<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by KappaZeta on Medium]]></title>
        <description><![CDATA[Stories by KappaZeta on Medium]]></description>
        <link>https://medium.com/@tetiana.shtym?source=rss-e0cd7c265a0------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*F-KsBgUUPDyR-r1zDevUvQ.png</url>
            <title>Stories by KappaZeta on Medium</title>
            <link>https://medium.com/@tetiana.shtym?source=rss-e0cd7c265a0------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 06:47:02 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@tetiana.shtym/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Farming Footprints: Mapping Tillage Practices with Earth Observation and AI]]></title>
            <link>https://medium.com/kappazeta/farming-footprints-mapping-tillage-practices-with-earth-observation-and-ai-dc43cac86e83?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/dc43cac86e83</guid>
            <category><![CDATA[sentinel]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agriculture-technology]]></category>
            <category><![CDATA[tillage]]></category>
            <category><![CDATA[earth-observation]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Fri, 13 Dec 2024 11:12:37 GMT</pubDate>
            <atom:updated>2025-02-12T15:02:22.001Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*_y7n5Qo3nwP6cdUH0JrXfg.jpeg" /><figcaption>Source: PixBay</figcaption></figure><p>At KappaZeta, we’ve been tackling an exciting challenge over the past year: detecting tillage practices from space. This work provides invaluable tools for farmers, paying agencies, and environmental advocates, enabling better decision-making for sustainable agricultural practices. If you’re new to the subject, our previous blog post offers a great introduction: <a href="https://medium.com/kappazeta/detecting-tillage-intensity-from-space-9821f60e5dcd">Detecting Tillage Intensity from Space</a>.</p><p>At the heart of our work is a bold mission: to give the agricultural industry the scalable, accurate tools it needs to detect tillage practices and unlock its full potential. This isn’t just about better tools — it’s about creating a foundation for compliance, sustainability, and a smarter future in farming.</p><p>In this blog post, we’ll take you on a journey through our process, challenges, and the exciting results we’ve achieved along the way.</p><p>Ready to see how it all turned out? Let’s dive in!</p><h3>Outcomes</h3><p>In this project, tillage models were developed using satellite time series data from Sentinel-1 (S1), Sentinel-2 (S2), and Landsat 8/9. Ground truth data was primarily collected from farmers and paying agencies (PAs) across the Baltic Sea region, with additional data generated by labelling satellite imagery time series.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Kgtx5SfliKQyHbw8Han-gw.png" /><figcaption>Example of conservation tillage label on time series</figcaption></figure><h4>Conventional tillage</h4><p>Conventional tillage includes practices such as harrowing, ploughing, and rototilling. In this project, the model incorporates data from the following countries: Estonia, Latvia, Lithuania, Germany, Poland, Denmark, Finland, and Sweden.</p><p>The 6-day and 12-day conventional tillage models utilize Sentinel-1, Sentinel-2, and Landsat features.</p><h4>Conventional tillage 6-day coherence model</h4><p>We developed two separate models for this use case: one for the Baltic region and another for the Nordic region.</p><p>For the model, we used data from Estonia (2018–2021), Latvia (2021), Lithuania (2021), Poland (2021), Germany (2021), Sweden (2021), Denmark (2021) and Finland (2021).</p><p>The table displays the performance of the 6-day coherence conventional tillage model across test sets from different countries. The results show that the model achieved an average F1-score of 91% across all test sets, with a recall of 92% and a precision of 90%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/701/1*ISdAkJzefpickqwxM28rOA.png" /><figcaption>Performance metrics of 6-day coherence conventional tillage model</figcaption></figure><p>The figure below displays a parcel in Germany, where the model successfully detected two conventional tillage events: the first around day of year (DOY) 110 and the second near DOY 280. Satellite imagery shows the field turning “yellow” during the first tillage event.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vc_aM_TluuWOUsTBhg30Vg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NsBWuEb1CkOpUrCWweXmIQ.png" /><figcaption>Example of a true positive detection in Germany 2021</figcaption></figure><p>The figure illustrates a parcel in Sweden (2021) where the model accurately detected two conventional tillage events: one at approximately DOY 85 and the other around DOY 225. The satellite images highlight that the field turned “yellow” following the second tillage event.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZMqmkrA_peTRWElgF7eojQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QOYMWq_wEi0g3hvnSdxMSw.png" /><figcaption>Example of a true positive detection in Sweden 2021</figcaption></figure><h4>Conventional tillage 12-day coherence model</h4><p>For the model, we used data from Estonia (2018–2023), Latvia (2021), Lithuania (2021), Poland (2021), Germany (2021), Finland (2021), Sweden (2021) and Denmark (2021).</p><p>The model demonstrated strong performance, with an average F1-score surpassing 93%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/701/1*Ou3jEzriixkwea866qKI4Q.png" /><figcaption>Performance metrics of 12-day coherence conventional tillage model</figcaption></figure><p>The figure below presents a parcel in Estonia from 2021, where the model accurately recognized a tillage event.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0qDQLk59UCSsxcOv7Iwcmg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dlB_Hwrd4aRbX73czy8-Eg.png" /><figcaption>Figure 8: Example of true positive detection in Estonia 2021</figcaption></figure><h4>Conservation tillage</h4><p>This model is designed to identify conservation tillage events using Sentinel-1 (6-day and 12-day coherence) and Sentinel-2 time series data. Conservation tillage practices include no/zero-tillage, reduced tillage, zone tillage, and ridge tillage.</p><h4>Conservation tillage 6-day coherence</h4><p>For the model, we used data from Estonia (2018–2021), Latvia (2020–2021), Lithuania (2021), Poland (2021), Germany (2021), Finland (2021), Sweden (2021) and Denmark (2019–2021).</p><p>The model’s average F1-score across all test sets is 82%, with standout results for Sweden, Latvia, and Lithuania.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/702/1*IeobieG37ZMHiZa1K0gxlA.png" /><figcaption>Performance metrics of 6-day coherence conservation tillage model</figcaption></figure><p>The figure below depicts a parcel in Denmark from 2021, where the model accurately detected a conservation tillage event. It can be seen that crop residue cover (CRC) was detected on a parcel for approximately 50 days as indicated by the length of the detection window.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rPnuJiiWwzrpy005UKOcjg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GHtZeQ3b1CvK8vlWWhXCag.png" /><figcaption>Example of true positive in Denmark 2021</figcaption></figure><h4>Conservation tillage 12-day coherence model</h4><p>For the model we used data from Estonia (2018–2023), Latvia (2019–2022), Lithuania (2021), Poland (2021), Germany (2021), Finland (2020–2022), Sweden (2021–2022) and Denmark (2018–2022).</p><p>The model achieved an average F1-score of 84% across all test sets. Overall, precision and recall trends fluctuate, with certain countries maintaining strong performance while others exhibit year-to-year variability.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*AIqMKXLnwhJ9rXhfI8LKtA.png" /><figcaption>Performance metrics of 12-day coherence conservation tillage model</figcaption></figure><p>The figure demonstrates a parcel in Finland 2021 where the model correctly identified the conservation tillage event around DOY 210.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V_U5ekKzM1kuSh6FOTiIAA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CP4ys5UeljxqydUZVwK4Fg.png" /><figcaption>Example of true positive in Finland 2021</figcaption></figure><h3>Positive Lessons Learned</h3><ol><li>Variations in agricultural practices and environmental conditions between regions reduced model accuracy; region-specific models offer better performance.</li><li>Gaps in optical satellite data, caused by cloud cover, hindered the detection of some tillage events. However, the integration of Sentinel-1 SAR data helps to address this issue.</li><li>A lack of high-quality ground truth data for conservation tillage limited model accuracy; addressing this ensures reliable outcomes.</li></ol><h3><strong>Key Benefits for End Users</strong></h3><p>Our satellite-based models deliver real-world impact tailored to your needs:</p><ul><li><strong>Simplified Management:</strong> Automated detection reduces the burden of manual inspections by agricultural paying agencies.</li><li><strong>Improved Compliance:</strong> Accurately identify whether farmers meet sustainable tillage requirements.</li><li><strong>Environmental Impact:</strong> Promote practices that protect soil and reduce greenhouse gas emissions.</li><li><strong>Policy Alignment:</strong> Align monitoring with policy frameworks like the EU’s Common Agricultural Policy (CAP), especially the <a href="https://agriculture.ec.europa.eu/common-agricultural-policy/income-support/eco-schemes_en">Eco-schemes</a> goal under the EU Green Deal, which supports farmers in adopting practices that minimise the negative impact of agriculture on the environment and climate and helps them evolve towards more sustainable farming models.</li></ul><h4>Acknowledgments</h4><p>We want to thank the European Space Agency (ESA) for supporting, advising, and funding the project.</p><p>If you have any questions or collaboration ideas, we would love to hear from you! Please feel free to reach out to Catherine via email at <em>catherine.odera@kappazeta.ee.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dc43cac86e83" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/farming-footprints-mapping-tillage-practices-with-earth-observation-and-ai-dc43cac86e83">Farming Footprints: Mapping Tillage Practices with Earth Observation and AI</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Detecting tillage intensity from space]]></title>
            <link>https://medium.com/kappazeta/detecting-tillage-intensity-from-space-9821f60e5dcd?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/9821f60e5dcd</guid>
            <category><![CDATA[tillage]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agriculture-technology]]></category>
            <category><![CDATA[earth-observation]]></category>
