Global Horizontal Irradiance Forecasts Research Articles

Evaluation of WRF-DART (ARW v3.9.1.1 and DART Manhattan release) multiphase cloud water path assimilation for short-term solar irradiance forecasting in a tropical environment

Abstract. Numerical weather prediction models tend to underestimate cloud presence and therefore often overestimate global horizontal irradiance (GHI). The assimilation of cloud water path (CWP) retrievals from geostationary satellites using an ensemble Kalman filter (EnKF) led to improved short-term GHI forecasts of the Weather Research and Forecasting (WRF) model in midlatitudes in case studies. An evaluation of the method under tropical conditions and a quantification of this improvement for study periods of more than a few days are still missing. This paper focuses on the assimilation of CWP retrievals in three phases (ice, supercooled, and liquid) in a 6-hourly cycling procedure and on the impact of this method on short-term forecasts of GHI for Réunion Island, a tropical island in the southwest Indian Ocean. The multilayer gridded cloud properties of NASA Langley's Satellite ClOud and Radiation Property retrieval System (SatCORPS) are assimilated using the EnKF of the Data Assimilation Research Testbed (DART) Manhattan release (revision 12002) and the advanced research WRF (ARW) v3.9.1.1. The ability of the method to improve cloud analyses and GHI forecasts is demonstrated, and a comparison using independent radiosoundings shows a reduction of specific humidity bias in the WRF analyses, especially in the low and middle troposphere. Ground-based GHI observations at 12 sites on Réunion Island are used to quantify the impact of CWP DA. Over a total of 44 d during austral summertime, when averaged over all sites, CWP data assimilation has a positive impact on GHI forecasts for all lead times between 5 and 14 h. Root mean square error and mean absolute error are reduced by 4 % and 3 %, respectively.

Forecast Horizon and Solar Variability Influences on the Performances of Multiscale Hybrid Forecast Model

The tropical insular region is characterized by a large diversity of microclimates and land/sea contrasts, creating a challenging solar forecasting. Therefore, it is necessary to develop and use performant and robustness forecasting techniques. This paper examines the predictive performance of a novel solar forecasting approach, the multiscale hybrid forecast model (MHFM), as a function of several parameters. The MHFM model is a technique recently used for irradiance forecasting based on a hybrid autoregressive (AR) and neural network (NN) model combined with multiscale decomposition methods. This technique presents a relevant performance for 1 h ahead global horizontal irradiance forecast. The goal of this work is to highlight the strength and limits of this model by assessing the influence of different parameters from a metric error analysis. This study illustrates modeling process performance as a function of daily insolation conditions and testifies the influence of learning data and test data time scales. Several forecast horizon strategies and their influence on the MHFM performance were investigated. With the best strategy, a rRMSE value from 4.43 % to 10.24 % was obtained for forecast horizons from 5 min to 6 h. The analysis of intra-day solar resource variability showed that the best performance of MHFM was obtained for clear sky days with a rRMSE of 2.91 % and worst for cloudy sky days with a rRMSE of 6.73 % . These works constitute an additional analysis in agreement with the literature about influence of daily insolation conditions and horizons time scales on modeling process.

Post-processing of NWP forecasts using ground or satellite-derived data through kernel conditional density estimation

Global horizontal irradiance (GHI) forecasts by numerical weather prediction (NWP) often contain model-led bias. There is thus a strong consensus on using post-processing techniques, such as model output statistics (MOS), to correct such errors. As opposed to the conventional parametric methods, this article considers a nonparametric approach for post-processing, namely, kernel conditional density estimation (KCDE). Essentially, KCDE constructs a relationship between the bias error (difference between the NWP-based GHI forecast and measurement) and NWP output variables, such as clear-sky index, zenith angle, air temperature, humidity, or surface pressure. Hence, when a new set of explanatory variables becomes available, the conditional expectation of the bias error can be estimated. Since the ground-based GHI measurements are not available everywhere, the possibility of using satellite-derived GHI data to correct NWP forecasts is also explored. In the case study, two years of GHI forecasts made using the North American Mesoscale forecast system are corrected using both ground-measured and satellite-derived GHI references. As compared to Lorenz's fourth-degree polynomial MOS, additional 10%–16% (using ground-measured GHI) and 5%–13% (using satellite-derived GHI) reductions in the forecast error are observed at 7 test stations across the continental United States.

