Applications in scientific computing operate with high-volume numerical data and the occupied space should be reduced. Traditional compression algorithms cannot provide sufficient compression ratio for such kind of data. We propose a lossless algorithm of delta-compression (a variant of predictive coding) that packs the higher-order differences between adjacent data elements. The algorithm takes into account varying domain (typically, time) steps. The algorithm is simple, it has high performance and delivers high compression ratio for smoothly changing data. Both lossless and lossy variants of the algorithm can be used. The algorithm has been successfully applied to the output from a simulation application that uses a solver of ordinary differential equations.