Analysis of Changes in Rice Harvest Timing Using Deep Learning- Based Climate Prediction

Analysis of Changes in Rice Harvest Timing Using Deep Learning- Based Climate Prediction | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Analysis of Changes in Rice Harvest Timing Using Deep Learning- Based Climate Prediction Sung Kook Kim¹, Jae Hyun Park¹ This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7354648/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study predicts future harvest dates for Samkwang rice across 12 major cultivation regions in South Korea using a deep learning-based time-series approach. A Long Short-Term Memory (LSTM) model was trained on daily average temperature data spanning 1995 to 2024 and subsequently used to forecast regional temperatures for the period 2025–2040. Harvest dates were estimated based on the day cumulative temperature after heading reached 1,150°C, in accordance with official agronomic guidelines. Prediction accuracy was evaluated using the Mean Absolute Error (MAE) for each region. The results indicate a general advancement in harvest dates, attributed to accelerated heat accumulation under ongoing climate warming. However, regional variations were observed: northern regions exhibited more delayed and variable harvest patterns, whereas southern regions demonstrated greater temporal stability. Unlike previous studies confined to specific areas, this research incorporates a broad latitudinal dataset, offering a generalized predictive model to support adaptive agricultural strategies in the face of climate change. Climate change Deep learning Harvest timing LSTM Rice cultivation Temperature prediction Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction 1.1 Background Rice is one of the world’s three major staple crops and serves as the primary food source for over half of the global population. In South Korea, it is the most extensively cultivated crop, covering approximately 697,714 hectares as of 2025 (KOSTAT) [ 1 ]. Although this represents a 1.5% decrease from the previous year, rice continues to play a critical role in national food security and the agricultural economy. However, rising temperatures have raised growing concerns regarding their effects on rice development. Recurrent abnormal weather events and the broader impacts of global warming have been reported to directly influence rice growth, emphasizing the urgent need for effective mitigation technologies [ 1 ]. Over the past two decades, the average annual temperature in South Korea has increased by 2.2°C—from 12.7°C in 2000 to 14.9°C in 2024 [ 8 ]—highlighting the intensification of climate change. In response, artificial intelligence (AI)-based technologies have gained attention, particularly deep learning, which has demonstrated powerful performance across diverse domains, including recent advances in meteorological forecasting [ 2 ]. This study applies deep learning techniques to time-series weather data to address climate-related challenges in agriculture. Daily average temperature data from 1995 to 2024 were collected from 12 representative rice-growing regions spanning a wide latitudinal range in South Korea. A Long Short-Term Memory (LSTM) network was employed for analysis. LSTM, a type of recurrent neural network (RNN), addresses the limitations of traditional RNNs—especially their inability to retain long-term dependencies—through the use of mechanisms such as forget gates and output gates [ 5 ]. This study aims to predict future temperature trends in selected regions using a time-series deep learning model and based on these projections, to estimate and analyze region-specific shifts in optimal rice harvest dates. 1.2 Literature Review Over the past 40 years, the average temperature in South Korea has increased by approximately 1.26°C, resulting in an extended rice cultivation period and a delay of about 3 to 5 days in transplanting schedules. These climatic shifts have directly affected rice production, and with further temperature increases and more frequent extreme weather events anticipated, the need for technological adaptation has become critical. Key strategies proposed in response include the development of heat-tolerant rice varieties and the adjustment of heading and transplanting schedules [ 1 ]. Moreover, there is growing concern over the delayed occurrence of optimal heading dates and declining grain quality, particularly due to rising temperatures during the ripening stage. Recent studies indicate that the optimal heading period has shifted later compared to the past, and the risk of grain quality deterioration from heat stress has increased. To ensure stable rice production, it is recommended that transplanting and heading schedules be systematically adjusted according to the specific characteristics of each cultivar and region [ 4 ]. As a technological response to climate change, deep learning-based temperature forecasting models have gained prominence. For instance, an LSTM (Long Short-Term Memory) model applied to 24-hour-ahead temperature prediction in the Gwangju region demonstrated superior accuracy compared to a traditional Multi-Layer Neural Network (MLNN), as measured by lower Mean Absolute Error (MAE) values. This study also emphasized the importance of ongoing model optimization and comparative evaluation with other time-series models that incorporate seasonal variability [ 5 ]. In another study, a Gated Recurrent Unit (GRU)-based model was used to predict the heading date of Ilpum rice using daily average temperature data from Jeongeup (2010–2024). By applying time-series preprocessing and incorporating GRU layers with Dropout, Batch Normalization, and Dense layers, the model effectively mitigated overfitting and improved training efficiency. The results suggested that average temperatures during the heading period from 2025 to 2034 are expected to remain within a stable range, comparable to previous years. The study underscored the potential of enhancing deep learning methodologies to further improve predictive performance [ 6 ]. As highlighted in studies [ 1 ] and [ 4 ], long-term increases in temperature and the growing frequency of extreme weather events have already begun to affect rice cultivation periods, heading dates, and transplanting schedules—posing significant risks to both yield and grain quality. In response, the application of deep learning techniques