Time Series Analysis of Environmental Data by using ARIMA and LSTM models for Precision Agriculture

Abstract Appropriate utilization of water and its distribution within the fields for irrigation purpose within the agricultural domain is of paramount importance. It not only helps in retaining the natural resource, but also eliminates various environmental risks associated, such as crop damage, soil fertility loss, etc. This estimation is can be predicted in advance after evaluating all the associated agricultural parameters mainly soil moisture, humidity, luminosity, and atmospheric pressure thoroughly. Farmers often practice water distribution within the fields without evaluating its precise requirement. With an aim to procure this natural resource and evaluate its distribution as per the requirement of the fields, the present work endorses evaluation of agricultural parameters through Autoregressive Integrated Moving Average (ARIMA) and Long short-term memory (LSTM) models based on Time Series analysis. Both these models predicts the values based on following two moving average methods, Simple Moving Average (SMA) and Exponential Moving Average (EMA). For obtaining values related to agricultural parameters, Libelium's Waspmote Plug & Sense, a hardware device has been incorporated within the study. The evaluated results depicted variation in results, in contrast to the actual practices begin followed by farmers. The intermediate diminution in error rates attained by ARIMA were found within the range of 74% - 77% in contrast to LSTM, indicating its preeminence. Based on the existing results, future possible values for various attributes can be estimated and on the basis of estimation, distribution of water for irrigating the fields can be scheduled according to requirement, resulting in better fields yields.