Spatiotemporal Evolution of Ecosystem Service Value in Caohai National Nature Reserve

Wetland ecosystems are important regulators of global climate change.Studying the spatiotemporal changes and driving mechanisms of their ecosystem service values(ESV)is beneficial for the sustainable development of wetlands.This paper uses the equivalent factor method,based on land use changes,to reveal the spatiotemporal evolution of the ecosystem service value in the Caohai National Nature Reserve(CNNR).The results show the following:①from 2000 to 2020,there was a significant decrease in the core zone s arable land area,with an increase in forest and water areas.Construction land mainly increased in the experimental area,and the grassland area showed a fluctuating change of first increasing and then decreasing;②in 2000,2010,and 2020,the ecosystem service value of the study area was 302 million,296 million,and 325 million yuan,respectively,showing a trend of fluctuating growth,with the value of wetland ecosystems playing a dominant role;③regulatory services are the main contributors to the ecosystem service value in the study area,with a contribution rate of 60%.Hydrological regulation is the ecosystem function with the highest value in wetland ecosystem services,contributing more than 35%to the ESV in all three periods;④in terms of spatial distribution,the core zone s ecosystem service value is dominant.Looking at the total ecosystem service value of the region,the core zone>the experimental area>the buffer zone.In terms of ESV per unit area,the core zone(89000 yuan/hm^(2))is significantly higher than the buffer zone(39100 yuan/hm^(2))and the experimental area(15800 yuan/hm^(2)).The study can provide a basis for research and spatial management of ecosystem services in wetland ecosystems and nature reserves.

卷积神经网络和近红外光谱的土壤pH值预测

土壤pH值是影响土壤养分转化和土壤肥力的关键因素,使用近红外光谱技术对土壤pH值进行检测可为土壤资源的开发利用提供重要依据。卷积神经网络作为深度学习在人工智能方面的典型算法,由于其结构具备“局部感知,权值共享”的能力,因此不仅能够对复杂的光谱数据进行特征抽取,还能够减少网络的训练参数,提高网络的运算效率。将卷积神经网络用于近红外光谱的建模分析,并提出一种基于一维卷积的卷积神经网络和近红外光谱的土壤pH值预测方法。网络由Python语言调用Tensorflow工具包搭建而成,其结构由输入层、卷积层、池化层以及全连接层四部分组成。以欧洲统计局在2008年—2012年开展的土地利用及覆盖面积统计调查所收集的矿物质土壤光谱样本数据集为研究对象,为消除光谱中存在的基线漂移,提高信噪比,对原始可见光近红外光谱(400~2500 nm)进行一阶导数和Savitzky-Golay平滑处理。在模型训练过程中,随机选取15000个样本作为训练集,剩余的2272个样本作为测试集,探讨不同的卷积层个数及训练迭代次数对模型性能的影响,并采用ReLU激活函数及Adam优化器防止模型出现梯度消失现象,提高模型的稳定性,之后通过分析模型的拟合优度和运算成本确定模型的最佳性能,最后将网络模型与传统的BP和PLSR模型进行对比。结果显示,当模型迭代次数为2500次,卷积层个数为4层时,模型达到最佳状态,模型对训练集的均方误差从1.898降到了0.097;模型对测试集的拟合优度为0.909,分别比BP和PLSR模型高0.117和0.218。使用卷积神经网络可以对土壤近红外光谱的内部特征信息进行抽取,从而实现对大面积土壤pH值的高效准确预测。CNNR模型可对农作物的合理栽种及精准施肥提供指导,从而达到土壤结构稳定和可持续发展的目的。基于卷积神经网络的近红外光谱回归方法也可以推广到其他土壤信息研究。