.K.M. Saiful. Collaborator: Bangladesh University of Engineering and Technology

Team leadership for this consultancy requires knowledge of the underlying science and
primary technologies required to reach the objectives and complete the tasking, along with
management capability, and teaching experience and technology transfer for conducting
effective workshops on developed technologies (the latter called out in the RFP as required
outputs). Team Leader Hopson has over 15 years experience both with the science and
technologies required to reach the objectives and tasking, having been the primary architect of
the CFAB river flow forecasting sytem (Hopson and Webster 2010) specified in Objective (i),
as well as the engineer of the operational system implementing of the modeling components,
which provided Bangladesh with operational forecasts of severe flooding at 1- to 10-day lead-
times from 2003-2008 (in 2009 these technologies were transferred to South Asian partners
for sustainable operations). He recently also conducted three World-Bank funded training
workshops in India to engineers and scientists working on long-lead river forecasting; as such
he will be responsible for Task 8 above.

Expertise with satellite precipitation estimates is also required for this consultancy (Task 3).
Hopson also has experience in operationally implementing and ingesting satellite
precipitation estimates from NOAA (“CMORPH”) and NASA (“TRMM 3B42”) into the
CFAB system. With colleagues Prof Mekonnen Gebremichael (UCLA) and Dr. Feyera Aga
Hirpa (EU Joint Research Council) investigated multiple satellite precipitation products for
hydrologic applications for East Africa (Hirpa et al. 2010). As such, Hopson (with assistance
from a post-doc) will be responsible for Task 3 above.

The consultancy also requires experience with operational ensemble forecast systems. In
addition to the CFAB system, Hopson has expertise working with ensemble weather forecasts
from multiple centers (provided via the Thorpex-Tigge project, see Section B.1.2), providing
the World Health Organization and African National Health Agencies with operational
ensemble humidity forecasts over the meningitis belt of Africa to inform meningitis
transmission models about optimal allocation of vaccine, which is important for Task 4 for
which he will be responsible for. Both Dr. Jason Knievel (ATEC Team Leader, Section
B.1.6) and Hopson (supporting scientist) have experience providing multi-model forecasts to
the US Army with operational ensemble weather forecasts across the U.S. via the ATEC
project, and also Hopson via the multi-hydrologic model of the CFAB system. As such,
Knievel will provide supervision on the operational aspects of this consultancy, and both
Hopson and Knievel will be responsible for generating the multimodel forecasts given in Task
5.

Implementation of improved river routing for the CFAB model is also called out for in the
consultancy. Dr. David Yates will provide supervision for this tasking, given his expertise in
implementing improved routing technology in the official released version of the NCAR
WRF-Hydro modeling system, as called out for in Task 5.
Selection No. 1172405
Technical Proposal
Section D

Via Objective (ii) and Task 6, the consultancy also requires expertise with generating
remotely-sensed river width variations (as surrogates for discharge variations), for which
DFO Director Brakenridge has extensive knowledge (Brakenridge et al. 2005, 2007). The
consultancy also requires (Objective (ii) and Task 6) knowledge of utilizing these
measurements for flood foreasting purposes. Related to this, Hopson and Brakenridge tested
the potential of remotely-sensed river widths derived from microwave imagery data to
forecast Ganges and Brahmaputra river floods downstream of the imagery locations (Hirpa et
al 2012), and will be responsible for this tasking. Dr. Hacker is renowned expert in data
assimiliation, and he will provide supervision for the incorporation of DFO measurements
into the CFAB modeling system, as called out for in Task 6.

With respect to Objective (iii) and Task 7, Brakenridge at the DFO has archived past
microwave and optical sensor imagery of actual inundation extent, that could be linked to
CFAB ensemble river discharge forecasts by selecting images with past similar analogues of
river flow compared to current forecasts. Analyst Dumont has expertise in generating
effective web-based visualizations, and will also be utilized for generating displays of these
products. Both Brakenridge, Hopson, and Dumont will be responsible for this tasking

With respect to Objective (iv), Collaborator Dr. Charon Birkett will be responsible for
providing the analysis and reporting required to meet this objective (in consultation with
Team Leader Hopson). Dr. Birkett is the Principal Investigator of the G-REALM program,
the Global Reservoir and Lake Monitor, which is a NASA/USDA funded program that ingests
raw altimetric data sets and delivers water-level variations products for the world’s largest
lakes and reservoirs. The technique of deriving lake water-level variations is similar for rivers,
floodplains and wetland regions, and the G-REALM system would be used to output products
for identified river channel crossings to test the potential of satellite radar altimetry data for
monitoring river heights for operational flood forecasting purposes.

