Background: Climate-smart agriculture (CSA) addresses the challenge of meeting the growing demand for food, fibre and fuel, despite the changing climate and fewer opportunities for agricultural expansion on additional lands. CSA focuses on contributing to economic development, poverty reduction and food security; maintaining and enhancing the productivity and resilience of natural and agricultural ecosystem functions, thus building natural capital; and reducing trade-offs involved in meeting these goals. Current gaps in knowledge, work within CSA, and agendas for interdisciplinary research and science-based actions identified at the 2013 Global Science Conference on Climate-Smart Agriculture (Davis, CA, USA) are described here within three themes: (1) farm and food systems, (2) landscape and regional issues and (3) institutional and policy aspects. The first two themes comprise crop physiology and genetics, mitigation and adaptation for livestock and agriculture, barriers to adoption of CSA practices, climate risk management and energy and biofuels (theme 1); and modelling adaptation and uncertainty, achieving multifunctionality, food and fishery systems, forest biodiversity and ecosystem services, rural migration from climate change and metrics (theme 2). Theme 3 comprises designing research that bridges disciplines, integrating stakeholder input to directly link science, action and governance.

Groundwater pumping chronically exceeds natural recharge in many agricultural regions in California. A common method of recharging groundwater -when surface water is available -is to deliberately flood an open area, allowing water to percolate into an aquifer. However, open land suitable for this type of recharge is scarce. Flooding agricultural land during fallow or dormant periods has the potential to increase groundwater recharge substantially, but this approach has not been well studied. Using data on soils, topography and crop type, we developed a spatially explicit index of the suitability for groundwater recharge of land in all agricultural regions in California. We identified 3.6 million acres of agricultural land statewide as having Excellent or Good potential for groundwater recharge. The index provides preliminary guidance about the locations where groundwater recharge on agricultural land is likely to be feasible. A variety of institutional, infrastructure and other issues must also be addressed before this practice can be implemented widely.

To date, subreach-scale variations in flow width and bed elevation have rarely been included in channel classifications. Variability in topographic features of rivers, however, in conjunction with sediment supply and discharge produces a mosaic of channel forms that provides unique habitats for sensitive aquatic species. In this study we investigated the utility of topographic variability attributes (TVAs) in distinguishing channel types and dominant channel formation and maintenance processes in montane and lowland streams of the Sacramento River basin, California, USA. A stratified random survey of 161 stream sites was performed to ensure balanced sampling across groups of stream reaches with expected similar geomorphic settings. For each site surveyed, width and depth variability were measured at baseflow and bankfull stages, and then incorporated in a channel classification framework alongside traditional reach-averaged geomorphic attributes (e.g., channel slope, width-to-depth, confinement, and dominant substrate) to evaluate the significance of TVAs in differentiating channel types. In contrast to more traditional attributes such as slope and contributing area, which are often touted as the key indicators of hydrogeomorphic processes, bankfull width variance emerged as a first-order attribute for distinguishing channel types. A total of nine channel types were distinguished for the Sacramento Basin consisting of both previously identified and new channel types. The results indicate that incorporating TVAs in channel classification provides a quantitative basis for interpreting nonuniform as well as uniform geomorphic processes, which can improve our ability to distinguish linked channel forms and processes of geomorphic and ecological significance.

16The extent and timing of many river ecosystem functions is controlled by the interplay of 17 streamflow dynamics with the river corridor shape and structure.

Hydrologic and geomorphic classifications have gained traction in response to the increasing need for basin‐wide water resources management. Regardless of the selected classification scheme, an open scientific challenge is how to extend information from limited field sites to classify tens of thousands to millions of channel reaches across a basin. To address this spatial scaling challenge, this study leverages machine learning to predict reach‐scale geomorphic channel types using publicly available geospatial data. A bottom‐up machine learning approach selects the most accurate and stable model among ∼20,000 combinations of 287 coarse geospatial predictors, preprocessing methods, and algorithms in a three‐tiered framework to (i) define a tractable problem and reduce predictor noise, (ii) assess model performance in statistical learning, and (iii) assess model performance in prediction. This study also addresses key issues related to the design, interpretation, and diagnosis of machine learning models in hydrologic sciences. In an application to the Sacramento River basin (California, USA), the developed framework selects a Random Forest model to predict 10 channel types previously determined from 290 field surveys over 108,943 two hundred‐meter reaches. Performance in statistical learning is reasonable with a 61% median cross‐validation accuracy, a sixfold increase over the 10% accuracy of the baseline random model, and the predictions coherently capture the large‐scale geomorphic organization of the landscape. Interestingly, in the study area, the persistent roughness of the topography partially controls channel types and the variation in the entropy‐based predictive performance is explained by imperfect training information and scale mismatch between labels and predictors.