Detecting Weather Fronts from Satellite Observations: What Does Reanalysis Add?

Ioannis Dravilas; Miruna Stoicescu; Roope Tervo; Carlos Horn; Oriol Hinojo Comellas; Sina Montazeri; Andrea Meraner; Ana Lucic

cs.LGNov 5, 2026

Detecting Weather Fronts from Satellite Observations: What Does Reanalysis Add?

Authors: Ioannis Dravilas, Miruna Stoicescu, Roope Tervo, Carlos Horn, Oriol Hinojo Comellas, Sina Montazeri, Andrea Meraner, Ana Lucic

Organizations: University of Amsterdam · EUMETSAT

Abstract

Weather fronts are transition regions between air masses with contrasting thermodynamic properties and are associated with mid-latitude precipitation and severe weather. We investigate machine learning (ML) methods for detecting weather fronts and quantify the contribution of reanalysis data to performance. We compare two configurations of the same ML model for pixel-wise frontal classification over the North Atlantic and Europe: (1) using only geostationary infrared satellite observations, and (2) additionally using 28 ERA5 reanalysis fields. The satellite-only model achieves a neighbourhood critical success index at a 150 km tolerance (CSI@150) of $0.402$ , which lies within the range of agreement between analyses from different meteorological services. The satellite $+$ ERA5 model achieves $0.575$ , an improvement of $0.173$ . Our analysis shows that low-level winds are the least predictable from the satellite representation ( $R^2 = 0.23$ , compared with $0.88$ for temperature), while removing winds or moisture causes the largest decreases in detection performance.

Figures & tables

Figure 1: Analysis of ERA5 variable groups. Top: linear predictability (

R^{2}

) from the frozen satellite-only encoder (best stage) and gain over a randomly initialised encoder, with 95% CIs. Near-sfc: 1000/925 hPa and 10 m fields. Dashed: positive control. Bottom: drop in CSI@150 when a group is replaced by values from another date.

	Overall	Cold	Warm	Occluded
Operational ASII product (satellite $+$ NWP)	0.143	0.156	0.109	0.163
Closest human analysis (Met Éireann)	0.501	0.543	0.454	0.508
OBS (satellite-only)	0.402 [0.389, 0.414]	0.450	0.359	0.398
HYB (satellite $+$ ERA5)	0.575 [0.563, 0.585]	0.615	0.532	0.578
ERA5 improvement $\Delta$	+0.173 [0.163, 0.183]	+0.164	+0.174	+0.180

Table 1: CSI@150 on the 2024 test year (Met Office ground truth). Overall is the mean over the three front types; brackets give 95% CIs. All per-class improvements exclude zero (

P(\Delta\leq 0)<10^{-3}

Figure 2: Annual frontal frequency (2024, all front types, smoothed at ${\sim}150$ km for display).

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Variable group	$R^{2}$ (max over stages)	$\Delta$ vs. random init.	$\Delta$ vs. raw input
Temperature	0.883	$+0.188$	$+0.177$
Moisture	0.600	$+0.254$	$+0.244$
Mass	0.585	$+0.282$	$+0.234$
Winds	0.232	$+0.181$	$+0.189$

Appendix

Table 2: Predictability of ERA5 variables from the frozen OBS encoder, by variable group (2024 test year).

R^{2}

is the maximum over encoder stages. The

\Delta

columns show the improvement over the randomly initialised encoder and the raw satellite input. All improvements exclude zero (

P(\Delta\leq 0)<10^{-3}

). 95% confidence intervals for

R^{2}

and the improvement over the random encoder are shown in Figure 1 .

	OBS + winds	OBS + moisture	OBS + temp. & mass
CSI@150	0.520 [0.506, 0.530]	0.540 [0.527, 0.550]	0.534 [0.521, 0.545]
$\Delta$ over OBS	+0.118 [0.108, 0.126]	+0.138 [0.129, 0.147]	+0.132 [0.122, 0.142]

Appendix

Table 3: Single-group addition experiments. CSI@150 of the satellite-only model augmented with one ERA5 variable group, and the paired improvement over OBS (2024 test year, 95% paired block-bootstrap confidence intervals). All three improvements have

P(\Delta\leq 0)<10^{-3}

Figure 3: Example on one 2024 test timestamp (22 November, 00:00 UTC). Top left: Met Office analysis. Top right: satellite $+$ ERA5 model (HYB). Bottom left: satellite-only model (OBS). Bottom right: the same OBS prediction after a morphological closing that bridges small gaps and removal of small regions, applied for visualisation only. All reported metrics use the unprocessed predictions. Cold, warm, and occluded fronts are shown in blue, red, and purple over the infrared background.

Tolerance (km)	50	100	150	200	250
$\Delta$ CSI	$+0.116$	$+0.160$	$+0.173$	$+0.172$	$+0.165$
95% CI	[0.109, 0.124]	[0.151, 0.169]	[0.163, 0.183]	[0.162, 0.183]	[0.155, 0.177]

Appendix

Table 4: Paired ERA5 improvement in CSI (HYB

-

OBS) by tolerance, 2024 test year, with 95% paired block-bootstrap confidence intervals. All exclude zero (

P(\Delta\leq 0)<10^{-3}

Remote Sensing

Remote sensing uses satellite, aerial, and ground-based observations to measure and map Earth and other planetary surfaces.

See all 540 papers

SATFRONT: Detecting Atmospheric Fronts with Satellite Data using Machine Learning
Oct 30, 2026Ioannis Dravilas, Miruna Stoicescu, Roope Tervo +5Climate Science Remote Sensing
University of Amsterdam, Amsterdam, The Netherlands · EUMETSAT, Darmstadt, Germany
FuXiWeather2: Learning accurate atmospheric state estimation for operational global weather forecasting
Mar 16, 2026Xiaoze Xu, Xiuyu Sun, Songling Zhu +5Weather Forecasting Data Assimilation
Shanghai Academy of Artificial Intelligence for Science, Shanghai, & 200232, China. · Artificial Intelligence Innovation and Incubation Institute, Fudan University, Shanghai & 200433, China. · FuXi Intelligent Computing Technology Co., Ltd., Shanghai & 200233, China.
Skillful high-resolution weather forecasting independent of physical models
May 27, 2026Pengcheng Zhao, Siqi Xiang, Weixin Jin +10Weather Forecasting
1Microsoft Corporation. · University of Cambridge, Cambridge, United Kingdom. · 3The Alan Turing Institute, London, United Kingdom.

Oct 30, 2026J/K move · Enter open · S save

SATFRONT: Detecting Atmospheric Fronts with Satellite Data using Machine Learning

Ioannis Dravilas, Miruna Stoicescu, Roope Tervo, Carlos Horn+4

University of Amsterdam, Amsterdam, The Netherlands · EUMETSAT, Darmstadt, Germany

Climate Science Remote Sensing

Open paper AskPDF arXiv

Atmospheric fronts are associated with much of the high-impact weather in the mid-latitudes, yet their identification and classification remain largely manual and subjective. Existing deep learning approaches rely primarily on gridded reanalysis or forecast data, limiting their applicability to satellite observations available for time-critical applications. We introduce SATFRONT, a machine learning framework for detecting cold, warm, and occluded fronts over Europe from geostationary satellite observations, with reanalysis data as an optional additional input. The model is trained on frontal analyses from five European national meteorological services and evaluated against UK Met Office analyses for 2024. Using satellite observations alone, SATFRONT achieves a Critical Success Index (CSI) of 0.428 at a 150 km tolerance, within the range of agreement between analyses from different meteorological services (0.277-0.501) and higher than that of an operational rule-based front detector that uses satellite data (0.133). Adding ERA5 reanalysis increases the CSI to 0.591. ERA5 substantially improves the placement of individual fronts, although its effect on their annual spatial distribution varies by front type. These results demonstrate that cold, warm, and occluded fronts can be detected directly from geostationary satellite observations, providing a basis for automated frontal analyses at the high temporal frequency required for time-critical applications.

Abstract

Remote Sensing

Explore similar work

SATFRONT: Detecting Atmospheric Fronts with Satellite Data using Machine Learning

FuXiWeather2: Learning accurate atmospheric state estimation for operational global weather forecasting

Skillful high-resolution weather forecasting independent of physical models