            <category><![CDATA[sentinel-2]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Mon, 03 Jun 2024 06:39:34 GMT</pubDate>
            <atom:updated>2025-06-05T11:56:48.651Z</atom:updated>
            <content:encoded><![CDATA[<p>Written by <em>Catherine Odera, EO project manager</em> <em>at KappaZeta</em></p><p>Tillage is the turning of soil during land preparation for farming, and this necessary agricultural practice comes with a price. The more intense the tillage, the more the carbon stored in the soil is released into the atmosphere. Tillage practices also influence soil health, affecting erosion rates and the distribution of nutrients. KappaZeta is on a mission to detect different types of tillage by combining satellite imagery and Artificial Intelligence (AI). The information from detecting, classifying, and monitoring tillage practices can then be used to reward farmers for low-impact tillage practices.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Lo0DEGFG45HNNNAmhml5IA.jpeg" /><figcaption>Image source: Pixbay</figcaption></figure><h4><strong>Conventional tillage and conservation tillage</strong></h4><p>Broadly, tillage practices can be categorized into conservation and conventional tillage. Conservation tillage is a method that maintains permanent or semi-permanent soil coverage, thereby promoting soil health and minimizing erosion. Practices under this category, such as no-tillage, mulch tillage, and ridge tillage, aim to enhance soil conservation and improve crop yields. These methods ensure minimal soil disturbance, preserve soil structure, enhance water retention, and support biodiversity. In contrast, conventional tillage is known for its intensive soil disturbance. This widespread method can negatively impact biodiversity, disrupt soil nutrient balance, and increase GHG emissions. The adverse effects on soil structure and function emphasize the importance of monitoring and exploring sustainable alternatives, i.e., conservation tillage.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f2myWrtOgcFNMRn54S99cg.jpeg" /><figcaption>Images: The left image shows an example of no-till farming, the right image is an example of conventional tillage. Left image source: <a href="https://www.no-tillfarmer.com/articles/530-vertical-tillage-a-gateway-tool-to-or-away-from-conservation">https://www.no-tillfarmer.com/articles/530-vertical-tillage-a-gateway-tool-to-or-away-from-conservation</a>. Right image source: <a href="https://www.kverneland.co.nz/news/ploughing-benefits-most-soils">https://www.kverneland.co.nz/news/ploughing-benefits-most-soils</a></figcaption></figure><h4><strong>Can tillage intensity be detected from space?</strong></h4><p>Earth observation, coupled with AI, provides a powerful tool for identifying and classifying tillage practices. It aids in analyzing and capturing changes in land surface characteristics such as residue presence and field patterns. Satellite sensors, e.g., Sentinel-2 and Landsat, along with their derived indices, are instrumental in tillage detection.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/790/1*eUVgax4IUDBWKldf8DG-lA.jpeg" /><figcaption>Image source: <a href="https://www.e-resident.gov.ee/blog/posts/doing-business-from-a-distance/">https://www.e-resident.gov.ee/blog/posts/doing-business-from-a-distance/</a></figcaption></figure><p>Some examples of important Sentinel-2 and Landsat-based indices are highlighted below:</p><ul><li>Normalized Difference Vegetation Difference (<strong>NDVI)</strong></li><li>Normalized Difference Tillage Index (<strong>NDTI</strong>)</li><li>Crop Residue Cover Index (<strong>CRCI</strong>)</li><li>Crop Residue Cover (<strong>CRC</strong>)</li><li>Normalized Difference Index 5 (<strong>NDI5</strong>)</li><li>Normalized Difference Index 7 (<strong>NDI7</strong>)</li><li>Normalized Difference Residue Index (<strong>NDRI</strong>)</li><li>Simulated Cellulose Absorption Index (<strong>BI1</strong>)</li><li>Simulated Lignin Cellulose Absorption (<strong>BI2</strong>)</li></ul><p>The indices, derived using various Landsat and Sentinel-2 spectral bands, are essential in highlighting vegetation health, soil differences, and cellulose degradation, aiding in the identification and differentiation of tillage practices. Moreover, the integration of Synthetic Aperture Radar (SAR) data, such as Sentinel-1, enables continuous monitoring regardless of weather conditions, capturing surface changes indicative of tillage activities.</p><p>While Earth observation offers valuable information, it faces challenges, especially in regions with small field sizes or where distinguishing between crop residues and soil is difficult. However, the integration of AI algorithms and advanced data fusion models has significantly improved tillage detection accuracy, facilitating more precise tillage mapping.</p><p>In conclusion, the integration of Earth observation technologies and AI algorithms holds great potential for effectively identifying, classifying, and monitoring tillage practices. This comprehensive understanding is crucial for promoting sustainable agricultural management and conservation efforts, ultimately contributing to improved agricultural and climate policies.</p><h4>C<strong>onservation tillage feasibility study by KappaZeta</strong></h4><p>As the spectral signature of conservation tillage practices has not been extensively researched, a feasibility study was undertaken to assess the possibility of detecting different tillage practices using various vegetation indices mentioned in scientific papers. Once the most suitable vegetation indices were identified through the feasibility study, the subsequent step would be developing a conservation tillage detection model.</p><p>To ensure that the observed patterns remained consistent and not overly dependent on a particular season, data from multiple years was incorporated into the analysis. However, some seasonal influences cannot be entirely ruled out. A total of 8766 data points sampled in Finland (2020–2021), Estonia (2018–2022), Denmark (2022), Sweden (2021–2022), Latvia(2021–2022) and Lithuania (2022) were used in the study. Data from Poland and Northern Germany was not included as we are still working on acquiring data from both countries.</p><p>Upon concluding the feasibility study, we discovered that it is possible to distinguish conventional tillage from conservation tillage.</p><h4><strong>Next steps and call for contributions</strong></h4><p>Future work includes improving the conventional tillage detection model and assessing the model’s performance in Estonia, Latvia, Lithuania, Finland, Sweden, Denmark, Poland, and Northern Germany. A conservation tillage detection model is also currently in development.</p><p>The main factor limiting the development of very accurate tillage detection models, especially for detecting conservation tillage, is the insufficient amount of trustworthy ground truth data available. We’re grateful for any helpful information related to ground reference data for different tillage practices.</p><p>If you have any questions or collaboration ideas, we would love to hear from you! Please feel free to reach out to Catherine via email at <em>catherine.odera@kappazeta.ee.</em></p><p><em>This Post was first published on </em><a href="https://kappazeta.ee/blog/detecting-tillage-intensity-from-space/"><em>https://kappazeta.ee/blog/detecting-tillage-intensity-from-space/</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9821f60e5dcd" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/detecting-tillage-intensity-from-space-9821f60e5dcd">Detecting tillage intensity from space</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Five new satellite analytics tools for agriculture]]></title>
            <link>https://medium.com/kappazeta/five-new-satellite-analytics-tools-for-agriculture-645540208327?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/645540208327</guid>
            <category><![CDATA[agriculture]]></category>
            <category><![CDATA[sentinel-2]]></category>
            <category><![CDATA[earth-observation]]></category>
            <category><![CDATA[sentinel-1]]></category>
            <category><![CDATA[agroinsurance]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Wed, 31 Jan 2024 14:04:04 GMT</pubDate>
            <atom:updated>2025-06-05T11:56:25.758Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Written by Tanel Kobrusepp, product manager</em></p><h4>Introduction to KappaZeta’s new AI-based Agricultural Solutions</h4><p>KappaZeta’s latest contribution to agricultural technology involves a set of five innovative products based on AI and machine-learning, each uniquely leveraging Sentinel-1 and Sentinel-2 satellite data. These services are designed to address key aspects of agricultural management and analysis, catering to the needs of scientists, government officials, and professionals in farming, farm management and insurance. The suite includes tools for Crop Type Detection, enabling precise identification of various crop types; Parcel Delineation for accurate land mapping; Seedling Emergence Detection to monitor early crop growth; Farmland Damaged Area Delineation for assessing areas affected by adverse events and Ploughing and Harvesting Events Detection to track critical farming activities. These tools collectively aim to enhance agricultural practices through data-driven insights, fostering more informed decision-making in the field of agriculture.</p><h4>Crop Type Detection</h4><p>The Crop Type Detection service developed by KappaZeta effectively leverages Sentinel satellite data to accurately identify various crop types across extensive agricultural regions. For crop insurance companies, this tool aids in risk assessment by providing detailed information about the crops insured, allowing for more accurate risk profiling and premium calculation. Government agencies can utilize this data for agricultural policy planning and monitoring, ensuring resources are appropriately allocated. Additionally, the tool assists in environmental monitoring, as understanding crop distribution is key in assessing ecological impacts and land use planning. For the agricultural market at large, this information helps in forecasting supply and demand trends, crucial for market stability and pricing strategies. Thus, the Crop Type Detection service offers practical benefits across multiple facets of the agricultural industry.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XQurmys5Fb5dYRrVvbDsbQ.png" /><figcaption><em>Figure 1. Example of the Crop Type Detection service.