Preliminary assessment of two spatio-temporal forecasting technics for hourly satellite-derived irradiance in a complex meteorological context

This paper examines two spatio-temporal approaches for short-term forecasting of global horizontal irradiance using gridded satellite-derived irradiances as experimental support. The first approach is a spatio-temporal vector autoregressive (STVAR) model combined with a statistical process for optimum selection of input variables. The second is an existing operational cloud motion vector (CMV) model. An evaluation of the predictive performance of these models is presented for a case study area in the Caribbean Islands. This region is characterized by a large diversity of microclimates and land/sea contrasts, creating a challenging solar forecasting context. Using scaled persistence as a reference, we benchmark the performance of the two spatio-temporal models over an extended 220 × 220 km domain, and for three specific, climatically distinct locations within this domain. We also assess the influence of intra-day solar resource variability on model performance. Finally, we present preliminary evidence that a blend of CMV and STVAR forecasts leads to improved accuracy under all conditions.

A Physics-based Smart Persistence model for Intra-hour forecasting of solar radiation (PSPI) using GHI measurements and a cloud retrieval technique

Short-term solar forecasting models based solely on global horizontal irradiance (GHI) measurements are often unable to discriminate the forecasting of the factors affecting GHI from those that can be precisely computed by atmospheric models. This study introduces a Physics-based Smart Persistence model for Intra-hour forecasting of solar radiation (PSPI) that decomposes the forecasting of GHI into the computation of extraterrestrial solar radiation and solar zenith angle and the forecasting of cloud albedo and cloud fraction. The extraterrestrial solar radiation and solar zenith angle are accurately computed by the Solar Position Algorithm (SPA) developed at the National Renewable Energy Laboratory (NREL). A cloud retrieval technique is used to estimate cloud albedo and cloud fraction from surface-based observations of GHI. With the assumption of persistent cloud structures, the cloud albedo and cloud fraction are predicted for future time steps using a two-stream approximation and a 5-min exponential weighted moving average, respectively. Our model evaluation using the long-term observations of GHI at NREL’s Solar Radiation Research Laboratory (SRRL) shows that the PSPI has a better performance than the persistence and smart persistence models in all forecast time horizons between 5 and 60 min, which is more significant in cloudy-sky conditions. Compared to the persistence and smart persistence models, the PSPI does not require additional observations of various atmospheric parameters but is customizable in that additional observations, if available, can be ingested to further improve the GHI forecast. An advanced technology of cloud forecast is also expected to improve the future performance of the PSPI.

Surface solar radiation forecasts by advecting cloud physical properties derived from Meteosat Second Generation observations

A surface solar radiation forecast algorithm is developed using cloud physical properties from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) on board of the Meteosat Second Generation (MSG) geostationary satellite. The novelty of the algorithm is the derivation of cloud motion vectors using cloud physical properties. The solar radiation forecast products include global horizontal irradiance (GHI) and direct normal irradiance (DNI). The forecast horizon is 0–4 h at a 15 min temporal resolution. The forecast is currently tested for the Netherlands at a spatial resolution of about 4 km × 6 km. Compared to measurements of the Baseline Surface Radiation Network (BSRN) site of Cabauw, the root mean square error (RMSE) is about 31–44% for GHI and 59–100% for DNI at a forecast horizon of 2 h. For a forecast horizon of 15 min, the RMSE is 22–24% for GHI and 43–61% for DNI. The correlation coefficients between the forecasts and BSRN measurements are similar for GHI and DNI, and decrease from about 0.8–0.9 at 15 min to 0.45–0.75 at 2 h. The SEVIRI forecast outperforms the HARMONIE numerical weather prediction model forecast in the first 2–3 h. The quality of the forecast depends on the sky conditions: for clear-sky indices larger than 0.7, the SEVIRI GHI forecast is better than both smart persistence and the HARMONIE forecast in the first 4 h.

