2.2 Data sources and quality control
Daily meteorological dataset of five parameters including mean temperatures (°C), rainfall (mm), relative humidity (%), wind speed (km day−1), and sunshine (h day−1) were obtained from the Bangladesh Meteorological Department (BMD). Though BMD has 43 weather stations throughout the country, the majority of the stations are unevenly distributed, some of them are newly equipped and few stations have no longer datasets. In this study, we used 18 weather stations dataset that measures meteorological parameters essential for calculating ETref (Fig. 1). The problem arises when missing data exist. If the missing data was observed in any year >2%, the data of that particular year was omitted. When the data comprised of >2% missing values (Rubin et al., 2007), the supplied estimated data might have inaccurate, subsequently inaccurate outputs may provide. Rahman and Islam (2019) also employed the threshold level for showing missing precipitation in Bangladesh. Other studies applied to the threshold of 2% for data quality control purposes (Vieira, 2017). After initial scrutiny, the availability of weather datasets at various sites was observed to range between 30 to 48 years. We used 37 years of daily climatic records during 1980–2017. It is noted that for a good representation of the data stations were selected which are uniformly scattered all over the country. However, quality control of the dataset was primarily undertaken thoroughly by checking namely, positive values of parameters, for example, Tmin is lower than Tmax, and humidity is less than 100%. Time series and homogeneity tests of the dataset were conducted to exhibit any anomaly in the dataset (Islam et al. 2020b). Besides, the quality appraisal of the dataset of the selected stations was performed using a subjective double mass curve procedure and the objective Standard Normal Homogeneity Test (SNHT) (Fig. S1). Each dataset was detected consistently to fit for quantitative appraisal. The SNHT and K-S tests among various subsets of datasets depicted no significant difference in those datasets for all the parameters at the 95% confidence level (Table S1). Hence, the quality of the datasets was achieved suitable for further appraisal.