in agricultural meteorology has received growing attention in recent years. Study [ 5 ] demonstrated that Long Short-Term Memory (LSTM) models outperform traditional Multi-Layer Neural Networks (MLNNs) in temperature forecasting, while study [ 6 ] confirmed the feasibility of Gated Recurrent Unit (GRU)-based models for phenological predictions in rice. Despite these advancements, most existing studies remain geographically limited, reducing the generalizability of their findings. Even within South Korea, rice-growing environments and seasonal patterns exhibit considerable variation across different latitudes [ 7 ]. To effectively capture and predict this regional variability, it is essential to construct a comprehensive dataset that spans multiple cultivation zones. In response to this need, this study aims to develop a generalized and scalable climate dataset and to build a nationwide model capable of predicting optimal rice harvest dates across diverse rice-growing regions in South Korea. 2. Materials and Methods 2.1 Data Collection In this study, Samkwang rice—one of the most widely cultivated cultivars in South Korea—was selected as the target crop for analysis. Twelve representative rice-growing regions were chosen based on their latitudinal distribution and the prevalence of Samkwang cultivation: Haenam, Cheorwon, Yeongcheon, Chungju, Yangpyeong, Andong, Busan, Gwangju, Geochang, Seosan, Icheon, and Cheongju. Meteorological data were sourced from the Korea Meteorological Administration’s open data portal. Daily average temperature records for each region, covering the period from 1995 to 2024, were collected and used for deep learning-based analysis. To process these time-dependent data, a Long Short-Term Memory (LSTM) model was employed, given its robustness in learning from long sequential inputs—an area where conventional Recurrent Neural Networks (RNNs) often underperform. After training the model, its predictive accuracy was evaluated independently for each region. Using the trained LSTM, regional average temperatures were forecasted through 2040. These temperature projections were then used to estimate rice harvest dates, incorporating latitudinal variation to assess regional differences in harvest timing. To determine optimal harvest dates, the study referred to the official Samkwang rice cultivation manual [10], which defines the ideal harvest point as the date on which the cumulative temperature after heading reaches a specified threshold. Table 1 presents the daily average temperature data (°C) for the Cheongju region from January 1, 1995, to December 31, 2024. These data served as a primary input for both the climate change analysis and the training of the deep learning-based predictive model. 2.2 Deep Learning Architecture In this study, a deep learning model was developed to forecast future daily average temperatures using time-series meteorological data. The model architecture was based on a Long Short-Term Memory (LSTM) network, which is well-suited for capturing long-range dependencies in sequential data. To exploit temporal patterns in both forward and backward directions, a Bidirectional LSTM architecture was employed. The network consisted of two Bidirectional LSTM layers, with 64 units in the first layer and 32 units in the second. To reduce overfitting, a Dropout layer with a rate of 0.2 was inserted between the recurrent layers. Following the second LSTM layer, Batch Normalization was applied to enhance training stability and accelerate convergence. A Dense layer was used as the output layer to generate a single predicted value representing the daily average temperature. The model was trained using the Mean Absolute Error (MAE) loss function and optimized with the Adam algorithm, with a learning rate of 0.0005. To further prevent overfitting and ensure optimal performance, the EarlyStopping technique was applied, terminating training when the validation loss did not improve over a specified number of epochs. Table 2 presents the architecture of the implemented deep learning model. 2.3 Model Evaluation The deep learning model used in this study was based on a Long Short-Term Memory (LSTM) architecture. Its predictive accuracy was evaluated using the Mean Absolute Error (MAE) metric. Figure 1 through 4 compare the predicted daily average temperatures with the observed values for four representative regions: Geochang, Seosan, Icheon, and Cheongju. The results demonstrate a strong alignment between predicted and actual values over time, indicating the effectiveness of the model. Table 3 presents the MAE values for each of the 12 rice-growing regions analyzed in this study. The results confirm that the LSTM-based model predicts temperature trends with a relatively high level of accuracy. Furthermore, incorporating a MAE margin around the predicted values allows for the construction of a confidence interval for the temperature forecasts. This approach improves the interpretability and credibility of the predictions by accounting for the model’s average error range. The figures illustrate MAE performance across a subset of the selected regions, chosen to represent a broad latitudinal range. This visualization facilitates a clear comparison of prediction accuracy among regions. 3. Results and Discussion 3.1 Results In this study, the optimal harvest periods for Samkwang rice were projected through the year 2040 for 12 regions across South Korea, selected based on their latitudinal distribution. According to the Samkwang Rice Cultivation Manual published by the Rural Development Administration [10], the harvest date is defined as the day on which the accumulated temperature after heading reaches 1,150°C. Based on this criterion, harvest timing was estimated, and a range of minimum to maximum dates was proposed using a MAE interval to account for prediction uncertainty. Figure 5 illustrates the projected annual variation in Samkwang rice harvest dates from 2025 to 2040 across the 12 study regions. The analysis revealed a general trend toward progressively earlier harvest dates in all regions, attributable to the accelerated accumulation of temperature under climate warming. Additionally, higher-latitude regions exhibited relatively later harvest dates compared to those at lower latitudes. In most regions, interannual variability in harvest timing tended to decrease and converge around the year 2035, suggesting