Search

A evapotranspiração depende da interação entre variáveis meteorológicas (radiação solar, temperatura do ar, precipitação, umidade relativa do ar e velocidade do vento) e condições fitossanitárias das culturas agrícolas. É complexo construir medidas confiáveis de evapotranspiração devido aos elevados custos para implantação de técnicas micrometeorológicas, além de dificuldades na operação e manutenção dos equipamentos necessários. O propósito desta pesquisa foi modelar a evapotranspiração de referência (ETo) por meio de técnicas de machine learning em dados climáticos de 30 estações meteorológicas automáticas do Planalto Ocidental Paulista, Estado de São Paulo, Brasil, no período de 2013-2017. Uma comparação do desempenho estatístico entre as técnicas utilizadas foi realizada onde constatou-se melhor desempenho do modelo EToMLP4 (rRMSE = 0.62%), seguido por EToANFIS4 (rRMSE = 0.75%), EToSVM4 (rRMSE = 1.19%) e EToGRNN4 (rRMSE = 11.05%). Medidas de performance da base de validação evidenciam que os modelos propostos são aptos à estimativa da evapotranspiração de referência com destaque para a técnica MPL. Palavras-chave: evapotranspiração; modelagem matemática; aprendizagem de máquina. Machine learning models applied in the estimation of reference evapotranspiration from the Western Plateau of Paulista ABSTRACT: Evapotranspiration depends on the interaction between meteorological variables (solar radiation, air temperature, precipitation, relative humidity and wind speed) and phytosanitary conditions of agricultural crops. It is complex to build reliable evapotranspiration measurements due to the high costs of implementing micrometeorological techniques, in addition to difficulties in the operation and maintenance of the necessary equipment. The purpose of this research was to model the reference evapotranspiration through machine learning techniques in climatic data from 30 automatic weather stations in the Planalto Ocidental Paulista, State of São Paulo, Brazil, in the period 2013-2017. A comparison of the statistical performance between the techniques used was carried out, where the best performance of the EToMLP4 model (rRMSE = 0.62%), followed by EToANFIS4 (rRMSE = 0.75%), EToSVM4 (rRMSE = 1.19%) and EToGRNN4 (rRMSE = 11.05 %). Performance measures of the validation base show that the proposed models are able to estimate the reference evapotranspiration, with emphasis on the MPL technique. Keywords: evapotranspiration; modeling; machine learning.

Para que o estabelecimento de plantio de cacaueiro seja economicamente viável, é necessário distinguir o regime das chuvas e relacionar com a demanda hídrica da cultura, para definir quando e quanto irrigar. Sendo assim, o presente trabalho tem por objetivo relacionar a precipitação com as necessidades hídricas de cacaueiro no município de Altamira e Medicilândia. Foram utilizados dados meteorológicos da rede de observação do Instituto Nacional de Meteorologia (INMET), de posse dos dados de evapotranspiração de referencia (ET0) foi possível estimar a evapotranspiração da cultura (ETC), multiplicando a ET0 pelo coeficiente de cultura (Kc) considerando as diferentes fases do cacaueiro. Foi relacionado as variáveis meteorológicas com a necessidades climáticas das plantas de cacau; e ainda, observado se há períodos críticos com relação as necessidades hídricas da cultura durante os meses do ano. A precipitação total anual do município de Altamira e Medicilândia está acima da demanda hídrica anual do cacaueiro, porém existe uma forte sazonalidade em decorrência da má distribuição pluviométrica, podendo prejudicar o bom desenvolvimento e crescimento da planta. Durante os últimos sete meses do ano, a demanda hídrica mensal do cacaueiro na fase inicial é maior que a precipitação. Uma possível solução para resolver essa diferença, seria com a introdução de manejos e tratos culturais com tecnologia da irrigação para adequar essa diferença. Estudos futuros são necessários para se verificar a viabilidade econômica da instalação de sistema de irrigação. Com isso, seria possível potencializar a produção de cacau e assim contribuir para o desenvolvimento da agricultura local.

