This study compares the multivariate predictions of daily temperature, temperature range, precipitation, surface wind and solar radiation of a single‐model analogue approach with an original multi‐model analogy over 12 regions in Europe and Maghreb. Both approaches are based on two‐level analogue models where atmospheric predictors are either dynamic or thermodynamic. In the multi‐model approach, independent analogue models with predictand‐specific predictors are used. In the single‐model one, a unique analogue model and its associated set of predictors is applied to all predictands. Testing numerous large‐scale predictors, we first identify the best predictor sets for each modelling strategy. Those obtained for the single‐model approach are significantly different from those of the predictand‐specific models. This is especially the case for local temperature and wind speed. Both methods perform similarly for precipitation, temperature range and radiation. We next assess the ability of both approaches to simulate physically coherent multivariate weather scenarios. With the single‐model method, weather scenarios are obtained for each prediction day from observations sampled simultaneously on one analogue day. The physical consistency between variables is thus automatically fulfilled each day. This allows the single‐model method to reproduce well the observed inter‐predictand correlations, especially the significant correlations between radiation and precipitation and between radiation and temperature range. These results suggest a hybrid analogue model using a single‐model for radiation, temperature range and precipitation, combined with a univariate approach for wind. Two options are proposed for temperature for which either the predictand‐specific method or a single‐model approach with an additional correction are conceivable. This hybrid approach leads to a possible compromise between reasonable univariate prediction skills and realistic inter‐predictands correlations, both classically required for many impact studies.

No abstract

Various techniques exist to estimate stream nitrate loads when measured concentration data are sparse. The inherent uncertainty associated with load estimation, however, makes tracking progress toward water quality goals more difficult. We used high‐frequency, in situ nitrate sensors strategically deployed across the agricultural state of Iowa to evaluate 2016 stream concentrations at 60 sites and loads at 35 sites. The generated data, collected at an average of 225 days per site, show daily average nitrate‐N yields ranging from 12 to 198 g/ha, with annual yields as high as 53 kg/ha from the intensely drained Des Moines Lobe. Thirteen of the sites that capture water from 82.5% of Iowa's area show statewide nitrate‐N loading in 2016 totaled 477 million kg, or 41% of the load delivered to the Mississippi–Atchafalaya River Basin (MARB). Considering the substantial private and public investment being made to reduce nitrate loading in many states within the MARB, networks of continuous, in situ measurement devices as described here can inform efforts to track year‐to‐year changes in nitrate load related to weather and conservation implementation. Nitrate and other data from the sensor network described in this study are made publicly available in real time through the Iowa Water Quality Information System.