that the projected rate of temperature increase may stabilize by 2040. Notably, southern regions exhibited a relatively narrow range of harvest date variation throughout the analysis period, indicating a greater likelihood of maintaining stable harvest schedules. This implies that temperature accumulation in southern areas is less sensitive to climate change, resulting in reduced variability and potentially more stable rice cultivation conditions in the future. Figure 6 visualizes the average predicted harvest dates from 2025 to 2040 across the 12 study regions. The spatial distribution of harvest timing is clearly aligned with latitude. The trend line in the graph indicates that average harvest dates occur progressively earlier as latitude decreases, from northern to southern regions. These findings provide empirical evidence of a strong correlation between rice harvest timing, regional latitude, and average temperature. It can therefore be inferred that latitudinal temperature differences are likely to persist across South Korea through at least 2040. 3.2 Discussion This study employed a deep learning–based time-series forecasting approach to analyze long-term temperature trends and predict the harvest period of Samkwang rice across 12 major cultivation regions in South Korea. Using meteorological data from 1995 to 2024, a LSTM model was developed to forecast daily average temperatures for each region from 2025 to 2040. Based on these forecasts, the harvest date was defined as the day on which the accumulated temperature after heading reached 1,150°C. The results revealed a general trend toward earlier harvest dates across all regions, suggesting that rising temperatures are accelerating the accumulation of thermal units. Regions at higher latitudes (i.e., northern areas) exhibited relatively later harvest dates, whereas southern regions showed more stable harvest timing with smaller interannual variation. These findings highlight the critical role of latitude in determining rice growth and harvest periods and suggest that regional temperature differences driven by latitude are likely to persist in the future. The predictive performance of the LSTM model was evaluated using the MAE, which ranged from 1.4°C to 1.7°C in most regions, indicating a relatively high level of accuracy in capturing actual temperature trends. Unlike previous studies confined to specific local areas, this study incorporated a diverse set of regions spanning a broad latitudinal range, enabling a more comprehensive and generalized analysis. As a result, the predicted rice harvest periods more accurately reflect regional climatic and phenological variations. These findings are expected to provide a valuable reference for the development of future agricultural strategies in response to climate change. Declarations Funding This research received no external funding. Author Contribution Sung Kook Kim contributed to the writing of the manuscript and performed the deep learning analysis, while Jae Hyun Park prepared the figures and tables. References Seo MC, Kim JH, Choi KJ, Lee YH, Sang WG, Cho HS, Cho JI, Shin P, Baek JK (2020) Review on adaptability of rice varieties and cultivation technology according to climate change in Korea. Korean J Crop Sci 65:327–338. https://doi.org/10.7740/kjcs.2020.65.4.327 Song SG, Kim SC, Hong SG, Cho MS (2017) Understanding deep learning for weather and climate forecasting research. Proc. Korean Meteorol. Soc. Conf. 2017 Autumn: 313–314 Choi HJ, Ko SG, Lim JY (2021) Deep learning–based time–series wind speed data imputation algorithm using temperature data. Proc. Korean Inst. Inf. Sci. Eng. Conf. 2021: 1993–1995 Jo SR, Shim GM, Heo JN, Kim YS, Kang MG (2022) Changes in optimal heading date and ripening environment of rice under climate change. Proc. Korean Meteorol. Soc. Conf. 2022 Spring: 101 Yoon JW, Jeon MG (2017) Temperature forecasting model by using deep learning technology based on LSTM. Proc. Korean Inst. Elect. Eng. Conf. 2017 Autumn: 912–915 Kim SK, Kang SH (2025) Analysis of changes in rice heading dates using deep learning–based temperature prediction. Korean J Crop Sci 70:51–56. https://doi.org/10.7740/kjcs.2025.70.2.051 Lu PL, Yu Q, Wang E, Liu JD, Xu SH (2008) Effects of climatic variation and warming on rice development across South China. Clim Res 36:79–88. https://doi.org/10.3354/cr00729 Korea Meteorological Administration (KMA) (2025) KMA Data Portal. KMA, Seoul, Korea. Available at: https://data.kma.go.kr (accessed Jan. 10, 2025) Chungcheongnam–do Agricultural Research and Extension Services (2025) Samkwang rice cultivation manual. Chungcheongnam–do Agricultural Research and Extension Services, Yesan, Korea. Available at: https://cnnongup.chungnam.go.kr/view.jsp?FileDir=/B0014&SystemFileName=11_20130213092808_0_116.pdf&ftype=pdf&FileName=%EC%82%BC%EA%B4%91%EB%B2%BC%20%EC%9E%AC%EB%B0%B0%20%EB%A7%A4%EB%89%B4%EC%96%BC.pdf (accessed Jan. 10, 2025) Tables Table 1. Sample of daily average temperature data (Cheongju, 1995–2024) Date Average Temperature 1995-01-01 -5.4℃ 1995-01-02 -5.1℃ 1995-01-03 -1.7℃ 1995-01-04 -0.2℃ ... ... 2024-12-31 0℃ Table 2. LSTM deep learning architecture Layer Configuration Input Layer Time-series weather data (window × feature) 1st Recurrent Layer 64units, return_sequences=True Dropout Layer 1 Dropout (rate = 0.2) Batch Normalization 1 BatchNormalization() 2nd Recurrent Layer 32 units Dropout Layer 2 Dropout (rate = 0.2) Batch Normalization 2 BatchNormalization() Output Layer Dense(1unit–predicted mean temperature) Table 3. Evaluation of prediction accuracy using MAE for each region Region MAE(Mean Absolute Error) Chungju 1.57℃ Seosan 1.58℃ Geochang 1.47℃ Icheon 1.66℃ Haenam 1.50℃ Cheorwon 1.72℃ Yeongcheon 1.58℃ Yangpyeong 1.57℃ Andong 1.58℃ Busan 1.39℃ Gwangju 1.49℃ Cheongju 1.76℃ Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7354648","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":512786253,"identity":"63bbda6a-fb46-4f72-813a-f23b72929615","order_by":0,"name":"Sung Kook Kim¹","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/klEQVRIiWNgGAWjYNCCAwwMEsyMDQc/VAA5zMwNRGphZz54WOIMSAsjsVr42ZIP8LaBeAS0yM/IPfyC4YyNnGQzj8EByXm10fztQC0/Krbh1GJwIy/NguFGmrE0M1BL4bbjuTMOMzYw9py5jVuLRI6ZAcOHw4nzQFoktx3LbQBqYWZsw61FfgayFt45x3LnE9LCcCPH+AHDjcOJs5nZEg7wNtTkbiCkxeDMGzOGhDNpxpLNzAcOSxw7kLsRqOUgPr/It+cYf/hwzEZO4vzB5o8faupy550/fPDBjwo8DmNgYJNIQHAOg8kD+NQDAfMHJE4dAcWjYBSMglEwEgEAQttf2gr6KyYAAAAASUVORK5CYII=","orcid":"","institution":"¹My