A Hybrid GA–MLPNN Model for One-Hour-Ahead Forecasting of the Global Horizontal Irradiance in Elizabeth City, North Carolina

The use of photovoltaics is still considered to be challenging because of certain reliability issues and high dependence on the global horizontal irradiance (GHI). GHI forecasting has a wide application from grid safety to supply–demand balance and economic load dispatching. Given a data set, a multi-layer perceptron neural network (MLPNN) is a strong tool for solving the forecasting problems. Furthermore, noise detection and feature selection in a data set with numerous variables including meteorological parameters and previous values of GHI are of crucial importance to obtain the desired results. This paper employs density-based spatial clustering of applications with noise (DBSCAN) and non-dominated sorting genetic algorithm II (NSGA II) algorithms for noise detection and feature selection, respectively. Tuning the neural network is another important issue that includes choosing the hidden layer size and activation functions between the layers of the network. Previous studies have utilized a combination of different parameters based on trial and error, which seems to be inefficient in terms of accurate selection of the desired features and also tuning of the neural network. In this research, two different methods—namely, particle swarm optimization (PSO) algorithm and genetic algorithm (GA)—are utilized in order to tune the MLPNN, and the results of one-hour-ahead forecasting of the GHI are subsequently compared. The methodology is validated using the hourly data for Elizabeth City located in North Carolina, USA, and the results demonstrated a better performance of GA in comparison with PSO. The GA-tuned MLPNN reported a normalized root mean square error (nRMSE) of 0.0458 and a normalized mean absolute error (nMAE) of 0.0238.

Short-Term Solar Irradiance Forecasts Using Sky Images and Radiative Transfer Model

In this paper, we propose a novel forecast method which addresses the difficulty in short-term solar irradiance forecasting that arises due to rapidly evolving environmental factors over short time periods. This involves the forecasting of Global Horizontal Irradiance (GHI) that combines prediction sky images with a Radiative Transfer Model (RTM). The prediction images (up to 10 min ahead) are produced by a non-local optical flow method, which is used to calculate the cloud motion for each pixel, with consecutive sky images at 1 min intervals. The Direct Normal Irradiance (DNI) and the diffuse radiation intensity field under clear sky and overcast conditions obtained from the RTM are then mapped to the sky images. Through combining the cloud locations on the prediction image with the corresponding instance of image-based DNI and diffuse radiation intensity fields, the GHI can be quantitatively forecasted for time horizons of 1–10 min ahead. The solar forecasts are evaluated in terms of root mean square error (RMSE) and mean absolute error (MAE) in relation to in-situ measurements and compared to the performance of the persistence model. The results of our experiment show that GHI forecasts using the proposed method perform better than the persistence model.

Nowcasting solar irradiance using an analog method and geostationary satellite images

Accurate forecasting of Global Horizontal Irradiance (GHI) is essential for the integration of the solar resource in an electrical grid. We present a novel data-driven method aimed at delivering up to 6 h hourly probabilistic forecasts of GHI on top of a localized solar energy source. The method does not require calibration to adapt to regional differences in cloud dynamics, and uses only one type of data, covering Europe and Africa. It is thus suited for applications that require a GHI forecast for solar energy sources at different locations with few ground measurements. Cloud dynamics are emulated using an analog method based on 5 years of hourly images of geostationary satellite-derived irradiance, without using any numerical prediction model. This database contains both the images to be compared to the current atmospheric observation and their successors at one or more hours of interval. The physics of the system is emulated statistically, and no numerical prediction model is used. The method is tested on one year of data and five locations in Europe with different climatic conditions. It is compared to persistence (keeping the last observation frozen), ensemble persistence (generating a probabilistic forecast using the last observations) and an adaptive first order vector autoregressive model. As an application, the model is downscaled using ground measurements. In both cases, the analog method outperforms the classical statistical approaches. Results demonstrate the skill of the method in emulating cloud dynamics, and its potential to be coupled with a forecasting algorithm using ground measurements for operational applications.