GIS and visualization capabilities are also critical for consultancy technology developments
and outputs (see Task 1 in particular). NCAR GIS Program team leader Olga Wilhelmi’s
team has extensive expertise in delineating river catchments, in developed GIS tools for
mapping numerical weather prediction outputs to river catchment areal domains (subtask of
Task 1), and visualization of GIS-derived outputs (see Section B.1.10 and B.1.11). She and
her team will be responsible for Tasks 1 and assist with Task 7.

Analyst Dumont also has a proven track record of scripted automated downloads and
archiving of data, and he and Hopson (with assistance from a postdoc) will be responsible for
Task 2. As well, expertise in processing and archiving of the multimodel ensemble weather
forecasts (Thorpex-TIGGE) is important, and here we will utilize the computer engineering
expertise of NCAR’s Douglass Schuster working with NCAR’s Computational Information
Systems Laboratory (CISL) (Doug was responsible for maintaining NCAR’s Thorpex-TIGGE
archive from 2007-2015), which is important for Task 4.

Computational demands in generating operational production of forecasts provides its own


challenges. For this, we will utilize the system administration expertise of NCAR software
Selection No. 1172405
Technical Proposal
Section D
engineer John Exby, as well as a devoted computer server (purchased through previous
project funding): Dell PowerEdge R820 with an 8-core CPU configured with 128GB system
RAM and an 8TB RAID.

Finally, with our Collaborator A.K.M. Saiful Islam, Professor at the Bangladesh University
of Engineering and Technology (BUET) and the Institute of Water and Flood Management
(IWFM), with his collaborations at the Bangladesh Flood Forecasting and Warning Center
(FFWC) and the Bangladesh Meteorological Department (BMD), we have the organizational
capacity to disseminate the technical and skill improvements derived from this project for
institutions in Bangladesh as well (called out for in the RFP, for “the need for continuity of
downstream work”).

Coordination of Group Effort


We will have project team meetings twice a month and more frequently as needed, such as in
preparing for workshops via Skype or conference call that will include key researchers on the
project. There will be more frequent sub-team meetings as appropriate.

Collaborations and Partnerships


We view the organization of our undertaking as an interconnected collaborative partnership,
beginning with the collaborative partnership which is our interdisciplinary research team at
the NCAR, the DFO (Brakenridge), and the University of Maryland (Birkett). The next scale
of partnership involves our international partners, including organizations engaged in
generating forecasts, such as the Flood Forecasting and Warning Center of Bangladesh
(Islam, BUET and FFWC). These partners are currently engaged with regional and global
institutions in a manner that will allow other regions at risk of similar flood hazards to
capitalize on improved flood forecasting. Finally, the next scale will be the partnership of
those who engage in the project’s two workshops—including technical experts (scientists and
engineers), decision makers and policymakers, with potential of directly reaching vulnerable
area residents. As such, our work is intended to reach beyond our initial technical institutional
partnerships, to create a dense network of expanded connections among people attempting to
mitigate the flood threat across different scales throughout the area.