Paul Alternative School","correspondingAuthor":true,"prefix":"","firstName":"Sung","middleName":"Kook","lastName":"Kim¹","suffix":""},{"id":512786257,"identity":"f63c0c16-f343-41e5-babc-94ae7a9e7a0b","order_by":1,"name":"Jae Hyun Park¹","email":"","orcid":"","institution":"¹My Paul Alternative School","correspondingAuthor":false,"prefix":"","firstName":"Jae","middleName":"Hyun","lastName":"Park¹","suffix":""}],"badges":[],"createdAt":"2025-08-12 10:38:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7354648/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7354648/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":91115784,"identity":"1e09441b-a763-432c-9f48-c6bfc4cd0ba1","added_by":"auto","created_at":"2025-09-11 17:26:13","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":214136,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of observed and predicted monthly average temperatures in Cheongju from 1995 to 2024. The blue line indicates actual temperature data, while the red line represents LSTM-based predictions. For visualization, daily predictions were aggregated to monthly means.\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-7354648/v1/bb0f509d336951ba613fb051.png"},{"id":91114098,"identity":"febd8d79-a102-4903-9ca8-44c368e3dae5","added_by":"auto","created_at":"2025-09-11 17:02:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":139415,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of observed and predicted monthly average temperatures in Seosan from 1995 to 2024. The blue line indicates actual temperature data, while the red line represents LSTM-based predictions. For visualization, daily predictions were aggregated to monthly means.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7354648/v1/1c75ebe51740d70895a76887.png"},{"id":91114101,"identity":"6fed80e5-1fa8-485e-b69e-2738e4680a70","added_by":"auto","created_at":"2025-09-11 17:02:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":204555,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of observed and predicted monthly average temperatures in Geochang from 1995 to 2024. The blue line indicates actual temperature data, while the red line represents LSTM-based predictions. For visualization, daily predictions were aggregated to monthly means.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7354648/v1/96e697fc3f1e7550cb4d685c.png"},{"id":91114102,"identity":"fbc4ecf4-4ec4-495f-8d95-c6224b4806d5","added_by":"auto","created_at":"2025-09-11 17:02:12","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":138143,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of observed and predicted monthly average temperatures in Icheon from 1995 to 2024. The blue line indicates actual temperature data, while the red line represents LSTM-based predictions. For visualization, daily predictions were aggregated to monthly means.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7354648/v1/a5f464a32e9b82bcbf5b8b5a.png"},{"id":91114107,"identity":"034a6f68-9bcf-4818-b37e-741ed4e9b236","added_by":"auto","created_at":"2025-09-11 17:02:12","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":55082,"visible":true,"origin":"","legend":"\u003cp\u003eAnnual variation in projected rice harvest dates across 12 major cultivation regions in South Korea from 2025 to 2040. Each line represents the predicted harvest date for a specific region. A general trend of earlier harvest timing is observed in several regions due to accelerated temperature accumulation.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7354648/v1/8deeec7280a2e75ad24def97.png"},{"id":91115094,"identity":"7f67d11b-4f5c-413d-ae50-aa020cde26b7","added_by":"auto","created_at":"2025-09-11 17:18:12","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":46395,"visible":true,"origin":"","legend":"\u003cp\u003eRegional average rice harvest dates from 2025 to 2040. Each point represents the mean harvest date for a specific region during the period. A downward linear trend indicates that southern regions generally exhibit earlier harvest dates compared to northern regions.\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-7354648/v1/f52f423fc23f10bd2f84c651.png"},{"id":92124522,"identity":"8b126150-7da6-45c4-af4d-e9217939b1e1","added_by":"auto","created_at":"2025-09-25 00:46:16","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1235200,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7354648/v1/bad7d9b3-8f6b-40a2-84a5-5ffd145d3b1e.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Analysis of Changes in Rice Harvest Timing Using Deep Learning- Based Climate Prediction","fulltext":[{"header":"1. Introduction","content":"\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e\u003ch2\u003e1.1 Background\u003c/h2\u003e\u003cp\u003eRice is one of the world\u0026rsquo;s three major staple crops and serves as the primary food source for over half of the global population. In South Korea, it is the most extensively cultivated crop, covering approximately 697,714 hectares as of 2025 (KOSTAT) [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Although this represents a 1.5% decrease from the previous year, rice continues to play a critical role in national food security and the agricultural economy. However, rising temperatures have raised growing concerns regarding their effects on rice development. Recurrent abnormal weather events and the broader impacts of global warming have been reported to directly influence rice growth, emphasizing the urgent need for effective mitigation technologies [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eOver the past two decades, the average annual temperature in South Korea has increased by 2.2\u0026deg;C\u0026mdash;from 12.7\u0026deg;C in 2000 to 14.9\u0026deg;C in 2024 [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u0026mdash;highlighting the intensification of climate change. In response, artificial intelligence (AI)-based technologies have gained attention, particularly deep learning, which has demonstrated powerful performance across diverse domains, including recent advances in meteorological forecasting [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. This study applies deep learning techniques to time-series weather data to address climate-related challenges in agriculture. Daily average temperature data from 1995 to 2024 were collected from 12 representative rice-growing regions spanning a wide latitudinal range in South Korea. A Long Short-Term Memory (LSTM) network was employed for analysis. LSTM, a type of recurrent neural network (RNN), addresses the limitations of traditional RNNs\u0026mdash;especially their inability to retain long-term dependencies\u0026mdash;through the use of mechanisms such as forget gates and output gates [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThis study aims to predict future temperature trends in selected regions using a time-series deep learning model and based on these projections, to estimate and analyze region-specific shifts in optimal rice harvest dates.