The hydrological cycle is receiving increasing attention both as an essential natural resource for humans and ecosystems and as a critical component controlling the earth’s climate system. Better understanding of the water cycle and its interaction with changing climate will require improved monitoring of the various water fluxes and storages in hydrological processes. River discharge is a unique component reflecting an integrated hydrological signal over larger regions. Existing in situ monitoring solutions to monitor discharge are often considered too expensive and the difficulties in data sharing are viewed as insurmountable obstacles, which has led to growing interest in finding an alternative. This paper argues that in situ monitoring is far less expensive than claimed and the obstacles are not necessarily as insurmountable as often stated and a conscious effort to revitalize in situ monitoring will be needed. This paper demonstrates that there is no substitute for in situ discharge monitoring, but there should be a synergy between in situ monitoring and remote sensing since they are truly complementary. This paper primarily focuses on river discharge, but the conclusions are relevant for a host of other earth observations (particularly water quality) that would greatly benefit from a reconsidered balance between in situ and remote sensing observations.
Growing concerns about water availability for both humans and healthy ecosystems are driving increased attention to reliable assessments of freshwater resources (Falkenmark and Biswas 1995; Falkenmark 1998; Gleick 1993; Postel et al. 1996). The hydrological cycle is important not only from a water management perspective, but water also plays a pivotal role in controlling climate. Accurate information about the amount of water circulating in the water cycle is vital for improving our understanding of the climate system (Rodda 1998).
Among the various components of the hydrological cycle, river discharge plays a special role. River systems are not only a major source of freshwater in many parts of the world, but are a very accurately measured element of the water and energy cycle where adequate in situ monitoring is in place (Hagemann and Dümenil 1998; Gutowski et al. 1997; Grabs et al. 1996). While the majority of the large global watersheds are reasonably well instrumented considering the number of operating stations and the catchment area they cover (Hannah et al. 2010), the lack of data sharing and the resulting gaps (primarily in developing countries), along with the steady decline in observational networks (Vörösmarty et al. 2002; Shiklomanov et al. 2002; Lanfear and Hirsch 1999), has led to growing interest in alternative solutions to discharge monitoring (Alsdorf and Lettenmaier 2003; Alsdorf et al. 2007; NRC 2007).
The purpose of our paper is to clarify the significance of discharge measurements and to explain some of the key characteristics that make river discharge well suited for in situ monitoring but a difficult target for remote sensing. We provide a brief inventory of existing monitoring capabilities and describe the challenges in maintaining up-to-date global discharge data archives. We contrast these challenges with several remote sensing solutions presented in the scientific literature over the last decade. We recognize that our list of remote sensing examples is incomplete, but we think it is adequate to highlight the challenges that remote sensing solutions will face.
While our paper concentrates on river discharge monitoring, its implications to the state of global in situ monitoring can be viewed much more broadly. The decline of observing stations as seen in other ground monitoring networks [e.g., precipitation (Stokstad 1999) and air temperature and water quality (Zhulidov et al. 2000)] is alarming particularly in an era of elevated interest in climate change. For example, the Russian water quality monitoring network of about 4000 sites in the mid-1970s covered approximately 1200 water systems (Zhulidov et al. 2000). By the late 1990s the monitoring network had declined to between 1699 and 1928 sites, depending on the information source. River discharge is not the only earth system component that needs to be measured on the ground. Remote sensing and in situ monitoring are largely complementary and that can be demonstrated for discharge. Places that are challenging for in situ monitoring are often ideal for remote sensing and vice versa. Remote sensing in conjunction with in situ observations, in a data assimilation framework, will provide the fullest picture of the state of our planet (Neal et al. 2009).
