{"id":3095,"date":"2026-03-25T16:05:14","date_gmt":"2026-03-25T16:05:14","guid":{"rendered":"https:\/\/research.reading.ac.uk\/met-darc\/?p=3095"},"modified":"2026-03-25T16:05:14","modified_gmt":"2026-03-25T16:05:14","slug":"teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses","status":"publish","type":"post","link":"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/","title":{"rendered":"Teaching a Neural Network to Smooth the Ocean model reanalyses"},"content":{"rendered":"
\t\t\t

by Sukun Cheng, March 2026<\/p>\n<\/div>\n\t\t<\/div>

The core problem<\/h3>\t\t\t

To understand historical ocean changes, researchers produce reanalyses that reconstruct past states by combining numerical models with historical observations. Over the last century, observation quality and quantity have shifted drastically. The sparse, shallow ship-deployed instruments of the early 1900s were eventually supplemented by eXpendable BathyThermograph (XBTs)<\/a> in the 1970s, which finally penetrated the thermocline. A definitive step-change occurred around 2004 with the Argo programme<\/a>, providing near-global, deep-ocean profiling. This density shift creates a fundamental consistency problem: distinguishing genuine climate signals from mere changes in our measurement capabilities (see figure 1).<\/p>\n

In modern ocean reanalyses, data assimilation (DA) regularly updates the model via increments to pull the state toward observations. During long data-sparse periods, the model drifts significantly; when rare observations are finally assimilated, they trigger spurious increment pulses.<\/p>\n

\"\"
Figure 1 Number of observations from various observation systems in North Atlantic region (30~60\u00b0N, )<\/figcaption><\/figure>\n<\/div>\n\t\t<\/div>

Methodological approach<\/h3>\t\t\t

Traditional post-processing algorithms struggle with this temporal discontinuity. A time-dependent smoother developed by Dong et al. (2021) corrects current increments by applying an exponential decay to future increments. However, its reliance on a rigid, uniform damping parameter fails to adapt to local observation density or the highly variable uncertainty between sparse and data-rich eras.<\/p>\n

We are investigating an AI smoother to override this rigid parameterization, drawing direct inspiration from video generation technologies that interpolate continuous motion from sparse, isolated frames. Following the core philosophy of Dong et al. (2021)\u2014learning from future increment “experience” to revise the target time increment, given that the spatial error covariance has already been estimated by the DA system\u2014a Temporal Convolutional Network (TCN) is trained to learn the temporal structure of ocean increments. This allows the network to redistribute erratic corrections physically, demonstrating strong adaptive capability. A window of raw increments and local ensemble spread (representing uncertainty) is fed into the TCN. Because the near-surface behaves dynamically differently from the deep thermocline, a Feature-wise Linear Modulation (FiLM) layer adapts the network’s behavior based on depth. Figure 2 provides a flowchart of AI smoothing process.<\/p>\n

\"\"
Figure 2 Flowchart: a Temporal Convolutional Network (TCN) with Feature-wise Linear Modulation (FiLM) depth conditioning produces a gated blend of the AI-smoothed and raw signals, followed by a conservation correction.<\/figcaption><\/figure>\n

Crucially, the smoother learns a gating weight (between 0 and 1) that blends the AI-smoothed signal with the original raw increment. During data-sparse eras, the gate opens, trusting the AI to redistribute corrections. During data-rich eras, it dynamically closes to stay conservative and focus on smoothing the high-frequency noise inherent in the raw increments. An explicit physical conservation constraint guarantees that the total heat content is strictly preserved across multi-year integration periods.<\/p>\n<\/div>\n\t\t<\/div>

Current findings<\/h3>\t\t\t

Results from the mid-thermocline (300\u2013800 m) and the deep ocean (800\u20132000 m) definitively prove the AI fundamentally outperforms the Dong baseline. The AI smoother reduces high-frequency noise by an order of magnitude while improving low-frequency signal retention, driving the correlation of multi-year variability from 0.88 to 0.99. Spurious pulses in the pre-Argo period are largely eliminated.<\/p>\n