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e1.2 Literature Review\u003c/h2\u003e\u003cp\u003eOver the past 40 years, the average temperature in South Korea has increased by approximately 1.26\u0026deg;C, resulting in an extended rice cultivation period and a delay of about 3 to 5 days in transplanting schedules. These climatic shifts have directly affected rice production, and with further temperature increases and more frequent extreme weather events anticipated, the need for technological adaptation has become critical. Key strategies proposed in response include the development of heat-tolerant rice varieties and the adjustment of heading and transplanting schedules [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eMoreover, there is growing concern over the delayed occurrence of optimal heading dates and declining grain quality, particularly due to rising temperatures during the ripening stage. Recent studies indicate that the optimal heading period has shifted later compared to the past, and the risk of grain quality deterioration from heat stress has increased. To ensure stable rice production, it is recommended that transplanting and heading schedules be systematically adjusted according to the specific characteristics of each cultivar and region [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAs a technological response to climate change, deep learning-based temperature forecasting models have gained prominence. For instance, an LSTM (Long Short-Term Memory) model applied to 24-hour-ahead temperature prediction in the Gwangju region demonstrated superior accuracy compared to a traditional Multi-Layer Neural Network (MLNN), as measured by lower Mean Absolute Error (MAE) values. This study also emphasized the importance of ongoing model optimization and comparative evaluation with other time-series models that incorporate seasonal variability [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn another study, a Gated Recurrent Unit (GRU)-based model was used to predict the heading date of Ilpum rice using daily average temperature data from Jeongeup (2010\u0026ndash;2024). By applying time-series preprocessing and incorporating GRU layers with Dropout, Batch Normalization, and Dense layers, the model effectively mitigated overfitting and improved training efficiency. The results suggested that average temperatures during the heading period from 2025 to 2034 are expected to remain within a stable range, comparable to previous years. The study underscored the potential of enhancing deep learning methodologies to further improve predictive performance [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAs highlighted in studies [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e] and [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], long-term increases in temperature and the growing frequency of extreme weather events have already begun to affect rice cultivation periods, heading dates, and transplanting schedules\u0026mdash;posing significant risks to both yield and grain quality. In response, the application of deep learning techniques in agricultural meteorology has received growing attention in recent years. Study [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] demonstrated that Long Short-Term Memory (LSTM) models outperform traditional Multi-Layer Neural Networks (MLNNs) in temperature forecasting, while study [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] confirmed the feasibility of Gated Recurrent Unit (GRU)-based models for phenological predictions in rice.\u003c/p\u003e\u003cp\u003eDespite these advancements, most existing studies remain geographically limited, reducing the generalizability of their findings. Even within South Korea, rice-growing environments and seasonal patterns exhibit considerable variation across different latitudes [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. To effectively capture and predict this regional variability, it is essential to construct a comprehensive dataset that spans multiple cultivation zones.\u003c/p\u003e\u003cp\u003eIn response to this need, this study aims to develop a generalized and scalable climate dataset and to build a nationwide model capable of predicting optimal rice harvest dates across diverse rice-growing regions in South Korea.\u003c/p\u003e\u003c/div\u003e"},{"header":"2. Materials and Methods","content":"\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Data Collection\u003c/h2\u003e\u003cp\u003eIn this study, Samkwang rice\u0026mdash;one of the most widely cultivated cultivars in South Korea\u0026mdash;was selected as the target crop for analysis. Twelve representative rice-growing regions were chosen based on their latitudinal distribution and the prevalence of Samkwang cultivation: Haenam, Cheorwon, Yeongcheon, Chungju, Yangpyeong, Andong, Busan, Gwangju, Geochang, Seosan, Icheon, and Cheongju.\u003c/p\u003e\u003cp\u003eMeteorological data were sourced from the Korea Meteorological Administration\u0026rsquo;s open data portal. Daily average temperature records for each region, covering the period from 1995 to 2024, were collected and used for deep learning-based analysis. To process these time-dependent data, a Long Short-Term Memory (LSTM) model was employed, given its robustness in learning from long sequential inputs\u0026mdash;an area where conventional Recurrent Neural Networks (RNNs) often underperform.\u003c/p\u003e\u003cp\u003eAfter training the model, its predictive accuracy was evaluated independently for each region. Using the trained LSTM, regional average temperatures were forecasted through 2040. These temperature projections were then used to estimate rice harvest dates, incorporating latitudinal variation to assess regional differences in harvest timing.