可解释的深度学习识别了淡水有害藻类繁殖的模式和驱动因素
Shengyue Chen1,2, Jinliang Huang1, Jiacong Huang3
1Fujian Key Laboratory of Coastal Pollution Prevention and Control, Xiamen University, Xiamen, 361102, China.
Environmental science and ecotechnology
|February 3, 2025
概括
使用长短期记忆 (LSTM) 的新型深度学习模型有效预测淡水生态系统中的有害藻类繁殖 (HAB). 水温是关键的驱动因素,该模型显示出强大的区域可转移性,以改善HAB管理.
科学领域:
- 环境科学 环境科学
- 生态生态学 生态生态学
- 数据科学数据科学数据科学
背景情况:
- 在全球范围内,有害藻类繁殖 (HAB) 正在增加,威胁到淡水生态系统.
- 理解HAB机制受到区域特异性和有限数据的阻碍.
- 传统模型很难将HAB动态概括和预测事件.
研究的目的:
- 为 HAB 建模开发一种可解释的深度学习方法.
- 确定HAB的关键驱动因素及其区域差异.
- 提高HAB预测准确度,特别是在数据稀缺的地区.
主要方法:
- 利用长期短期记忆 (LSTM) 模型与解释技术.
- 在中国的102个地点在三年内应用了这种方法来模拟藻类密度.
- 员工转移学习,以改善预测在不足测量的地区.
主要成果:
- 在捕捉每日藻类动态 (NSE 0.48-0.95) 中,LSTM 实现了高精度.
- 水温被确定为主要的HAB驱动因素,敏感度有区域差异.
- 转移学习在超过75%的测量不良站点中提高了预测准确度.
结论:
- 可解释的深度学习 (LSTM) 有效地解决了 HAB 建模中的区域特异性和数据限制.
- 准确的预测和驾驶员识别为缓解HAB提供了可操作的见解.
- 该方法支持在全国和区域范围内有效管理淡水生态系统.
相关概念视频
00:54What are Biogeochemical Cycles?
31.0K
The most common elements in organic molecules, carbon, hydrogen, oxygen, nitrogen, sulfur, and phosphorus, are only available in the ecosystem in limited amounts. Therefore, these nutrients must be recycled through both biotic and abiotic components of the ecosystem, in processes generally called biogeochemical cycles.
31.0K
02:41Responses to Drought and Flooding
10.6K
Water plays a significant role in the life cycle of plants. However, insufficient or excess of water can be detrimental and pose a serious threat to plants.
10.6K
01:06Primary Production
23.5K
The total amount of energy acquired by primary producers in an ecosystem is called gross primary production (GPP). However, of this energy, producers use some for metabolic processes, and some is lost as heat, decreasing the amount of energy available to the next trophic level. The remaining usable amount of energy is called the net primary productivity (NPP). In terrestrial ecosystems, NPP is driven by climate, while light penetration and nutrient availability drive NPP in aquatic ecosystems.
23.5K