This work describes the application of models to estimate the transmitted fraction of direct solar irradiation into normal incidence ( K t b ) as a function of the atmospheric transmissivity (Kt) and the insolation ratio ( n / N ). In the first model, the values of K t b and Kt in the hourly (h) and daily (d) partitions were correlated using polynomial regression. In the second model, K t b and n / N in the daily partition were correlated through linear regression. The Adaptive Neuro-Fuzzy Inference System (ANFIS) is used to estimate the obtained values K t b . The performance of ANFIS1 (as a function of Kt hourly), ANFIS2 (as a function of daily Kt), ANFIS3 (as a function of daily n / N ), and ANFIS4 (as a function of the ratio of Kt and daily n / N ) are compared to statistical models. Two databases, defined as a typical year and an atypical year, are adopted to validate the models and ANFIS. The models were validated comparing the estimate and measurements using the statistical indicators rMBE and rRMSE. The statistical models as a function of Kt resulted in hourly average ( K t b h ): 〈 r M B E 〉 = − 4.08 ± 2.18 % and 〈 r R M S E 〉 = 25.97 ± 0.90 % ; daily ( K t b d ): 〈 r M B E 〉 = − 4.93 ± 4.89 % and 〈 r R M S E 〉 = 22.06 ± 2.06 % . The model as a function of the insolation ratio resulted in a daily average ( K t b d ): 〈 r M B E 〉 = − 3.01 ± 2.62 % and 〈 r R M S E 〉 = 24.15 ± 0.30 % . The rMBE of models using ANFIS are lower than those of statistical models. The rRMSE value using ANFIS1 model was 〈 24.45 % 〉 , while using ANFIS2, ANFIS3, and ANFIS4 models, they resulted on 〈 16.47 % 〉 , 〈 19.60 % 〉 , and 〈 14.60 % 〉 in average, respectively, lower than the statistical models. ANFIS's performance is superior to statistical models.

Air temperature is a meteorological variable that influences the climate in the world. The availability of air temperature data is of concern in Brazil, particularly in the State of Mato Grosso do Sul (MS), since most weather stations are concentrated on the country's coast. Thus, the study aimed to develop models to estimate the average monthly and annual air temperatures (maximum and minimum) for the site of the State of MS. The linear multiple regression technique is adopted in this study. Temperature data from 1978 to 2018 were used, corresponding to 78 meteorological stations on the website of the State of MS. Geographical coordinates (latitude, longitude and altitude) were used as predictor variables for the models, and monthly and annual extreme temperatures (Tmax, Tmin) models were fitted. The regression models used in the study were statistically tested (α ≤ 0.01). The models of mean annual Tmin and mean annual Tmax obtained adjusted determination coefficients (R2adj) of 81.2% and 74.9%, respectively. The monthly average temperature models showed adjusted coefficients of determination between 0.69 and 0.90 for Tmax and from 0.71 to 0.86 for Tmin. Another method used to validate our results, the digital elevation model for the State of MS, obtained through a Shuttle Radar Topography Mission radar image. The obtained results fitted well with these of the annual and monthly models for extreme temperatures. The temperature models used in the study are duly suitable to predict air temperature in all sites in the State of MS.

This present study aims to evaluate the stratospheric ozone that was continuously measured during 2016 over Campo Grande, the capital of South Mato Grosso state, Brazil. To determine the best-adjus...

Resumo No presente trabalho, a comparação de três dos principais métodos de estimativa da radiação solar foi realizada: modelos empíricos, Redes Neurais Artificiais (RNA) e Máquina de Vetores de Suporte (SVM). Quatro modelos empíricos considerados clássicos foram calibrados e validados para a estimativa da radiação solar UV horária em Botucatu, Estado de São Paulo, Brasil. Mantendo os modelos empíricos como referências em relação a precisão e variáveis de entrada utilizadas, os desempenhos da RNA e da SVM foram avaliados. Por meio dos parâmetros estatísticos Erro Médio Padrão (MBE) e Erro Médio Absoluto (MAE) foi confirmada a superioridade da SVM em relação à RNA e aos modelos empíricos. Comparada à RNA, a SVM foi capaz de gerar melhores resultados utilizando um número menor de variáveis de entrada. Dentre todos os métodos de estimativa testados, a SVM utilizando o conjunto de variáveis de entrada {UV0, KT} é considerada a melhor alternativa devido ao pequeno número de variáveis de entrada utilizadas e pela relativa acurácia. Abstract In the present paper, the comparison of three of the main estimation methods of solar radiation was performed: empirical models, Artificial Neural Network (ANN) and Support Vector Machine (SVM). Four classical empirical models were calibrated and validated in order to estimate hourly solar UV data in Botucatu, São Paulo State, Brazil. Taken the empirical models as reference of accuracy and set for input variables, the performance of ANN and SVM were assessed. Through the statistical parameters Mean Bias Error (MBE) and Mean Absolute Error (MAE) was confirmed the superiority of the SVM over the ANN and empirical models. The SVM is capable to generate better results than ANN using a less number of input variables. Among all estimation methods, SVM using the set of input variables {UV0, KT} is considered the best alternative due to the smaller number of input variables and relative precision.

