Flood prediction in the future: Recognizing hydrologic issues in anticipation of the Global Precipitation Measurement mission



[1] The planned Global Precipitation Measurement (GPM) mission beckons hydrologists as an opportunity to improve flood prediction capability for medium to large river basins, especially in the underdeveloped world where in situ precipitation gauge networks are sparse. However, before the potential of GPM can be realized, there are a number of hydrologic issues that must be addressed. In particular, we argue that unless there is a shift in paradigm, the conventional assessment frameworks and metrics for estimation of precipitation from satellite sensors will remain inadequate for hydrologic purposes such as flood prediction. We also argue that greater emphasis must be placed on development of hydrologically relevant precipitation estimation algorithms and that this will require involvement of a broader cross section of the hydrologic community.

1. Introduction

[2] The Global Precipitation Measurement (GPM) mission is a constellation of satellites consisting of a fleet of passive microwave (PMW) sensors (Figure 1a), augmented by a single precipitation radar (PR), which is expected to be launched by 2012. It will provide high-resolution global precipitation products (i.e., snow and rainfall) with temporal sampling rates ranging from three to six hours, and spatial resolution of 25–100 km2 [Smith et al., 2006; see also http://gpm.gsfc.nasa.gov]. For the hydrologist, GPM will facilitate a quantum leap over current state-of-the-art satellite precipitation products with coverage over most of the global land areas. The global nature of coherent and more accurate satellite precipitation products anticipated from GPM should offer hydrologists tremendous opportunities to improve flood monitoring in medium to large river basins especially in those (mostly underdeveloped or remote) areas where rainfall is abundant but in situ measurement networks are inadequate. In a recent article, Hossain and Katiyar [2006] discussed the potential implications of GPM to improve flow monitoring across political boundaries for flood prone nations in international river basins (IRBs). Their survey identified the lack of treaties for real-time rainfall data sharing among riparian nations as the main impediment to longer range flood forecasting. They argued that the scientific community needs to prioritize the development of parsimonious assessment frameworks that can efficiently utilize satellite rainfall data and hence rapidly prototype satellite-based forecasting systems in anticipation of GPM. While the hydropolitical dimension of flood prediction that will be afforded by GPM seems reasonably well recognized, it is the hydrologic potential that remains relatively less well understood, and is the focus of this short paper.

Figure 1a.

Constellation of GPM satellites. The larger satellite on the left represents the core with a radar on board, while the rest carry polar orbiting PMW sensors (from http://gpm.gsfc.nasa.gov).

[3] We believe it is important therefore that the hydrologic prediction community, many of whom are unfamiliar with precipitation remote sensing techniques, become more involved so that the community can improve flood monitoring in the vast portions of the globe where in situ networks are sparse (Figure 1b). In this article, we identify key hydrologic issues that we believe should be addressed by the hydrologic community as the launch of GPM approaches. We further argue that conventional assessment frameworks and metrics for characterizing satellite precipitation data are inadequate for assessing the flood monitoring potential of GPM. As a way forward, we present some of the options available to hydrologists and propose possible roadmaps that could lead to better exploitation of GPM data to improve dramatically the state of the art of flood prediction in the future.

Figure 1b.

Global distribution of in situ rainfall gauges showing the sparse and unevenness in the underdeveloped world (from http://www.cpc.noaa.gov).

2. Overview of the Global Precipitation Measurement Mission

[4] The heritage of GPM can be traced to the 1970s when infrared (IR) radiometers on geostationary satellites were first launched to provide high-resolution (in space and time) sampling [Griffith et al., 1978]. While geostationary IR sensors have substantial advantages in that they provide essentially time-continuous observations, a major deficiency is that the quantity being sensed (cloud top temperature), is only indirectly related to precipitation [Huffman et al., 2001]. Subsequently, space-borne passive microwave (PMW) radiometers evolved as a more credible alternative a decade later. PMW sensors work on the principle that naturally emitted radiation in the microwave wavelengths is modulated by the atmospheric hydrometeors. This technique suffers from two drawbacks. First, antenna requirements make geostationary sensing technologically infeasible, and dictate that such sensors must be flown in low Earth orbit. This means that PMW observations at a given point on Earth's surface will be instantaneous, with revisit times typically once or twice a day. Second, the background microwave emissivity over land is highly heterogeneous, thereby making the contribution from the hydrometeors to the total passive signal sensed by PMW sensors difficult to distinguish. For this reason, PMW sensors generally work much better over the oceans, where the background emissivity is much better known. Nevertheless, PMW sensors are still more accurate under most conditions for precipitation estimation over land than their IR counterparts.

