Search for: All records

Deep Evidential Regression (DER) places a prior on the original Gaussian likelihood and treats learning as an evidence acquisition process to quantify uncertainty. For the validity of the evidence theory, DER requires specialized activation functions to ensure that the prior parameters remain non-negative. However, such constraints will trigger evidence contraction, causing sub-optimal performance. In this paper, we analyse DER theoretically, revealing the intrinsic limitations for sub-optimal performance: the non-negativity constraints on the Normal Inverse-Gamma (NIG) prior parameter trigger the evidence contraction under the specialized activation function, which hinders the optimization of DER performance. On this basis, we design a Non-saturating Uncertainty Regularization term, which effectively ensures that the performance is further optimized in the right direction. Experiments on real-world datasets show that our proposed approach improves the performance of DER while maintaining the ability to quantify uncertainty. 

Here, we explicitly define a half-cell reaction approach for pH calculation using the electrode couple comprised of the solid-state chloride ion-selective electrode (Cl-ISE) as the reference electrode and the hydrogen ionselective ion-sensitive field effect transistor (ISFET) of the Honeywell Durafet as the hydrogen ion (H+)-sensitive measuring or working electrode. This new approach splits and isolates the independent responses of the Cl-ISE to the chloride ion (Cl−) (and salinity) and the ISFET to H+ (and pH), and calculates pH directly on the total scale (pHEXT total) in molinity (mol (kg-soln)−1) concentration units. We further apply and compare pHEXT total calculated using the half-cell and the existing complete cell reaction (defined by Martz et al. (2010)) approaches using measurements from two SeapHOx sensors deployed in a test tank. Salinity (on the Practical Salinity Scale) and pH oscillated between 1 and 31 and 6.9 and 8.1, respectively, over a six-day period. In contrast to established Sensor Best Practices, we employ a new calibration method where the calibration of raw pH sensor timeseries are split out as needed according to salinity. When doing this, pHEXT total had root-mean squared errors ranging between ±0.0026 and ±0.0168 pH calculated using both reaction approaches relative to pHtotal of the discrete bottle samples (pHdisc total). Our results further demonstrate the rapid response of the Cl-ISE reference to variable salinity with changes up to ±12 (30 min)−1. Final calculated pHEXT total were ≤±0.012 pH when compared to pHdisc total following salinity dilution or concentration. These results are notably in contrast to those of the few in situ field deployments over similar environmental conditions that demonstrated pHEXT total calculated using the Cl-ISE as the reference electrode had larger uncertainty in nearshore waters. Therefore, additional work beyond the correction of variable temperature and salinity conditions in pH calculation using the Cl-ISE is needed to examine the effects of other external stimuli on in situ electrode response. Furthermore, whereas past work has focused on in situ reference electrode response, greater scrutiny of the ISFET as the H+-sensitive measuring electrode for pH measurement in natural waters is also needed. 

The impacts of the interdecadal variability of the Pacific and the Atlantic Oceans on precipitation over the Central Andes during the austral summer (December-January-February, DJF) are investigated for the 1921–2010 period based on monthly gridded precipitation data and low-pass filtered time series of the Niño 4 index (IN4), the Niño 1 + 2 index with Niño 3.4 index removed (IN1+2 * ), Atlantic Multidecadal Oscillation (AMO), and Interdecadal Pacific Oscillation (IPO) indices, and the three first rotated principal components of the interdecadal component of the sea surface temperature (SST) anomalies over the Atlantic Ocean. A rotated empirical orthogonal function (REOF) analysis of precipitation in the Central Andes (10°S–30°S) yields two leading modes, RPC1 and RPC2, which represent 40.4% and 18.6% of the total variance, respectively. REOF1 features a precipitation dipole between the northern Bolivian and the Chilean Altiplano. REOF2 also features a precipitation dipole, with highest negative loading over the southern Peruvian Andes. The REOF1 positive phase is associated with moisture transport from the lowlands toward the Bolivian Altiplano, induced by upper-level easterly wind anomalies over the Central Andes. At the same time conditions tend to be dry over the southern Peruvian Andes. The positive phase of REOF2 is related to weakened moisture transport, induced by upper-level westerly wind anomalies over Peru. The IPO warm phase induces significant dry anomalies over the Bolivian Altiplano, albeit weaker than during the IN4 warm phase, via upper-level westerly wind anomalies over the Central Andes. No significant relationship was found between Central Andean precipitation and the AMO on interdecadal timescales.