The most significant and influential meteorological element in environmental conditions and human activities is precipitation. The objective of this study was to adjust eight probability distributions to monthly, seasonal and annual rainfall data in the Pantanal of Mato Grosso do Sul, Brazil, using a time series of data (1983-2013) by the National Meteorological Water Agency (ANA). The performance evaluation of different probability distribution models was assessed by the quality of fit of the selected probability distributions for precipitation data. Quality tests as chi-square, Kolmogorov-Smirnov (KS) and Anderson-Darling (AD), the information criteria as Akaike (AIC) and the Bayesian criterion (BIC) were used. Then the mean root square error (RMSE) and the coefficient of determination (R2) were applied. The analyzes were made monthly, annually and by seasons. The 3-parameter Lognormal distribution performs the best for all twelve months and provides the best-fit to the monthly rainfall data. Thus characterizing a dry period that runs from May to September and a rainy period between the months of October and April, it was observed that the 3-parameter Lognormal distribution has best adjustment for spring and summer, and for winter and autumn the 2-parameter Gamma and 3-parameter Gamma distribution performed better. For annual observations, the function that best fits is 3-parameter Weibull distribution. The most significant and influential meteorological element in environmental conditions and human activities is precipitation. The objective of this study was to adjust eight probability distributions to monthly, seasonal and annual rainfall data in the Pantanal of Mato Grosso do Sul, Brazil, using a time series of data (1983-2013) by the National Meteorological Water Agency (ANA). The performance evaluation of different probability distribution models was assessed by the quality of fit of the selected probability distributions for precipitation data. Quality tests as chi-square, Kolmogorov-Smirnov (KS) and Anderson-Darling (AD), the information criteria as Akaike (AIC) and the Bayesian criterion (BIC) were used. Then the mean root square error (RMSE) and the coefficient of determination (R2) were applied. The analyzes were made monthly, annually and by seasons. The 3-parameter Lognormal distribution performs the best for all twelve months and provides the best-fit to the monthly rainfall data. Thus characterizing a dry period that runs from May to September and a rainy period between the months of October and April, it was observed that the 3-parameter Lognormal distribution has best adjustment for spring and summer, and for winter and autumn the 2-parameter Gamma and 3-parameter Gamma distribution performed better. For annual observations, the function that best fits is 3-parameter Weibull distribution.

O trabalho descreve a obtencao e a validacao de equacoes para estimar as fracoes solares global (KG NIR ) e difusa (K d NIR) do espectro infravermelho em funcao das fracoes solares global (Kt) e difusa (K d ) do espectro total. As medidas das irradiacoes global e difusa foram do periodo de 2003 a 2006 em Botucatu/SP/Brasil. Na modelagem estatistica, os valores medios de KG NIR e K d NIR foram equacionados em funcao de Kt por meio de regressao linear e polinomial, respectivamente. Similarmente, os valores medios de K d NIR em funcao de K d por meio de regressao polinomial. Os indices estatisticos de validacao (rMBE e rRMSE) foram utilizados para avaliar o desempenho dos modelos. Os coeficientes de determinacao obtidos para as tres correlacoes em funcao de Kt e K d foram superiores R 2 = 0,960. Os indicativos estatisticos obtidos na validacao: KG NIR (rMBE = 2,8% e rRMSE = 3,8%) e K d NIR (rMBE = 1,0% e rRMSE = 13,7%) mostram que as equacoes podem ser utilizadas nas estimativas de KG NIR e K d NIR em funcao de Kt, respectivamente, com precisao e exatidao. Similarmente, K d NIR (rMBE = 0,5% e rRMSE = 10,7%) na estimativa de K d NIR em funcao de K d .