[5] In 1997, the Tropical Rainfall Measuring Mission (TRMM), the first space-borne active microwave (AMW) precipitation radar (TRMM PR), was launched [Simpson et al., 1996]. While radar generally is the most accurate remote sensing technique for precipitation estimation, radar technology is expensive, and TRMM PR has limited spatial coverage (at latitudes between about 35 S and 35 N) with a sampling frequency about once per day. Therefore the constellation of PMW sensors (three currently orbiting as part of the Defense Meteorological Satellite Program (DMSP), and a fourth, AMSR-E, flying on board the NASA Aqua research satellite) continue to represent a middle ground between IR sensors and TRMM PR in terms of sampling frequency, accuracy, and global coverage.

[6] GPM is now being planned as a global constellation of low Earth orbiting satellites (some of them existing) carrying various PMW sensors [Smith et al., 2006]. It will essentially be an expansion of the TRMM mission, which would provide near-global coverage of land areas, and would formally incorporate a means of combining precipitation radar with PMW sensors to optimize sampling and retrieval accuracy. The GPM Core satellite will be similar in concept to the TRMM satellite, and will house a dual-frequency precipitation radar of improved accuracy (an advancement over TRMM PR) and a TMI-like PMW sensor (Figure 1a). Through this configuration, GPM aims to provide coherent global precipitation products with temporal resolution ranging from 3 to 6 hours and spatial resolution in the range 25–100 km2 [Smith et al., 2006; see also http://gpm.gsfc.nasa.gov].

3. Key Issues Associated With GPM Data

[7] To evaluate the potential of satellite precipitation data for flood prediction, it is important to understand that hydrologic prediction errors will arise from three major sources (the retrieval algorithm used, sampling uncertainty, and the hydrological modeling system). These error sources are all intimately linked and are difficult to separate. To date, most assessments of satellite precipitation estimation errors have focused primarily on sampling uncertainty which arises from the relatively infrequent intervals at which PMW sensors pass over any point. In most of these assessments, precipitation retrieval algorithm uncertainty is assumed to be a negligible component of the total rainfall error budget [see, e.g., Steiner et al., 2003; Astin, 1997]. While in some cases a focus on sampling errors can provide a basis for estimating a lower limit to runoff prediction errors, we believe that the hydrologic community must provide a more coherent approach to assessing the implications of all major sources of errors in flood prediction. For the space-timescales that are required to represent surface hydrologic variability in river basins, retrievals errors in particular are a critical concern that cannot be ignored.

[8] Even though approaches to addressing hydrologic prediction uncertainty [e.g., Beven and Binley, 1992] are as old as efforts to characterize uncertainty in remote sensing estimates of precipitation [North and Nakamoto, 1989], both efforts have evolved more or less independently. Two satellite rainfall algorithms with similar bias and root mean squared error (RMSE) can have much different error propagation properties when used in hydrologic models [Lee and Anagnostou, 2004]. Thus hydrologists who are interested in exploiting satellite precipitation data are left with inadequate metrics to communicate to the data producers the criteria that would determine the effectiveness of a particular remotely sensed precipitation data source for flood prediction needs.

[9] In order to foster more effective communication between hydrologists and satellite precipitation data producers that we believe is necessary for improved space-based flood monitoring in the future, we suggest the following characteristics for an assessment framework: (1) The framework should function as a filter wherein the hydrological implications of fine-scale components of the satellite precipitation estimation structure can be explicitly determined by coupling it with a hydrological/land surface model. (2) The framework should be modular in design with the capability to allow uncertainty assessment of any satellite rainfall algorithm at various hydrologic scales. (3) The metrics of the framework should be such that their hydrologic implications are physically interpretable by the data producers and thus provide better focus to the development of next generation algorithms in anticipation of GPM.

[10] We do not intend for these criteria to serve as a panacea that will resolve scale-dependent satellite rainfall uncertainties. Rather, our focus is on providing a realistic framework that will help to assess the potential for improving flood monitoring using GPM products.

4. A Path Forward

[11] In the search for effective frameworks and metrics, we believe that hydrologists should build upon existing techniques and gradually modify them to meet the emerging needs of GPM cited above. At a minimum, there are currently two noteworthy approaches that seem to have potential that the hydrologic community could exploit. These are (1) downscaling/disaggregation techniques and (2) benchmark-dependent techniques. It is important for hydrologists to understand the paradigm behind these techniques so that the strengths may be appropriately harnessed toward the development of GPM-compatible metrics and frameworks. Each of these techniques can potentially lead to a successful roadmap for the hydrologist engaged in flood monitoring for ungauged river basins as discussed below.