</em></figcaption></figure><h4>Parcel Delineation</h4><p>The Parcel Delineation, a part of KappaZeta’s array of tools, focuses on the critical task of accurately mapping agricultural land parcels. Utilizing images from Sentinel satellites, the tool provides detailed and precise outlines of farm plots. This product is particularly valuable for Earth observation data analysis companies, as it enhances their ability to develop downstream applications, especially in scenarios where existing parcel boundaries are not readily available. By providing detailed and accurate delineations, the product enables these companies to generate more refined insights into land use, crop health, and environmental monitoring. Government entities benefit from this in land use planning and policy implementation, ensuring fair and efficient allocation of resources and compliance with agricultural regulations. Beyond its operational value, accurate parcel delineation is also a step towards more responsible and sustainable agricultural practices, as it helps in better understanding and managing land resources.</p><h4>Seedling Emergence Detection</h4><p>The Seedling Emergence Detection service addresses a critical early stage in the agricultural cycle. This service effectively identifies the emergence of seedlings across various crop types. This early detection capability is invaluable for farmers and agronomists, enabling them to swiftly assess germination success and take timely actions if needed, such as re-sowing or adjusting crop management practices.</p><p>By knowing the exact emergence dates, insurance providers can better evaluate the vulnerability of winter crops to early-season adversities such as frost, or pest attacks, leading to more accurate premium calculations, portfolio management and efficient claim management.</p><p>For Earth observation companies, this service adds a crucial layer of data, enhancing their ability to provide comprehensive agricultural analyses and insights. In contexts where early growth stages are critical for yield prediction and risk assessment, this tool provides a significant advantage. Furthermore, the Seedling Emergence Detection assists in fine-tuning irrigation and fertilization plans, contributing to more efficient and sustainable farming methods. Its utility extends to research and policy planning, offering data that can inform studies on crop development and agricultural strategies.</p><h4>Farmland Damage Assessment</h4><p>The Farmland Damaged Area Delineation specializes in mapping areas within agricultural lands that have sustained damage by adverse events such as extreme weather, pests, or disease outbreaks. For farmers, this tool is invaluable for quickly pinpointing affected areas, enabling them to implement targeted responses such as reallocating resources, adjusting irrigation, or applying specific treatments to the damaged zones. In the realm of crop insurance, this service provides crucial data for the prompt and accurate processing of claims, offering objective, verifiable evidence of the extent and location of damage, without even going to the field. This service is additionally vital in disaster response and management, assisting government agencies in efficiently directing resources and aid to the most affected regions. Furthermore, the data gathered by this product can contribute to long-term agricultural planning and environmental monitoring, helping to understand patterns of damage and inform future mitigation strategies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fDwBMvmoHuV4Cx0zi1uH_A.png" /><figcaption>Figure 2. <em>Example of the Farmland Damage Assessment service. Crop type: Winter Barley, field size: 36.06ha, damaged area: 9.19ha (25.49%).</em></figcaption></figure><h4>Detecting Ploughing and Harvesting Events</h4><p>The Ploughing and Harvesting Events’ date Detection service is a critical tool for monitoring key agricultural activities. For crop insurance companies, this information is essential in assessing the timing and methods of farming practices, which are integral factors in risk assessment and claim verification. This tool also plays a significant role in compliance with agricultural policies. Specifically, for government agricultural paying agencies, the detection of ploughing events is mandatory under the CAP2020 policy. The product’s ability to provide accurate and timely data ensures that these agencies can effectively monitor and enforce compliance with agricultural policies. Additionally, cultivation data offers valuable insights into farming patterns and their environmental impacts, assisting in the development of more sustainable farming practices which is a key component of carbon farming project monitoring.</p><h4>Empowering the Future of Agriculture with AI and Satellite Data</h4><p>KappaZeta’s innovative suite of AI and machine learning-based tools, utilizing Sentinel-1 and Sentinel-2 satellite data, represents a significant advancement in agricultural technology. These tools address key areas of agricultural management and analysis, catering to a diverse range of users including scientists, government officials, and professionals in farming, farm management, and insurance.</p><p>Overall, these tools collectively enhance data-driven decision-making in agriculture, leading to more efficient and sustainable practices. They demonstrate the pivotal role of advanced satellite analytics in transforming modern agriculture.</p><p>The prototypes for all five services were developed during the project “Satellite monitoring-based services for the insurance sector — CropCop”, supported by the European Regional Development Fund and Enterprise Estonia.</p><p><em>This Post was first published on </em><a href="https://kappazeta.ee/blog/five-new-satellite-analytics-tools-for-agriculture/"><em>https://kappazeta.ee/blog/five-new-satellite-analytics-tools-for-agriculture/</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=645540208327" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/five-new-satellite-analytics-tools-for-agriculture-645540208327">Five new satellite analytics tools for agriculture</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Adventures in the realm of Synthetic NDVI]]></title>
            <link>https://medium.com/kappazeta/adventures-in-the-realm-of-synthetic-ndvi-2dd52ba842f3?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/2dd52ba842f3</guid>
            <category><![CDATA[ndvi]]></category>
            <category><![CDATA[sentinel-2]]></category>
            <category><![CDATA[earth-observation]]></category>
            <category><![CDATA[generative-adversarial]]></category>
            <category><![CDATA[sentinel-1]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Thu, 21 Sep 2023 12:10:06 GMT</pubDate>
            <atom:updated>2025-06-05T11:55:26.380Z</atom:updated>
            <content:encoded><![CDATA[<p>Written by <em>Karl Hendrik Tamkivi, Machine Learning Intern in KappaZeta</em></p><p>Modern agriculture is increasingly merging with technological advancements, especially in the area of remote sensing. With climate conditions becoming more unpredictable, there is a clear need for resilient agricultural monitoring systems. Manual checks of crop damages or specific agricultural activities can be tedious and inefficient, prompting both farmers and crop insurance companies to lean towards automated detection systems. This not only aids in identifying crop failures early but also assists in tracking different growth stages. At the same time, agricultural paying agencies are looking for automated ways to monitor field activities, such as ploughing and harvesting, in order to increase the cost-effectiveness of their work. A promising solution to these varied needs is the development of synthetic vegetation indices, notably the synthetic NDVI (Normalized Difference Vegetation Index).</p><p>Crafting reliable generative deep learning models comes with its challenges. One key challenge is integrating SAR (Synthetic Aperture Radar) and optical (RGB and NIR) remote sensing data, especially from satellites like Sentinel-1 and Sentinel-2 of the European Union and the European Space Agency (ESA). While this data is freely available, using it effectively requires precise methodologies. The performance of these GAN (Generative Adversarial Network) models largely depends on the right feature selection. For SAR, the choice between 6-day and 12-day temporal baselines for interferometric coherence is crucial. Here, KappaZeta’s expertise comes into play, making the complex task of processing SAR data much more manageable. Historically, KappaZeta has developed models for both coherence durations. The 6-day models have shown good results in capturing swift changes in crop growth, especially for non-cereal crops, while the 12-day models were better suited for predicting events like crop decline as the season ends.</p><p>The existing set of generative models would have been sufficient enough if Sentinel-1B would not have had a mission-ending electrical anomaly in December of 2021. This incident made the 6-day coherence model unfeasible, at least until the launch of Sentinel-1C, leading to a significant question: Could the existing 12-day model be adapted and improved to function as an all-encompassing synthetic NDVI generative model, bridging the gap of the two previous models? This was the starting ground to my expedition into the world of synthetic vegetation indices and to everything else this captivating field had to offer.</p><p>Generative Adversarial Networks (GANs), used to generate artificial images of synthetic vegetation index values, are often perceived as rather complex neural networks to design. A primary source of this complexity stems from the dual structure involving two intertwined networks — a generator and a discriminator — that are trained simultaneously, leading to intricate training dynamics. Achieving equilibrium between the generator and the discriminator, where neither outperforms the other, often necessitates meticulous tuning, a balance that can be challenging to achieve. GANs are also notorious among machine learning engineers for the inherent difficulties in assessing the quality and diversity of generated outputs. Such assessments are typically non-trivial, often demanding subjective or indirect measures, complicating quantitative performance evaluation. To demystify this enormous complexity for myself, I resolved to dissect the problem from three distinct perspectives: model architecture, data manipulation, and performance evaluation. The subsequent sections provide a brief recap of my explorations and some intriguing findings within these domains.</p><p>We always aim for our models to generate precise and accurate outputs and this often requires delving into complex neural networks that can learn intricate patterns beyond human comprehension. With GANs, the myriad of model architecture design possibilities presents a challenge and it’s hard to predict which designs will be effective and which might end up being counterproductive. When dealing with geospatial remote sensing data, especially in the context of agricultural parcels and making estimations about their conditions, it’s crucial for the model to produce accurate results while maintaining good spatial awareness and the ability to generate visually compelling, believable outputs. However, the task of measuring and assessing these attributes is tauntingly complex, not even mentioning the difficulty of the pursuit of solutions to attain them.