A evapotranspiração é um fenômeno físico que promove a complexa transferência de água para a atmosfera por meio da relação entre o balanço hídrico climatológico, evaporação da água da superfície e transpiração das culturas agrícolas. É um parâmetro importante para otimização da gestão de recursos hídricos e planejamento da irrigação. Obter medidas confiáveis de evapotranspiração é uma tarefa complexa, pois depende de variáveis não disponíveis em algumas localidades. O objetivo deste trabalho foi aplicar a técnica multivariada de Análise de Componentes Principais para preenchimento de dados faltantes e propor modelos mais simples na estimação da evapotranspiração de referência para o Planalto Ocidental Paulista, comparando-os com o modelo de Penman-Monteith. Foi aplicado um procedimento baseado na Análise de Componente Principais para reconstrução da base de dados meteorológicos do período de 2013 a 2017 de 30 estações meteorológicas automáticas do Planalto Ocidental Paulista, localizado a noroeste do Estado de São Paulo, Brasil. Posteriormente foi realizada uma análise exploratória das variáveis climáticas para verificar o agrupamento de variáveis climáticas mais relevantes nos processos físicos da evapotranspiração. Esses agrupamentos foram a base para a construção de diferentes modelos de estimação da evapotranspiração de referência por meio de Regressões Múltiplas. Os resultados mostraram os melhores desempenhos dos modelos EToRLM4 (rRMSE = 5,23%) e EToRNLM4 (rRMSE = 6,39%). Os valores dos indicativos estatísticos da base de validação de RLM4 e RNLM4 indicam que ambos os modelos de regressões múltiplas podem ser utilizados na estimativa da evapotranspiração de referência.

Global solar irradiation (Hg) is a variable of great importance for different applications. Despite the great advance in the latter, Hg measurements are not readily available in many places. So, several models with different input variables have been developed and used to estimate Hg worldwide. Estimating Hg from models that use air temperature offers an interesting alternative in the absence of measurements. Four groups of models were used in this study: empirical models group (M) with ten models, linear regression models group (RM) with six models, artificial neural network (ANN) models, and support vector machine (SVM) models groups with six models for each one. All the methods take into account different air temperature combinations as input variables (Tmax, Tmin, ΔT, and Tmed). These models were applied to Campo Grande, Brazil, with 10 years data (2005–2017) in order to adjust and evaluate the Hg. The models were evaluated based on four statistical indexes: RMBE, RRMSE, MAPE, and WAI. The models were adjusted and validated using statistical techniques. The results show that the air temperature is an important entry point. Estimates with empirical models perform well (RMBE between 2.84 and 8.68%; RRMSE between 17.78 and 22.93%; MAPE between 17.23 and 23.31%; WAI between 0.85 and 0.90), with errors less than or equal to ANN models (RMBE between 13.94 and 22.67%; RRMSE between 25.37 and 34.95%; MAPE between 29.26 and 49.51%; WAI between 0.51 and 0.81) and SVM (RMBE between 5.80 and 10.15%; RRMSE between 18.87 and 30.34%; MAPE between 19.49 and 42.14%; WAI between 0.55 and 0.884). The regression models had errors with variations similar to the empirical and SVM models. The results show the need to adjust the local coefficients and further studies to further consolidate the models.

Made available in DSpace on 2021-06-25T11:11:04Z (GMT). No. of bitstreams: 0 Previous issue date: 2021-06-01 The interest of the direct normal irradiation (DNI) estimation is important in the evaluation of the solar potential and, consequently, for data correction and expansion of the historical series. In this study, a review of the performance of 16 models of radiative transfer was performed. These models are used to estimate DNI on a clear day in Botucatu/SP region located in Brazil. The revised models are categorized into two classes: simple models (11 models: ASH, MAJ, ALLEN, GH, P1, HLJ, FU, KU, H1, IP, and INC) and complex models (five models: BIRD, IQ, MRM5, P2, and YANG). The evaluation methodology used here is consistent with the literature. The input parameters were estimated and a statistical analysis using relative-mean-bias-error (rMBE), root-mean-square-error (rRMSE), and mean absolute percentage error (MAPE) indicators were performed to validate those models. The results indicate that the complex models (that require more atmospheric inputs) generally performed better than simpler models. Despite the consistent limitations in the use of estimated parameters, the performance of the models can be considered satisfactory. The best performances are highlighted for models MRM5 and YANG. Simple models ASH and IP performed similar to complex models. These results were confirmed using frequency distribution and the cumulative frequency analysis. These results are important for engineers of solar systems to use the best model and select the most suitable locations for installing a small or large solar plant. Faculty of Agronomic Engineering UFPA, Rua 9 Coronel Jose Porfırio, n° 2515 Saõ Paulo State University UNESP Federal University of Mato Grosso do Sul Department of Mechanical Engineering University of M'Sila Saõ Paulo State University-UNESP Saõ Paulo State University UNESP Saõ Paulo State University-UNESP

O conhecimento do potencial solar é considerado uma informação essencial no dimensionamento e simulação do desempenho de sistemas que convertem radiação solar em energia elétrica ou calor disponível em fluídos, bem como em estudos agrícolas voltados ao manejo de irrigação e avaliação da produtividade potencial. Nem sempre as informações sobre a radiação solar incidente estão disponíveis e, por isso, torna-se importante o desenvolvimento de métodos capazes de estimar com acurácia a radiação solar. O presente estudo avaliou o desempenho de diferentes métodos de estimativa da radiação solar global (RG) em ambiente subtropical úmido brasileiro, a fim de determinar qual deles apresenta maior acurácia. As bases de dados meteorológicos utilizadas foram registradas pela estação meteorológica automática pertencente ao Instituto Nacional de Meteorologia (INMET) e instalada em Iguape, município localizado na porção paulista do Vale do Ribeira. Modelos empíricos foram calibrados e tomados como referência na avaliação do desempenho da Máquina de Vetores de Suporte (SVM) utilizando diferentes conjuntos de variáveis de entrada. Dentre os métodos testados, a SVM utilizando o conjunto de varáveis de entrada {R0, Tmax, Tmin, es max, es min} apresentou o melhor desempenho.Palavras-chave: modelos empíricos; Vale do Ribeira; SVM. ESTIMATING DAILY GLOBAL RADIATION FROM COMMON METEOROLOGICAL VARIABLES IN A HUMID SUBTROPICAL ENVIRONMENT ABSTRACT: Knowledge of solar potential is essential in the design and simulation of systems that convert solar radiation in electric power or available heat in fluids, as well as in agricultural studies focused on irrigation management and potential productivity assessment. Oftentimes, information about solar radiation are not available and, therefore, it becomes important the development of methods for obtain reliable data of solar radiation. The present study evaluated the performance of different methods of estimating global solar radiation (RG) in a Brazilian humid subtropical environment in order to determine which is more accurate. The meteorological databases used were registered by the automatic meteorological station belonging to the National Institute of Meteorology (INMET) and installed in Iguape, a municipality located in the São Paulo portion in the Ribeira Valley. Empirical models were calibrated and taken as reference in the performance evaluation of the Support Vector Machine (SVM) using different sets of input variables. Among the tested methods, the SVM using the set of input variables {R0, Tmax, Tmin, esmax, es min} presented the best performance.Keywords: empirical models; Ribeira Valley, SVM.