River discharge is the most accurately measured component of the water cycle (Hagemann and Dümenil 1998; Gutowski et al. 1997; Grabs et al. 1996) using traditional in situ instruments. River discharge, which is the integral of flow velocities within the river’s cross section, is rarely measured directly (only during field campaigns) and the key challenge of discharge monitoring is to relate the observable flow conditions (e.g., flow heights or width) to actual discharge (Bjerklie et al. 2004). When discharge is monitored on the ground, flow height is the measured flow property. The stream-gauging program of the U.S. Geological Survey (USGS) strives to produce discharge data, where 95% of the daily discharge values are within 10% of the true discharge (Novak 1985). While hydrometeorological agencies aim at the 10% accuracy the discharge error can be as large as 40+% (Di Baldassarre and Montanari 2009). The high error is often quoted by those who advocate remote sensing alternatives for discharge monitoring as a shortcoming for in situ monitoring, so it is worthwhile to emphasize that such errors are the exceptions and not the norm.
Discharge is a unique signal in the sense that it aggregates excess water originating upstream, resulting in an observation that is representative not only of the location where the measurement was taken but of a larger area. Because the upstream area increases gradually between major confluences of larger tributaries, discharge along river channels changes less for any given time than the sudden jumps exhibited where rivers merge. The spatial steadiness and persistence of discharge along river reaches is beneficial for both in situ and remote sensing because it allows the selection of monitoring sites according to where the measurement is more convenient. For instance, in situ surveys for calibrating discharge gauges are often carried out at different sections of a river than where the actual gauges are located. Alsdorf et al. (2007) argued that the traditional one-dimensional, point-based view is appropriate in well-defined channels, but it is insufficient in complex riverine environments where river channels interact with surrounding floodplains and wetlands. While instrumenting such complex terrains with any sensor (ground based or remote sensing) is indeed difficult (potentially impossible), these complex river reaches are typically bounded by more confined river channels upstream and downstream, where the in- and outflows can be measured more easily.
The gradual change in discharge along river reaches also dictates the preference to monitor incoming tributaries rather than discharge along the main stem before and after confluences, where the discharge increment between the upstream and downstream location could be well within the desired 10% value of computed discharge relative to true discharge.
Unlike spatial variations in discharge along river channels, temporal variations are much more dynamic. Particularly, smaller rivers can respond to intensive precipitation events very rapidly. Flash floods could cause severe flooding in a matter of hours that are likely to remain unnoticed by satellites with infrequent revisit periods. Modern dataloggers virtually eliminate the limitations on water level observational frequency, allowing for the full recording of the flood-wave propagation at a particular river section. In contrast, the satellite platforms under consideration for discharge monitoring tend to have a weekly or biweekly ground-track repeat cycle. While overlapping swaths can allow satellite observations more frequently than the ground-track revisit time, the resulting compromise comes as a trade-off in off-nadir viewing angle that could severely limit the quality of the remote sensing data. Figure 1 shows the degradation of annual discharge estimates as a function of undersampling compared to the discharge calculated from the full daily records. Clearly, proper discharge monitoring will require sampling frequencies well above the revisiting frequency of typical low-orbit satellites. Alternatively, satellite systems to monitor discharge will need to operate in a constellation that would be prohibitively expensive. A better approach is the use of general-purpose satellites that are not necessarily optimized for monitoring rivers but fill in the temporal gaps.
Perhaps the biggest challenge in discharge monitoring is calibration of the observable characteristics of river flow. In the case of in situ monitoring, river stage height is the observed flow property that is normally measured at a few-centimeter (USGS aims for 3 mm) accuracy (Hirsch and Costa 2004). This can be contrasted with the anticipated 50 cm or more error in individual radar echo of the wide swath altimeter envisioned for the Surface Water on Ocean Topography (SWOT) mission (Andreadis et al. 2007; Alsdorf et al. 2007). The large error in individual radar measurements will be reduced to a few centimeters’ accuracy when averaged over larger water surface. The SWOT science requirement specification aims at <10-cm accuracy that could drop to a few centimeters over large rivers (≥100 m). Smaller rivers have less surface area and will be monitored less accurately. However, this is just the opposite of the actual monitoring needs for rivers. Figure 2 shows the degradation of discharge estimate from stage heights (for 750 USGS discharge gauges) as a function of height measurement accuracy. We have to emphasize that an accurate stage height observation is more critical for smaller rivers with less discharge than for large rivers, since a few-centimeter change in a small river will reflect a large change in discharge, while the same change in stage height on a large river corresponds to a much smaller change in discharge in relative terms.
