Page 228 - Special Topic Session (STS) - Volume 2

STS486 Alessandro F. et al.

Change detection and harmonisation of
atmospheric large spatiotemporal series
Alessandro Fassò , Hsin-Cheng Huangy , Igor Valli , Fabio Madonnaz
2
3
1
1
1 Department of Management, Infomation and Production Engineering, University of Bergamo,
Bergamo, Italy
2 Institute of Statistical Science, Academia Sinica, Taipei, Taiwan
3 CNR-IMAA, C.da S. Loja, Tito Scalo, PZ, Italy.

Abstract
The purpose of this paper is to discuss the change detection and
harmonisation process for temperature profiles from the Integrated Global
Radiosonde Archive (IGRA) which consists of global radiosonde observations
dating back to 1905. Harmonisation methods developed for radiosonde have
a long history, see for example Haimberger et al. (2012), Thorne et al. (2011)
and Sherwood et al. (2008). In this paper, we propose a locally stationary 4D
geostatistical model to compute fitted values. The residuals are then used as
input to a fused LASSO approach capable of identifying changes.

1. Introduction
The purpose of this paper is to discuss the change detection and
harmonisation process for temperature profiles from the Integrated Global
Radiosonde Archive (IGRA).
IGRA consists of radiosonde and pilot balloon observations at over 2,700
globally distributed stations. The earliest data date back to 1905, and recent
data become available in near real time. Observations are available at standard
and variable pressure levels, fixed- and variable-height wind levels, and the
surface and tropopause. Variables include pressure, temperature, geopotential
height, relative humidity, dewpoint depression, wind direction and speed, and
elapsed time since launch.
Harmonisation methods developed for radiosonde have a long history, see
for example Haimberger et al. (2012), Thorne et al. (2011) and Sherwood et al.
(2008).
In this paper, the focus is on undocumented changes in single stations. In
particular permanent step changes, temporary step changes and impulses or
outliers are considered.
The idea is to use a simple locally stationary 4D geostatistical model to
compute fitted values. The residuals are then used as input to a fused LASSO
model capable of identifying all the above changes.

217 | I S I W S C 2 0 1 9