Solar-power-generation forecasting tools are essential for microgrid stability, operation, and planning. The prediction of solar irradiance (SI) usually relies on the time series of SI and other meteorological data. In this study, the considered microgrid was a combined cold- and power-generation system, located in Tahiti. Point forecasts were obtained using a particle swarm optimization (PSO) algorithm combined with three stand-alone models: XGboost (PSO-XGboost), the long short-term memory neural network (PSO-LSTM), and the gradient boosting regression algorithm (PSO-GBRT). The implemented daily SI forecasts relied on an hourly time-step. The input data were composed of outputs from the numerical forecasting model AROME (Météo France) combined with historical meteorological data. Our three hybrid models were compared with other stand-alone models, namely, artificial neural network (ANN), convolutional neural network (CNN), random forest (RF), LSTM, GBRT, and XGboost. The probabilistic forecasts were obtained by mapping the quantiles of the hourly residuals, which enabled the computation of 38%, 68%, 95%, and 99% prediction intervals (PIs). The experimental results showed that PSO-LSTM had the best accuracy for day-ahead solar irradiance forecasting compared with the other benchmark models, through overall deterministic and probabilistic metrics.

Systems biology promises to revolutionize medicine, yet human wellbeing is also inherently linked to healthy societies and environments (sustainabilit

Traditionally, the modeling of the water vapor contents of the atmosphere is done through the estimation of precipitable water (PW)--the integrated value of the mass of water vapor over a vertical column expressed in millimeter equivalent height. This modeling method is justified by the fact that, to a high degree of approximation, the atmosphere can be seen as the stacking of horizontal layers, including water vapor, over distances larger than the height of the tropopause. Nevertheless, the cycle of the water vapor, the most prevalent of the greenhouse gases in the atmosphere, has a highly turbulent regime both in time and in space that the variations in PW cannot fully embrace. In this article, we explore the modeling method, as a series expansion in time and space, of the slant wet delays (SWDs) from one GPS receiver, as an extension of the usual modeling in zenith wet delays (ZWDs) and then PW values. In the first part, we assess, from a metrological point of view, the derivation of the SWDs computed from GPS carrier phase measurements, in the case of a very humid location, the tropical island of Tahiti, for a typical sample over the wet and dry seasons. In the second part, we introduce the series expansion of the SWDs, as seen from one GPS receiver, in terms of trigonometric functions of time and spherical harmonics of elevation and azimuth. This allows us to infer time and space correlations for the SWDs that are unreachable through the modeling of ZWD values alones. In a third part, to show that our approach also includes the zenith case, we make a comparison between the modeled SWDs in the zenith direction with wind velocities from a ground weather station and radiosonde soundings (RSs). The three main conclusions from our data case are: first, the SWDs are correlated in time by about four days, and in space, with an angular correlation distance of about 20°, both for the dry and wet seasons; second, the postfit residuals are almost uncorrelated with the SWDs from a temporal and spatial point of view, but with a diurnal component; third, there is a weak correlation between the SWDs and wind velocity, the pattern depends on the season.

Intraseasonal and diurnal variability of precipitation over Tahiti, French Polynesia, are investigated with the use of wind regimes as intermediary tools. Four wind regimes have been determined over a large domain around Tahiti in the wet season. It has been shown that some phases of the Madden–Julian Oscillation (MJO) trigger some regimes and undermine some others. The diurnal cycle of precipitation in Tahiti is composed of two rainfall maxima. A frequent maximum at midnight but with a low value. An afternoon maximum, less frequent but with a higher value. We have noticed that the afternoon rainfall maximum, which is a reasonable consequence of the sea breeze circulation, is smoothen on the windward side when vigorous wind regimes obstruct the development of the sea breeze. We have also found that daily mean precipitation is significantly increased during MJO Phases 7 and 8, and reduced during Phases 2 and 3. Phase 8 is the wettest phase on Tahiti, suggesting the convective envelope is overlooking the island. Lastly, we have provided evidence that the diurnal cycle of rainfall is significantly affected by the MJO. Indeed, both the midnight and afternoon maxima increase during MJO Phase 7. The most quiescent wind regimes are triggered during Phase 7, as a result from the propagation of the westerly wind anomaly, which allow the development of the sea breeze circulation and the increase of the afternoon peak. For MJO Phase 8, however, the above relationships break down, as the diurnal cycle is not enhanced by the predominant quiescent wind regimes. We assume that the MJO convective envelope covers the island and stays several days, preventing the solar heating of the surface, hence the development of the land/sea breeze circulation, and foremost leads to higher amounts of rainfall as the Phase 8 is the wettest phase on Tahiti.

Since Bevis first proposed Global Positioning System (GPS) meteorology in 1992, the precipitable water (PW) estimates retrieved from Global Navigation Satellite System (GNSS) networks with high accuracy have been widely used in many meteorological applications. The proper estimation of GNSS PW can be affected by the GNSS processing strategy as well as the local geographical properties of GNSS sites. To better understand the impact of these factors, we compare PW estimates from two nearby permanent GPS stations (THTI and FAA1) in the tropical Tahiti Island, a basalt shield volcano located in the South Pacific, with a mean slope of 8% and a diameter of 30 km. The altitude difference between the two stations is 86.14 m, and their horizontal distance difference is 2.56 km. In this paper, Bernese GNSS Software Version 5.2 with precise point positioning (PPP) and Vienna mapping function 1 (VMF1) was applied to estimate the zenith tropospheric delay (ZTD), which was compared with the International GNSS Service (IGS) Final products. The meteorological parameters sourced from the European Center for Medium-Range Weather Forecasts (ECMWF) and the local weighted mean temperature ( $T_m$ ) model were used to estimate the GPS PW for three years (May 2016 to April 2019). The results show that the differences of PW between two nearby GPS stations is nearly a constant with value 1.73 mm. In our case, this difference is mainly driven by insolation differences, the difference in altitude and the wind being only second factors.