</p><p>In networks that employ convolutional layers and also deconvolutions, like GANs typically do, one strategy to at least theoretically enhance spatial awareness is by enlarging the kernel sizes within those layers. Not only can this adjustment improve spatial sensitivity, but it can also mitigate the visual checkerboard artefacts commonly observed in outputs from models that utilise deconvolutional layers. It is rarely the case that a theoretically posited solution translates seamlessly into practical efficacy, yet that’s precisely what happened with the increased kernel size in this context (Fig. 1). However, seemingly straightforward solutions like this seldom come without trade-offs. Enlarging the kernel size can significantly inflate the model’s memory footprint and elongate training durations — considerations of utmost importance when resources are constrained.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rFzucIUVXcymuP9_Hefhwg.png" /><figcaption>Figure 1. From left to right: Original Image: The initial, untouched image. Initial Model Output: Notice regular pattern distortions and lack of spatial detail. Enhanced Model Output: Result from an improved model with adjusted kernel size and updated interpolation techniques. Notice the clearer and more spatially accurate representation</figcaption></figure><p>A well-suited loss function often holds the key to achieving outstanding results and efficiently optimising a model. However, discovering the right one is seldom straightforward. In the realm of generative models, it’s beneficial to incorporate loss functions that evaluate image-specific attributes such as luminance, contrast, and structure. These attributes help in gauging the quality of the output image in relation to the actual target image. When integrated with traditional loss functions like L1 or L2, the results can potentially be further enhanced. The application of an image-centric loss function notably improved predictions of NDVI values in these experiments, registering enhancements both metrically and visually. Yet, if our primary concern revolves around the agricultural parcels — which often represent just a small fraction of the entire image — why prioritise the whole image? Integrating masking techniques within the loss function enables the model to zone in on these specific agricultural plots, sidelining extraneous areas. This approach is particularly valuable considering that changes in index values within the parcels exhibit different temporal patterns compared to their surroundings. By masking these surrounding areas, the model hones in on the parcels with greater accuracy. However, it’s a double-edged sword: while predictions for the parcels improve, those expecting a visually congruent comparison between the model’s output and the original image might be left wanting. Models trained with such a specialised loss function tend to falter when predicting areas outside of the parcels.</p><p>If there ever was a god of machine learning, it would probably declare: ‘Data is at the core of it all!’ Truly grasping the nuances of data, which can often be more complex than anticipated, is the foundation of any successful model that is built upon it. Geospatial remote sensing data brings its own set of challenges, some of which one would never even think about when working with classical tabular datasets. But after days and weeks of agony it becomes evident that these challenges can also be turned into assets. Considering satellite pass direction during the model training process is a prime example of this phenomenon. Satellites, as they orbit the Earth, can capture the same location from different directions across their various passes. Depending on whether the satellite is on an ascending or descending pass when taking an image, there can be slight variations in the angular patterns of the output. These minute differences might escape the human eye, but neural networks excel at detecting them. Remarkably, integrating information about the input image’s direction can significantly boost a model’s accuracy and computational efficiency as well. The key takeaway here is that sometimes, challenges can be cleverly repurposed into strengths.</p><p>In standard image classification problems, evaluating neural networks’ performance is usually rather straightforward with metrics like accuracy or the F1-score. However, when the focus shifts to continuous scales typical of different vegetation indices for example, evaluation nuances arise. Common metrics like Mean Absolute Error (MAE) or Root Mean Squared Error (RMSE) become essential and often satisfy the needs, but in remote sensing, there is more than just numeric accuracy to consider. Assessing the visual “goodness” and spatial integrity of the output images becomes paramount. This involves looking at the texture and smoothness of the image, its consistency with adjacent areas, its boundary delineations etc. Beyond the visual and spatial aspects, computational efficiency is a crucial aspect as well — the time taken for the model to process, its memory usage and activation patterns are all indicators of its efficiency. In essence, remote sensing brings forth multifaceted challenges that require a combination of quantitative metrics, qualitative assessments and often rather non-standard evaluation methods.</p><p>Fundamental probability theory asserts that predicting an exact value from a continuous distribution is impossible. This insight underscores the notion that while we should strive to minimise the error in predicting each pixel’s NDVI value, achieving absolute accuracy is just unattainable. Fortunately, in many contexts, exactitude isn’t even necessary. Often, synthetic NDVI serves as an alert system, signalling anomalies such as suboptimal wheat growth or a grassland not being mowed by a certain date. This leads to the idea of grouping pixels according to their index values or even further, we can look at how these values change compared to older images of the same location, much like a traffic-light system indicates status. This grouping makes it simpler to compare a model’s results with the actual image, helping differentiate between high-performing and subpar models. The objective, then, is to ensure that the class distributions in the two images align as closely as possible (Fig. 2). Furthermore, if we want to emphasise the importance of minimising significant errors, it’s beneficial to analyse the proportion of pixels with substantial disagreements between the model’s output and target image. This rather pragmatic simplification laid the groundwork for the Disagreement Index (DI), which emerged as a pivotal visual output evaluation metric in various experiments and helped to assess the model’s ability to predict the direction of change (Fig. 3). Pursuing an all-encompassing solution is often elusive, especially in the beginning stages. Thus, strategic simplifications, despite their imperfections, can pave the path toward ultimate objectives.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_7O1KILjN-O_pC4NcgT20A.png" /><figcaption>Figure 2. Upper row: NDVI value changes compared to a historical image (not shown) for both the actual image (left) and model generated image (right). Bottom row: Agreement map between the images in the upper row.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GzzUuCaQ2trz4dax06BNUw.png" /><figcaption>Figure 3 . Agricultural parcels on the axis of DI and share of big disagreement pixels. Black dashed line represents the parcel ‘ideal zone’.</figcaption></figure><p>Topics described in previous paragraphs make up just a small share of experiments carried out during my two-month stay at KappaZeta and even smaller share of all of the possible aspects to further test and develop about the synthetic NDVI models in the future. One is for certain — some talented and dedicated people are making sure that KappaZeta will be the leading provider of reliable synthetic vegetation index services in the forthcoming years!</p><p><em>This Post was first published on </em><a href="https://kappazeta.ee/blog/adventures-in-the-realm-of-synthetic-ndvi/"><em>https://kappazeta.ee/blog/adventures-in-the-realm-of-synthetic-ndvi/</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2dd52ba842f3" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/adventures-in-the-realm-of-synthetic-ndvi-2dd52ba842f3">Adventures in the realm of Synthetic NDVI</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A story of starting a satellite mission]]></title>
            <link>https://medium.com/kappazeta/a-story-of-starting-a-satellite-mission-ab40f188c038?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/ab40f188c038</guid>
            <category><![CDATA[earth-observation]]></category>
            <category><![CDATA[remote-sensing]]></category>
            <category><![CDATA[satellite-technology]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Thu, 06 Apr 2023 09:27:20 GMT</pubDate>
            <atom:updated>2023-04-06T09:52:52.218Z</atom:updated>
            <content:encoded><![CDATA[<p>Written by <em>Kaupo Voormansik, CEO of KappaZeta</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DYlodGWpOcjTgTL3FHxqsA.png" /><figcaption>The mission concept: constellation of passive receiver satellites formation flying with Sentinel-1.</figcaption></figure><h3><strong>How it began</strong></h3><p>It all started in November 2021 when TalTech space programme manager Rauno Gordon convened all the interested parties to discuss the idea of a new „Estonian satellite“. We met with the other enthusiasts on several occasions to discuss ideas — the common denominator being the desire to develop a mission useful for Estonia. Soon enough I was able to convince everyone that building a new satellite to observe just Estonia is not the way to go. Instead, we should think about how to use satellite remote sensing to benefit the world. Most of the other participants at the meeting seemed to agree.</p><p>We had been thinking of the idea of enriching the Sentinel-1 dataset for a long time. At first, we hoped ESA would do it by adding the height dimension using accompanying missions to Sentinel-1 and by making the next generation fully polarimetric. Discussions on this topic showed that ESA was not planning to do it themselves but would be willing to support third parties who develop such a mission. OK, great — I pitched this idea at the next „Estonian satellite“ meeting and it was accepted quite well. Adding a small and relatively inexpensive companion to enhance the data of a large and important mission seemed both feasible and meaningful. After one or two more meetings, I realized that in order to make it actually happen, I have to take the lead. To stand up and say that we (i.e KappaZeta) will start organising it, and others will then join too. We had what was necessary to get the snowball rolling — systematic knowledge about SAR, personal experience from developing ESTCube-1, an understanding of the Earth observation market and about the importance of the dataset to be developed. As the last and very important component, we also had a decent network within the Estonian and other European countries’ space technology and remote sensing communities, to bring the entire vision into reality. Today’s KappaZeta could not do it alone, but with help from friends — absolutely!