This remains an active research problem. The near-surface (0\u201320 m) and deep abyss (below 2000 m) present distinct challenges. Surface increments lack temporal structure and resemble white noise, forcing us to drastically narrow the network’s temporal receptive field to prevent overfitting. Conversely, the abyss suffers from extremely slow dynamics and near-zero observations. To isolate these physical signals from noise, we plan to leverage the CIGAR CS<\/a> project\u2019s 1-degree resolution models, using a 40-member ensemble<\/strong> to explicitly map uncertainty\u2014a critical advantage over deterministic high-resolution models.<\/p>\n<\/div>\n\t\t<\/div>

Broader impact<\/h3>\t\t\t
\n

A temporally consistent reanalysis is essential for oceanographers studying critical climate change indicators, ensuring observed shifts are physical rather than observational artefacts. Operational services stand to gain immediate product quality improvements. More broadly, this demonstrates how machine learning can function as an intelligent, physics-constrained post-processor embedded directly within the DA workflow, building upon traditional knowledge rather than blindly replacing it.<\/p>\n<\/div>\n<\/div>\n\t\t<\/div>

References<\/h3>\t\t\t

Dong et al. 2021: Improved high resolution ocean reanalyses using a simple smoother algorithm.\u00a0JAMES. <\/em>\u00a0https:\/\/doi.org\/10.1029\/2021MS002626<\/a><\/strong><\/p>\n<\/div>\n\t\t<\/div><\/div><\/div><\/div>","protected":false},"excerpt":{"rendered":"

by Sukun Cheng, March 2026 To understand historical ocean changes, researchers produce reanalyses that reconstruct past states by combining numerical models with historical observations. Over the last century, observation quality…Read More ><\/a><\/p>\n","protected":false},"author":87,"featured_media":3104,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"__cvm_playback_settings":[],"__cvm_video_id":"","footnotes":"","_links_to":"","_links_to_target":""},"categories":[1],"tags":[69,65,74,72,73],"class_list":["post-3095","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorised","tag-ai","tag-data-assimilation","tag-ocean-heat-content","tag-ocean-reanalysis","tag-temporal-convolutional-network"],"acf":[],"yoast_head":"\nTeaching a Neural Network to Smooth the Ocean model reanalyses - DARC<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/\" \/>\n<meta property=\"og:locale\" content=\"en_GB\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Teaching a Neural Network to Smooth the Ocean model reanalyses - DARC\" \/>\n<meta property=\"og:description\" content=\"by Sukun Cheng, March 2026 To understand historical ocean changes, researchers produce reanalyses that reconstruct past states by combining numerical models with historical observations. Over the last century, observation quality...Read More >\" \/>\n<meta property=\"og:url\" content=\"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/\" \/>\n<meta property=\"og:site_name\" content=\"DARC\" \/>\n<meta property=\"article:published_time\" content=\"2026-03-25T16:05:14+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/research.reading.ac.uk\/met-darc\/wp-content\/uploads\/sites\/48\/2026\/03\/pexels-elliotconnor-7082802.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1280\" \/>\n\t<meta property=\"og:image:height\" content=\"960\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Alison Fowler\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Alison Fowler\" \/>\n\t<meta name=\"twitter:label2\" content=\"Estimated reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"4 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/\",\"url\":\"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/\",\"name\":\"Teaching a Neural Network to Smooth the Ocean model reanalyses - DARC\",\"isPartOf\":{\"@id\":\"https:\/\/research.reading.ac.uk\/met-darc\/#website\"},\"datePublished\":\"2026-03-25T16:05:14+00:00\",\"dateModified\":\"2026-03-25T16:05:14+00:00\",\"author\":{\"@id\":\"https:\/\/research.reading.ac.uk\/met-darc\/#\/schema\/person\/cb510f269e533e68b4761c346208ce0c\"},\"breadcrumb\":{\"@id\":\"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/#breadcrumb\"},\"inLanguage\":\"en-GB\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/research.reading.ac.uk\/met-darc\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Teaching a Neural Network to Smooth the Ocean model reanalyses\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/research.reading.ac.uk\/met-darc\/#website\",\"url\":\"https:\/\/research.reading.ac.uk\/met-darc\/\",\"name\":\"DARC\",\"description\":\"\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/research.reading.ac.uk\/met-darc\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"en-GB\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/research.reading.ac.uk\/met-darc\/#\/schema\/person\/cb510f269e533e68b4761c346208ce0c\",\"name\":\"Alison Fowler\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-GB\",\"@id\":\"https:\/\/research.reading.ac.uk\/met-darc\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/fe96bd6f6770c8b6fb6af28d52b3db79f69fcc2df450a9a6665bc9a13f4ab226?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/fe96bd6f6770c8b6fb6af28d52b3db79f69fcc2df450a9a6665bc9a13f4ab226?s=96&d=mm&r=g\",\"caption\":\"Alison Fowler\"},\"url\":\"https:\/\/research.reading.ac.uk\/met-darc\/author\/a-m-fowlerreading-ac-uk\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Teaching a Neural Network to Smooth the Ocean model reanalyses - DARC","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/","og_locale":"en_GB","og_type":"article","og_title":"Teaching a Neural Network to Smooth the Ocean model reanalyses - DARC","og_description":"by Sukun Cheng, March 2026 To understand historical ocean changes, researchers produce reanalyses that reconstruct past states by combining numerical models with historical observations. Over the last century, observation quality...Read More >","og_url":"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/","og_site_name":"DARC","article_published_time":"2026-03-25T16:05:14+00:00","og_image":[{"width":1280,"height":960,"url":"https:\/\/research.reading.ac.uk\/met-darc\/wp-content\/uploads\/sites\/48\/2026\/03\/pexels-elliotconnor-7082802.jpg","type":"image\/jpeg"}],"author":"Alison Fowler","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Alison Fowler","Estimated reading time":"4 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/","url":"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/","name":"Teaching a Neural Network to Smooth the Ocean model reanalyses - DARC","isPartOf":{"@id":"https:\/\/research.reading.ac.uk\/met-darc\/#website"},"datePublished":"2026-03-25T16:05:14+00:00","dateModified":"2026-03-25T16:05:14+00:00","author":{"@id":"https:\/\/research.reading.ac.uk\/met-darc\/#\/schema\/person\/cb510f269e533e68b4761c346208ce0c"},"breadcrumb":{"@id":"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/#breadcrumb"},"inLanguage":"en-GB","potentialAction":[{"@type":"ReadAction","target":["https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/research.reading.ac.uk\/met-darc\/2026\/03\/25\/teaching-a-neural-network-to-smooth-the-ocean-model-reanalyses\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/research.reading.ac.uk\/met-darc\/"},{"@type":"ListItem","position":2,"name":"Teaching a Neural Network to Smooth the Ocean model reanalyses"}]},{"@type":"WebSite","@id":"https:\/\/research.reading.ac.uk\/met-darc\/#website","url":"https:\/\/research.reading.ac.uk\/met-darc\/","name":"DARC","description":"","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/research.reading.ac.uk\/met-darc\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"en-GB"},{"@type":"Person","@id":"https:\/\/research.reading.ac.uk\/met-darc\/#\/schema\/person\/cb510f269e533e68b4761c346208ce0c","name":"Alison Fowler","image":{"@type":"ImageObject","inLanguage":"en-GB","@id":"https:\/\/research.reading.ac.uk\/met-darc\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/fe96bd6f6770c8b6fb6af28d52b3db79f69fcc2df450a9a6665bc9a13f4ab226?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/fe96bd6f6770c8b6fb6af28d52b3db79f69fcc2df450a9a6665bc9a13f4ab226?s=96&d=mm&r=g","caption":"Alison Fowler"},"url":"https:\/\/research.reading.ac.uk\/met-darc\/author\/a-m-fowlerreading-ac-uk\/"}]}},"cc_featured_image_caption":{"caption_text":"","source_text":"","source_url":""},"_links":{"self":[{"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/posts\/3095","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/users\/87"}],"replies":[{"embeddable":true,"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/comments?post=3095"}],"version-history":[{"count":10,"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/posts\/3095\/revisions"}],"predecessor-version":[{"id":3121,"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/posts\/3095\/revisions\/3121"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/media\/3104"}],"wp:attachment":[{"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/media?parent=3095"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/categories?post=3095"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/research.reading.ac.uk\/met-darc\/wp-json\/wp\/v2\/tags?post=3095"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}