Photovoltaic yield prediction using an irradiance forecast model based on multiple neural networks

In order to develop predictive control algorithms for efficient energy management and monitoring for residential grid connected photovoltaic systems, accurate and reliable photovoltaic (PV) power forecasts are required. A PV yield prediction system is presented based on an irradiance forecast model and a PV model. The PV power forecast is obtained from the irradiance forecast using the PV model. The proposed irradiance forecast model is based on multiple feed-forward neural networks. The global horizontal irradiance forecast has a mean absolute percentage error of 3.4% on a sunny day and 23% on a cloudy day for Stuttgart. PV power forecasts based on the neural network irradiance forecast have performed much better than the PV power persistence forecast model.

Conditional summertime day-ahead solar irradiance forecast

We investigated the accuracy of numerical weather prediction (NWP)-based global horizontal irradiance (GHI) and clear-sky index forecasting over southern Nevada. Accurate forecasts of solar irradiance are required for electric utilities to economically integrate substantial amounts of solar power into their power generation portfolios. Solar irradiance forecasting can enhance the value of renewable energy by anticipating fluctuations in these variable resources. Summertime cloud variability depends largely on the combination of tropical and extratropical synoptic-scale forcing, most of which is observable, predictable, and highly related to the North American Monsoon moisture surge events. We used high-resolution realtime NWP output based on the weather and research forecasting (WRF) model to study the ability of the model to deliver day-ahead GHI and clear-sky index forecasts for a the National Renewable Energy Laboratory (NREL)-University of Nevada site, located in Las Vegas, Nevada. High-resolution forecast products were obtained from the Desert Research Institute (DRI) archived real-time numerical weather forecasting products. Results showed the importance of developing a site-specific seasonal and weather-dependent model output statistics (MOS) approach to improving forecast accuracy, which removes the bias and reduces the overall relative root-mean – square error (rRMSE) of GHI by as much as 6%, when compared to the uncorrected model output; improving forecast accuracy is obtained by adding information that relates regional-scale circulation patterns driving cloudiness, hence irradiance variability to the target area. We show the seasonal dependence of the NWP forecast accuracy and demonstrate that intelligent weather functions provide a pathway to improve accuracy of solar forecasts further.

Assimilating observations to simulate marine layer stratocumulus for solar forecasting

Integration of solar energy forecasts into the electric network is becoming essential because of the continually increasing penetration level of solar energy. Three-dimensional numerical weather prediction (NWP) models predict the weather based on the current weather conditions (called initialization) and simulate the ensuing atmospheric processes. The accuracy of forecasts therefore depends, in part, on the accuracy of the model initializations. Data assimilation is recognized as the most widely used technique to improve the initialization into NWP models. In this study, meteorological observations from the surface and upper-air in-situ networks over the southern California coast are assimilated into the advanced research version of the Weather Research and Forecasting (WRF) model using a three dimensional variational data assimilation technique (3DVAR). A single observation test was conducted to tune-up the length scale and variance scale along with the regional domain-dependent background error statistics. A customized version of 3DVAR data assimilation was deployed with two sets of cyclic data assimilation with 6-h and 1-h assimilation windows along with the cold-start mode. The cyclic data assimilation experiments consistently outperformed the cold-start data assimilation and WRF for intra-day Global Horizontal Irradiance (GHI) and Clear Sky Index (CSI) forecast. Hourly cyclic assimilation showed the highest forecast skill score against ground measurements and satellite measurements. Even at the coastal stations with more challenging meteorological conditions, the hourly cyclic assimilation consistently outperformed the 24-h persistence forecast. The average (mean of four case studies) hourly cyclic data assimilation showed the highest forecast skill score in GHI and CSI intra-day forecast with reference to 24-h persistence forecast up to 39.4% and 40.7% respectively at the coastal stations. The spatial distributions of GHI biases estimated against SolarAnywhere satellite measurements showed that the hourly cyclic assimilation consistently improved the stratocumulus cloud coverage, thickness, and life time over the coastal region, but biases are still present further inland.