</p><p>If you’re interested in the technical content of the planned mission, please see the abstract from the Finnish Satellite Workshop Remote Sensing seminar: <a href="https://drive.google.com/file/d/1rhppDdSWLjU7CXeanmLWo1Vl4xLE3h8_/view">https://drive.google.com/file/d/1rhppDdSWLjU7CXeanmLWo1Vl4xLE3h8_/view</a>.</p><p>In the beginning of the story, I referred to the „Estonian satellite“ using quotation marks, but despite all our national pride, we shouldn’t emphasize its nationality. The more countries that consider the Sentinel-1 accompanying satellite constellation to be their own, the better. Satellite remote sensing is inherently global and international cooperation its key to success, helping to avoid unnecessary fragmentation and duplication, and helping to progress further together. A satellite that is suitable for gathering data about Estonia is just as capable of doing the same for all other countries of the world.</p><h3><strong>Gathering feedback</strong></h3><p>By now, we’ve been preparing the mission for about a year without any dedicated budget and we’ve started involving the first investors. This has been exciting for the most part, but not without smaller drawbacks. The first surprising experience was the unbelievably supportive reception of the idea by ESA experts at informal discussions. We received many practical suggestions for both work organisation and involving international partners.</p><p>To sum up the feedback received so far, it seems that successful people and organisations are willing to help and encourage our ambitions. They want us to succeed and are willing to provide good advice and sometimes their own effort and personal contacts. The ones who have been critical of the mission concept seem to be the ones with less success. But we’re not going to focus on them. What was the Asian saying — see no evil, hear no evil, speak no evil?</p><p>One of the best experiences from the preparatory phase happened last autumn. The founder and CEO of a rapidly growing Estonian tech company with a valuation of hundreds of millions, finds an hour from his busy schedule for us. Within five minutes we „reach the same wavelength“, followed by an incredibly intense exchange of information. He quickly grasps what we want to do, we receive practical tips and contacts to help us implement the mission. I leave with a great feeling — he sincerely wants us to succeed and will cheer us on along the way.</p><h3><strong>Some interesting stories and criticism</strong></h3><p>One of the most interesting critical remarks came from a good colleague: „Your undertaking has too much of a community feel to it“. I would say that’s completely intentional. KappaZeta is a do-good company with a mission and that’s no secret. We’re convinced that an open approach will be successful, as opposed to being afraid, trying to hide, building walls wherever possible, keeping our work and plans a secret. The Copernicus programme, which the new mission would complement, is a success largely thanks to its open philosophy.</p><p>All interested parties need to know that the door for discussing cooperation is open, be that for benefitting from the planned data or for collaborating on satellite development. Experience from the first Estonian satellite project ESTCube-1 taught me many lessons that led to this conclusion. I’m glad that I never started criticising the new „aliens“ who had just joined the team and didn’t seem very reliable at first sight. Many of them later surprised the entire team, proving to be thorough and careful engineers that could always be relied upon. The second lesson was that each person plays their own role and not every person has to be very profoundly involved in development. For example, one ESTCube-1 member never stood out with great engineering results but brought a friend to the team who became one of the core developers. All of this demonstrates the power of community, which is enormous. Everyone needs to feel welcome to join and contribute. This is especially important in the beginning, when the entire competence to complete a mission doesn’t already exist within the team.</p><p>More criticism on the same note: „It doesn’t matter that you have a lot of friends!“. It matters tremendously! One person alone, however smart, and capable, could not get a lot done. Looking at the lively and supportive cooperation between Estonian startups, it seems that most people have understood this already long ago. From our experiences, we can thank the SuperAngel accelerator, ESA BIC Estonia, the Estonian Founders Society, and the Estonian Space Office.</p><p>To sum all of this up — there’s still a lot of work ahead, but there aren’t any fundamental obstacles preventing the mission from becoming a reality. The only thing that might make us reconsider our plans, would be if a better alternative technology were to appear. As long as our planned mission (bistatic SAR interferometry using accompanying satellites) provides the best combination of high accuracy, global coverage, and a reasonable price for measuring the global forest biomass, we will be working onwards.</p><p>Today, we have verbal agreements to raise 2,35 million euros in investments. Hopefully, we can publish official news about it soon. I extend my sincere gratitude to all our team members, advisors, the future investors, and the constructive critics.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ab40f188c038" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/a-story-of-starting-a-satellite-mission-ab40f188c038">A story of starting a satellite mission</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[SNDVI — Synthesizing NDVI images from SAR)]]></title>
            <link>https://medium.com/kappazeta/sndvi-synthesizing-ndvi-images-from-sar-d6f281c82482?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/d6f281c82482</guid>
            <category><![CDATA[sars]]></category>
            <category><![CDATA[earth-observation]]></category>
            <category><![CDATA[ndvi]]></category>
            <category><![CDATA[sndvi]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Wed, 08 Feb 2023 10:02:54 GMT</pubDate>
            <atom:updated>2025-06-05T11:54:43.739Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>SNDVI — </strong><em>Synthesizing NDVI images from SAR)</em></h3><p>If you rely on optical satellite imagery for vegetation monitoring, clouds are an unwanted phenomenon. For simpler applications using parcel statistics, it may be straightforward to interpolate missing band or index values. However, for image-dependent use-cases, including field management zoning or crop damage detection amongst others, cloudy days simply mean little to no meaningful data.</p><p>In this article we share results from applying generative modeling to synthesize NDVI images that reasonably account for within-field changes in agricultural parcels (<em>Figure 1)</em>. We modified the <strong><em>pix2pix</em></strong> generative architecture and used multi-temporal inputs that include complementary radar (SAR) data to synthesize NDVI to up to 0.07 MAE on 70.000+ evaluated parcels.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EPqlHtgLs4YXvEbQ8S2ZNQ.png" /><figcaption><strong><em>Figure 1.</em></strong><em> Cloudy NDVI image and SNDVI alternative (See more examples on our </em><a href="https://map.kappaone.eu/ard_sndvi_demo/"><em>demo map</em></a><em>)</em></figcaption></figure><h4><strong>Modeling Approach</strong></h4><p>We train a modified <strong><em>pix2pix</em></strong> architecture on 512 × 512 px sub-tiles of <strong>aligned</strong> SAR and optical imagery covering (mostly) agricultural parcels from areas in Northern Estonia. Model inputs are a recent SAR image +- 2 days from the NDVI image to be synthesized, whilst historical SAR and optical image sub-tiles are (max) 30 days from target NDVI. A recent variation to this approach, adds <strong>recent</strong> optical inputs (RGB, NIR, NDVI) to the SAR image, applying our in-house cloud mask (<a href="https://github.com/kappazeta/km_predict"><em>KappaMask</em></a>) to predict NDVI for occluded areas only. This yields better performance for larger heterogenous areas.</p><h4><strong>Evaluating Synthetic NDVI Imagery</strong></h4><p>To validate synthesized images, we initially rely on full-reference image metrics including SSIM, PSNR and MAE, comparing target NDVI with synthesized NDVI (SNDVI) sub-tiles, as shown in Table 1.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3fPw3qaeZWCcrJkg4hgGyA.png" /><figcaption><strong>Table 1.</strong> Sub-tile image quality assessment for unseen test area (NW Estonia)</figcaption></figure><p>However, focusing more on within-parcel changes, we compare how well the absolute NDVI change for an individual parcel in consecutive target images has been captured by corresponding predicted images, and the (image histogram) correlation between real and predicted NDVI changes. For vegetation monitoring, the applicability of NDVI can be linked to monitoring crop phenology, and we began exploring this use-case with a broad categorization of observed changes w.r.t NDVI increase (crop growth) or decrease (crop decline/maturity).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8Olpz7gN6D02gESfUQBrIQ.png" /><figcaption><strong><em>Figure 2.</em></strong><em> Comparing Historical Changes </em>(<em>hmae=abs(hndvi− ndvi))(hmae=abs(hndvi− ndvi)</em>) <em>and prediction errors </em>(mae= abs(sndvi− ndvi))(mae= abs(sndvi− ndvi))</figcaption></figure><p><em>Figure 2</em> describes the overall change in NDVI (between historical and target NDVI), compared with the MAE prediction error between SNDVI and true/target NDVI for observed parcels. In summary, NDVI synthesis is more accurate for crop growth events, and less accurate for crop decline and the sudden changes (including, plowing, harvesting) that influences these events.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*edBKwuUvn_x8sb_wNX_Jjw.png" /><figcaption><strong>Figure 3. </strong>Comparing heterogenous parcel changes between HNDVI, SNDVI, NDVI subtiles. (Historical inputs lag: 17 days, NDVI date: June 8, 2018)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*36KqO80fyTUMBXnD_TWIww.png" /><figcaption><strong>Figure 4.