\u003c/p\u003e\u003cp\u003eTo determine optimal harvest dates, the study referred to the official Samkwang rice cultivation manual [10], which defines the ideal harvest point as the date on which the cumulative temperature after heading reaches a specified threshold. Table\u0026nbsp;1 presents the daily average temperature data (\u0026deg;C) for the Cheongju region from January 1, 1995, to December 31, 2024. These data served as a primary input for both the climate change analysis and the training of the deep learning-based predictive model.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Deep Learning Architecture\u003c/h2\u003e\u003cp\u003eIn this study, a deep learning model was developed to forecast future daily average temperatures using time-series meteorological data. The model architecture was based on a Long Short-Term Memory (LSTM) network, which is well-suited for capturing long-range dependencies in sequential data. To exploit temporal patterns in both forward and backward directions, a Bidirectional LSTM architecture was employed.\u003c/p\u003e\u003cp\u003eThe network consisted of two Bidirectional LSTM layers, with 64 units in the first layer and 32 units in the second. To reduce overfitting, a Dropout layer with a rate of 0.2 was inserted between the recurrent layers. Following the second LSTM layer, Batch Normalization was applied to enhance training stability and accelerate convergence. A Dense layer was used as the output layer to generate a single predicted value representing the daily average temperature.\u003c/p\u003e\u003cp\u003eThe model was trained using the Mean Absolute Error (MAE) loss function and optimized with the Adam algorithm, with a learning rate of 0.0005. To further prevent overfitting and ensure optimal performance, the EarlyStopping technique was applied, terminating training when the validation loss did not improve over a specified number of epochs. Table\u0026nbsp;2 presents the architecture of the implemented deep learning model.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Model Evaluation\u003c/h2\u003e\u003cp\u003eThe deep learning model used in this study was based on a Long Short-Term Memory (LSTM) architecture. Its predictive accuracy was evaluated using the Mean Absolute Error (MAE) metric. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e through \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e compare the predicted daily average temperatures with the observed values for four representative regions: Geochang, Seosan, Icheon, and Cheongju. The results demonstrate a strong alignment between predicted and actual values over time, indicating the effectiveness of the model.\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;3 presents the MAE values for each of the 12 rice-growing regions analyzed in this study. The results confirm that the LSTM-based model predicts temperature trends with a relatively high level of accuracy.\u003c/p\u003e\u003cp\u003eFurthermore, incorporating a MAE margin around the predicted values allows for the construction of a confidence interval for the temperature forecasts. This approach improves the interpretability and credibility of the predictions by accounting for the model\u0026rsquo;s average error range. The figures illustrate MAE performance across a subset of the selected regions, chosen to represent a broad latitudinal range. This visualization facilitates a clear comparison of prediction accuracy among regions.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results and Discussion","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Results\u003c/h2\u003e\u003cp\u003eIn this study, the optimal harvest periods for Samkwang rice were projected through the year 2040 for 12 regions across South Korea, selected based on their latitudinal distribution. According to the Samkwang Rice Cultivation Manual published by the Rural Development Administration [10], the harvest date is defined as the day on which the accumulated temperature after heading reaches 1,150\u0026deg;C. Based on this criterion, harvest timing was estimated, and a range of minimum to maximum dates was proposed using a MAE interval to account for prediction uncertainty. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e illustrates the projected annual variation in Samkwang rice harvest dates from 2025 to 2040 across the 12 study regions. The analysis revealed a general trend toward progressively earlier harvest dates in all regions, attributable to the accelerated accumulation of temperature under climate warming. Additionally, higher-latitude regions exhibited relatively later harvest dates compared to those at lower latitudes. In most regions, interannual variability in harvest timing tended to decrease and converge around the year 2035, suggesting that the projected rate of temperature increase may stabilize by 2040. Notably, southern regions exhibited a relatively narrow range of harvest date variation throughout the analysis period, indicating a greater likelihood of maintaining stable harvest schedules. This implies that temperature accumulation in southern areas is less sensitive to climate change, resulting in reduced variability and potentially more stable rice cultivation conditions in the future.\u003c/p\u003e\u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e visualizes the average predicted harvest dates from 2025 to 2040 across the 12 study regions. The spatial distribution of harvest timing is clearly aligned with latitude. The trend line in the graph indicates that average harvest dates occur progressively earlier as latitude decreases, from northern to southern regions. These findings provide empirical evidence of a strong correlation between rice harvest timing, regional latitude, and average temperature. It can therefore be inferred that latitudinal temperature differences are likely to persist across South Korea through at least 2040.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Discussion\u003c/h2\u003e\u003cp\u003eThis study employed a deep learning\u0026ndash;based time-series forecasting approach to analyze long-term temperature trends and predict the harvest period of Samkwang rice across 12 major cultivation regions in South Korea. Using meteorological data from 1995 to 2024, a LSTM model was developed to forecast daily average temperatures for each region from 2025 to 2040. Based on these forecasts, the harvest date was defined as the day on which the accumulated temperature after heading reached 1,150\u0026deg;C. The results revealed a general trend toward earlier harvest dates across all regions, suggesting that rising temperatures are accelerating the accumulation of thermal units. Regions at higher latitudes (i.e., northern areas) exhibited relatively later harvest dates, whereas southern regions showed more stable harvest timing with smaller interannual variation. These findings highlight the critical role of latitude in determining rice growth and harvest periods and suggest that regional temperature differences driven by latitude are likely to persist in the future. The predictive performance of the LSTM model was evaluated using the MAE, which ranged from 1.4\u0026deg;C to 1.7\u0026deg;C in most regions, indicating a relatively high level of accuracy in capturing actual temperature trends. Unlike previous studies confined to specific local areas, this study incorporated a diverse set of regions spanning a broad latitudinal range, enabling a more comprehensive and generalized analysis. As a result, the predicted rice harvest periods more accurately reflect regional climatic and phenological variations. These findings are expected to provide a valuable reference for the development of future agricultural strategies in response to climate change.\u003c/p\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThis research received no external funding.\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eSung Kook Kim contributed to the writing of the manuscript and performed the deep learning analysis, while Jae Hyun Park prepared the figures and tables.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSeo MC, Kim JH, Choi KJ, Lee YH, Sang WG, Cho HS, Cho JI, Shin P, Baek JK (2020) Review on adaptability of rice varieties and cultivation technology according to climate change in Korea. Korean J Crop Sci 65:327\u0026ndash;338. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7740/kjcs.2020.65.4.327\u003c/span\u003e\u003cspan address=\"10.7740/kjcs.2020.65.4.327\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSong SG, Kim SC, Hong SG, Cho MS (2017) Understanding deep learning for weather and climate forecasting research. Proc. Korean Meteorol. Soc. Conf. 2017 Autumn: 313\u0026ndash;314\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChoi HJ, Ko SG, Lim JY (2021) Deep learning\u0026ndash;based time\u0026ndash;series wind speed data imputation algorithm using temperature data. Proc. Korean Inst. Inf. Sci. Eng. Conf. 2021: 1993\u0026ndash;1995\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJo SR, Shim GM, Heo JN, Kim YS, Kang MG (2022) Changes in optimal heading date and ripening environment of rice under climate change. Proc. Korean Meteorol. Soc. Conf. 2022 Spring: 101\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYoon JW, Jeon MG (2017) Temperature forecasting model by using deep learning technology based on LSTM. Proc. Korean Inst. Elect. Eng. Conf. 2017 Autumn: 912\u0026ndash;915\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim SK, Kang SH (2025) Analysis of changes in rice heading dates using deep learning\u0026ndash;based temperature prediction. Korean J Crop Sci 70:51\u0026ndash;56. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7740/kjcs.2025.70.2.051\u003c/span\u003e\u003cspan address=\"10.7740/kjcs.2025.70.2.051\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLu PL, Yu Q, Wang E, Liu JD, Xu SH (2008) Effects of climatic variation and warming on rice development across South China. Clim Res 36:79\u0026ndash;88. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3354/cr00729\u003c/span\u003e\u003cspan address=\"10.3354/cr00729\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKorea Meteorological Administration (KMA) (2025) KMA Data Portal. KMA, Seoul, Korea. Available at: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://data.kma.go.kr\u003c/span\u003e\u003cspan address=\"https://data.kma.go.kr\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (accessed Jan. 10, 2025)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChungcheongnam\u0026ndash;do Agricultural Research and Extension Services (2025) Samkwang rice cultivation manual. Chungcheongnam\u0026ndash;do Agricultural Research and Extension Services, Yesan, Korea. Available at: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://cnnongup.chungnam.go.kr/view.jsp?FileDir=/B0014\u0026amp;SystemFileName=11_20130213092808_0_116.pdf\u0026amp;ftype=pdf\u0026amp;FileName=%EC%82%BC%EA%B4%91%EB%B2%BC%20%EC%9E%AC%EB%B0%B0%20%EB%A7%A4%EB%89%B4%EC%96%BC.pdf\u003c/span\u003e\u003cspan address=\"https://cnnongup.chungnam.go.kr/view.jsp?FileDir=/B0014\u0026amp;SystemFileName=11_20130213092808_0_116.pdf\u0026amp;ftype=pdf\u0026amp;FileName=%EC%82%BC%EA%B4%91%EB%B2%BC%20%EC%9E%AC%EB%B0%B0%20%EB%A7%A4%EB%89%B4%EC%96%BC.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (accessed Jan. 10, 2025)\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1.\u0026nbsp;\u003c/strong\u003eSample of daily average temperature data (Cheongju, 1995\u0026ndash;2024)\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 40.4255%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDate\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59.5745%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAverage Temperature\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 40.4255%;\"\u003e\n \u003cp\u003e1995-01-01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59.5745%;\"\u003e\n \u003cp\u003e-5.4℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 40.4255%;\"\u003e\n \u003cp\u003e1995-01-02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59.5745%;\"\u003e\n \u003cp\u003e-5.1℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 40.4255%;\"\u003e\n \u003cp\u003e1995-01-03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59.5745%;\"\u003e\n \u003cp\u003e-1.7℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 40.4255%;\"\u003e\n \u003cp\u003e1995-01-04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59.5745%;\"\u003e\n \u003cp\u003e-0.2℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 40.4255%;\"\u003e\n \u003cp\u003e...\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59.5745%;\"\u003e\n \u003cp\u003e...