Measuring discharge itself is a tedious and difficult task that can be done only via intensive field surveys that measure the cross-sectional profile and the flow velocity distribution (Rantz et al. 1982). The most expensive elements of discharge monitoring on the ground are indeed these field surveys that are used to establish rating curves for relating river stage height to discharge. The discharge rating curves often take the shape of a power function because of the tendency of river channels to also follow a power function shape (w = ahb, where w is river width and h is river depth) (Dingman 2007). Stage height appears to be the most robust tracking variable, compared to surface flow velocity or river width (Bjerklie et al. 2003), which is due to the general tendency of river channels to form a U-shaped cross section (where exponent b < 1) that increases in depth more rapidly than in width. The calibration of the discharge rating curve, which is the key to accurate discharge monitoring, has to be repeated on a regular basis to account for the stability of the river channel, vegetation growth, and ice formation. Remote sensing solutions without similarly rigorous calibration will have compromised measurement accuracy.
The integrated nature of river discharge makes in situ measurement very efficient for monitoring large areas with relatively few discharge gauging stations. For instance, the 70 largest river basins in the world (the dominant majority already monitored) capture over 50% of the river discharge to oceans; therefore’ operating discharge gauges at the river mouth of these basins would allow us to monitor a significant portion of the water cycle. On the other hand, expanding the coverage is increasingly difficult because disproportionally more gauges are needed to increase the monitoring coverage over smaller and smaller basins. For instance, the top 2500 river basins—based on catchment area—would barely capture 85% of global discharge to the oceans. Based on the 30′ resolution, simulated gridded network (STN30) (Vörösmarty et al. 2000), the number of basins exceeding 150 000 km2 catchment area that Gravity Recovery and Climate Experiment (GRACE) satellites can evaluate (Rodell and Famiglietti 1999; Syed et al. 2010) is somewhere between 100 and 110, and <500 gauges can be evenly spaced (with roughly the same interstation area) over those basins exceeding multiplies of the GRACE minimum basin threshold. In other words, GRACE’s river basin monitoring capabilities are at best the equivalent of operating 500 discharge gauges. Operating 5000 gauges distributed evenly over the 98 × 106 km2 actively contributing continental lands with ~20 000-km2 interstation area (Fekete et al. 2002) would be the equivalent of having discharge observations at a 1.5° × 1.5° (longitude × latitude) spatial resolution, which is around the resolution of state-of-the-art global circulation models.
A recent paper by Hannah et al. (2010) showed that the discharge monitoring infrastructure needed for global monitoring is largely in place, particularly in developed countries (Fig. 3). Large gaps in Saharan Africa, the Arab Peninsula, and Tibet are clearly due to the lack of rivers. The monitoring gaps in India, China, Indonesia, and South America are likely attributable to local data policies rather than the lack of infrastructure. Remaining gaps in Africa, central Asia, and at high latitudes are more likely due to insufficient monitoring infrastructure. Data from a large portion of this existing monitoring network is already available from individual hydrometeorological agencies. For instance, the U.S. Geological Survey distributes discharge records for over 9230 gauges in the United States on a real-time basis and their archive database contains data from over 25 300 gauges. Both the real-time and the archive discharge records are accessible through the National Water Information System (http://waterdata.usgs.gov/nwis). Environment Canada distributes discharge data in near–real time (http://www.wateroffice.ec.gc.ca/index_e.html) for over 1700 of the 2847 gauges in operation (Hannah et al. 2010). Data for an additional 5607 inactive stations are stored in the national Hydroclimatological Data Retrieval Program (HYDAT) database (Hannah et al. 2010). The European Water Archive (EWA) consists of data from over 3800 stations in 29 countries. While EWA is not as accessible at the moment as its North American counterparts, the tendency is that these databases will eventually be opened for wider audience.
