# [SVC46-P01] Research on observation method of volcanic ash accumulation using spaceborne SAR: Quantitative extraction of decorrelation by volcanic ash-fall

Keywords:volcanic ash, SAR, coherence

Volcanic ash often causes heavy disaster due to debris flow. To mitigate such disaster, rapid observation of the volcanic ash distribution and forecast of debris flow from such observation are important. However, rapid observation of the volcanic ash distribution has been carried out mainly by on-site survey, and it is difficult to observe it rapidly. Then development of more rapid observation method are required. As one of solutions resolving such problem, utilization of satellite remote sensing technique is expected. Then, we are researching on observation method of volcanic ash distribution using spaceborne SAR in “Cross-ministerial Strategis Innovation Promotion Program (SIP)-2nd season” as one of development items. In this development, we use InSAR coherence analysis. In InSAR results, clear fringes are seen if the coherence is high. On the other hand, noise is dominant if the coherence is low, and fringe becomes unclear. Generally, loss of coherence (decorrelation) is caused by difference of radar scatter at the pixel for InSAR pair. In a result from an interferometric pair including the eruption occurrence, decorrelation due to radar scatter change caused by volcanic ash accumulation is generally seen around the crater. If the relation between decorrelation and the amount of volcanic ash accumulation becomes clear quantitatively, the amount of volcanic ash accumulation will be obtained from InSAR coherence analysis. Then, in this development, we estimate decorrelation only due to volcanic ash accumulation, using a temporal coherence change model without volcanic ash effect constructed from InSAR coherence time-series analysis. In this presentation, we mention about such an attempt of quantitative extraction of decorrelation by volcanic ash-fall.

Coherence decreases with time due to a change in radar scattering due to a change in ground surface coverage and shape, a change in water content, and so on. We assume such temporal change the exponential function “g = Aexp(Bt) + C”, where g is the coherence, t is the observation interval of the interference pair, and A, B, and C are constants. Fitting this function to observed coherence in Kuchinoerabu-jima by PALSAR-2, large residuals exceeding 0.2 were obtained. Such large residuals are remarkably in dense vegetation area, and therefore, it must have been caused by mis-modeling of decorrelation due to temporal change of the vegetation effect. Since vegetation varies with the seasons, we assume that its component fluctuate annually. Then we modified the model to “g = Aexp(Bt){Csin(2 pai t)+Dcos(2 pai t)+Esin(4 pai t)+Fcos(4 pai t)}+G”, and temporal change of coherence could be explained with residual of less than 0.1. However, fitness of the model for time-series of coherence that reference SAR data is different was not good. Therefore it may be necessary to consider the time variation of the time constant of the exponential function.

Coherence decreases with time due to a change in radar scattering due to a change in ground surface coverage and shape, a change in water content, and so on. We assume such temporal change the exponential function “g = Aexp(Bt) + C”, where g is the coherence, t is the observation interval of the interference pair, and A, B, and C are constants. Fitting this function to observed coherence in Kuchinoerabu-jima by PALSAR-2, large residuals exceeding 0.2 were obtained. Such large residuals are remarkably in dense vegetation area, and therefore, it must have been caused by mis-modeling of decorrelation due to temporal change of the vegetation effect. Since vegetation varies with the seasons, we assume that its component fluctuate annually. Then we modified the model to “g = Aexp(Bt){Csin(2 pai t)+Dcos(2 pai t)+Esin(4 pai t)+Fcos(4 pai t)}+G”, and temporal change of coherence could be explained with residual of less than 0.1. However, fitness of the model for time-series of coherence that reference SAR data is different was not good. Therefore it may be necessary to consider the time variation of the time constant of the exponential function.