Global water resources knowledge gaps

Abstract: The stationarity of hydrological systems is dead. Has our hydrological/water resources knowledge well transformed to address this change? By using publications indexed in the Web of Science database since 1900, we aim to investigate the global development of water resources knowledge at river basin scale from a system science perspective. Water resources knowledge development in a river basin is defined as a complex system involving the co-evolutionary dynamics of scientific disciplines and management issues. It is found that: 1) centralised and legacy-inclined water resources knowledge structures 10 dominated major river basins in the world; 2) links between water resources knowledge structure and the management issues it addressed are increasingly homogenised; and 3) cross-disciplinary collaborations have remained largely unchanged and collaborations with social sciences have been very limited. In conclusions, the stationarity of the water resources knowledge system persists. A shift of water resources knowledge development to cope with the rapidly changing hydrological systems and associated management issues is urgently needed. 15


Data analysis
The change point detection method by the "changepoint" package in R (https://cran.r-110 project.org/web/packages/changepoint/index.html) was used to identify different temporal periods of development in the Water Resource discipline. It calculated the abrupt changes in mean and variances of the total number of articles published in time and decomposed by different management issues from a spatial perspective. The change point detection rather than the trend detection method (e.g. Mann-Kendall test) was used because it focuses on identifying the abrupt changes of publications in time (Jaiswal et al., 2015). The interconnections between different management issues were represented by summing the 115 weightings from all issue networks for the 95 river basins.
For each of the 95 river basins, the knowledge structure was firstly measured by the degree and closeness of the Water Resources in the overall disciplinary networks. The agglomerative hierarchical clustering (AHC) using the "factoextra" package from R (https://cran.r-project.org/web/packages/factoextra/index.html) was conducted to cluster river basins based https://doi.org/10.5194/hess-2021-137 Preprint. Discussion started: 24 March 2021 c Author(s) 2021. CC BY 4.0 License. on these structural features. The clustering was performed based on the Euclidean distances and the Ward's 120 agglomerative criterion (Murtagh & Legendre, 2014) for the normalised degree and closeness values (between 0 and 1) of Water Resources. The number of suitable groups was determined to maximise the sum of square errors between different groups and minimise the errors within groups (refer to Appendix B for more details). Based on the clustering results, we identify the thresholds dividing between different knowledge groups. Rivers with < 0.25 for both normalised degree and closeness values were considered to have minimal impact in the knowledge system, and thus categorized into the structure 125 with "limited development". For rivers with, on average > 0.5 normalised degree values were considered the central rivers contributing to knowledge development in water resources, thus identified as "centralised development"; whereas rivers with, on average > 0.5 normalised closeness values were considered to have developed confined knowledge and thus "isolated development". For the remaining rivers with average normalised degree > closeness values were considered to be more reliant on central river basins' knowledge development and thus "legacy-inclined development"; while those rivers with normalised 130 closeness < degree tended to develop regional-specific, confined knowledge and "innovative-inclined development".
Then, groups of river basins with different knowledge structures were mapped with the corresponding management issues of focus to discuss the relationship between them. Finally, the collaborations of the Water Resources with other disciplines were analysed with the links between Water Resources and the other disciplines in the WoS. The top 10 most collaborated discipline were also mapped with the corresponding issues of focus to identify the collaborations that should be strengthened. 135

Temporal and spatial distribution of the Water Resources publications by management issues
The earliest publication year on Water Resources for the 95 mostly published river basins was in 1970, and accumulated to over 10,000 publications in total in 2017. As shown in Figure 1a, three development periods were identified. Before 1993, the number of articles published annually were very limited (fewer than 250 publications), with the top 3 issues of focus being 140 water pollution and treatment (64 publications), surface water and groundwater management (48), and sedimentation and erosion (28). Annual publications began to take off since the 1990s, with an increment of about 10 times. During this second period (1994 -2005), water pollution and treatment (626) continued to be the focus of studies in these rivers, followed by surface water and groundwater management (388) and water policy (257). Articles on Water Resources continued to increase during the most recent period (2006 -2017), although the rate has slowed down (3 times from the previous period). Surface 145 water and groundwater management (1610) and water pollution and treatment (1228) continued to be the centres of focus, with studies on water policy, climate variability and change, sedimentation and erosion, and ecological degradation and restoration gaining momentums (each with over 550 publications).
The strongest connections among these issues occurred between water pollution and treatment and ecological degradation and restoration for all 3 periods (Figure 1b), which indicate the strong impacts of water quality on the ecosystems. This was 150 followed by the respective connections of these two issues with surface water and groundwater management, forming a triangular issue of focus related to different biophysical aspects in the eco-hydrological systems. There were also increasing connections among the erosion and sedimentation, water pollution and treatment, and ecological degradation and restoration, forming the second key triangular focus related to hydro-morphology. A new triangular focus among climate variability and change, water pollution and treatment, and surface water and groundwater management appeared since the 1994 -2005 period, 155 indicating the emergence of integrated research issues around climatic impacts. In the following 2006 -2017 period, increasing interests were observed among water policy, water pollution and treatment, and ecological degradation and restoration, the third major triangular focus. The development of water policy was also extended to the climate variability and change, and the erosion and sedimentation issues during this period. More interests in water policy on other issues indicate the shift of water https://doi.org/10.5194/hess-2021-137 Preprint. Discussion started: 24 March 2021 c Author(s) 2021. CC BY 4.0 License. resource management to water demand management and integrated governance. Meanwhile, linking all other issues, it is 160 demonstrated that knowledge in surface water and groundwater is central to water resources management.
The spatial distributions of publications on Water Resources indicated great diversity among river basins around the globe ( Figure 1c). River basins located in the North America and southeast Asia had most publications in all time. The top five are the Yellow River, the Yangtze River, the Mississippi River, the Murray-Darling Basin, and the Colorado River. Different research preferences were also demonstrated in different river basins. For example, the Yellow River and the Yangtze River 165 received the most focus on surface water and groundwater management, whereas research on the Mississippi River focused on water pollution and treatment and the Murray-Darling River on water policy. Among all river basins, over 38% received most publications on water pollution and treatment issue, 53% of which were located in North America. Over 28% rivers focused on the surface water and groundwater management issue, 46% of which were located in Asia. River basins in Europe (54%) were also most focused on water pollution and treatment. Among the limited number of rivers in South America, Africa, 170 Antarctica and Oceania identified (12% of 95 rivers), the focus was on surface water and groundwater management, ecological degradation and restoration, and water policy.

