GANs-LSTM Model for Soil Temperature Estimation From Meteorological: A New Approach

Abstract

Soil temperature (T s) is a vital meteorological parameter for ecological, physical and biological research. T s estimation is very important for a variety of fields and presents great challenges because the relevant areas have complex characteristics, human activities, and a nonlinear nature between T s and its environmental factors. Hence, a novel model based on long short-term memory (LSTM), is proposed here as an alternative data-intelligence tool. The proposed model designs a novel function that combines LSTM loss with an adversarial term for enhancing the correlation between T s and its environmental factors. For this purpose, we designed a novel function that combines LSTM loss with an adversarial term. The adversarial term encourages the LSTM model to estimate T s , which cannot be distinguished from the observed T s by an adversarial model. In this study, the proposed model is trained using meteorological information from two stations in China (Changbai Mountain and Haibei). The model is constructed using hourly input variables, including air temperature (T a), wind speed (W), relative humidity (RH), solar radiation (SR), and vapor pressure deficit(VPD), while the objective variable is the T s measurement at a 5 cm depth for the period 2003-2005. Different statistical evaluation criteria, including the root mean square error (RMSE), mean absolute error (MAE), Nash-Sutcliffe model efficiency coefficient (NS), Willmott index of agreement(WI) and the Legates and McCabe index (LMI), have been employed to assess the model performance. Through experimentation, the proposed model generally performs superior to the other seven state-of-the-art estimation models (autoregressive integrated moving average model, linear regression, backpropagation neural networks, support vector regression, extreme learning machine, eXtreme gradient boosting and long short-term memory) in T s estimation at a 5 cm depth over the Changbai Mountain and Haibei stations. For this case, the most accurate performance is attained for T s estimation at a 5 cm depth, with the highest values of NS = 0.915, WI = 0.978, and LMI = 0.686, and the lowest values of the relative RMSE and MAE being 2.276 and 1.796, respectively. In accordance with the present results, it is concluded that the proposed model can serve as an alternative approach for estimating T s , while ensuring that an appropriate combination of meteorological inputs is applied to yield an optimal model. INDEX TERMS Data-driven model, soil temperature, LSTM, adversarial training.

A Novel Multichannel Long Short-Term Memory Method With Time Series for Soil Temperature Modeling

“…The advantage of this feature makes LSTM the predictive model that is most widely applied in earth sciences. For example, Li et al [17] integrated GANs with the LSTM network for Ts estimation, and Zhang et al [33] used dropout [34] to avoid LSTM in overfitting for a water table depth estimation task. The above methods perform the estimation without understanding the physical processes of the estimation domain, and the LSTM can capture the correlation between input data and output data automatically.…”

“…However, there are many practical limitations due to the parameterization of physical processes or unresolved scale issues [10]. Recently, machine learning methods and time series methods have been used for estimations in the earth sciences [4,[11][12][13][14][15][16], especially for soil temperature fields [15][16][17][18][19]. The workhorses of classical time series methods are autoregressive (AR) models such as the autoregressive integrated moving average (ARIMA) class of models and their numerous variants.…”

“…ANNs have a strong selflearning ability due to the flexibility of their architecture. Hence, they are widely applied in Ts estimation [17,[21][22][23][24]. For example, Kim et al [25] applied an ANFIS and an MLPNN in daily Ts estimation.…”

“…Zhang et al [33] used the dropout method [34] to avoid LSTM in overfitting for a water table depth estimation task and showed that the proposed model can contribute to a strong learning ability on time series data. Li et al [17] integrated adversarial learning in generative adversarial networks [35] with an LSTM network for Ts estimation and noted that the estimation model based on LSTM outperforms existing machine learning methods. The LSTM models applied in this study, have been successfully applied in Ts estimation due to the ability to model the deterministic part of the time series data.…”

A New Embedded Estimation Model for Soil Temperature Prediction

“…e extreme learning machine is used to predict the soil temperature for improving the accuracy by Feng et al [15]. Furthermore, LSTM has also received attention from researchers [16].…”