In hydrology, modeling streamflow remains a challenging task due to the limited availability of basin characteristics information such as soil geology and geomorphology. These characteristics may be noisy due to measurement errors or may be missing altogether. To overcome this challenge, we propose a knowledge-guided, probabilistic inverse modeling method for recovering physical characteristics from streamflow and weather data, which are more readily available. We compare our framework with state-of-the-art inverse models for estimating river basin characteristics. We also show that these estimates offer improvement in streamflow modeling as opposed to using the original basin characteristic values. Our inverse model offers 3\% improvement in R$^2$ for the inverse model (basin characteristic estimation) and 6\% for the forward model (streamflow prediction). Our framework also offers improved explainability since it can quantify uncertainty in both the inverse and the forward model. Uncertainty quantification plays a pivotal role in improving the explainability of machine learning models by providing additional insights into the reliability and limitations of model predictions. In our analysis, we assess the quality of the uncertainty estimates. Compared to baseline uncertainty quantification methods, our framework offers 10\% improvement in the dispersion of epistemic uncertainty and 13\% improvement in coverage rate. This information can help stakeholders understand the level of uncertainty associated with the predictions and provide a more comprehensive view of the potential outcomes.

Reliable prediction of river floods in the first 72 hours can reduce harm because emergency agencies have sufficient time to prepare and deploy for help at the scene. Such river flood prediction models already exist and perform relatively well in most high-income countries. But, due to the limited availability of data, these models are lacking in low-income countries. Here, we offer the first global river flood prediction framework based on the newly published Caravan dataset. Our framework aims to serve as a benchmark for future global river flood prediction research. To support generalizability claims we include custom data evaluation splits. Further, we propose and evaluate a novel two-path LSTM architecture (2P-LSTM) against three baseline models. Finally, we evaluate the generated models on different locations in Africa and Asia that were not part of the Caravan dataset.

![Figure][1] Most of Victoria and New South Wales in Australia experienced the Millennium Drought. River levels dropped, and reservoirs were at a fraction of their capacity. In 2010, the Lake Eildon reservoir in Victoria was at 29% capacity. PHOTO: ASHLEY COOPER/CONSTRUCTION PHOTOGRAPHY/AVALON/GETTY IMAGES Ecological systems can switch into qualitatively different states after small perturbations ([ 1 ][2]). Climate change and anthropic activities are dominant drivers of such ecological shifts, which affect the ability of these systems to recover from future disturbance ([ 2 ][3]). Such finite resilience in complex and dynamic natural systems has been predicted and documented ([ 3 ][4]). However, whether abrupt transitions have occurred or will occur in the water cycle is an unsolved problem ([ 4 ][5]). The possibility for river basins to achieve thresholds at which tiny perturbations alter their state and lead to chronic water scarcity or excessive water bears substantial implications for the sustainable use of water resources in extreme climate conditions ([ 5 ][6]). On page 745 of this issue, Peterson et al. ([ 6 ][7]) demonstrate that river systems exhibit a finite resilience to perturbations and that climate may indeed drive river basins between alternative states. To assess multistability in hydrological systems, Peterson et al. statistically investigated annual and seasonal precipitation and runoff records of more than 160 river basins in Victoria, Australia, before, during, and after the Millennium Drought (2001–2009), the worst drought ever recorded for southeast Australia ([ 7 ][8]). River basins are networks of channels and ridges that convey water to a common outlet. The studied river basins had neither major reservoirs nor water extractions, and runoff changes were not correlated to remotely sensed changes of Earth's surface cover. Seven years after the meteorological drought ended (that is, when dry weather ceased to dominate), more than a third of the river basins had not returned to previous runoff conditions, and most of them showed no signs of recovering soon. Conversely, basins that did shift back to a normal runoff state had shown warning signs in the 3 years