5:15 PM - 6:30 PM
[STT36-P05] Progress on developing a GNSS-based InSAR atmospheric delay correction model
Keywords:InSAR, delay correction, GNSS
Correcting the InSAR atmospheric delay noise is one of the problem for InSAR researches. In teh last year, Kinoshita (2020, GSJ) proposed a new InSAR atmospheric delay model by use of the GNSS zenith total delay (ZTD) and its horizontal gradient data, in which we referred the method proposed by Arief and Heki (2020). The preliminary result of our proposed model showed a better performance compared with a traditional correction approaches such as the GACOS global mode and the JMA's MSM based model. Here we investigated parameter sensitivities of our model in terms of the grid spacing and the use of the delay gradient.
We at first investigated the model sensitivity in the grid spacing. As a default, we have set it to 5 km spacing, although this setting may not be the best value. In this experiment, we changed the grid spacing ranging from 2 km to 30km while other parameters were fixed. The result showed that the InSAR phase reduction was most significant with the 4 km grid spacing. Even if the grid spacing were set to be finer such as 2 km and 3 km, the delay correction effectiveness slightly worsened. This would indicate the limitation of the resolvable ability of using the GNSS ZTD and gradient.
We next investigated the sensitivity to the GNSS station density. In this experiment, we performed model simulations under conditions that reduced the number of GNSS stations. Other parameters were fixed as a default and the grid spacing was set to 5 km as being the same as the original model run. The result showed the monotonical decrease of the phase reduction ability with decreasing the number of GNSS stations as expected. An interesting feature was that the decreasing ratio of the model correction ability done not worsened significantly until the GNSS station density was reduced to 20 % of the original number (average distance between stations was aproximately 50km). This result indicated that our proposed model would have an applicability for regions where only small numbers of GNSS stations are installed like south-eastern Asia and Africa.
In addition, we compared the model's correction ability with that of the model without using delay gradients. Contrary to our expectation, the InSAR phase reduction was slightly better in the model without using gradients (corrected phase standard deviation was 25.62 mm) than the model with using gradients (25.75 mm). This difference may be not significant statistically, but the cause of this may be 1) an overfitting to the observed gradients for the small scale (an order of kilometers) in the inversion, 2) an inadequacy of the smoothing hyper-parameter value that was set manually, or 3) inappropriate value of the scale height that we used the scale height value of 2.5 km which is for atmospheric water vapor.
In the presentation, we will discuss our model characteristics shown above.
We at first investigated the model sensitivity in the grid spacing. As a default, we have set it to 5 km spacing, although this setting may not be the best value. In this experiment, we changed the grid spacing ranging from 2 km to 30km while other parameters were fixed. The result showed that the InSAR phase reduction was most significant with the 4 km grid spacing. Even if the grid spacing were set to be finer such as 2 km and 3 km, the delay correction effectiveness slightly worsened. This would indicate the limitation of the resolvable ability of using the GNSS ZTD and gradient.
We next investigated the sensitivity to the GNSS station density. In this experiment, we performed model simulations under conditions that reduced the number of GNSS stations. Other parameters were fixed as a default and the grid spacing was set to 5 km as being the same as the original model run. The result showed the monotonical decrease of the phase reduction ability with decreasing the number of GNSS stations as expected. An interesting feature was that the decreasing ratio of the model correction ability done not worsened significantly until the GNSS station density was reduced to 20 % of the original number (average distance between stations was aproximately 50km). This result indicated that our proposed model would have an applicability for regions where only small numbers of GNSS stations are installed like south-eastern Asia and Africa.
In addition, we compared the model's correction ability with that of the model without using delay gradients. Contrary to our expectation, the InSAR phase reduction was slightly better in the model without using gradients (corrected phase standard deviation was 25.62 mm) than the model with using gradients (25.75 mm). This difference may be not significant statistically, but the cause of this may be 1) an overfitting to the observed gradients for the small scale (an order of kilometers) in the inversion, 2) an inadequacy of the smoothing hyper-parameter value that was set manually, or 3) inappropriate value of the scale height that we used the scale height value of 2.5 km which is for atmospheric water vapor.
In the presentation, we will discuss our model characteristics shown above.