Verification of ECMWF and ECMWF/MACC's global and direct irradiance forecasts with respect to solar electricity production forecasts

The successful electricity grid integration of solar energy into day-ahead markets requires at least hourly resolved 48 h forecasts. Technologies as photovoltaics and non-concentrating solar thermal technologies make use of global horizontal irradiance (GHI) forecasts, while all concentrating technologies both from the photovoltaic and the thermal sector require direct normal irradiances (DNI). The European Centre for Medium-Range Weather Forecasts (ECMWF) has recently changed towards providing direct as well as global irradiances. Additionally, theMACC (Monitoring Atmospheric Composition & Climate) near-real time services provide daily analysis and forecasts of aerosol properties in preparation of the upcoming European Copernicus programme. The operational ECMWF/IFS (Integrated Forecast System) forecast system will in the medium term profit from the Copernicus service aerosol forecasts. Therefore, within the MACC-II project specific experiment runs were performed allowing for the assessment of the performance gain of these potential future capabilities. Also the potential impact of providing forecasts with hourly output resolution compared to three-hourly resolved forecasts is investigated. The inclusion of the new aerosol climatology in October 2003 improved both the GHI and DNI forecasts remarkably, while the change towards a new radiation scheme in 2007 only had minor and partly even unfavourable impacts on the performance indicators. For GHI, larger RMSE (root mean square error) values are found for broken/overcast conditions than for scattered cloud fields. For DNI, the findings are opposite with larger RMSE values for scattered clouds compared to overcast/broken cloud situations. The introduction of direct irradiances as an output parameter in the operational IFS version has not resulted in a general performance improvement with respect to biases and RMSE compared to the widely used Skartveit et al. (1998) global to direct irradiance conversion scheme. Cloudy situations and especially thin ice cloud cases are forecasted much better with respect to biases and RMSE, but large biases are introduced in clear sky cases. When applying the MACC aerosol scheme to include aerosol direct effects, an improvement especially in DNI biases is found for cloud free cases as expected. However, a performance decrease is found for water cloud cases. It is assumed that this is caused by the lack of an explicit modelling of cloud-aerosol interactions, while other meteorological forcings for cloud processes like the temperature field are modified by the aerosols.

Comparison of global horizontal irradiance forecasts based on numerical weather prediction models with different spatio‐temporal resolutions

AbstractIn the framework of the IEA SHC Task 46 “Solar Resource Assessment and Forecasting” we compare solar global horizontal irradiance forecasts based on numerical weather predictions for a variety of different models. These include direct model output of several numerical weather prediction models, a rapid update cycle model assimilating satellite derived cloud products as well as radar data, the multi model ensemble prediction system GLAMEPS, and two MOS systems. In order to allow for a transparent and comparable analysis of the different methods we have set up a joint, consistent framework of evaluation. As a basis for the comparisons we have compiled a common data set of hourly measured solar irradiance values for Denmark, Germany, and Switzerland. Local and regional forecasts are analyzed with respect to different properties. In particular we show that spatial and temporal averaging effects have a strong impact on the root mean square error when comparing solar irradiance forecasts of numerical weather prediction models with different output resolutions. Furthermore, we investigate a new approach to evaluate the models' ability to represent and forecast solar irradiance and cloud variability. The benefit of high resolution mesoscale models in this respect is demonstrated. Copyright © 2016 John Wiley & Sons, Ltd.

