Enhancing cotton irrigation with distributional actor–critic reinforcement learning
Accurate predictions of irrigation’s impact on crop yield are crucial for effective decision-making. However, current research predominantly focuses on the relationship between irrigation events and soil moisture, often neglecting the physiological state of the crops themselves. This study introduce...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-02-01
|
| Series: | Agricultural Water Management |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S0378377424005304 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1841557018411794432 |
|---|---|
| author | Yi Chen Meiwei Lin Zhuo Yu Weihong Sun Weiguo Fu Liang He |
| author_facet | Yi Chen Meiwei Lin Zhuo Yu Weihong Sun Weiguo Fu Liang He |
| author_sort | Yi Chen |
| collection | DOAJ |
| description | Accurate predictions of irrigation’s impact on crop yield are crucial for effective decision-making. However, current research predominantly focuses on the relationship between irrigation events and soil moisture, often neglecting the physiological state of the crops themselves. This study introduces a novel intelligent irrigation approach based on distributional reinforcement learning, ensuring that the algorithm simultaneously considers weather, soil, and crop conditions to make optimal irrigation decisions for long-term benefits. To achieve this, we collected climate data from 1980 to 2024 and conducted a two-year cotton planting experiment in 2023 and 2024. We used soil and plant state indicators from 5 experimental groups with varying irrigation treatments to calibrate and validate the DSSAT model. Subsequently, we innovatively integrated a distributional reinforcement learning method—an effective machine learning technique for continuous control problems. Our algorithm focuses on 17 indicators, including crop leaf area, stem leaf count, and soil evapotranspiration, among others. Through a well-designed network structure and cumulative rewards, our approach effectively captures the relationships between irrigation events and these states. Additionally, we validated the robustness and generalizability of the model using three years of extreme weather data and two consecutive years of cross-site observations. This method surpasses previous irrigation strategies managed by standard reinforcement learning techniques (e.g., DQN). Empirical results indicate that our approach significantly outperforms traditional agronomic decision-making, enhancing cotton yield by 13.6% and improving water use efficiency per kilogram of crop by 6.7%. In 2024, our method was validated in actual field experiments, achieving the highest yield among all approaches, with a 12.9% increase compared to traditional practices. Our research provides a robust framework for intelligent cotton irrigation in the region and offers promising new directions for implementing smart agricultural decision systems across diverse areas. |
| format | Article |
| id | doaj-art-65ebddfc0b0043269922c0610a417a08 |
| institution | Kabale University |
| issn | 1873-2283 |
| language | English |
| publishDate | 2025-02-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Agricultural Water Management |
| spelling | doaj-art-65ebddfc0b0043269922c0610a417a082025-01-07T04:16:41ZengElsevierAgricultural Water Management1873-22832025-02-01307109194Enhancing cotton irrigation with distributional actor–critic reinforcement learningYi Chen0Meiwei Lin1Zhuo Yu2Weihong Sun3Weiguo Fu4Liang He5School of Computer Science and Technology, Xinjiang University, Urumqi, 830017, ChinaSchool of Agricultural Engineering, Jiangsu University, Zhenjiang, 212013, ChinaSchool of Agricultural Engineering, Jiangsu University, Zhenjiang, 212013, ChinaSchool of Agricultural Engineering, Jiangsu University, Zhenjiang, 212013, ChinaSchool of Agricultural Engineering, Jiangsu University, Zhenjiang, 212013, ChinaSchool of Computer Science and Technology, Xinjiang University, Urumqi, 830017, China; Department of Electronic Engineering, and Beijing National Research Center for Information Science and Technology, Tsinghua University, Beijing, 100084, China; Corresponding author at: Department of Electronic Engineering, and Beijing National Research Center for Information Science and Technology, Tsinghua University, Beijing, 100084, China.Accurate predictions of irrigation’s impact on crop yield are crucial for effective decision-making. However, current research predominantly focuses on the relationship between irrigation events and soil moisture, often neglecting the physiological state of the crops themselves. This study introduces a novel intelligent irrigation approach based on distributional reinforcement learning, ensuring that the algorithm simultaneously considers weather, soil, and crop conditions to make optimal irrigation decisions for long-term benefits. To achieve this, we collected climate data from 1980 to 2024 and conducted a two-year cotton planting experiment in 2023 and 2024. We used soil and plant state indicators from 5 experimental groups with varying irrigation treatments to calibrate and validate the DSSAT model. Subsequently, we innovatively integrated a distributional reinforcement learning method—an effective machine learning technique for continuous control problems. Our algorithm focuses on 17 indicators, including crop leaf area, stem leaf count, and soil evapotranspiration, among others. Through a well-designed network structure and cumulative rewards, our approach effectively captures the relationships between irrigation events and these states. Additionally, we validated the robustness and generalizability of the model using three years of extreme weather data and two consecutive years of cross-site observations. This method surpasses previous irrigation strategies managed by standard reinforcement learning techniques (e.g., DQN). Empirical results indicate that our approach significantly outperforms traditional agronomic decision-making, enhancing cotton yield by 13.6% and improving water use efficiency per kilogram of crop by 6.7%. In 2024, our method was validated in actual field experiments, achieving the highest yield among all approaches, with a 12.9% increase compared to traditional practices. Our research provides a robust framework for intelligent cotton irrigation in the region and offers promising new directions for implementing smart agricultural decision systems across diverse areas.http://www.sciencedirect.com/science/article/pii/S0378377424005304Distributional reinforcement learningIrrigation decisionDSSAT modelAgricultural managementCotton irrigation |
| spellingShingle | Yi Chen Meiwei Lin Zhuo Yu Weihong Sun Weiguo Fu Liang He Enhancing cotton irrigation with distributional actor–critic reinforcement learning Agricultural Water Management Distributional reinforcement learning Irrigation decision DSSAT model Agricultural management Cotton irrigation |
| title | Enhancing cotton irrigation with distributional actor–critic reinforcement learning |
| title_full | Enhancing cotton irrigation with distributional actor–critic reinforcement learning |
| title_fullStr | Enhancing cotton irrigation with distributional actor–critic reinforcement learning |
| title_full_unstemmed | Enhancing cotton irrigation with distributional actor–critic reinforcement learning |
| title_short | Enhancing cotton irrigation with distributional actor–critic reinforcement learning |
| title_sort | enhancing cotton irrigation with distributional actor critic reinforcement learning |
| topic | Distributional reinforcement learning Irrigation decision DSSAT model Agricultural management Cotton irrigation |
| url | http://www.sciencedirect.com/science/article/pii/S0378377424005304 |
| work_keys_str_mv | AT yichen enhancingcottonirrigationwithdistributionalactorcriticreinforcementlearning AT meiweilin enhancingcottonirrigationwithdistributionalactorcriticreinforcementlearning AT zhuoyu enhancingcottonirrigationwithdistributionalactorcriticreinforcementlearning AT weihongsun enhancingcottonirrigationwithdistributionalactorcriticreinforcementlearning AT weiguofu enhancingcottonirrigationwithdistributionalactorcriticreinforcementlearning AT lianghe enhancingcottonirrigationwithdistributionalactorcriticreinforcementlearning |