Relationship between researched management issues and structural development of the Water Resources discipline 215
As shown in Figure 3, water pollution was the most prominent theme for rivers with the knowledge structure as "limited development", which comprised over 60% of the 95 rivers during the 1970 -1993 period. The rivers with the centralised knowledge structure tended to focus on the issues related to surface water and groundwater management (e.g. the Colorado

Cross-disciplinary collaborations of the Water Resources discipline 240
Collaborations of the Water Resources discipline with other disciplines remained relatively stable in time (Figure 4a).
Environmental Science remained as the top one in all 3 periods although the percentage in total publications reduced from 23% to 19%. It belonged to the category of life science and biomedicine, which also comprised over 50% of all collaborations

Discussion and conclusion 275
Using the academic publications indexed in the WoS database between 1900 -2017 as the data source, this study investigated the development of water resources knowledge at the global river basins scale, key findings and the identified knowledge gaps on the Water Resources discipline are summarised below.
Firstly, investigation of "new" river basin phenomena emerging from management issues should be encouraged. This is evidenced by: 1) There appeared increasing inter-connections among four major triangular issues of focus: water pollution -280 ecological degradation -surface water and groundwater management, sedimentation and erosion -water pollution -ecological degradation, and water pollution -ecological degradation -water policy. It implies that integrated research has been playing the dominant role (Figure 1). 2) The energy, sediment and water fluxes influenced by human activities and the impacts of climate change in most river basins have been well recognised (Ayllón et al., 2018). It may be the time to ask what the "new" river basin phenomena emerging from the interactions of current management issues are for preventive action to avoid future 285 water crisis.
Secondly, the spatial diversity of Water Resources research should be encouraged and homogenization of the structure-issue links should be avoided. This is evidenced by that the development of water resources knowledge is more and more centralized in several major river basins such as the Mississippi River, the Great Lakes, and the Yangtze River, with increasing river basins that have legacy-inclined knowledge structure and are under the influences of these centralized river basins (Figure 2). While 290 the domination of centralised or legacy-inclined river basins could have strong diffusive power, there is risk of knowledge redundancy that could hinder innovation and potential waste of research resources (Makri et al., 2010). This also implies that the diversity as an important feature of a good system structure is missing (Allen et al., 2016). Furthermore, the tendency of centralised or legacy-inclined river basins focusing on the same management issues (Figure 3) further increases the risk of homogenization and reducing the resilience (capacity) of water resources knowledge systems to address problems arising from 295 the abruptly changing environment (i.e. the new phenomena).
Thirdly, there is urgent need to strengthen collaborations with social sciences. This is evidenced by collaborations of with the Water Resources discipline have been overwhelmingly dominated by Environmental sciences, Multidisciplinary geosciences, Marine & Freshwater biology, Ecology, and Environmental Engineering, whereas collaborations with social sciences remained nearly null in all time (Figure 4). The earth system enters the Anthropocene when human impacts on the natural environment 300 are prominent (Lewis & Maslin, 2015;Steffen et al., 2011), wherein societal processes have been an indispensable part of understanding changes in hydrological systems. Without the collaboration from social sciences, the Water Resources discipline will not have sufficient capacity to address water management issue in this era. The effectiveness of those water policies examined with limited involvement of social sciences are in doubt, indirectly augmenting current water governance crisis. In addition, this implies that constant calls on interdisciplinary research in the past decades (e.g. (Caldas et al., 2015;Gleick, 305 2000) and funding agencies' proposals (e.g. NERC in the UK, NSF in the US and ARC in Australia) continue to struggle in enabling cross-disciplinary collaborations (Xu et al., 2015), despite that some new sub-disciplines such as socio-hydrology are developing toward this direction. More knowledge structure driven rather than output-driven research development strategies should be encouraged.
To conclude, the stationarity of the hydrological systems is dead (Milly et al., 2008), but the stationarity of the water resources 310 knowledge system persist. This knowledge system is characterised by a highly centralized and legacy-inclined knowledge structure, homogenized structure-issue links, and unchanging disciplinary collaborations with limited contributions from social sciences. We echo with Sivapalan and Blöschl (2017) that the Water Resources knowledge system should be at the end of euphoria and need a rapid shift after such an extended period of stasis to cope with rapidly changing hydrological systems and associated management issues.

Appendix B 325
The agglomerative hierarchical clustering (AHC) was conducted on R version 4.0.2 (2020-06-22) using the "factoextra" package (https://cran.r-project.org/web/packages/factoextra/index.html). The optimum number of clustering group was determined using the Elbow method, which intends to minimise the total sum of squares , which is calculated as Eq. B.1, where is any structural value, and <= >>>> is the average value in cluster q: