Le système de failles de Haiyuan qui borde le plateau du Tibet au Nord-Est est un système majeur sénestre. Au cours du dernier siècle, deux grands séismes (M~8) ont rompu ce système de failles: le séisme de Haiyuan en 1920 et le séisme de Gulang en 1927. A l'aide d'interférométrie radar à synthèse d'ouverture, nous analysons les variations spatiales et temporelles de la déformation intersismique au travers de la faille de Haiyuan, dans une zone étendue (150x150 km2) qui couvre l'extrémité Ouest de la rupture de 1920 et la lacune sismique de Tianzhu. Avec une approche dite en Small Baseline, nous traitons cinq séries temporelles d'images SAR, acquises par le satellite Envisat le long de tracks descendantes et ascendantes pendant la période allant de 2003 à 2009. Les cartes de vitesse moyenne de déformation dans la ligne de visée du satellite ainsi obtenues sont cohérentes avec un mouvement sénestre au travers de la faille et montrent des variations latérales du gradient de vitesse dans la zone de faille. Nous inversons ces cartes de vitesse moyenne en LOS pour obtenir le taux de chargement à court terme en profondeur et la distribution du glissement dans la partie sismogène le long du plan de faille. Le taux de chargement en profondeur est d'environ 5mm/an. Les sections de faille ayant rompu en 1920 et une grande partie de la lacune sismique de Tianzhu sont bloquées en surface. Entre ces deux sections, un segment de 35 km de long, qui montre une forte activité micro-sismique, glisse de manière asismique avec un taux de glissement horizontal qui atteint presque 5 mm/an. Cependant, le taux de glissement asismique le long de la partie sismogène varie le long du plan de faille et atteint localement des taux supérieurs au chargement tectonique, suggérant des variations temporelles du glissement asismique. La comparaison de profils moyens de vitesse parallèle à la faille issus de données InSAR sur les périodes 1993-1998 (données ERS) et 2003-2009 suggèrent une migration vers la surface du glissement asismique sur une période de 20~ans. Une analyse en séries temporelles des données Envisat, en appliquant un lissage temporel, montrent une accélération du taux de glissement asismique pendant l'année 2007. Cette accélération est précédée et a probablement été déclenchée par un séisme de magnitude 4.7 au sein même du glissement asismique. Enfin, nous étudions la relation entre l'évolution spatio-temporelle du glissement asismique en surface et la rugosité de la trace de la faille à l'aide d'une analyse multi-échelle. Nous montrons que les propriétés élastiques de la croûte cassante contrôlent la rugosité de la faille, qui exerce à son tour un contrôle sur la distribution de glissement asismique en surface. Le glissement asismique est fait de spasmes qui interagissent les uns avec les autres en suivant une loi d'échelle similaire à la loi de Gutenberg-Richter pour les séismes. / The Haiyuan fault system is a major left-lateral fault system bounding the tibetan plateau to the north-east. Two M~8 earthquakes ruptured that fault system in the past hundred years: the 1920, Haiyuan earthquake and the 1927, Gulang earthquake. Here, we use Synthetic Aperture Radar interferometry to explore the spatial and temporal variations of the interseismic deformation across the Haiyuan fault, over a broad (150x150 km2) area covering the 1920 rupture zone and the millennial Tianzhu seismic gap. Using a small baseline approach, we process five SAR images time series acquired by the Envisat satellite along descending and ascending orbits, spanning the 2003-2009 period. The resulting mean Line-Of-Sight velocity maps are, in overall, consistent with left-lateral motion across the fault and reveal lateral variations of the velocity gradient in the near fault zone. We invert these mean LOS velocity maps for the short-term loading rate on the fault plane at depth and for the shallow slip distribution along the seismogenic part of the fault. The short-term loading rate is about 5 mm/yr. The shallow sections of the fault, that ruptured in 1920 and the most part of the Tianzhu seismic gap are currently locked. In between, a 35 km-long section, that experiences a strong micro-seismic activity, is creeping at a mean horizontal rate of almost 5 mm/yr. However, the shallow creep rate varies along the fault strike and locally reaches values higher than the deep loading rate. This suggests temporal fluctuations of the observed aseismic slip. The comparison of InSAR-derived averaged profiles of the fault parallel velocity, spanning the 1993-1998 (ERS data) and 2003-2009 periods, suggests an upward migration of the creep over the 20 years-long observation period. A time series analysis on the Envisat dataset using a temporal smoothing reveals a creep rate increase during the year 2007. This rate increase follows and may have been triggered by a M4.7 earthquake that occurred on the creeping patch. We finally investigate the relationship between the spatio-temporal evolution of the surface creep and the roughness of the surface fault trace with a multiscale analysis. We show the control of the elastic properties of the brittle crust on the fault roughness, that in turn exerts a direct control on the surface aseismic slip distribution. The aseismic slip is made of locally interacting bursts that follow a scaling law similar to the Gutenberg-Richter law for earthquakes.

