Air Temperature Sensor Estimation on Automatic Weather Station Using ARIMA and MLP

Case Study on AWS Serang

Abstract

Surface meteorological quantities are now measured by Automatic Weather Station (AWS). AWS Serang records weather parameters minutely in Banten Province of Indonesia. Air temperature sensor is one instrument of this system. This study aims to design an air temperature sensor estimator model using ARIMA and Artificial Neural Network (ANN) as solution for avoiding loss data. Air temperature sensor on AWS Serang data in August of 2022 period is segmented into training, validating and testing sections. Based on criterion calculation, ARIMA (1,1,5) is simulated. It obtains not more than 0.12 of RMSE,  0.052⁰C of MAE, 0.193% of MAPE and 0.194% of SMAPE. Meanwhile, three different models of MLP ANN for air temperature estimator is also simulated. Input variables include air temperature, relative humidity and solar radiation intensity. Roy model has highest accuracy level for MLP ANN algorithm with 0.048 of RMSE, 0.026⁰C for MAE, 5% of MAPE and 4.83% of SMAPE.  Overall, ARIMA (1,1,5) is better than Roy MLP ANN model in estimating air temperature sensor data on AWS Serang. Nonetheless, both models are properly fulfilling WMO (World Meteorological Organization) accuracy requirements for air temperature measurement.