4.1. Downscaling/Disaggregation Path

[12] The basic approach for downscaling of precipitation is based on the concept of scaling in rainfall, or, relating the properties associated with the rainfall process at one scale to a finer-resolution scale [Margulis and Entekhabi, 2001]. Techniques for generation of high-resolution rainfall from coarse resolution satellite data have perhaps the longest heritage of use in hydrology. One aspect of the disaggregation technique that hydrologists could more effectively address are the implications of assumptions of spatial and temporal covariance structure (which may not always be appropriate at the resolutions downscaled). This is especially important for satellite-based estimates of precipitation that are well known to increase in error complexity at finer scales [Hossain and Anagnostou, 2006]. While there is a need to understand the spatiotemporal resolution to which satellite products can realistically be disaggregated, the downscaling approach is a path that hydrologists could more conveniently utilize given the long history of work already accomplished in this area.

4.2. Benchmark-Dependent Path

[13] This path is based on a ‘benchmark’, i.e., precipitation fields of higher accuracy and resolution that are the closest representation of “true” surface precipitation process (i.e., ground validation), and stochastic space-time formulations to corrupt these benchmark precipitation fields to mimic ensembles of satellite-like estimates. Thus the implications of the role played by specific components of the error structure of satellite precipitation can be assessed in hydrologic prediction and consequently communicated to data producers as feedback for better algorithm development for surface hydrology.

[14] Two specific examples of a benchmark-dependent technique are noteworthy. The first is the Two-Dimensional Satellite Rainfall Error Model (SREM2D [Hossain and Anagnostou, 2006]) that has been developed in response to the limitation of currently used bidimensional error metrics (such as bias and error variance) in distinguishing the hydrologic error propagation in runoff across scales and satellite rainfall algorithms. SREM2D decomposes the complex error structure of satellite rainfall estimates into 9 parameters at the scales that are of interest to the surface hydrologist. A limitation associated with this approach is the issue of regional dependency of error metrics and consequently the global applicability of the framework over regions lacking in benchmark (ground validation) data [Hossain and Anagnostou, 2006]. While this is a legitimate concern, significant work has already been accomplished on global classification of precipitation systems and finding similarities with regions with benchmark (validation) precipitation data [Petersen and Rutledge, 2002]. Hence there exists a possibility that SREM2D metrics may one day be amenable for transfer to ungauged basins based on regime similarity.

[15] The second example of a benchmark-dependent technique is a methodology that derives conditional probability distribution functions of satellite precipitation on a pixel-by-pixel basis from a benchmark [Bellerby and Sun, 2005]. This technique is based on the assumption that uncertainty of satellite precipitation data is frequently difficult to characterize using scalar measures of additive error, given the complexities associated in delineation of rainy and nonrainy areas by passive sensors. Each generated trace of satellite precipitation fields represents an equiprobable realization that is consistent with the original satellite data while containing a random element commensurate with the uncertainty in that field. Although this technique is relatively nascent in comparison with SREM2D or downscaling methods described earlier, it has a major advantage that it is not overly dependent on ground validation data, and can use satellite sensor data of higher physical consistency instead.

5. Closing Remarks

[16] We have identified some key issues for hydrologists eagerly anticipating a new era in flood monitoring with GPM data. We believe that the development of effective assessment frameworks and metrics for satellite precipitation data will largely dictate the utility and continual improvement of GPM products for flood monitoring. Greater emphasis now needs to be placed on the critical evaluation of existing approaches toward the search for frameworks and error metrics that are appropriate for GPM in flood prediction.

[17] It is important for the hydrologic community to understand that in the current budget climate at NASA, the fact that GPM is an approved mission does not necessarily mean that it will fly [Zielinski, 2005]. Several NASA earth science missions have been cancelled recently (most notably for hydrologists the HYDROS soil moisture mission) despite strong recommendations of the National Research Council (NRC) Decadal Review of Earth Science and Applications from Space [NRC, 2005] to the contrary. The support of the hydrologic community will be critical to assure that the same fate does not befall GPM. It has now become clear that NASA is rebalancing its priorities, such as focusing more on manned space exploration [Zielinski, 2006]. This refocusing is causing a good deal of uncertainty regarding NASA's future in earth sciences. We view the evolution of global hydrology as one of the most important trends in hydrology in recent decades [Eagleson, 1986], and the availability of global observations, many of which are only feasible from space, must be a key element of this subdiscipline. For all of these reasons, strong and vocal support of missions like GPM is, we believe, central to the continued health of an area that is increasingly important to hydrology.


[18] The authors wish to sincerely thank the continued support and critique of the Editor while preparing this manuscript. Constructive comments received from three anonymous reviewers are also acknowledged.