基于LSTM模型的草原土壤状态预测

Grassland Soil Condition Prediction Based on LSTM Model

摘要: 中国是一个资源大国，拥有丰富的草地。这些草原生态系统是保持我国生态平衡的关键防线，同时为经济发展提供稳固的基础。然而，随着畜牧业的迅速发展，草地退化问题日益突出，甚至在一些地区呈现沙漠化趋势。面对这一情况，迫切需要提供科学合理的草地管理方式。因此，对土壤状态的准确预测对于草原的可持续保护和合理开发至关重要。本文通过对历年统计数据的分析，首先通过Softmax逻辑回归模型，结合处理后的数据，得到六种土壤状态与不同放牧策略的四分类模型，建立了不同放牧策略对草原土壤状态影响的数学模型。然后使用LSTM模型并基于多年份同放牧强度和放牧小区的土壤不同状态下的数据来进行训练，预测得到2023年的同条件下的土壤在不同状态下的数据。再利用沙漠化程度指数预测模型和数据来确定不同放牧强度下监测点的沙漠化程度数值。最后使用有机量、含水量、叶面积指数等指标代替用于衡量土壤状态的指标，有土壤肥力变化、土壤湿度、植被覆盖等，综合2014~2022年的土壤数据，使用LSTM模型预测得到2024年同月的土壤状态数据。

Abstract: China is a resource-rich country with abundant grassland areas. These grassland ecosystems serve as crucial barriers to maintaining the ecological balance in the country while providing a stable foundation for economic development. However, with the rapid expansion of livestock farming, the issue of grassland degradation has become increasingly prominent, even manifesting desertifica-tion trends in certain regions. Faced with this situation, there is an urgent need to establish scien-tifically sound grassland management approaches. Therefore, accurate prediction of soil conditions is crucial for the sustainable protection and rational development of grasslands. This paper analyz-es historical statistical data and employs a Softmax logistic regression model to establish a four-class model for the impact of different grazing strategies on six soil conditions. This model is based on processed data, providing insights into the influence of various grazing strategies on grassland soil conditions. Subsequently, an LSTM model is utilized and trained on data from multi-ple years with consistent grazing intensities and different soil conditions in grazing zones, to predict soil conditions in 2023 under the same conditions. Utilize a Desertification Severity Index Predic-tion Model and data to determine the degree of desertification at monitoring points under different grazing intensities. Finally, employing indicators such as organic matter, moisture content, and leaf area index as substitutes for metrics measuring soil conditions like soil fertility changes, soil mois-ture, and vegetation cover, the LSTM model, incorporating comprehensive soil data from 2014 to 2022, predicts soil condition data for the same months in 2024.