References
Arulampalam, M. S., Maskell, S., Gordon, N., & Clapp, T. (2002). A tutorial on particle
filters for online nonlinear/non-Gaussian Bayesian Tracking. IEEE Transactions on
Signal Processing, 50, 174-188.
Brakenridge, G. R., Nghiem, S. V., Anderson, E., & Chien, S. (2005). Space-based
measurement of river runoff. EOS, Transactions of the American Geophysical Union,
86(19), 185-188.
Brakenridge, G. R., Nghiem, S. V., Anderson, E., & Mic, R. (2007). Orbital microwave
measurement of river discharge and ice status. Water Resources Research, 43,
W04405. doi: 10.1029/2006WR005238
CEGIS. (2006). Sustainable end-to-end climate/flood forecast application through pilot
projects showing measurable improvements. CEGIS Base Line Report, 78 pp.
Selection No. 1172405
Technical Proposal
Section D
Clark, M. P., Rupp, D. E., Woods, R. A., Zheng, X., Ibbitt, R. P., Slater, A. G., Schmidt, J.,
Uddstrom, M. J. (2008). Hydrocological data assimilation with the ensemble Kalman
filter: Use of streamflow observations to update states in a distributed hydrological
model. Advanced Water Resource, 31, 1309-1324.
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 Sensing of
Environment, 131, 140-151. doi: DOI:10.1016/j.rse.2012.11.013
Hopson, T. M., & Webster, P. J. (2010). A 1-10-Day ensemble forecasting scheme for the
major river basins of Bangladesh: Forecasting severe floods of 2003-07. Journal of
Hydrology, 11, 618-641.
Huffman, G. J., R. F. Adler, S. Curtis, D. T. Bolvin, and E. J.Nelkin (2005). Global rainfall
analyses at monthly and 3-hr time scales. Measuring Precipitation from Space:
EURAINSAT and the Future, V. Levizzani, P. Bauer, and J. F. Turk, Eds., Springer,
722 pp.
Huffman, G. J., Bolvin, D. T., Nelkin, E. J., Wolff, D. B., Adler, R. F., Gu, G., . . . Stocker, E.
F. (2007). The TRMM multisatellite precipitation analysis (TMPA): Quasi-global,
multiyear, combined-sensor precipitation estimates at fine scales. Journal of
Hydrometeorology, 8(1), 38-55.
Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie (2004). CMORPH: A method that
produces global precipitation estimates from passive microwave and infrared data at
high spatial and temporal resolution. J. Hydrometeor., 5, 487–503.
Kleuskens, M., Westerhoff, R. S., & Huizinga, J. (2011). Operational flood mapping: A pilot
study in the Mekong Area. Paper presented at the Proceedings of the International
Symposium of Remote Sensing of the Environment, Sydney, Australia.
Lee, H. S., Seo, D. J., Lui, Y., Koren, V., McKee, P., & Corby, R. (2012). Variatinal
assimilation of streamflow into operational ditributed hydologic models: Effect of
spatiotemporal scale adjustment. Hydrology and Earth System Sciences, 16, 2233-
2251. doi: 10.2194/hess-16-2233-2012
Montanari, M., Hostache, R., Matgen, P., Schumann, G., Pfister, L., & Hoffman, L. (2009).
Calibration and sequential updating of a coupled hydrolic-hydrualic model using
remote sensing-derived water stages. Hydrology and Earth System Sciences, 13, 367-
380. doi: 10.5194/hess-13-367-2009
Moradkhani, H., Hsu, K., Gupta, H. V., & Sorooshian, S. (2005). Uncertainty assessment of
hydrologic model states and parameters: Sequential data assimilation using particle
filter. Water Resources Research, 41, W05012. doi: 10.1029/2004WR003604
Seo, D. J., Cajina, L., Corby, R., & Howieson, T. (2009). Automatic state updating for
operational streamflow forecasting via variational data assimilation. Journal of
Hydrology, 367, 255-275.
Song, X. Y., & Lee, S. Y. (2004). Bayesian analysis of two-level nonlinear structural equation
models with continuous and polytomous data. British Journal of Mathematical and
Statistical Psychology, 57, 29-52.
Syvitski, J. P. M., & Brakenridge, G. R. (2013). Causation and avoidance of catastrophic
flooding along the Indus river, Pakistan. GSA Today, 23(1). doi:
10.1130/GSATG1165A.1131
Selection No. 1172405
Technical Proposal
Section D
Webster, P. J. (2013). Improve weather forecasts for the developing world. Nature, 493, 17-
19.
Webster, P. J., Jian, J., Hopson, T. M., Hoyos, C. D., Agudelo, P., Chang, H.-R., . . . Subbiah,
A. R. (2010). Extended-range probabilistic forecasts of Ganges and Brahmaputa
floods in Bangladesh. Bulletin of the American Meteorological Society, 91, 1493-
1514.
Weerts, A. H., & El Serafy, G. Y. H. (2006). Particle filtering and ensemble Kalman filtering
for state updating with hydrological conceptual rainfall-runoff models. Water
Resources Research, 42, W09403. doi: 10.1029/2005WR004093
Westerhoff, R. S., Huizinga, J., Kleuskens, M., Burren, R., & Casey, S. (2010). Operational
satellite-based flood mapping using the Delft-FEWS System. In Proceedings of the
ESA Living Planet Symposium, Bergen, Norway, June 28-July 2, 2010. Retrieved Jan
10, 2013, from http:/kennisonline.deltares.nl/product/22381
Selection No1172405
Technical Proposal
Annex 3

You might also like