Applicability of Agmerra for Gap-Filling of Afghanistan in-situ Temperature and Precipitation Data

Climate change (CC) is one of the most important concerns for mankind in the current century. Increasing CO2 concentration and the proof of the greenhouse effect theory in which the type and composition of atmospheric gases which influence the earth temperature, are among undeniable facts makes the future climate change more possible. Impacts of Global warming on hydrological cycles and precipitation patterns would be more prominent in arid and semi-arid regions of the earth. For the arid and semi-arid nature and the poverty more fraction of Afghanistan suffer from, it is likely that the impacts of CC on the country will be more intense. This is while there is no credible and reliant research addressing the impacts of CC on agriculture and food security sector of Afghanistan. Studying the impacts of CC on agriculture, future changes in agroclimatic indices and application of crop growth simulation models intensively require a precise and adequate sets of meteorological data. Because of many reasons, Afghanistan's historical meteorological data coverage is really weak. In this research the applicability of AgMERRA as a gauge-satellite based dataset for filling the Afghanistan in-situ meteorological gaps is evaluated via goodness of fit measures, patterns of seasonal changes and the probability distribution functions.

This study is conducted on four major stations of Afghanistan (Kabul, Herat, Mazar Sharif and Qandahar in the east, west, north and south of the country, respectively) (Fig. 1 and table 1) which had the best in-situ meteorological data coverage. Observed Maximum (Tmax) and Minimum temperature (Tmin) and precipitation (PRCP) data is collected via Afghanistan Meteorological Authority (AMA) or other sources. AgMERRA database downloaded with .nc4 format and extracted with R statistical software or Panoply ver. 4.8.4, dependently. Then five goodness of fit (GOF) measures (RMSE, NRMSE, MBE, R2 and d) are calculated according to the equations 1 to 5. There are different norms and indices to measure the validity of a models, some based on Pearson correlation coefficient (R and R2) which indicate the degree of correlation between observed and predicted data but have some amounts of sensitivity to extreme values (outliers). Although, many other measures are considered to overcome the weaknesses but it is hard to distinguish the best.

The results of this research indicated the good potency, effectiveness and ability of AgMERRA for gap-filling of in-situ meteorological data and producing spatiotemporal data series. Several studies in this area have almost the same results. It is reported that AgMERRA is the most applicable dataset for reflecting precipitation data comparing with ERA-Interim, ERA-Interim/Land and JRA-55 datasets. Comparisons via NRMSE shows great (>10%) and good (>20%) amounts in all stations and temporal scales. Among other stations, Mazar Shrif showed the best conformity between AgMERRA and observed data, while Kabul station had the weakest, probably due to complex topographic situation of the Kabul airport station. The amounts of R2 for predicting temperature (Tmax and Tmin) were more than 0.86 in daily, 14-days and monthly temporal scales. The lowest amount of the coefficient of determination was obtained at Qandahar station for Tmean in daily temporal scale (R2=0.8) and the highest amount obtained for daily Tmax at Mazar Sharif station (R2=0.947). R2 for daily PRCP were inadequate, but increasing to adequate amounts in 14-days and monthly temporal scales. The highest spatiotemporal amount of Tmax,Tmin and Tmean was obtained in daily scale and the lowest amount was obtained for Tmean (1.8 and 0.9, respectively). The Index of agreement (d), also had adequate amounts for 14-days and monthly PRCP (>0.87). The amount of MBE for precipitation in Herat, Mazar Sharif and Kabul stations were negative, while it was positive in Qandahar station with a hot and dry climate. AgMERRA could show a good compliance with changes of observed seasonal patterns, however, some amount of over and under-estimates are obvious especially for Kabul station. This compliance with in-situ observed patterns was acceptable for daily temporal scale, although AgMERRA was unable to predict some of the fluctuations in probability distribution composition (with the range of 1 °C), especially fot Tmax and Tmin, but fot Tmean the fluctuations simulated well.

According to the results of the study, AgMERRA showed an acceptable potency to simulate the in-situ meteorological data in four major studied stations of Afghanistan. According to the stochastic nature of PRCP, the variable showed the weakest results in daily temporal scale but acceptable in 14-days and monthly. Given the weak coverage of in-situ meteorological data of Afghanistan, AgMERRA could be a valid dataset for producing well scaled spatiotemporal data series to be used in agroclimatic, CC and crop growth modeling studies.