Land surface energy exchange simulation based on combined fuzzy sets and neural network approach

Principal variables are land/water and air temperatures and moisture, solar downward irradiance , albedo, reflected and emitted radiances, heat fluxes into ground and atmosphere, latent and eddy heat fluxes . Its values are mainly determined by individual optical, physical and chemical properties of a given homogeneous land/water pattern. The pattern scale , depending heavily on landscape pecularities, could variates from several meters to several kilometres (or even to tens or hundreds km in desert areas). However, there two weak features of these data implementation in conventional numerical models. First issue is that all mentioned variables with the exception of air temperature and moisture are not synoptically observed and become available for users with one month delay at least. Second one is due to substantial departures between observed and modelled diurnal values of these variables (see: Betts et al, Q.J.Roy.Met.Soc.,119, p.975-1001, 1993). Hourly meteorological data (standard data, radiative budget and heat balance variables)of ten year length for several observational stations of Russian north-west region had been used. The data set of simultaneous observations ( for all variables and stations ) allow to determine main features of joint diurnal distributions by means of the fuzzy set constructions. The couples: air temperature and humidity, pressure and humidity correlation amount to 0.4-0.6 and describe only its linear relationships. Corresponding fuzzy set consistent rates jump up to 0.8-0.9 due to appropriate non-linear link description. Two stationary and two transitional diurnal modes were introduced to simulate simplest diurnal cycles. They describe the conservation of warm and cold weather patterns and possible transitions from one of these states to another. Fuzzy logic model was invented for weather pattern recognition, based on hourly observational data. Implementation of neural networks (back propagation algorithm) allow to perform several modelling experiments. Determination of neural network hidden layers and nodes are investigated. Novel feature of our approach was implementation of fuzzy set technique for constructing of the input and output nodes. Problem of optimum neural network configuration (composition and number of nodes in hidden layers ) is discussed. Some experiments on evolutionary modelling of individual variables and several variable groups had been performed for 7 months (217 days : 1986-1992, July). Our aim was to proceed the rate of agreement or disagreement in temporal “travelling” of state vector(s) from one fuzzy set to another due to neural network algorithm simulation. Obtained results indicate close relationship of atmospheric pressure on one hand and solar radiation, air humidity and temperature, on other hand. Agreement rate in evolutionary simulation two-cluster experiments for groups included solar radiation, pressure, air temperature and humidity exceed 0.9 threshold. Using pressure observations or forecast output at any day and several previous days one could derive the next day estimation of radiation, cloudiness and humidity, having 90 % hit rate in two-cluster case. Hence, combined fuzzy logic& neural nerwork simulation model could supply weather and climate models with more accurate data on surface meteorological variables impacted on principal climate parameters .