This study describes correlations between global, diffuse and direct daily solar radiation of the total spectra and global, diffuse and direct daily solar radiation of the infrared spectra (NIR). The database was obtained in the period from 2003 to 2006 in Botucatu / SP / Brazil. Through statistical analyses, values of the three total and seasonal average irradiations in both spectra were determined, while a climate analysis evaluated the effects of cloudiness, water vapor and aerosols on the seasonality of the irradiations. Linear regression equations (y = a*X) were proposed to estimate global (HGNIR), diffuse (HdNIR) and direct (HDNIR) irradiations of the infrared spectrum as a function of the global (HG), diffuse (Hd) and direct (HD) irradiations of the total spectrum in all sky cover conditions. Determination coefficients (R2) close to 1 show that the irradiations of the global and NIR spectra are statistically well correlated. The angular coefficients quantify the values of incident infrared fractions as: KGNIR = (HGNIR / HG) = 46.1%; KdNIR = (HdNIR / Hd) = 38.2% and KDNIR = (HDNIR / HD) = 51.1%. Statistical indicators in the validation: MBE less than 4.6%; RMSE equal to 4.3%, 17.8% and 7.3% and Willmott's “d” values higher than 0.978, respectively, indicate precision, accuracy and high agreement on the adjustment between estimated and measured values for the three equations. In relation to values of the annual fractions, the seasonal analysis of the estimate equations shows that the KGNIR value is lower in the summer (44.8%) and spring (45.1%) and higher in the autumn (47.6%) and winter (48.0%); KdNIR value is the same in the summer (38.2%) and higher in the spring (40.1%) and lower in the autumn (34.9%) and winter (33.9%); KDNIR value is lower in the summer (49.0%) and spring (49.3%) and higher in the autumn (52.7%) and winter (53.9%). The analysis of the estimate equations considering sky covers shows that KGNIR value is lower in cloudy sky (I), partially cloudy sky with prevalence of diffuse sky (II) and partially cloudy sky with prevalence of clear sky (III) in 43.4%, 45.0% and 45.8% respectively, and higher in clear cover (IV) 46.9%; KdNIR value is higher in sky covers I and II: 41.0% and 40.7%, respectively and lower in sky covers III and IV: 35.9% and 25.0%, respectively; KDNIR value is a little higher in sky covers I and IV, both with 51.4% and a little lower in sky covers II and III both with 50.6%.

In this study, attenuation of direct normal solar irradiance (DNI) in Botucatu / Sao Paulo, an area under the influence of local and adjacent agricultural burning, is expressed using the Linke's turbidity factor (TL) in the period from 1996 to 2008. Two methodologies represented as TL Dj and TL Li were used. Temporal variability (hourly average for the season and monthly average) is presented. Turbidity was correlated with wind speed and air temperature. Frequency distribution and cumulative frequency are analyzed to determine turbidity predominance levels in the local atmosphere. Optical depth information of aerosols at 550 nm (AOD 550nm ) and water vapor were obtained by the Terra satellite using the MODIS sensor. The highest degree of DNI transmission is observed in the morning. Close to solar noon, transmission is smaller (greatest TL value). Diurnal TL variability is more evident in the hot period than in the cold period. May and June were the months of lowest DNI attenuation (highest atmospheric transparency). The highest DNI attenuation occurs in spring (TL Dj = 4.22 ± 0.05 and TL Li = 4.65 ± 0.06) and summer (TL Dj = 4.27 ± 0.14 and TL Li = 4.69 ± 0.15). Wind speed and air temperature were positively correlated with TL. In > 28% of hours of clear sky, turbidity exceeded the value of 4.0. The region of Botucatu seems to be influenced by water vapor and aerosols from different origins. This study concludes that these factors significantly reduce DNI incidence on the surface, with higher atmospheric transparency in the cold period and lower atmospheric transparency in the warm period.

This study evaluates the estimation of hourly and daily normal direct irradiation (H b ) using machine learning techniques (ML): Artificial Neural Network (ANN) and Support Vector Machine (SVM). Time series of different meteorological variables measured over thirteen years in Botucatu were used for training and validating ANN and SVM. Seven different sets of input variables were tested and evaluated, which were chosen based on statistical models reported in the literature. Relative Mean Bias Error (rMBE), Relative Root Mean Square Error (rRMSE), determination coefficient (R 2 ) and “d” Willmott index were used to evaluate ANN and SVM models. When compared to statistical models which use the same set of input variables (R 2 between 0.22 and 0.78), ANN and SVM show higher values of R 2 (hourly models between 0.52 and 0.88; daily models between 0.42 and 0.91). Considering the input variables, atmospheric transmissivity of global radiation (kt), integrated solar constant (H sc ) and insolation ratio (n/N, n is sunshine duration and N is photoperiod) were the most relevant in ANN and SVM models. The rMBE and rRMSE values in the two time partitions of SVM models are lower than those obtained with ANN. Hourly ANN and SVM models have higher rRMSE values than daily models. Optimal performance with hourly models was obtained with ANN4 h (rMBE = 12.24%, rRMSE = 23.99% and “d” = 0.96) and SVM4 h (rMBE = 1.75%, rRMSE = 20.10% and “d” = 0.96). Optimal performance with daily models was obtained with ANN2 d (rMBE = −3.09%, rRMSE = 18.95% and “d” = 0.97) and SVM2 d (rMBE = 0.60%, rRMSE = 19.39% and “d” = 0.97). ANN and SVM models improved H b estimations as compared with other results from the literature. SVM has better performance than ANN to estimate H b , and it should be the first option of choice.