\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 40.4255%;\"\u003e\n \u003cp\u003e2024-12-31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59.5745%;\"\u003e\n \u003cp\u003e0℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2. \u0026nbsp;\u003c/strong\u003eLSTM deep learning architecture\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"462\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35.7143%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLayer\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64.2857%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eConfiguration\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35.7143%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInput Layer\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64.2857%;\"\u003e\n \u003cp\u003eTime-series weather data (window \u0026times; feature)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35.7143%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1st Recurrent Layer\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64.2857%;\"\u003e\n \u003cp\u003e64units, return_sequences=True\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35.7143%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDropout Layer 1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64.2857%;\"\u003e\n \u003cp\u003eDropout (rate = 0.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35.7143%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBatch Normalization 1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64.2857%;\"\u003e\n \u003cp\u003eBatchNormalization()\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35.7143%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e2nd Recurrent Layer\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64.2857%;\"\u003e\n \u003cp\u003e32 units\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35.7143%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDropout Layer 2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64.2857%;\"\u003e\n \u003cp\u003eDropout (rate = 0.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35.7143%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBatch Normalization 2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64.2857%;\"\u003e\n \u003cp\u003eBatchNormalization()\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35.7143%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOutput Layer\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64.2857%;\"\u003e\n \u003cp\u003eDense(1unit\u0026ndash;predicted mean temperature)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3.\u0026nbsp;\u003c/strong\u003eEvaluation of prediction accuracy using MAE for each region\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRegion\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMAE(Mean Absolute Error)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eChungju\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.57℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSeosan\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.58℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGeochang\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.47℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eIcheon\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.66℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHaenam\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.50℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCheorwon\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.72℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYeongcheon\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.58℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYangpyeong\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.57℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAndong\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.58℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBusan\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.39℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGwangju\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.49℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 41.3105%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCheongju\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58.6895%;\"\u003e\n \u003cp\u003e1.76℃\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Climate change, Deep learning, Harvest timing, LSTM, Rice cultivation, Temperature prediction","lastPublishedDoi":"10.21203/rs.3.rs-7354648/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7354648/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study predicts future harvest dates for Samkwang rice across 12 major cultivation regions in South Korea using a deep learning-based time-series approach. A Long Short-Term Memory (LSTM) model was trained on daily average temperature data spanning 1995 to 2024 and subsequently used to forecast regional temperatures for the period 2025\u0026ndash;2040. Harvest dates were estimated based on the day cumulative temperature after heading reached 1,150\u0026deg;C, in accordance with official agronomic guidelines. Prediction accuracy was evaluated using the Mean Absolute Error (MAE) for each region. The results indicate a general advancement in harvest dates, attributed to accelerated heat accumulation under ongoing climate warming. However, regional variations were observed: northern regions exhibited more delayed and variable harvest patterns, whereas southern regions demonstrated greater temporal stability. Unlike previous studies confined to specific areas, this research incorporates a broad latitudinal dataset, offering a generalized predictive model to support adaptive agricultural strategies in the face of climate change.\u003c/p\u003e","manuscriptTitle":"Analysis of Changes in Rice Harvest Timing Using Deep Learning- Based Climate Prediction","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-11 17:02:08","doi":"10.21203/rs.3.rs-7354648/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bb5a76bb-713d-4365-966a-a3365b87d0ce","owner":[],"postedDate":"September 11th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-09-25T00:38:09+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-11 17:02:08","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7354648","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7354648","identity":"rs-7354648","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}