A neural network MODIS-CERES narrowband to broadband conversion

Monitoring changes of radiative energy budget at the top of the atmosphere requires satellite measurements with good instrument stability and very high accuracy. With climate monitoring in mind, improving instrument stability has been one of the most important objectives of CERES instrument designs. This study is aimed at assessing such instrument stability objectively by investigating the distributions of radiative fluxes ftom satellite measurements of well defined atmospheric objects. With months of CERES observations on TRMM and TERRA, we have developed a strategy to demonstrate the month-to-month stability of CERES instruments by studying the distributions of deep convective cloud albedos as well as outgoing longwave fluxes. Our study shows that both shortwave and longewave radiative flux distributions of CERES measurements for such objects are practically identical ftom month to month. It also shows very little differences among TRMM CERES, TERRA CERES FM I and FM2. This study also intends to provide an objective stability analysis of narrowband instrument, such as MODIS, in terms of energy budget. First, we develop a neural network model which performs MODIS narrowband radiances to CERES broadband radiance conversions. Then, we study the monthly mean flux distributions derived from the narrowband-broadband conversion. Collocated MODIS and CERES cross track scanning data are used for training the multi-level feedforward back-propagating neural network model with special noise handling characteristics. Other information, such as theoretical radiative transfer calculations are adopted for improving the design of the neural network model as well as filling missing data at certain viewing angular geometry.