Evaluating the spatio-temporal performance of sky-imager-based solar irradiance analysis and forecasts

Abstract. Clouds are the dominant source of small-scale variability in surface solar radiation and uncertainty in its prediction. However, the increasing share of solar energy in the worldwide electric power supply increases the need for accurate solar radiation forecasts. In this work, we present results of a very short term global horizontal irradiance (GHI) forecast experiment based on hemispheric sky images. A 2-month data set with images from one sky imager and high-resolution GHI measurements from 99 pyranometers distributed over 10 km by 12 km is used for validation. We developed a multi-step model and processed GHI forecasts up to 25 min with an update interval of 15 s. A cloud type classification is used to separate the time series into different cloud scenarios. Overall, the sky-imager-based forecasts do not outperform the reference persistence forecasts. Nevertheless, we find that analysis and forecast performance depends strongly on the predominant cloud conditions. Especially convective type clouds lead to high temporal and spatial GHI variability. For cumulus cloud conditions, the analysis error is found to be lower than that introduced by a single pyranometer if it is used representatively for the whole area in distances from the camera larger than 1–2 km. Moreover, forecast skill is much higher for these conditions compared to overcast or clear sky situations causing low GHI variability, which is easier to predict by persistence. In order to generalize the cloud-induced forecast error, we identify a variability threshold indicating conditions with positive forecast skill.

A case study of outlier event on solar irradiance forecasts from the two NWPs with different horizontal resolutions

Photovoltaic (PV) power generation is directly effected by global horizontal irradiance (GHI) and has also large variations in spatial and/or temporal scales. For a safety control of an energy management system (EMS), a day-ahead forecast or several hour forecast of solar irradiance by a numerical weather prediction model (NWP) becomes important for a control of reserve capacity (thermal power generation, etc.). In particular, a large forecast error of PV power and/or GHI forecasts has to be prevented in the EMS. The Japan Meteorological Agency (JMA) developed two NWPs with different horizontal resolutions. First one is a mesoscale model with horizontal grid spacing of 5 km and second one is a local forecast model with that of 2 km. The two NWPs have been used as an operational model in JMA. In this study, GHI forecasts obtained from the two models are validated and conducted a case study for large forecast error (outlier events) case of GHI.

Regional and seasonal characteristics of global horizontal irradiance forecasts obtained from the Japan Meteorological Agency mesoscale model

To obtain accurate forecasts of photovoltaic power generation, the use of forecast datasets of meteorological elements from numerical prediction models, specifically global horizontal irradiance (GHI), is necessary. This study seeks to validate, and therefore improve GHI forecasts. Ground-based data from Japan Meteorological Agency (JMA) stations are used in a JMA mesoscale model (MSM) during the time period from 2008 to 2012 and temporal and spatial characteristics of forecast errors are analyzed. Statistical monthly evaluations show that associated errors vary between seasons, with monthly GHI mean bias error values ranging from −60 to +45W/m2 and root mean square errors (RMSEs) ranging from 95 to 170W/m2. Mapping of forecast errors show that underestimation of GHI forecast values and large RMSE values are significant in the southern part of Japan (a subtropical region located along the Pacific Ocean), particularly during summers. In winter, overestimation of GHI forecasts is found throughout the entire Japanese archipelago. The frequency of different cloud type occurrences over the Japanese islands indicate that regional and seasonal variations in cloud types are related to relatively large GHI forecast errors. High-level cirrus clouds, mid-level altocumulus, and low-level stratus are often observed during summer, when forecasted values are underestimated, and during winter, when values are overestimated.