The Global Runoff Data Centre (GRDC) has made tremendous progress since it was established to collect and archive discharge data. Figure 4 shows the increase in GRDC data holdings between 2001 and 2011 as more and more stations were added to the data archive. A prominent feature in the number of stations by year is the sharp decline seen after the late ‘80s. While the magnitude of the decline mostly reflects the time lag between observation and entering the data into the GRDC database, it is also caused by lack of reporting and discontinuing operation. Forty-eight countries out of the 156 for which data are available in the GRDC have not shared data updates after 1984. Lanfear and Hirsch (1999) reported a steady decline in USGS gauges, while Shiklomanov et al. (2002) noted widespread similar trends in other parts of the world such as the former Soviet Union and Canada. Apparently, the number of monitoring stations peaked in the 1980s (Rodda 1998) as a response to growing concerns about population growth and its impact on the environment (Hannah et al. 2010; Rodda 1998), but as focus shifted toward climate change the commitment to continued operation of in situ monitoring networks diminished. Traditional in situ discharge monitoring is considered expensive despite being on par with the cost of operating remote sensing sensors. USGS spends $10,000 per station, which is matched by the individual states to yield a total of $20,000 per station in operation costs. Environment Canada’s operating costs are very similar to those in the United States. Operating 5000 stations globally would yield a $100 million (U.S. dollars) annual budget, which is comparable to the $450 million estimated price tag for the 3 years of SWOT or GRACE-II missions (NRC 2007).
In 2005, the GRDC proposed a baseline network of river discharge stations on the world’s largest rivers near their outlets to the oceans. These stations, a subset of gauging stations around the world operated by the respective countries, collectively form a new Global Climate Observing System (GCOS)/Global Terrestrial Observing System (GTOS) baseline network: the Global Terrestrial Network for River Discharge (GTN-R) (WMO 2010). Data from these approximately 400 stations would capture about 70% of the freshwater fluxes to the oceans. All of these stations have reported at some stage in the past and most of them are operating today. This network is now being adjusted in consultation with national hydrological services, but funds have not yet been secured for proper management of the GTN-R network and for the required refurbishment and upgrading of the infrastructure and data transmission and management. The 2010 Update of the Implementation Plan for the Global Observing System for Climate in Support of the UNFCCC (WMO 2010) recognizes the lack of global river monitoring and states that an annual budget not exceeding $10 million would be sufficient for the operation of GTN-R. A similar, one-time investment will be needed to assess national needs for river gauges in support of impact assessment and adaptation and to consider the adequacy of those networks. Additional funds would then be needed to cover the operational costs of the stations.
Complementing discharge measurements with water quality monitoring is more expensive, both in terms of equipment and human resources. For example, the Texas Commission on Environmental Quality (the state’s regulatory agency) alone had a $480.7 million operating budget for the 2007 fiscal year (http://www.tceq.state.tx.us), which is nearly half the entire gross domestic product of the Republic of Guinea-Bissau in western Africa (https://www.cia.gov/library/publications/the-world-factbook/geos/pu.html). There is a need to develop effective, low-cost, and field-robust technologies to obtain these data, while at the same time supporting the development and implementation of community-based monitoring programs such as the miniSASS in South Africa1—a low technology, scientifically reliable, and robust technique to monitor water quality in rivers and streams (Graham et al. 2004).