The Angstrom–Prescott model called M1 together with ten modified versions, all based on the sunshine duration were adjusted to estimate the daily global and monthly averaged solar irradiation for some sites in the hinterland of Alagoas State in the eastern coast of the Northeastern Brazil. The models were adjusted with meteorological data from 2007 to 2010 and their skills were analyzed using: the Mean Bias Error, Root Mean Square Error and Willmott’s Index of Agreement. The results indicate that the fitted coefficients depend on the geographical coordinates, altitude and local microclimate with 15% differences among the coefficients and estimates. The largest errors are observed in the regions with more cloudiness. Mean Bias Error and Root Mean Square Error for the daily evaluation of models M1, M9 and M11 were similar, with high values of Willmott’s Index. The daily estimates obtained with models M1 and M11 did not differ more than 5%. Models M9 and M11 showed a better performance than that of M1 on a monthly basis. Finally, models M1 and M11 yielded the best results and due to their efficiency and simplicity are recommended to estimate the daily and monthly solar irradiation where sunshine duration data are available.

Made available in DSpace on 2018-12-11T17:12:09Z (GMT). No. of bitstreams: 0 Previous issue date: 2017-07-01 Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) This study describes the comparative study of different methods for estimating daily global solar irradiation (H): Angstrom-Prescott (A-P) model and two Machine Learning techniques (ML) – Support Vector Machine (SVM) and Artificial Neural Network (ANN). The H database was measured from 1996 to 2011 in Botucatu/SP/Brazil. Different combinations of input variables were adopted. MBE, RMSE, d Willmott, r and r2 statistical indicators obtained in the validation of A-P and SVM and ANN models showed that: SVM technique has better performance in estimating H than A-P and ANN models. A-P model has better performance in estimating H than ANN. Department of Rural Engineering - FCA UNESP Agriculture College - UFPA Experimental Campus – UNESP Department of Rural Engineering - FCA UNESP Experimental Campus – UNESP

The present study assesses the performance of nine empirical models: the models of Bristow & Campbell and Hargreaves & Samani (together with their modified versions) in estimating the daily and monthly solar irradiation using just extraterrestrial solar irradiation and air temperature extremes (maximum and minimum) as input data. Two schemes to calculate the air temperature amplitudes (Δ T 1 and Δ T 2 ) were used. The data used in this study cover the period from 2007 to 2009 and were collected at eight solarimetric stations in Alagoas State (Northeastern Brazil); three are located in the interior, two in the hinterlands and three in the humid/coastal zones. Statistical parameters were used to evaluate the model performance. The estimates obtained with the Δ T 1 scheme are better than those using the Δ T 2 scheme for the interior (1.10%) and hinterlands (2.50%). The daily (0.160–0.201) and monthly (0.158–0.199) values of the coefficients of the original Hargreaves and Samani model did not show significant differences among them; this was not the case of Bristow and Campbell model. Have a special from the coastline (thermal amplitude, humidity and cloudiness) and altitude (bulk thermal capacity and optical depth of the atmosphere). On the daily basis, the original model of Hargreaves & Samani yields better estimates than those obtained with the Bristow & Campbell model: 2.30% (interior) and 5.20% (hinterlands). The latter had a better performance mainly for the sites along the humid/coastal zone (10.20%).