</strong> Comparing within-parcel changes (from sub-tile in <strong>Figure 3</strong>) between HNDVI, SNDVI, NDVI. (Crop: Spring barley, Historical inputs lag: 17 days, NDVI date: June 8, 2018)</figcaption></figure><h4><strong>Conclusion, limitations and future work</strong></h4><p>In this article, we introduced a generative approach to synthesizing NDVI images from multi-temporal SAR and historical optical data. We showed that the performance of our MTCGAN model is significantly better for predicting NDVI increase compared to decrease. For this decrease or crop removal/maturity, an important factor is the contrast between interferometric coherence in historical and target SAR images. After a crop removal event, at least 2 SAR images are required for the coherence to reflect the change, compared to crop growth with less drastic coherence changes. Accounting for this SAR limitation will be key in improving results for this case.</p><p>Another useful application for monitoring agencies and/or farmers may be zoning or detecting fields and areas within fields which demonstrate homogenous or heterogenous growth. Initial analyses show F1-score accuracies of 71% for low NDVI variance and 42% for high variance classes. Continuing work will evaluate crop-specific cases before concluding analyses.</p><p>For some examples of AI-generated images, visit our SNDVI demo map <a href="https://map.kappaone.eu/ard_sndvi_demo/">here</a>. In addition to SNDVI created for 6-days coherence, complementary rasters created from 12-days coherence will be added, and in the future, other AI-derived vegetation indices for crop monitoring.</p><p><strong>Acronyms</strong></p><p>Normalized Difference Vegetation Index (NDVI), Synthetic Aperture Radar (SAR), Synthetic NDVI (SNDVI), Historical NDVI (HNDVI), Mean Absolute Error (MAE), Historical MAE (HMAE), Structure Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR)</p><p><em>This Post was first published on </em><a href="https://kappazeta.ee/blog/sndvi-synthesized-ndvi-from-sar/"><em>https://kappazeta.ee/blog/sndvi-synthesized-ndvi-from-sar/</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d6f281c82482" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/sndvi-synthesizing-ndvi-images-from-sar-d6f281c82482">SNDVI — Synthesizing NDVI images from SAR)</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Make the globe cloud-free with KappaMaskv2]]></title>
            <link>https://medium.com/kappazeta/make-the-globe-cloud-free-with-kappamaskv2-a0b52b55ab0c?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/a0b52b55ab0c</guid>
            <category><![CDATA[earth-observation]]></category>
            <category><![CDATA[cloud-masking]]></category>
            <category><![CDATA[sentinel-2]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Wed, 04 Jan 2023 12:57:42 GMT</pubDate>
            <atom:updated>2025-06-05T11:54:02.748Z</atom:updated>
            <content:encoded><![CDATA[<p>Last year KappaZeta released KappaMask, an AI-based cloud masking processor for Sentinel-2. It was designed specifically for the Northern European terrestrial summer conditions and used U-Net as a cloud detector. You can read more here: <a href="https://www.mdpi.com/2072-4292/13/20/4100">https://www.mdpi.com/2072-4292/13/20/4100</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*uJnf4iN_LKoqM0Dc.png" /><figcaption>Sentinel-2 product obtained somewhere over Estonia</figcaption></figure><p>KappaMask’s impressive performance motivated its further development. We aimed to expand it to global conditions. That took quite a while but here it is (drumroll…)<strong> KappaMaskv2.</strong> In this blog post, we will briefly present to you the process of KappaMaskv2 development, including dataset compilation, model training, and the most exciting, performance results. Check out the full blogpost <a href="https://kappazeta.ee/blog/make-the-globe-cloud-free-with-kappamaskv2">here</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*3ZAZAvtyXeef3SWL.gif" /><figcaption>The overall overflow of KappaMaskv2</figcaption></figure><h3><strong>KappaSet development</strong></h3><p>KappaMaskv2 is an AI-based cloud and cloud shadow processor whose quality highly depends on the training dataset. There are only a few datasets publicly available. Therefore, KappaZeta went through deserts and oceans, mountains and rainforests to create the one and only, <strong>KappaSet, a novel and diverse cloud and cloud shadow dataset</strong>. Check out the dataset distribution below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*cX6lTkZdKMWhr-Qj.png" /><figcaption>KappaSet train distribution</figcaption></figure><p>KappaSet was generated using<strong> </strong>an<strong> </strong><a href="https://en.wikipedia.org/wiki/Active_learning"><strong>active learning</strong></a><strong> method</strong>, where <strong>sub-tiles with the most significant impact</strong> on the model performance were sorted out for labelling. It consists of <strong>9251</strong> labelled 512x512 sub-tiles, where 10% contain water bodies and around 7% include snow cover. <br>We genuinely hope KappaSet will benefit the research community and boost cloud masking development. <strong>KappaSet is available here: </strong><a href="https://zenodo.org/record/7100327">https://zenodo.org/record/7100327</a></p><h3>Model architecture</h3><p>The main improvement over the original KappaMask is the cloud detector architecture. We went <strong>from U-Net to DeepLabv3</strong>+. One might ask: why not use U-Net again if it was already accurate? First, it performed better on semi-transparent clouds. Secondly, it allowed us to reduce the size of our model which also reduced inference time.</p><h3>Performance results</h3><p>For a fair evaluation of any model, it is crucial to have <strong>a diverse and challenging</strong> test set. We ensured that the test set includes both similar and dissimilar sub-scenes seen in the training set. There are <strong>803 sub-tiles in the test set</strong>. You can find its distribution below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*SjbmZPhuVm2GL0JJ.png" /><figcaption>KappaSet test set distribution</figcaption></figure><p>KappaMaskv2 was compared to other cloud masking processors, including <strong>Sen2Cor</strong>, <strong>Fmask</strong>, <strong>MAJA</strong>, <strong>IdePix</strong> and <strong>S2Cloudless</strong>. In Figure below, you can see that KappaMaskv2 generally performs better in every class. Fmask prediction is close to the KappaMask prediction, except for water being classified as a cloud. Sen2Cor and IdePix correctly identified cumulus clouds when cloud shadows and semi-transparent clouds are underestimated. Sen2Cor example shows that some semi-transparent clouds and cloud shadows are detected as missing class.</p><p>S2Cloudless prediction can be defined as “close to the ground truth label” if semi-transparent clouds and cloud shadows are considered. The general picture shows that clouds are overestimated. However, if the semi-transparent class is counted as a cloud, the prediction looks more accurate. MAJA prediction is similar to S2Cloudless, although some pixels are misclassified as clear.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*Rzq8nTg-u1ttTHdb.png" /><figcaption>Comparison of L1C prediction output for a 512 × 512 pixels sub-tile in the test dataset. By design, Fmask and S2Cloudless do not have semi-transparent class included. S2Cloudless does not have a cloud shadow class.</figcaption></figure><h3>Summary</h3><ul><li>We compiled <strong>cloud and cloud shadow dataset</strong>, named <strong>KappaSet</strong>. It consists of <strong>9251</strong> labelled sub-tiles at 10 m resolution, from Sentinel-2 Level-1C products. We believe that KappaSet will benefit the research on cloud masking. <strong>KappaSet can be found </strong><a href="https://zenodo.org/record/7100327"><strong>here</strong></a><strong>.</strong></li><li>We presented <strong>KappaMaskv2</strong>, an AI-based cloud and cloud shadow processor for Sentinel-2, <strong>which operates at the global scale</strong>. It was compared to other cloud masking approaches on the carefully selected test set. The results showed that KappaMaskv2 significantly improved cloud shadow and semi-transparent cloud detection. <strong>KappaMaskv2 is available </strong><a href="https://github.com/kappazeta/km_predict"><strong>here</strong></a><strong>.</strong> <strong>Feel free to try it out and share your experience with us!</strong></li></ul><h4>Acknowledgements</h4><p>The activity AI-based tillage detection for improved agricultural and climate policies was carried out under a programme of, and funded by, the European Space Agency.</p><p><em>This Post was first published on </em><a href="https://kappazeta.ee/blog/make-the-globe-cloud-free-with-kappamaskv2/"><em>https://kappazeta.ee/blog/make-the-globe-cloud-free-with-kappamaskv2/</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a0b52b55ab0c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/make-the-globe-cloud-free-with-kappamaskv2-a0b52b55ab0c">Make the globe cloud-free with KappaMaskv2</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data splitting challenge]]></title>
            <link>https://medium.com/kappazeta/data-splitting-challenge-b2def7d6e8aa?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/b2def7d6e8aa</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[mowing-detection]]></category>
            <category><![CDATA[crop-classification]]></category>
            <category><![CDATA[data-pre-processing]]></category>
            <category><![CDATA[data-splitting]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Fri, 16 Oct 2020 00:00:56 GMT</pubDate>
            <atom:updated>2025-06-05T11:53:22.667Z</atom:updated>
            <content:encoded><![CDATA[<p>Almost any machine learning pipeline requires an input data split for training and validation purposes. However, ground truth collection is challenging and could be gathered from different sources. Various sources provide different confidence levels for the labels and in general it would be beneficial to test the model on the most confident samples, but also providing some part of it for training as well, keeping the class distributions as uniform as possible. We are facing the challenge of having unbiased data split with adjustable filters in different tasks and it feels that there is a need for a more general solution or brainstorming from the community.</p><h4>Mowing detection</h4><p>One of the examples where we meet the splitting challenge is the mowing detection task. The goal of mowing detection is to predict the time at which the grass on the field was cut. Thus, as part of our mowing detection project, each year we collect field books from farmers and field reports from the inspectors of the local paying agency. The received data is converted and reviewed manually, and some of the ground truth is produced from the manual labelling.</p><p>The labels would differ in trustworthiness, depending on the source (farmer field books, inspector field reports, any of the former with manually adjusted dates, or fully manual labelling). Since inspector field reports are the most reliable source, we would use most of them for the validation and test set. However, we would need at least some of them to be present in the training dataset as well. Additionally, each dataset is expected to have as well balanced classes distribution as possible, perhaps with additional filtering to randomly drop least trustworthy samples from the over-represented class.</p><p>Considering the aforementioned conditions, let’s say we would like to have 70% of the labelled data for training, 20% for validation and 10% for testing. For validation and testing, we would only use instances from inspector field reports and farmer field books with tweaked dates. For training, we would use data from all sources, including the ones from inspector field reports and tweaked field books which were left over from the test and validation datasets.</p><h4>Crop classification</h4><p>Another task we are dealing with is crop classification. We would like to detect the crop type of agricultural fields out of 28 possible classes. Similarly to the mowing detection we have different sources for labels, some of which have been provided by the local Agricultural Registers and Information Board, some from drone observations. For crop classification, class balance distribution plays the core role. In order to mitigate the issue of an unbalanced dataset, undersampling and oversampling can be used. Undersampling and oversampling should be available for the training subset, while for testing we would use some of the fields with labels of high confidence. Some of the classes might have a poor representation, due to which general split ratios might leave the validation or test dataset without any samples, whereas we need to ensure that all datasets have enough samples.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*3bGWchxAgq-hdw6r.JPG" /><figcaption>Image credit: Madis Ajaots</figcaption></figure><p>Thus, the requirements for splitting are the following. We would like to have 70% / 20% / 10% splits, ensuring that for smaller representing classes at least one instance is present in all sets. Additionally for the test set we would like to have the list of high confidence instances together with random leftover samples that added up to 10% of the whole data.</p><h4>Generic and configurable</h4><p>While such processing chains can be implemented, we have found it tricky to have it generic and configurable enough to cater for all sorts of projects with different (and sometimes rapidly changing) requirements.</p><h3>Current solution</h3><p>Currently we have separate implementations for mowing detection and crop classification, both of which take input parameters from a config file. The config file is basically python code and supports the definition of custom filter functions for datasets. For each dataset, the current solution invokes custom filters (if any) and then performs random sampling of data indices, leaving the rest of the samples for the next datasets. The samples which have been filtered out, are also left for the next datasets, for each dataset might have a different filter.</p><p>The reason why we prefer to use data / sample indices instead of data directly, is to have a layer of abstraction. This way the splitting logic could be agnostic of data type. It would not matter whether a single sample / instance is a raster image, an image time-series or time-series of parameter values which have been averaged over a pre-defined geometry.</p><p>For multiclass applications such as crop classification, data indices are sampled separately for each class within each dataset. The splitting also supports capping of samples for classes which are represented too well. However, if there are too few samples per class, a low threshold can be applied such that a different split ratio would be used. For instance, in the case of 70% training, 20% validation and 10% testing dataset with just 9 samples in one of the classes, we might end up with 7 samples in the training dataset, 2 samples in validation and 0 in the test set. To mitigate the issue, we could have the ratios adjusted to 40% training, 30% validation and 30% testing for classes with less than 100 samples.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Ic7PlHgmPq7n82pH.png" /></figure><h3>Ideas for future developments</h3><p>Instead of project-specific implementation of the splitting logic, we would prefer to have a generic framework for graph-based data splitting with support for cross-validation and bagging. Please let us know if there is such a framework already out there, or if there would be community interest in developing the framework.</p><p><em>This Post was first published on </em><a href="https://kappazeta.ee/blog/data-splitting-challenge/"><em>https://kappazeta.ee/blog/data-splitting-challenge/</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b2def7d6e8aa" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/data-splitting-challenge-b2def7d6e8aa">Data splitting challenge</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Towards the operational service of crop classification — Kappazeta]]></title>
            <link>https://medium.com/kappazeta/towards-the-operational-service-of-crop-classification-kappazeta-3a67e5bbe649?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/3a67e5bbe649</guid>
            <category><![CDATA[sars]]></category>
            <category><![CDATA[sentinel-2]]></category>
            <category><![CDATA[sentinel-1]]></category>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[crop-classification]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Mon, 12 Oct 2020 00:00:44 GMT</pubDate>
            <atom:updated>2025-06-05T11:52:55.942Z</atom:updated>
            <content:encoded><![CDATA[<h3>Towards the operational service of crop classification</h3><p>We are about to finish a R&amp;D project, where we developed and tested crop classification methodology specifically suited for Estonian agricultural, ecological and climatic conditions. We relied mostly on Sentinel-1 and -2 data and used neural network machine learning approach to distinguish 28 different crop types. Results are promising and the methodology is ready for operational service to automate another part of agricultural monitoring.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*l9GXylANjTMcXinj.JPG" /></figure><p>Using machine learning in crop type classification is not new, and definitely not a revolutionary breakthrough — already for decades different classifiers (Support Vector Machine, Decision Trees, Random Forest and many more) have been used in land cover classification. Recently also neural networks, the <em>wunderkind </em>of machine learning and image recognition, are widely used in crop discrimination. Satellite data, as the main input to classification models, has no serious alternatives, since our aim is to implement it on worldwide scale and in applications, which run near real time. So, why even get excited about another crop type classification study, which exploits same methods and datasets as tens of previous studies?</p><p>I can give you one reason. Estonia has been very successful in following European Commission (EC) guidelines and rules in modernizing the EU Common Agricultural Policy. In 2018 EC adopted new rules that allow to completely replace physical checks on farms with a system of automated checks based on analysis of Earth observation data. The same year <a href="https://www.pria.ee/en">Estonian Agricultural Registers And Information Board</a> (ARIB) launched the first nation-wide fully automated mowing detection system, which uses Sentinel-1 and Sentinel-2 data and where the prediction model inside the system is developed by KappaZeta. The system has been running for 3 years, it has significantly reduced the amount of on-site checks and increased the detection of non-compliances. In short — saved Estonian and EU taxpayers’ money. Automated crop discrimination is the next step in pursuing the above-mentioned vision and will probably become the foundation of all agricultural monitoring. With proved and tested methodology, it’s highly likely that Estonia will take this next step in the very near future and launch it again on the nationwide level. This is definitely a perspective to be excited about.</p><p>Now, let’s see how we tackled “the good old” crop classification task.</p><h4><strong>Input data</strong></h4><p>Although algorithms and methods are important to make a difference in prediction model performance, the training data is the most valuable player in this game. In Estonia all farmers who want to be eligible for subsidies need to declare crops online (field geometry + crop type label). This <a href="https://opendata.riik.ee/en/andmehulgad/p-llumassiivide-register/">open dataset</a> is freely accessible to everyone and has the permission to re-use and redistribute both commercial and non-commercial purposes. Since the crop type labels are defined by farmers and most of them are not double-checked by ARIB, there can be mistakes (according to ARIB’s estimations, less than 5%). Therefore, for additional validation we ran our own cluster analysis on time-series to filter out obvious outliers in each class.</p><p>After we had the parcels and labels, we calculated time-series of different satellite-based, plus some ground-based features (precipitation, average temperatures, soil type). When extracting features from satellite images there are two ways to go: pixel- or parcel-based extraction. We selected the latter and averaged pixel values over the parcel to obtain one numerical feature value per statistic for each point in time (see Figure 1).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*0nsTYgSNar68mEOQ.png" /><figcaption>Figure 1. An example time-series of one Sentinel-1 feature (cohvh — 6-day coherence in the VH polarization) for one parcel.</figcaption></figure><p>For Sentinel-1 images preprocessing we have developed our own <a href="https://kappazeta.ee/time-series">processing chain</a> to produce reliable time-series for several features. From the previous studies it’s known that features (channel values and indexes) from Sentinel-2 images combined with features from Sentinel-1 images (coherence, backscatter) give better classification results than any of these features separately.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/578/0*K5Ks2Fqf1wRADzE3.png" /><figcaption>Figure 2. The whole dataset can be imagined as a three-dimensional tensor with the feature parameters on one axis, parameter statistics on another, and date-time on the third axis.</figcaption></figure><p>We used data from 2018 and 2019 seasons (altogether more than 200 000 parcels) and aggregated all crop type labels into 28 classes which were defined by the need of the ARIB.