Recently

an adaptive method called Empirical mode decomposition (EMD) is proposed for signal analysis. It has attracted great deal of attention and been used in many areas successfully since its advent. However

when the signal is contaminated by noise

EMD suffers from the drawback of over decomposition and likely is affected by noise

which severely restricts its applications. In order to solve this problem

an improved version of EMD is proposed. During the first decomposition circle

the original cubic spline interpolation is replaced by the smoothing spline fitting

which can avoid the over decomposition problem and then reduce the disturbance of noise component. Simulations validate the improvement of the new proposed method. Moreover

two real climate data examples show the effective and superiority of the new method for real signals.