before recovery. Unlike hydrological models that assume an infinite resilience of river systems, these findings imply that river basins do not always recover from droughts and that returning to predrought runoff conditions is not just a matter of time. Rather, the onset of positive feedbacks in hydrological systems, such as increased transpiration in nonrecovered basins, may amplify the impacts of climate change and, alarmingly, reduce the probability of switching back to a normal runoff state. Among basins switching back to predrought conditions, runoff recovery was not simply explained by increased soil moisture or groundwater recharge, as estimated by exploring basin wetness during shifts between multiple states. Hydrological recovery may instead have been driven by complex interactions of vegetation and soil hydraulics, which demands further research. Although climate shifts are predicted to alter water processes over widespread regions ([ 8 ][9]), the existence of multiple equilibrium points in the water cycle has focused on small subcontinental scales. Positive feedbacks between precipitation and soil moisture have suggested the emergence of preferential states in soil moisture dynamics and the persistence of droughts in the state of Illinois in the United States ([ 9 ][10]). Bistability was also demonstrated in coupled salt-vegetation dynamics at the basin scale, where gradual changes of salt concentration in irrigation water led to abrupt and irreversible changes in land productivity ([ 10 ][11]). Bistability was also identified in urban settings, where more frequent floods, droughts, population growth, and competition for resources may challenge the resilience of water supply systems and lead worldwide cities toward poverty ([ 11 ][12]). The conclusion of Peterson et al. —that river basins may irreversibly enter a persistent water scarcity state after severe meteorological droughts—challenges the comforting assumption that water systems naturally tend to absorb disturbances. This emphasizes the necessity to change the way global water processes are conceptualized. Assessing the existence of multiple equilibrium points in a natural water system and accounting for feedback mechanisms will likely help improve understanding of the hydrological response of river basins in drying regions and the development of appropriate water management adaptation strategies to climate change. However, the intrinsic heterogeneity of water systems and the stochastic nature of meteorological forcing (such as temperature, precipitation, and wind) raise questions about the possibility that all river basins have multiple equilibrium points ([ 12 ][13]). Further research is needed to identify the critical tipping points at which river basins switch to alternative states as well as the early warning signals of change in their resilience. This requires long-term monitoring of river basins that not only experience climate disturbances up and beyond critical thresholds but also do not undergo notable land-use change. Unfortunately, screening dynamic water processes is an actual challenge, and substantial limitations in current monitoring systems hamper systematic basin-scale hydrological investigations. Observations through standard equipment are still inadequate to fully grasp natural processes ([ 13 ][14]). They offer limited spatial coverage and generally involve high maintenance costs, which hinder implementation in many parts of the world, such as remote environments and developing countries ([ 14 ][15]). Although classical hydrological observations have been consistently decreasing worldwide since the 1980s, the recent use of innovative and unintended technology (such as low-cost electronics and participatory sensing) is providing opportunities for sensing the water cycle at even higher spatiotemporal resolutions ([ 15 ][16]). Dense and accurate monitoring of the hydrological response of and within river basins coupled with advanced data interpretation are necessary steps toward disentangling the complex interactions between basin morphological and functional attributes and hydroclimatic drivers in a changing world. 1. [↵][17]1. M. Scheffer, 2. S. Carpenter, 3. J. A. Foley, 4. C. Folke, 5. B. Walker , Nature 413, 591 (2001). [OpenUrl][18][CrossRef][19][PubMed][20][Web of Science][21] 2. [↵][22]1. B. Walker, 2. C. S. Holling, 3. S. R. Carpenter, 4. A. Kinzig , Ecol. Soc. 9, (2004). 