Modelling net radiation at surface using “in situ” netpyrradiometer measurements with artificial neural networks

https://doi.org/10.1016/j.eswa.2011.04.231Get rights and content

Abstract

The knowledge of net radiation at the surface is of fundamental importance because it defines the total amount of energy available for the physical and biological processes such as evapotranspiration, air and soil warming. It is measured with net radiometers, but, the radiometers are expensive sensors, difficult to handle, that require constant care and also involve periodic calibration. This paper presents a methodology based on neural networks in order to replace the use of net radiometers (expensive tools) by modeling the relationships between the net radiation and meteorological variables measured in meteorological stations. Two different data sets (acquired at different locations) have been used in order to train and validate the developed artificial neural model. The statistical results (low root mean square errors and mean absolute error) show that the proposed methodology is suitable to estimate net radiation at surface from common meteorological variables, therefore, can be used as a substitute for net radiometers.

Highlights

Neural networks can be use to replace net radiometers by a model of the net radiation. ► The proposed methodology is suitable to estimate net radiation from meteorological variables. ► A sensitivity analysis has been done to obtain the importance of the each variable.

Introduction

Net radiation is a fundamental parameter that governs the climate of the lower layers of the atmosphere and it depends critically on the structure and composition of the atmosphere and the presence of clouds, in addition to surface features such as albedo, emissivity, temperature, humidity and thermal properties of the underlying soil. Thus net radiation is a fundamental quantity for analyzing the evolution of climate, from both local and global perspective. It is the driving force of physical and biological processes such as evapotranspiration, the latter being used to optimize the quality and yield of crops, water resources planning, weather forecasting, etc. (Bennie et al., 2008, Ji et al., 2009, Li et al., 2009). Despite its importance, the net radiation is measured only in a very few number of standard weather stations because net radiometers are expensive instruments and require constant care in the field, so that the net radiation measurements can be reliable. Hence, this quantity is difficult to obtain due to the cost of net pyrradiometers. This paper presents a methodology for modeling net radiation using artificial neural networks. After an initial period collecting data in order to train the network with real samples, the neural network model can be used as an estimator of net radiation samples for a given area without using net radiometers at all times. The strategy here is to train the neural network model using the net radiation collected “in situ” over a representative period and then use that model and not the net radiometer.

There are a large number of linear and nonlinear models who perform modeling of the net radiation at surface but using as input the incoming solar radiation or the net radiation components separately (downwelling shortwave radiation, reflected shortwave radiation, downwelling and upwelling longwave radiation). But the root of the problem remains; radiometers are needed to obtain these input variables to the model (Alados et al., 2003, Daughtry et al., 1990, Kohsiek et al., 2007). This problem can be avoided by using as input parameters, in the neural networks developed to model the net radiation at surface, the most common meteorological variables collected in the majority of weather stations, around the word, including those that send, daily, information to the Global Telecommunication System (GTS). The GTS is defining as: The co-ordinated global system of telecommunication facilities and arrangements for the rapid collection, exchange and distribution of observations and processed information within the framework of the World Weather Watch citegts. These variables are: wind speed, air temperature, atmospheric pressure and humidity.

The following sections describe the neural model used, the multilayer perceptron. After this, the datasets and the variables involved in the problem will be presented. Finally, we will present the results and conclusions obtained in the study.

Section snippets

Multilayer perceptron

The multilayer percepron (MLP) has been the neural network model used in this study. It consists of some individual process elements called neurons, which are arranged in a series of layers. Fig. 1 shows the structure of these neurons.

This neuron is constituted, in its first part, by a multiplier, which multiplies the inputs by a series of coefficients called synaptic weights. The objective of learning algorithm is to obtain the optimum values for the synaptic weights (Haykin, 2009). In the

Data sets

In order to validate our approach were used two data sets obtained from two surface areas with the same land use (vineyard crop) but with different land cover (vineyard and bare soil). Methodology and sensors employed to data collection are described below.

  • Data set 1 (FESEBAV). The first data set corresponds to data collected during the field campaign called FESEBAV 2007 (Field Experiment on Surface Energy Balance Aspects over the Valencia Anchor Station area) conducted from June 19th to

Results

In order to obtain the best neural network model, the models were trained using two hidden layers (by the Cybenko theorem it is known that two layers are necessary to establish the relationships between two data sets (Haykin, 2009)). The number of hidden neurons in each layer was varied from 2 to 20, these limits were imposed because, in all the tests, never reached the upper limits for the number of neurons using cross-validation. Moreover, since the learning algorithm is a local search

Conclusions

Net radiation measure is important for the analysis and study of climate, but the devices used to do this are very expensive and difficult to manage requiring further constant care in the field. This paper demonstrates the ability of neural models to replace the use of radiometers for the measurement of surface net radiation. Using neural models and conventional weather variables can be estimated net radiation with an acceptable error without using expensive and costly radiometers. A

References (12)

There are more references available in the full text version of this article.

Cited by (0)

This work was supported by the Programme Alβan, the European Union Programme of High Level Scholarships for Latin America, scholarship no. E05D058998BR-Antonio Geraldo Ferreira and also for Remote Sensing Techniques for the Observation of Environmental Parameters in the Valencia Community Autonomous Region for 2007-2009,contract of the Department for Environment, Water, Planning and Housing, General Directorate for Climate Change, Generalitat Valenciana, also this work was supported by the Spanish Ministerio de Educacion y Ciencia under grant TIN2007-61006:Aprendizaje por Refuerzo Aplicado en Farmacocinetica (RLA2AP).

View full text