[pt] A segmentação de falhas sísmicas apresenta uma tarefa desafiadora edemorada na geofísica, especialmente na exploração e extração de petróleo egás natural. Métodos de Aprendizado Profundo (Deep Learning) têm mostradoum grande potencial para enfrentar esses desafios e oferecem vantagens emcomparação com métodos tradicionais. No entanto, abordagens baseadas emAprendizado Profundo geralmente requerem uma quantidade substancial dedados rotulados, o que contradiz o cenário atual de disponibilidade limitadade dados sísmicos rotulados. Para lidar com essa limitação, pesquisadores têmexplorado a geração de dados sintéticos como uma solução potencial paradados reais não rotulados. Essa abordagem envolve treinar um modelo emdados sintéticos rotulados e, posteriormente, aplicar diretamente ao conjuntode dados real. No entanto, a geração de dados sintéticos encontra o problemade deslocamento de domínio devido à complexidade das situações geológicasdo mundo real, resultando em diferenças na distribuição entre conjuntosde dados sintéticos e reais. Para mitigar o problema de deslocamento dedomínio na detecção de falhas sísmicas, propomos uma nova abordagem queutiliza técnicas de Adaptação de Domínio Não Supervisionada ou UnsupervisedDomain Adaptation (UDA). Nossa proposta envolve o uso de um conjunto dedados sintéticos para treinamento do modelo e sua adaptação a dois conjuntosde dados reais disponíveis publicamente na literatura. As técnicas de UDAescolhidas incluem Maximum Mean Discrepancy (MMD), Domain-AdversarialNeural Networks (DANN) e Fourier Domain Adaptation (FDA). MMD eDANN visam alinhar características em um espaço de características comumde n dimensões, minimizando discrepâncias e aumentando a confusão dedomínio por meio do treinamento adversarial, respectivamente. Por outro lado,FDA transfere o estilo de amostras reais para sintéticas usando TransformadaRápida de Fourier. Para os experimentos, utilizamos uma versão menor doUNet e sua variante Atrous UNet, que incorpora camadas convolucionaisdilatadas em seu gargalo. Além disso, o DexiNed (Dense Extreme InceptionNetwork), um modelo do estado da arte para detecção de bordas, foi empregadopara fornecer uma análise mais abrangente. Além disso, estudamos a aplicaçãode ajuste fino ou fine-tuning em conjuntos de dados rotulados para investigarseu impacto no desempenho, pois muitos estudos o têm utilizado para reduziro deslocamento de domínio.Os resultados finais demonstraram melhorias significativas no desempenho de detecção de falhas ao aplicar técnicas de UDA, com aumento médio deaté 13 por cento em métricas de avaliação como Intersection over Union e F1-score.Além disso, a abordagem proposta obteve detecções mais consistentes de falhassísmicas com menos falsos positivos, indicando seu potencial para aplicações nomundo real. Por outro lado, a aplicação de ajuste fino não demonstrou ganhossignificativos no desempenho, mas reduziu o tempo de treinamento. / [en] Seismic fault segmentation presents a challenging and time-consuming task in geophysics, particularly in the exploration and extraction of oil and natural gas. Deep Learning (DL) methods have shown significant potential to address these challenges and offer advantages compared to traditional methods. However, DL-based approaches typically require a substantial amount of labeled data, which contradicts the current scenario of limited availability of labeled seismic data. To address this limitation, researchers have explored synthetic data generation as a potential solution for unlabeled real data. This approach involves training a model on labeled synthetic data and subsequently applying it directly to the real dataset. However, synthetic data generation encounters the domain shift problem due to the complexity of real-world geological situations, resulting in differences in distribution between synthetic and real datasets. To mitigate the domain shift issue in seismic fault detection, we propose a novel approach utilizing Unsupervised Domain Adaptation (UDA) techniques. Our proposal involves using a synthetic dataset for model training and adapting it to two publicly available real datasets found in the literature. The chosen UDA techniques include Maximum Mean Discrepancy (MMD), Domain-Adversarial Neural Networks (DANN), and Fourier Domain Adaptation (FDA). MMD and DANN aim to align features in a common n-dimensional feature space by minimizing discrepancy and increasing domain confusion through adversarial training, respectively. On the other hand, FDA transfers the style from real to synthetic samples using Fast Fourier Transform. For the experiments, we utilized a smaller version of UNet and its variant Atrous UNet, which incorporates Dilated Convolutional layers in its bottleneck. Furthermore, DexiNed (Dense Extreme Inception Network), a state-of-the-art model for edge detection, was employed to provide a more comprehensive analysis. Additionally, we studied the application of fine-tuning on labeled datasets to investigate its impact on performance, as many studies have employed it to reduce domain shift. The final results demonstrated significant improvements in fault detection performance by applying UDA techniques, with up to a 13 percent increase in evaluation metrics such as Intersection over Union and F1-score on average. Moreover, the proposed approach achieved more consistent detections of seismic faults with fewer false positives, indicating its potential for realworld applications. Conversely, the application of fine-tuning did not show a significant gain in performance but did reduce the training time.