Hyperspectral inversion of the RF model for soil salinity in oasis tillage layer based on optimal mathematics and wavelet transform
-
摘要:
土壤盐渍化会导致土壤结构恶化、作物减产,影响农业生态系统的可持续性,而其监测和管理已成为盐渍化地区亟待解决的问题。为精准识别盐渍化区域以优化农田管理,以新疆渭干河-库车河绿洲耕层土壤为研究对象,测定其光谱反射率并进行数学变换、连续小波变换(continuous wavelet transformation,CWT)、离散小波变换(discrete wavelet transformation,DWT),以及数学变换结合CWT和DWT处理,并基于Pearson相关分析提取与土壤盐分含量(soil salt content,SSC)密切相关的特征波段,从而构建随机森林(random forest,RF)模型定量估算SSC。结果表明:1)在350~
2450 nm范围内,不同盐渍化等级的土壤光谱反射率随SSC增加而升高,但达到极重度盐渍化时反射率降低。2)一阶微分变换、CWT、DWT以及一阶微分结合CWT和DWT处理均能显著提升光谱数据与SSC的相关性,其中DWT效果最优,相关系数最高达−0.621(P<0.01)。3)一阶微分结合CWT处理显著提升了模型的估测能力,以(1/R)′_CWT_28的全波段构建的估测模型最优,其验证集的决定系数为0.715,均方根误差为17.467 g/kg,相对分析误差为1.56。研究表明数学变换结合小波变换能提升模型的估测效果,可为SSC定量监测提供参考。Abstract:Soil salinization can often result in the continuous accumulation of soluble salt in the surface layer of soil, due mainly to natural and human activities. Saline soil has then posed a serious threat to the soil structure and crop yield in sustainable agriculture. It is very urgent to effectively monitor and manage the soil salinization in saline-alkali areas. Fortunately, hyperspectral imaging can be expected to estimate the soil salinity in recent years, due to its rich spectral information and high resolution. Among them, soil salt content (SSC) is one of the most important indicators to evaluate the degree of soil salinization. This study aims to estimate the soil salinity in the tillage layer of the oasis during hyperspectral imaging. The salinized areas were also precisely identified to optimize farmland management. A total of 193 soil samples were collected from the field in the Weigan-Kuqa River Oasis in Xinjiang, China. The salt content was then measured in the laboratory. Additionally, the spectrum of each soil sample was measured ten times and then averaged to obtain the spectral reflectance. The noisy spectral bands were also removed at the ends of the curves and two water absorption bands, followed by smoothing preprocessing using the Savitzky-Golay method. The spectral data was obtained to further transform using mathematical transformations, continuous wavelet transformation (CWT), and discrete wavelet transformation (DWT). According to Pearson correlation analysis, the spectral bands passing the significance test (P<0.01) were extracted as the SSC characteristic bands. A random forest model was then constructed to quantitatively estimate the SSC. The results indicate that: 1) There was a general consistency in the spectral curve shapes of soils with different salinization levels. There was a great variation in the patterns of reflectance with the wavelength and SSC. As the wavelength increased from the short (350 nm) to the long (2 450 nm), the reflectance increased in the range of 350~800 nm and then remained relatively constant in the range of 800~2 130 nm. While an absorption peak appeared around 2 130 nm corresponding to Na2SO4. After that, the reflectance started to decrease. Furthermore, the reflectance increased with the SSC in the entire range of wavelengths. There was also a positive correlation between them. But there was a decreasing trend at extreme levels of high salinization. 