3. [↵][23]1. R. M. May , Nature 269, 471 (1977). [OpenUrl][24][CrossRef][25][Web of Science][26] 4. [↵][27]1. G. Blöschl et al ., Hydrol. Sci. J. 64, 1141 (2019). [OpenUrl][28] 5. [↵][29]1. T. R. Ault , Science 368, 256 (2020). [OpenUrl][30][Abstract/FREE Full Text][31] 6. [↵][32]1. T. J. Peterson, 2. M. Saft, 3. M. C. Peel, 4. A. John , Science 372, 745 (2021). [OpenUrl][33][Abstract/FREE Full Text][34] 7. [↵][35]1. A. I. J. M. van Dijk et al ., Water Resour. Res. 49, 1040 (2013). [OpenUrl][36] 8. [↵][37]1. J. S. Caplan et al ., Sci. Adv. 5, eaau6635 (2019). [OpenUrl][38][FREE Full Text][39] 9. [↵][40]1. P. D'Odorico, 2. A. Porporato , Proc. Natl. Acad. Sci. U.S.A. 101, 8848 (2004). [OpenUrl][41][Abstract/FREE Full Text][42] 10. [↵][43]1. C. W. Runyan, 2. P. D'Odorico , Water Resour. Res. 46, W11561 (2010). [OpenUrl][44] 11. [↵][45]1. E. H. Krueger et al ., Earths Future 7, 1167 (2019). [OpenUrl][46] 12. [↵][47]1. T. J. Peterson, 2. A. W. Western , Water Resour. Res. 50, 2993 (2014). [OpenUrl][48] 13. [↵][49]1. A. K. Mishra, 2. P. Coulibaly , Rev. Geophys. 47, RG2001 (2009). [OpenUrl][50] 14. [↵][51]1. N. van de Giesen, 2. R. Hut, 3. J. Selker , Wiley Interdiscip. Rev. Water 1, 341 (2014). [OpenUrl][52] 15. [↵][53]1. F. Tauro et al ., Hydrol. Sci. J. 63, 169 (2018). [OpenUrl][54] [1]: pending:yes [2]: #ref-1 [3]: #ref-2 [4]: #ref-3 [5]: #ref-4 [6]: #ref-5 [7]: #ref-6 [8]: #ref-7 [9]: #ref-8 [10]: #ref-9 [11]: #ref-10 [12]: #ref-11 [13]: #ref-12 [14]: #ref-13 [15]: #ref-14 [16]: #ref-15 [17]: #xref-ref-1-1 "View reference 1 in text" [18]: {openurl}?query=rft.jtitle%253DNature%26rft.stitle%253DNature%26rft.aulast%253DScheffer%26rft.auinit1%253DM.%26rft.volume%253D413%26rft.issue%253D6856%26rft.spage%253D591%26rft.epage%253D596%26rft.atitle%253DCatastrophic%2Bshifts%2Bin%2Becosystems.%26rft_id%253Dinfo%253Adoi%252F10.1038%252F35098000%26rft_id%253Dinfo%253Apmid%252F11595939%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [19]: /lookup/external-ref?access_num=10.1038/35098000&link_type=DOI [20]: /lookup/external-ref?access_num=11595939&link_type=MED&atom=%2Fsci%2F372%2F6543%2F680.atom [21]: /lookup/external-ref?access_num=000171485700039&link_type=ISI [22]: #xref-ref-2-1 "View reference 2 in text" [23]: #xref-ref-3-1 "View reference 3 in text" [24]: {openurl}?query=rft.jtitle%253DNature%26rft.volume%253D269%26rft.spage%253D471%26rft_id%253Dinfo%253Adoi%252F10.1038%252F269471a0%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [25]: /lookup/external-ref?access_num=10.1038/269471a0&link_type=DOI [26]: /lookup/external-ref?access_num=A1977DW65000019&link_type=ISI [27]: #xref-ref-4-1 "View reference 4 in text" [28]: {openurl}?query=rft.jtitle%253DHydrol.%2BSci.%2BJ.%26rft.volume%253D64%26rft.spage%253D1141%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [29]: #xref-ref-5-1 "View reference 5 in text" [30]: {openurl}?query=rft.jtitle%253DScience%26rft.stitle%253DScience%26rft.aulast%253DAult%26rft.auinit1%253DT.%2BR.%26rft.volume%253D368%26rft.issue%253D6488%26rft.spage%253D256%26rft.epage%253D260%26rft.atitle%253DOn%2Bthe%2Bessentials%2Bof%2Bdrought%2Bin%2Ba%2Bchanging%2Bclimate%26rft_id%253Dinfo%253Adoi%252F10.1126%252Fscience.aaz5492%26rft_id%253Dinfo%253Apmid%252F32299944%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [31]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEyOiIzNjgvNjQ4OC8yNTYiO3M6NDoiYXRvbSI7czoyMjoiL3NjaS8zNzIvNjU0My82ODAuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9 [32]: #xref-ref-6-1 "View reference 6 in text" [33]: {openurl}?query=rft.jtitle%253DScience%26rft.stitle%253DScience%26rft.aulast%253DPeterson%26rft.auinit1%253DT.%2BJ.%26rft.volume%253D372%26rft.issue%253D6543%26rft.spage%253D745%26rft.epage%253D749%26rft.atitle%253DWatersheds%2Bmay%2Bnot%2Brecover%2Bfrom%2Bdrought%26rft_id%253Dinfo%253Adoi%252F10.1126%252Fscience.abd5085%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [34]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEyOiIzNzIvNjU0My83NDUiO3M6NDoiYXRvbSI7czoyMjoiL3NjaS8zNzIvNjU0My82ODAuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9 [35]: #xref-ref-7-1 "View reference 7 in text" [36]: {openurl}?query=rft.jtitle%253DWater%2BResour.%2BRes.%26rft.volume%253D49%26rft.spage%253D1040%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [37]: #xref-ref-8-1 "View reference 8 in text" [38]: {openurl}?query=rft.jtitle%253DScience%2BAdvances%26rft.stitle%253DSci%2BAdv%26rft.volume%253D5%26rft.issue%253D9%26rft.spage%253Deaau6635%26rft.epage%253Deaau6635%26rft.atitle%253DDecadal-scale%2Bshifts%2Bin%2Bsoil%2Bhydraulic%2Bproperties%2Bas%2Binduced%2Bby%2Baltered%2Bprecipitation%26rft_id%253Dinfo%253Adoi%252F10.1126%252Fsciadv.aau6635%26rft_id%253Dinfo%253Apmid%252F31535016%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [39]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6MzoiUERGIjtzOjExOiJqb3VybmFsQ29kZSI7czo4OiJhZHZhbmNlcyI7czo1OiJyZXNpZCI7czoxMjoiNS85L2VhYXU2NjM1IjtzOjQ6ImF0b20iO3M6MjI6Ii9zY2kvMzcyLzY1NDMvNjgwLmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ== [40]: #xref-ref-9-1 "View reference 9 in text" [41]: {openurl}?query=rft.jtitle%253DProc.