The steady decline of globally available discharge information and the difficulties in making such data widely available are broadly recognized (Alsdorf and Lettenmaier 2003; Alsdorf et al. 2007; Vörösmarty et al. 2002) and scientists are increasingly interested to respond by finding alternative solutions (Alsdorf and Lettenmaier 2003; Alsdorf et al. 2003, 2007). Satellite remote sensing is seen as an attractive alternative because of the near-global and contiguous coverage it can provide. On the other hand, the insufficient revisiting frequency of low-orbit satellites is widely recognized. Incorporating satellite measurements into data assimilation schemes (Alsdorf et al. 2007) using hydrological models driven by climate forcing is viewed as one of the possible solutions. Such an approach inevitably reduces the independence of the derived river discharge from the hydrological model calculation and, as a consequence, limits the utility of the discharge observation as calibration/validation.
Andreadis et al. (2007) presented an example of the envisioned data assimilation suitable for processing the swath altimetry. They applied the LISFLOOD-FP raster-based hydrodynamics model (Bates and DeRoo 2000) for a section of the Ohio River near Martins Ferry, Ohio (with a 60 000-km2 watershed area), driven by boundary water fluxes from the Variable Infiltration Capacity (VIC) model (Maurer et al. 2002) in an identical twin experiment. Truth set was developed by running LISFLOOD-FP with VIC forcing. “Synthetic” observations were generated by corrupting both the forcing data and the water depth. Additionally, they introduced a 25% negative bias to the VIC-simulated upstream discharge forcing. The “observed” water depth time series was generated by the Instrument Simulator from the Jet Propulsion Laboratory. The authors remained a bit vague about how their study made the leap from water level (potentially measured by the altimeter sensor) to water depth, which requires some assumptions about riverbed geometry and the displacement of the river bottom relative to the surrounding topography. Applying the ensemble Kalman filter (Evensen 1994) data assimilation of water depth in an 84-day simulation experiment, they were able to reduce the 23.2% RMSE in discharge introduced by the distortion of precipitation input to an RMSE of 10%, 12.1%, and 16.9% considering 8-, 16-, or 32-day satellite overpasses.
This experiment is not very informative in terms of the potential accuracy of the proposed data assimilation scheme in real applications, but it certainly highlights some of the main challenges. The accuracy largely hinges on the upstream discharge boundary driven by a rainfall runoff model. Although hydrological models driven by reliable precipitation have demonstrated capabilities to close the annual water budgets within 5% accuracy at continental scales (Fekete et al. 2002; Vörösmarty et al. 1998a), their performance could be significantly worse in smaller regions (Döll and Siebert 2002; Döll et al. 2003; Fekete et al. 2002). As a consequence, the initial boundary discharge might easily have errors much higher than 23.2%. Furthermore, a more appropriate metric would have been to report the reduction of the mean absolute error over the 84-day simulation period, since the 5%–10% accuracy of traditional discharge monitoring is normally achieved on individual observations. The actual accuracy of the measurements averaged over a longer period of time is likely to be significantly higher. One can conclude from this experiment that the accuracy of discharge estimates from data assimilation will not come close to traditional discharge monitoring. What may be more interesting is to use observed hydrographs in an assimilative capacity with models to perhaps correct model bias on the fly and incorporate SWOT measurements in real time to extend the recorded data in space.
Given the anticipated sensor capabilities of the swath altimeter envisioned for the SWOT mission, it is likely that the performance of the data assimilation scheme will deteriorate even further for smaller rivers. If one targeted rivers similar to Ohio at Martins Ferry,2 with 590 mm yr−1 annual runoff (about twice the 290 mm yr−1 global mean; Fekete et al. 2002) as a minimum target size, then operating the proposed data assimilation scheme would be the equivalent of monitoring basins twice as large (120 000 km2) on average, which is comparable to the minimum basin size for GRACE. SWOT specification aims to monitor rivers down to 100 m wide (which is about one-fourth of Ohio at Martins Ferry, based on a crude assessment using Google Maps; https://maps.google.com/).