Sete modelos empíricos (M1 a M7) baseados na temperatura do ar, foram ajustados e avaliados para estimar a irradiação solar global média horária mensal (Hgh) em quatro localidades do Estado de Alagoas, Brasil. Os dados foram medidos em estações solarimetricas automáticas situadas em Arapiraca (9°47’54,8”S e 37°36’23,4”W), Palmeira dos Índios (9°24’19,9”S e 36°39’22,7”W), Maceió (10°01’29,1”S e 36°16’29,1”W) e Coruripe (9°28’29,1”S e 35°49’43,6”W), entre 2007 e 2009. Para indicar o desempenho dos modelos utilizaram-se os indicativos estatísticos: desvio das médias (MBE), raiz quadrada do desvio quadrático médio (RMSE), “d” de Willmott e teste-t. Os coeficientes ajustados (β1 e β2) dos modelos foram dependentes da região e dos efeitos climáticos locais. Os coeficientes β2 dos M1, M2 e M3 foram significativos a 5%, enquanto que, o coeficiente β1 de M4 e M5 resultaram em comportamentos semelhantes. Os coeficientes β1 e β2 de M6 e M7 apresentaram elevadas oscilações. Com RMSE médio de 0,29 ± 0,15 MJ m-2h-1, o melhor ajuste para região de Arapiraca foi com M7. O melhor ajuste para as condições climáticas nas regiões de Palmeira dos Índios, Maceió e Coruripe foi com M4. Estes resultados mostram a eficiência dos modelos empíricos em prever Hgh. Palavras-chave: Modelos empíricos, Bristow e Campbell, Hargreaves e Samani, condições climáticas. MODELLING OF MONTHLY HOURLY GLOBAL SOLAR IRRADIATION (Hgh) TO FOUR SITES IN ALAGOAS/BRAZIL ABSTRACT In this paper was used seven empirical models (M1 to M7) based on temperature of air and they were fitted and validated to estimating the monthly hourly of global solar irradiation (Hgh) in four sites in Alagoas State, Brazil. The data were measured in automatic solarimetric stations situated in Arapiraca (9°47’54.8”S and 37°36’23.4”W), Palmeira dos Índios (9°24’19.9”S and 36°39’22.7”W), Maceió (10°01’29.1”S and 36°16’29.1”W) and Coruripe (9°28’29.1”S and 35°49’43.6”W), from 2007 to 2009. To indicate the models performance were used: Mean Bias Error (MBE), Root Mean Square Error (RMSE), “d” of Willmott and t-test. The fitted coefficients (β1 and β2) of models were dependents of region and local climatic effects. The coefficients β2 of M1, M2 and M3 were significantly different at the level of 5%. The β1 coefficient of M4 and M5 resulted in similar oscillations. The β1 and β2 coefficients of M6 and M7 presented high oscillations. With RMSE average of 0.29±0.15 MJ m-2 h-1, the best fitted model for the Arapiraca region was M7. The best fitted model for the climatic conditions in the region of Palmeira dos Índios, Maceió and Coruripe were M4. The M4 generated the best result among the evaluated models. These results showed the efficiency of the empirical models to predict Hgh. Keywords: Empirical models, Bristow and Campbell, Hargreaves and Samani, Climatic conditions.DOI: http://dx.doi.org/10.14583/2318-7670.v02n02a04

A statistical estimate model for the anisotropic diffuse fraction as a function of clearness index KT is proposed to estimate hourly, daily and monthly diffuse irradiations. Global, diffuse and direct solar irradiances were provided by the Laboratory of Solar Radiometry of Botucatu-UNESP (latitude 22.9° South, longitude 48.45° West, altitude 745 m). The period assigned for the study comprised the years 1996 to 2002. Global solar irradiance was measured by an Eppley PSP pyranometer, direct irradiance by an Eppley Nip pyrheliometer and diffuse irradiance by an Eppley PSP pyranometer under the Melo-Escobedo-Oliveira shadowring (radius of 40 cm and width of 10 cm). Isotropic and KT corrections were applied in diffuse irradiance. The proposed model was compared to classic models reported in the literature, with good results according to the MBE and RMSE statistical indicators for hourly, daily and monthly partitions, respectively. The results showed that the inclusion of KT correction improved the performance of the shadowring MEO according to MBE values for the three partitions: hourly, reduction from -7.21% to -1.74%; daily from -4.70% to 0.88% and monthly from -6.58% to -1.18%.