</p><h4><strong>Model architecture</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/305/0*wgsCJRB1YJtVcsxv.png" /><figcaption>Figure 3. Model architecture.</figcaption></figure><p>Due to the very unbalanced dataset we had to under-sample some classes and over-sample others for the training data. In small classes we used the existing time-series and added noise for data augmentation.</p><p>Model architecture was rather simple — input layer, flatten layer, three fully connected dense layers (two of them followed by batch normalization layer) and output (Figure 3). Our experiment with adding 1D CNN layer after input didn’t improve results significantly. More complicated ResNet (residual neural network) architecture increased training time by approx. 30%, but results were similar to a linear neural network.</p><h4><strong>Classification results</strong></h4><p>F1 score on validation set (9% of all dataset) was <strong>0.85</strong> and on test set (2% of all dataset) <strong>0.84</strong>. In 10 classes the recall was more than 0.9 and in 16 classes more than 0.8. See more from Figure 4 and 5.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/815/0*0Z7jql46e9v_vbWa.PNG" /><figcaption>Figure 4. Test set results.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*6kiEHvZCJgr9gqZo.png" /><figcaption>Figure 5. Normalized confusion matrix of the crop classification results (recall values).</figcaption></figure><h4><strong>Some features are more important than others</strong></h4><p>In a near-real time operative system our model and feature extraction would have to be as efficient as possible. For an R&amp;D project we could easily calculate 20+ features from satellite images, feed them all to the model and let the machines compute. But what if not all features are equally important?</p><p>They are not. We found that the 5 most important features are Sentinel-1 backscatter ( <em>s1_s0vh</em>, <em>s1_s0vv</em>), <em>NDVI</em>, <em>TC Vegetation</em> and <em>PSRI</em> from Sentinel-2. To our surprise, soil type and precipitation sum before satellite image acquisition had low relevance.</p><p>The 5 most important features played different role during the season — Sentinel-2 features were more important in the beginning and in the end of the season, while Sentinel-1 features had more effect during mid-season.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*DTSpsrmH95SBLJwa.png" /><figcaption>Figure 6. Importance of different features in crop classification, estimated using Random Forest.</figcaption></figure><h4><strong>What next?</strong></h4><p>This project was part of a much larger initiative called “National Program for Addressing Socio-Economic Challenges through R&amp;D. Using remote sensing data in favor of the public sector services.” Several research groups all over Estonia worked on prototypes to use remote sensing in fields like detecting forest fire hazard, mapping floods and monitoring urban construction. Now its up to Estonia’s public sector institutions to take the initiative and turn prototypes into operational services. With this work we have proved, that satellite-based crop classification in Estonia is possible, accurate enough and ready to be implemented as the next monitoring service for ARIB.</p><p><strong>If you are more interested about this study, our Sentinel-1 processing pipeline or machine learning expertise, then feel free to get in touch. We have the mentality to share not hide our experience and learn together on this exciting journey.</strong></p><p><em>This Post was first published on </em><a href="https://kappazeta.ee/blog/towards-the-operational-service-of-crop-classification/"><em>https://kappazeta.ee/blog/towards-the-operational-service-of-crop-classification/</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3a67e5bbe649" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/towards-the-operational-service-of-crop-classification-kappazeta-3a67e5bbe649">Towards the operational service of crop classification — Kappazeta</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Open access to ALOS-2 radar satellite data?]]></title>
            <link>https://medium.com/kappazeta/open-access-to-alos-2-radar-satellite-data-7d2a3c782e10?source=rss-e0cd7c265a0------2</link>
            <guid isPermaLink="false">https://medium.com/p/7d2a3c782e10</guid>
            <category><![CDATA[alos-2]]></category>
            <category><![CDATA[sars]]></category>
            <dc:creator><![CDATA[KappaZeta]]></dc:creator>
            <pubDate>Tue, 28 Apr 2020 00:00:52 GMT</pubDate>
            <atom:updated>2025-06-05T11:52:18.507Z</atom:updated>
            <content:encoded><![CDATA[<p>End of previous year (2019) Japan announced that the Japan Aerospace Exploration Agency (JAXA) will be providing open access to information and data from a suite of their radar satellites (original statement <a href="https://www.earthobservations.org/geoweek19.php?t=statements">here</a>). To be more specific, free and open access to the wide-swath observation data from the L-band radar satellites, ALOS (ALOS/AVINIR-2, PALSAR) and ALOS-2 (ALOS-2/ScanSAR) will be made available. The <a href="https://www.restec.or.jp/en/solution/product-alos-2">price</a> of ScanSAR images is at the moment around 700 euros.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*K0GJhlDUtvoJYT2Y.jpg" /><figcaption>ALOS-2 spacecraft in orbit (image credit: JAXA)</figcaption></figure><ul><li>Spotlight — spatial resolution 1x3 m, NESZ -24, swath 25 km.</li><li>Stripmap — spatial resolution 3–10 m, swath 30–70 km. Consist of Ultrafine (3 m), High sensitive (6 m) and Fine (10 m) modes.</li><li>ScanSAR — spatial resolution 60–100 m, swath 350–490 km.</li></ul><p>The Japanese space and satellite <a href="https://global.jaxa.jp/projects/sat/index.html">program</a> consist of two series of satellites — those used mainly for Earth observation and others for communication and positioning. There are 3 Earth Observation satellites in nominal phase, 3 in latter phase in operation and 3 more under development.</p><p>Greenhouse gases Observing SATellite-2 “IBUKI-2” (GOSAT-2) is measuring global CO 2 and CH 4 distribution of lower and upper atmosphere. Climate “SHIKISAI” (GCOM-C) satellite carries an optical sensor capable of multi-channel observation at wavelengths from near-UV to thermal infrared wavelengths (380nm to 12µm) to execute global, long-term observation of the Earth’s environment. Advanced Land Observing Satellite-2 “DAICHI-2” (ALOS-2) aims are to monitor disaster areas, cultivated areas and contribute to cartography.</p><p>ALOS-2, which is specifically interesting for radar enthusiasts, is a follow-on mission from the ALOS “DAICHI”. Launched in 2006, ALOS was one of the largest Earth observation satellites ever developed and had 3 different sensors aboard: PRISM (Panchromatic Remote-sensing Instrument for Stereo Mapping) for digital elevation mapping, AVNIR-2 (Advanced Visible and Near Infrared Radiometer type 2) for precise land coverage observation and PALSAR (Phased Array type L-band Synthetic Aperture Radar) for day-and-night and all-weather land observation. ALOS operations were <a href="https://global.jaxa.jp/press/2011/05/20110512_daichi_e.html">completed</a> in 2011, after it had been operated for over 5 years.</p><p>ALOS-2 was launched in 2014 and carries only radar instrument aboard. New optical satellite, ALOS-3, which will improve ground resolution by approx. three times from that of ALOS (2.5 to 0.8 m at nadir, wide-swath of 70 km at nadir), is already under development together with ALOS-4, which will take over from ALOS-2 to improve the functionality and performance.</p><p>Let’s come back to present day. The state-of-the-art L-band Synthetic Aperture Radar (PALSAR-2) aboard ALOS-2 have enhanced performance compared to its predecessor. It has a right-and-left looking function and can acquire data in three different observation modes:</p><ul><li>Wetlands and rapid deforestation regions in ScanSAR mode</li><li>Crustal deformation regions both in Stripmap Fine and ScanSAR mode</li><li>Polar regions both in Stripmap Fine and ScanSAR mode</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*g_VOU7-XHd-i5KbP.png" /><figcaption>PALSAR-2 specifications (images credit: JAXA)</figcaption></figure><p>Emergency observations have highest priority for ALOS-2, but for systematic observations <a href="https://www.eorc.jaxa.jp/ALOS/en/top/obs_top.htm">Basic Observation Scenario</a> (BOS) has been developed. This ensures spatial and temporal consistency at global scales and adequate revisit frequency. ALOS-2 BOS has separate plans for Japan and for the rest of the world, success rate for these acquisitions is 70–80 %.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*KuINapjveHh6jza2.png" /><figcaption>PALSAR-2 observation modes (images credit: JAXA)</figcaption></figure><p>Basic observations over Japan are mostly undertaken in Stripmap Ultrafine mode and sea ice observations during winter in ScanSAR mode.</p><p>Stripmap Fine and ScanSAR modes are used for global BOS. There are several areas of interest, where ALOS-2 is putting more focus, for example:</p><p>In addition to those special regions global land areas are observed in Stripmap Fine mode at least once per year.</p><p>We made a little experiment to test, how many acquisitions we get over city of Tartu per year. Here are the results (platform for viewing and ordering data is <a href="https://satpf.jp/spf_atl/?lang=en">here</a>):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*aQgeTkThIv47sgha.png" /><figcaption>Screenshot from Earth Observation Data Utilization Promotion Platform.</figcaption></figure><p>So, compared to Sentinel-1 radar-satellite, ALOS-2 acquisitions frequency is much lower over Europe, and its difficult to develop agriculture monitoring services only on this platform. For forestry and other environmental monitoring, where changes are not happening that often as in agriculture, ALOS-2 can be very useful due to its better spatial resolution than Sentinel-1. Being an L-band satellite it can also penetrate deeper into vegetation and provide information about the lower layers of the canopy. JAXA is already developing ALOS-4 with PALSAR-3 aboard, which will aim broader observation swath compared to the predecessor.</p><p><em>This Post was first published on </em><a href="https://kappazeta.ee/blog/open-access-to-alos-2-radar-satellite-data/"><em>https://kappazeta.ee/blog/open-access-to-alos-2-radar-satellite-data/</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7d2a3c782e10" width="1" height="1" alt=""><hr><p><a href="https://medium.com/kappazeta/open-access-to-alos-2-radar-satellite-data-7d2a3c782e10">Open access to ALOS-2 radar satellite data?</a> was originally published in <a href="https://medium.com/kappazeta">KappaZeta</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>