An alternative approach for estimating continental discharge has been proposed by Syed et al. (2009) based on storage change measurements estimated from gravity anomalies using GRACE combined with atmospheric water budgets from two global reanalysis products [the National Centers for Environmental Protection (NCEP)–National Center for Atmospheric Research (NCAR) (Kalnay et al. 1996; Kistler et al. 2001) and the European Centre for Medium-Range Forecasts (ECMWF) operational forecast analysis (http://www.ecmwf.int/products)]. The rather coarse monthly temporal resolution of their study was dictated by GRACE’s limitations to depict gravity anomalies at higher temporal frequencies. As a consequence, the storage changes in various compartments of the hydrological cycle—which tend to approach zero for longer temporal intervals—plays a minor role in establishing the water budget relative to the atmospheric water budget from the reanalysis products. The uncertainties in using the reanalysis products can be characterized by the differences in the annual discharge to oceans based on NCEP–NCAR (32 851 ± 1744 km3 yr−1) versus ECMWF (28 590 ± 1685 km3 yr−1). Nevertheless, the authors produce a combined estimate by averaging NCEP–NCAR and ECMWF, yielding a 30 354 ± 1212 km3 yr−1 annual discharge for the years 2003–05. This estimate is significantly lower than 36 000–40 000 km3 yr−1 from several other authors (Baumgartner and Reichel 1975; L’vovich et al. 1990; Shiklomanov 1998; Fekete et al. 2002) and lacks the convincing improvement over well-established water balance calculations (Döll and Siebert 2002; Haddeland et al. 2006; Vörösmarty et al. 1998b) based on observed precipitation and climate forcing, which are increasingly able to depict human alteration of the hydrological cycle (Döll and Siebert 2002; Haddeland et al. 2006; Hanasaki et al. 2006; Wisser et al. 2008, 2010a,b).
Syed et al. (2010) presented a similar analysis based on a water budget over oceans using precipitation from the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) (Xie and Arkin 1997) and the Global Precipitation Climatology Project (GPCP) (Adler et al. 2003) evaporation estimates from the Special Sensor Microwave Imager (SSM/I) (Wentz et al. 2007), the Objectively Analyzed Air–Sea Fluxes (OAFlux) project (http://oaflux.whoi.edu) (Yu and Weller 2007) from the Woods Hole Oceanographic Institute (WHOI), and the Hamburg Ocean Atmosphere Parameters and Fluxes from Satellite data (HOAPS) (Anderson et al. 2010). These were combined with measured changes in global ocean water mass based on altimeter [Ocean Topography Experiment (TOPEX)/Poseidon and Jason-1 satellite] observations of global mean sea level (GMSL) in conjunction with ocean temperature and salinity data (Nerem et al. 2006). While this analysis did not have the capability to provide any details about the discharge distribution over the continental landmass, overall it seemed to provide more realistic estimate (36 055 km3 yr−1) for the 1994–2006 period. Interestingly, the reported discharge values for the 2003–05 period are between 33 500 and 40 000 km3 yr−1 (averaging over 36 000 km3 yr−1), which was significantly more than the NCEP–NCAR/ECMWF/GRACE-based study from the same team discussed previously.
Given the significant challenges of using remote sensing to monitor discharge, one has to conclude that accurate monitoring of discharge can be achieved only on the ground. While in situ discharge measurements are very accurate and economically competitive compared to remote sensing solutions, they will never give complete global coverage. The best solution for ungauged basins (and the refinement of in situ discharge monitoring networks) remains the improvement of traditional hydrological modeling combined with data assimilation techniques utilizing a wide array of remote sensing data in conjunction with in situ observation. These types of approaches will take advantage of the remote sensing capabilities afforded by satellites such as GRACE or altimetry platforms and provide increased value to the existing suite of satellites, models, assimilation techniques, and measurements. Remote sensing already plays a pivotal role in providing improved land characterization, topography, precipitation forcing, soil moisture estimates, etc. Most of these observations are either difficult or impossible on the ground and remote sensing offers the only possible means for regular monitoring.
