What can we learn from weather apps user complaints?

<p>Meteored’s goal is to provide a weather forecast to an heterogeneous audience around the world through its websites and mobile apps.  Although the meteorological information is available through several products such as radar, satellite or weather field maps, most of the views are focused on checking the forecast for a specific location. Our weather forecast is built on the HRES model from ECWMF, which is post processed, spatially interpolated to the interested coordinates,  and, finally, summarized in several weather symbols. If any user doesn’t agree with the symbol that represents the forecast, she/he can select which symbol better represents its weather perception.</p><p>Using this simplification to validate forecast entails several challenges: 1) Spatial representativeness; there aren’t weather stations at each location where users demand to validate, 2) timing; sometimes there is lack of concurrency between a weather phenomenon and the user weather check,  3) user perception; same symbol can represent different weather for different users, and 4)  population density; most of the user complaints are focused on the most populated regions but this doesn't mean the performance is worse there.</p><p>Last year more than 374000 symbol suggestions were recorded from worldwide users, mainly from Europe and Southamerica. The percentage of complaints were 39% cloudy, 24% rainy, 21% suny, 8% storm, 5% snow and 3 % foggy. 16 % of the complaints happen when a cloudy symbol is shown but the user suggests a rainy symbol. Temporal series show more feedback during summer and slightly lower during March (maybe due to the pandemic). Complaints about snow significantly increased due to the historical event in Spain during January. From weather feedback, the straightforward question is: why most frequent complaining is about cloudiness? We can find several answers: there is an important error in the weather modelling, there is an error in the symbol representation, it is a frequent meteorology event or it is one of the most relevant in users daily life.</p><p>In order to understand how reliable the user’s feedback is, our forecast is compared against almost 10000 SYNOP observations, assessing 2 m temperature, 10 m wind speed, precipitation, fog and also, symbols. Preliminary results show a pronounced dependence of the bias with the orography with larger errors over some islands and over main mountain systems. This spatial variability for bias is smoothed in Meteored forecast due to biquadratic interpolation. However, the Meteored forecast has a diurnal cycle bias error with higher temperatures during the daytime and lower temperature at nighttime due to the temporal interpolation approach.  Regarding to the weather symbols validation is difficult to extract conclusion since failures and hits are hetereogenously distributed. In addition,  most discrepancies are related to fog although it has a low percentage of complaints. </p>

EDISON: An Edge-Native Method and Architecture for Distributed Interpolation

Spatio-temporal interpolation provides estimates of observations in unobserved locations and time slots. In smart cities, interpolation helps to provide a fine-grained contextual and situational understanding of the urban environment, in terms of both short-term (e.g., weather, air quality, traffic) or long term (e.g., crime, demographics) spatio-temporal phenomena. Various initiatives improve spatio-temporal interpolation results by including additional data sources such as vehicle-fitted sensors, mobile phones, or micro weather stations of, for example, smart homes. However, the underlying computing paradigm in such initiatives is predominantly centralized, with all data collected and analyzed in the cloud. This solution is not scalable, as when the spatial and temporal density of sensor data grows, the required transmission bandwidth and computational capacity become unfeasible. To address the scaling problem, we propose EDISON: algorithms for distributed learning and inference, and an edge-native architecture for distributing spatio-temporal interpolation models, their computations, and the observed data vertically and horizontally between device, edge and cloud layers. We demonstrate EDISON functionality in a controlled, simulated spatio-temporal setup with 1 M artificial data points. While the main motivation of EDISON is the distribution of the heavy computations, the results show that EDISON also provides an improvement over alternative approaches, reaching at best a 10% smaller RMSE than a global interpolation and 6% smaller RMSE than a baseline distributed approach.

Investigating the value of regional water isotope data on transit time and SAS modelling

<p>High nutrient concentrations despite mitigation measures and reduced inputs are a common problem in anthropogenically impacted catchments. To investigate how water and solutes of different ages are mixed and released from catchment storage to the stream, catchment-scale models based on water transit time from StorAge Selection functions (SAS) are a promising tool. Tracking fluxes of environmental tracers, such as stable water isotopes, allows to calibrate and validate these models. However, this requires collection of water samples with an adequate temporal and spatial resolution, while sampling in catchments at the management scale is often limited by the high costs of the instruments, maintenance and chemical analysis. Therefore, temporal and spatial interpolation techniques are needed. This study demonstrates how to deal with sparse tracer data in space and time, and evaluates if these data are valuable to constrain the subsurface mixing dynamics and transit time with SAS modelling. We simulated water isotope data in diverse sub-basins of the Bode catchment (Germany) and calibrated the SAS function parameters against the measured streamflow isotope data. We tested four different combinations of spatial and temporal interpolation of the measured precipitation isotope data. In terms of temporal interpolation, monthly oxygen isotopes in precipitation (δ<sup>18</sup>O<sub>P</sub>) collected between 2012 and 2015 were converted to a daily time step with a step function and sinusoidal interpolation. In terms of spatial interpolation, the model was tested with raw values of δ<sup>18</sup>O<sub>P</sub> collected at a specific sampling point and with δ<sup>18</sup>O<sub>P</sub> interpolated using kriging to gain the spatial pattern of precipitation. The effect of the spatial and temporal interpolation techniques on the modeled SAS functions was analyzed using different parameterizations of the SAS function (i.e., power law time-invariant, power law time-variant and beta law). The results show how tracer input data with different distribution in time and space affect the SAS parameterization and water transit time. Moreover, they reveal preference of the sub-basins to mobilize either younger or older water, which has implications on how water flows through a catchment and on the fate of solutes.</p>