%2BNatl.%2BAcad.%2BSci.%2BU.S.A.%26rft_id%253Dinfo%253Adoi%252F10.1073%252Fpnas.0401428101%26rft_id%253Dinfo%253Apmid%252F15184676%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [42]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NDoicG5hcyI7czo1OiJyZXNpZCI7czoxMToiMTAxLzI0Lzg4NDgiO3M6NDoiYXRvbSI7czoyMjoiL3NjaS8zNzIvNjU0My82ODAuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9 [43]: #xref-ref-10-1 "View reference 10 in text" [44]: {openurl}?query=rft.jtitle%253DWater%2BResour.%2BRes.%26rft.volume%253D46%26rft.spage%253DW11561%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [45]: #xref-ref-11-1 "View reference 11 in text" [46]: {openurl}?query=rft.jtitle%253DEarths%2BFuture%26rft.volume%253D7%26rft.spage%253D1167%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [47]: #xref-ref-12-1 "View reference 12 in text" [48]: {openurl}?query=rft.jtitle%253DWater%2BResour.%2BRes.%26rft.volume%253D50%26rft.spage%253D2993%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [49]: #xref-ref-13-1 "View reference 13 in text" [50]: {openurl}?query=rft.jtitle%253DRev.%2BGeophys.%26rft.volume%253D47%26rft.spage%253D2001RG%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [51]: #xref-ref-14-1 "View reference 14 in text" [52]: {openurl}?query=rft.jtitle%253DWiley%2BInterdiscip.%2BRev.%2BWater%26rft.volume%253D1%26rft.spage%253D341%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [53]: #xref-ref-15-1 "View reference 15 in text" [54]: {openurl}?query=rft.jtitle%253DHydrol.%2BSci.%2BJ.%26rft.volume%253D63%26rft.spage%253D169%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx

India, the world's second most populated nation, sees more than 20% of the global flood-related fatalities each year as overrun riverbanks sweep tens of thousands of homes with them. Two years ago, Google volunteered to help. In 2018, the company began its flood forecasting pilot initiative in Patna -- the capital of the Indian state of Bihar, which has historically been the most flood-prone region in the nation with over 100 fatalities each year -- to provide accurate real-time flood forecasting information to people in the region. The company's AI model analyzes historical flood data gleaned from several river basins in different parts of the world to make accurate prediction for any river basin. For this project, Google has not worked in isolation. Instead, it has collaborated with India's Central Water Commission, Israel Institute of Technology, and Bar-Ilan University.

Flooding is a destructive and dangerous hazard and climate change appears to be increasing the frequency of catastrophic flooding events around the world. Physics-based flood models are costly to calibrate and are rarely generalizable across different river basins, as model outputs are sensitive to site-specific parameters and human-regulated infrastructure. In contrast, statistical models implicitly account for such factors through the data on which they are trained. Such models trained primarily from remotely-sensed Earth observation data could reduce the need for extensive in-situ measurements. In this work, we develop generalizable, multi-basin models of river flooding susceptibility using geographically-distributed data from the USGS stream gauge network.

Flooding is a destructive and dangerous hazard and climate change appears to be increasing the frequency of catastrophic flooding events around the world. Physics-based flood models are costly to calibrate and are rarely generalizable across different river basins, as model outputs are sensitive to site-specific parameters and human-regulated infrastructure. In contrast, statistical models implicitly account for such factors through the data on which they are trained. Such models trained primarily from remotely-sensed Earth observation data could reduce the need for extensive in-situ measurements. In this work, we develop generalizable, multi-basin models of river flooding susceptibility using geographically-distributed data from the USGS stream gauge network. Machine learning models are trained in a supervised framework to predict two measures of flood susceptibility from a mix of river basin attributes, impervious surface cover information derived from satellite imagery, and historical records of rainfall and stream height. We report prediction performance of multiple models using precision-recall curves, and compare with performance of naive baselines. This work on multi-basin flood prediction represents a step in the direction of making flood prediction accessible to all at-risk communities.