Estimation of Daily Net Radiation in the Urmia Lake Basin under Clear-Sky Conditions Based on MODIS Data

Introduction
Daily average net radiation (DANR) is necessary for hydrological modeling, water resources management and planning, in particular for prediction of daily evapotranspiration (ET). Recently, some remote sensing based methods have been developed at watershed and regional scales to help researchers estimate DANR and ER under both clear and cloudy conditions. DANR is a critical variable linking estimates of instantaneous latent heat flux (satellite overpass time) from energy balance based models. Radiation modeling based on ground or station data is actually point scale estimation. On the other hands, application of geostatistical methods to produce radiation maps at watershed scale is often associated with remarkable errors due to the limited number of stations. In this study, net daily radiation maps were created based on newly developed geometric radiation model based on land and atmospheric MODIS data in the Urmia Lake Basin (ULB) under clear-sky condition for four selected dates in 2006 and 2007.
Materials and Methods
 In this study, the ULB located in northwestern Iran was selected as the study area that extending in latitude from around 35.67^° to 38.47^° N and in longitude from around 44.22^° to 47.89^° E. The basin has an area of approximately 52700 square kilometers. This study attempts to estimate the DANR maps for four selected days under clear sky. Long et al. (2010) developed a geometric model considering slope and aspect effects to improve parameterization schemes to estimate daily average net shortwave radiation (Rns) and daily average net longwave radiation (Rnl). In present study, the same methodology employed a characteristics model (slope, azimuth and elevation) based on daily MODIS data products and the global ASTER DEM. The DANR is total of the daily net short wavelength radiation and daily net long wavelength radiation. Long wave radiation was estimated using MODIS products including surface albedo, Land Surface Temperature (LST), atmospheric temperature, pressure and emissivity. A four-observation based method using Terra-MODIS and Aqua-MODIS LST data was applied to estimate daily average LST in order to improve the accuracy of estimations of Rnl. In the study, we applied the 24-hour incoming direct solar radiation (S_in24) measurements of Tabriz and Urmia synoptic stations for validation. The RMSE and MAE measures were used for model evaluation. In addition, due to the lack of ground-based Rnl observations, we utilized the FAO (Penman) method to generate corresponding ground data for model validation.
Results and discussion
 In order to compare the amount of incoming S_in24 on flat and slope surface, the whole study area was assumed as a flat and the S_in24 maps were produced. The differences between S_in24 on flat (S_in24,s) and sloping surfaces (S_in24,f) were calculated for all pixels. The histogram plot or frequency of S_in24,s - S_in24,f quantity at different intervals of slope showed that by increasing in slope degree, the frequency of pixels at all slope classes becomes greater. This reflects the importance of the slope for estimating S_in24. In addition, the values of S_in24 at flat are higher than sloping surfaces especially for where slopes < 30 degree. Aspect is another influencing factor on the amount of receiving and absorbed radiation. The RMSE and MAE for observed and estimated S_in24 values were 30.29 and 19w.m^-2, respectively. The models in average overestimate S_in24 about 10.5%. The results showed that the eastern and western aspects receive more daily net radiation during day and sunshine hours or day length has its impact on radiation. For Rnl estimates, the RMSE was 36w.m^-2
Conclusion  
In present study, the Sin24 was parameterized by taking into account the effect of terrain factors, such as slope, azimuth and elevation on direct solar radiation. The sunrise and sunset angles for a given sloping surface (pixel) are different in geometric radiation model. Results indicate that the model has the capability to characterize the variability in Sin24. The incoming solar radiation for flat surface is higher than sloping surface. We founded that southeast and southwest aspects receive greater net radiation. The DANR values decrease with an increase in elevation. The geometric model used in this research needs to be evaluated in different watersheds in Iran without data availability problem especially daily Rns, Rnl and LST measurements, a high number of meteorological stations with proper spatial distribution. This research was done under limited data in the ULB.