Forecasting of global horizontal irradiance by exponential smoothing, using decompositions

Time series methods are frequently used in solar irradiance forecasting when two dimensional cloud information provided by satellite or sky camera is unavailable. ETS (exponential smoothing) has received extensive attention in the recent years since the invention of its state space formulation. In this work, we combine these models with knowledge based heuristic time series decomposition methods to improve the forecasting accuracy and computational efficiency.In particular, three decomposition methods are proposed. The first method implements an additive seasonal-trend decomposition as a preprocessing technique prior to ETS. This can reduce the state space thus improve the computational efficiency. The second method decomposes the GHI (global horizontal irradiance) time series into a direct component and a diffuse component. These two components are used as forecasting model inputs separately; and their corresponding results are recombined via the closure equation to obtain the GHI forecasts. In the third method, the time series of the cloud cover index is considered. ETS is applied to the cloud cover time series to obtain the cloud cover forecast thus the forecast GHI through polynomial regressions. The results show that the third method performs the best among three methods and all proposed methods outperform the persistence models.

Streamline-based method for intra-day solar forecasting through remote sensing

This work presents an enhanced deterministic solar irradiance forecasting approach that relies on satellite images and ground measurements as inputs. The proposed method is based on a ground-truth improved satellite-to-irradiance model for the prediction of global horizontal irradiance (GHI). This approach relies on cloud tracking and advection with an optical flow algorithm. The application of the optical flow algorithm between two consecutive satellite image frames allows for the calculation of a vector field covering each pixel in the satellite image. This cloud motion vector (CMV) field determines the streamline passing through the location of interest. The estimated cloud advection along the quasi-steady streamline to the location of interest is than computed, and this information is translated into an irradiance forecast for the location of interest. In order to reduce the error associated with a linear satellite-to-irradiance model, a novel approach employing ground measurements is proposed. Additionally, decision heuristics are identified and implemented to issue a forecast based on CMV or persistence, depending on the current sky conditions. The overall method is tested for over 110days of operational 1-, 2- and 3-h ahead GHI forecasts, implemented and evaluated for San Diego, California. The continual forecasting skill of this method for the 110days ranges between 8% and 19% over persistence, depending on the forecast horizon. While previously proposed methods achieve similar skills, the completely deterministic approach combined with comparably low computation and implementation costs makes the proposed method suitable for applications with limited availability of data.

Improvement of the Japan Meteorological Agency Meso-Scale Model for the Forecasting the Photovoltaic Power Production: Modification of the Cloud Scheme

Power production of a photovoltaic (PV) power plant varies according to weather conditions. Therefore, it is important for the prediction of the PV power production to use the weather data (satellite data or numerical weather prediction, etc.). In our research group, for the day ahead forecasting, the output of the numerical weather model, Japan Meteorological Agency Meso-Scale Model (hereafter MSM) is used for the input of the PV power production model. From our previous research, the MSM forecast of the global horizontal irradiance (GHI) tends to be underestimated (overestimated) during summer (winter). Further investigation revealed that the error of the MSM GHI forecast is in a relation of the inverse correlation with the error of the MSM cloudiness forecast. So, in this study, to improve the MSM GHI forecast, the cloud scheme is modified to remove the error.The MSM is an operational, non-hydrostatic and regional model used for a short-range forecast (33hours). The model horizontal resolution is 5km mesh and the model vertical resolution is 50 levels. The current cloud scheme of the MSM has the seasonal error so that the parameter which dependent on the surface air temperature is introduced to represent the seasonality: if the surface air temperature is low (high), then the cloud production is accelerated (decelerated). Twelve cases (6 for winter and 6 for summer) are chosen for the analysis.The modified cloud scheme makes a success of the reduction of the MSM GHI forecast error. The daytime averaged root mean square error (RMSE) of the MSM GHI forecast for all cases is improved about 5% (from 120W m2 to 114W m2). The daytime averaged mean bias error (MBE) of the MSM GHI forecast for all cases is significantly reduced from -14.3W m2 to -5.13W m2. For each cases, although three of them are increased the RMSE (about 3W m2), the total trend are decreased the RMSE.