Neural network and state-space models for studying relationships among soil properties

Abstract

The study of soil property relationships is of great importance in agronomy aiming for a rational management of environmental resources and an improvement of agricultural productivity. Studies of this kind are traditionally performed using static regression models, which do not take into account the involved spatial structure. This work has the objective of evaluating the relation between a time-consuming and "expensive" variable (like soil total nitrogen) and other simple, easier to measure variables (as for instance, soil organic carbon, pH, etc.). Two important classes of models (linear statespace and neural networks) are used for prediction and compared with standard uni-and multivariate regression models, used as reference. For an oat crop cultivated area, situated in Jaguariuna, SP, Brazil (22º41' S, 47º00' W) soil samples of a Typic Haplustox were collected from the plow layer at points spaced 2 m apart along a 194 m spatial transect. Recurrent neural networks and standard state-space models had a better predictive performance of soil total nitrogen as compared to the standard regression models. Among the standard regression models the Vector Auto-Regression model had a better predictive performance for soil total nitrogen. Key words: soil attributes, prediction models, spatial transect, latent variables REDES NEURAIS E MODELOS DE ESPAÇO DE ESTADOS PARA O ESTUDO DA RELAÇÃO ENTRE PROPRIEDADES DO SOLORESUMO: O estudo da relação entre as propriedades do solo é de grande importância na área agronômica objetivando um manejo racional dos recursos naturais do meio ambiente e um aumento na produtividade agrícola. Tradicionalmente este estudo tem sido realizado usando modelos de regressão estática os quais não levam em consideração a estrutura espacial envolvida. Este trabalho teve o objetivo de avaliar a relação entre uma variável de determinação mais cara e demorada (por exemplo, nitrogênio total do solo) e outras de mais barata e rápida determinação (p.e., carbono orgânico do solo, pH, etc.). Duas importantes classes de modelos (espaço de estados linear e redes neurais) são usadas para predição e comparadas aos modelos de regressão uni-e multivariados aqui usados como referência. Para tal, em uma área experimental cultivada com aveia, situada em Jaguariúna, SP (22º41' S e 47º00' W), amostras de um solo classificado como Latossolo foram coletadas na camada arável ao longo de uma transeção espacial de 194 m, eqüidistantes de 2 m. Os modelos de rede neural recorrente e de espaço de estados padrão tiveram uma melhor performance preditiva da variável nitrogênio total do solo quando comparados aos modelos de regressão padrão. Entre os modelos de regressão padrão o Autoregressivo Vetorial teve um melhor desempenho preditivo da variável nitrogênio total do solo. Palavras-chave: atributos do solo, modelos de predição, transeção espacial, variáveis latentes

Temporal variability of soil water storage evaluated for a coffee field

“…Since the first half of last century, the problem of obtaining representative sampling of agricultural fields has led to the development of new sampling schemes. Initially, scientists based their strategies on classical statistical concepts, which were later complemented with geoestatistics and time-space series analyses, and more recently neural networks (Hills and Reynolds 1969;Mohanty and Mousli 2000;Western et al 2002;Timm et al 2006;Hu et al 2008).…”

Neural networks for predicting breeding values and genetic gains

“…Once the mean squared error of prediction reached an optimal level, training stopped and this led to the best estimates of the network coefficients (Haykin, 1998). This stop criterion is usually applied when the validation process of the ANN is obtained with one validation file, and has been used by several authors (Gianola et al, 2011;Timm et al, 2006;Ventura et al, 2012). Because in this work 120 validation files were used, an alternative was adopted in which the stop criterion was the maximum number of hits obtained by the ANN considering the 120 replicates.…”

Drought Forecasting Using Indices and Artificial Neural Networks for Upper Tana River Basin, Kenya-A Review Concept

“…The hidden neurons unit activation or the output values are fed back into the network as inputs (Figure 8). These models exhibit a dynamic characteristic when modeling data that depends upon different temporal and spatial resolutions [53]. The feedback links layers whose state vary with time, and has adjustable weights.…”