Remote sensing and in situ monitoring are very much complementary. In situ monitoring is more suited for smaller rivers (<50 m wide) while field surveys are more labor intensive for large rivers. In contrast, large rivers that have slower dynamics are ideal targets for remote sensing. In situ monitoring works best in well-confined river channels (Alsdorf et al. 2007) while remote sensing works better over braided rivers (Smith 1997). In situ discharge measurements are often difficult and even dangerous during high flows and flood events. Remote sensing has demonstrated value during floods.
Remote sensing undeniably has changed and improved dramatically our ability to monitor planet Earth. Numerous and diverse observations (land use, topography, cloud cover, etc.) would not be possible without satellite observations but satellites do have their limitations in certain applications and must be complemented with adequate in situ monitoring. Some components of the earth system need to be monitored on the ground. Our paper focused on river discharge, but a number of other variables (e.g., various inland water quality parameters, socioeconomic data, etc.) also require data collection on the ground. From an in situ monitoring perspective, the last 20 years can be viewed as two lost decades despite an increasing awareness about climate change. While our paper focused on the state of discharge monitoring, a similar decline in meteorological stations is apparent (particularly at high altitude and latitudes), threatening the integrity of scientific analysis. There is no doubt that remote sensing provides tremendous value in assessing hydrological state and forcing variables; however, the community needs to balance the in situ estimates of these values with nontraditional techniques such as remote sensing. Fluxes in the water cycle (e.g., flow, precipitation rates, and evaporation) are notoriously difficult to measure with remote sensing alone and a number of scientists have warned about the consequences of the decline in monitoring networks.
The main concern regarding in situ monitoring is the lack of cooperation between data providers and users at the international level. It is true that small underfunded organizations like the GRDC, Global Precipitation Climatology Centre (GPCC), International Groundwater Resources Assessment Centre (IGRAC), and Global Environment Monitoring System Water Programme (GEMS/Water), charged with collecting, archiving, and disseminating data (under the constrains of international agreements, while relying entirely on volunteer contributions from national agencies), have had limited success in assembling comprehensive monitoring records. The perceived incentive to deny data sharing is less prominent than the lack of incentives to share. GRDC, GEMS/Water, and other data centers are likely to have had much better success if they were in the position to offer financial assistance to data providers to repackage and ship their data on a regular basis.
Critics argue that such international cooperation cannot be achieved because of the conflicting interests among various countries. One has to wonder if this is indeed the case and if such a minimalist goal as sharing river discharge or other key in situ monitoring data is deemed to be impossible, how any serious commitment to assess and combat climate change is expected to succeed. Undeniably, establishing and maintaining an adequate global observing network as a combination of remote and in situ monitoring sensors should have been on the table long before negotiating the necessary actions it takes to mitigate climate change. In the information age, when inexpensive small sensors can be easily linked up to telemetering networks, operating ground observing networks should have received more attention (Hirsch and Costa 2004).
The degree of success in building in situ monitoring networks that transcend political boundaries should serve as a litmus test for the feasibility to foster international cooperation for larger and more complex goals. Even if our generation fails to act on reducing carbon emissions, establishing a detailed record of how climate changes are advancing is our minimum obligation to future generations. Relying on remote sensing observations only as demonstrated by the first 5 years of the Global Earth Observing System of Systems (GEOS) implementation, when robust in situ solutions are well established and cost effective, will be the modern-day equivalent of hunting for past climate signals in proxy data. The scientific community should therefore advocate more strongly for the sustained investment in earth system monitoring—both in situ and remote sensing.
We thank Dr. Harry Lins, Dr. Paul Bates, and the anonymous reviewers for their thorough and thoughtful comments. Particularly, the constructive criticisms from Dr. Bates helped to reshape our paper. We hope that our paper represents a common ground that all the reviewers and the scientific community will agree with.
The South African Scoring System (miniSASS) is a low-cost biomonitoring method to assess river water quality based on a few aquatic invertebrate groupings as surrogates for a more extensive suite of taxa.
USGS gauge 03111534 with 1194 m3 s−1 mean annual discharge from a 63 765-km2 catchment area according to the National Water Information System.