2) First-order differential transformation, CWT, DWT, as well as first-order differential combined with CWT and first-order differential combined with DWT, all significantly enhanced the sensitivity of the spectra to the SSC. Specifically, the DWT was performed the best among them. The highest correlation coefficient between the processed spectra and SSC reached −0.621 (P<0.01), indicating a negative correlation. The performance of the rest was ranked in the descending order of the first-order differential combined with CWT, first-order differential combined with DWT, CWT, and first-order differential transformation. 3) The first-order differential transformation, CWT, DWT, first-order differential combined with CWT, and first-order differential combined with DWT all improved the accuracy and stability of the model. The combination of first-order differential with CWT significantly enhanced the accuracy of SSC estimation. The optimal model used the full-band spectral of (1/R)'_CWT_28, with the coefficients of determination (R2) of 0.821 and 0.715 for the training and validation sets, respectively, the root mean square errors (RMSE) of 16.049 and 17.467 g/kg, respectively, and the relative predictive determinant (RPD) values of 1.50 and 1.70, respectively. As such, the effective SSC estimation was achieved in the study area. Therefore, the mathematical transformations with the wavelet transformation were better utilized to process the smoothed original spectral data. The sensitivity of spectra to SSC was significantly enhanced after processing. Compared with the mathematical or wavelet transformations alone, the combined approach was more effective in quantitatively monitoring the SSC. The finding can also provide a valuable reference to alleviate the soil structure deterioration for agricultural production and environmental protection
-
Keywords:
- soils /
- salinity /
- random forest /
- mathematical transformations /
- wavelet transformation
-
-
表 1 盐渍化各等级样点数量及盐分分布特征
Table 1 The number of samples and the characteristics of salt distribution for soils at each level
盐渍化等级
Salinization grade样点数量占比
Sample size proportion/%土壤盐分含量Soil salt content (SSC)/(g·kg−1) 变异系数
Coefficient of variation/%最小值
Minimum最大值
Maximum中位值
Medium平均值
Mean标准差
Standard deviation非盐渍化Non salinization(NS) 54.92 1.084 7.929 3.476 4.133 1.794 43.4 轻度盐渍化Light salinization(LS) 7.25 8.120 9.461 8.794 8.827 0.471 5.3 中度盐渍化Medium salinization(MS) 15.03 10.150 14.727 10.963 11.621 1.263 10.9 重度盐渍化Heavy salinization(HS) 2.59 16.051 19.607 18.089 17.986 1.395 7.8 极重度盐渍化Very heavy salinization(VHS) 20.21 20.826 207.615 71.400 77.584 46.897 60.5 表 2 特征波段数量与相关性值统计
Table 2 Statistics on number of feature bands (NFB)and correlation
变换
Transformation
NFB正相关极值
Positive extremes负相关极值
Negative extremes¯|r| R 643 0.227 −0.160 0.211 1/R 729 0.222 −0.222 0.206 lgR 680 0.225 −0.191 0.199 lg(1/R) 680 0.191 −0.225 0.199 R' 535 0.539 −0.552 0.259 (1/R)' 561 0.585 −0.502 0.261 lg'R 570 0.528 −0.577 0.262 lg'(1/R) 570 0.577 −0.528 0.262 R" 53 0.551 −0.532 0.250 (1/R)" 50 0.566 −0.564 0.255 lg"R 54 0.596 −0.557 0.250 lg"(1/R) 54 0.557 −0.596 0.250 表 3 SSC估算模型的训练集和验证集结果
Table 3 Training and validation results of the SSC estimation model
变量
Variable光谱处理方式
Spectral processing algorithm训练集Training set 验证集Validation set R2 RMSE/(g·kg−1) RPD R2 RMSE/(g·kg−1) RPD 全波段
Full bandR 0.528 26.267 1.06 0.414 26.335 1.06 (1/R)' 0.656 22.456 1.42 0.547 23.365 1.36 lg'R 0.587 24.418 1.15 0.521 23.009 1.23 lg'(1/R) 0.571 24.921 1.28 0.460 26.051 1.23 R_CWT_27 0.746 19.262 1.81 0.601 22.580 1.54 (1/R)'_CWT_28 0.821 16.049 1.70 0.715 17.467 1.56 lg'R_CWT_28 0.773 18.691 1.65 0.615 20.824 1.48 lg'(1/R)_CWT_28 0.863 14.079 2.36 0.665 19.916 1.67 R_DWT_H9 0.584 24.420 1.20 0.349 28.130 1.04 (1/R)'_DWT_H8 0.780 18.309 1.46 0.588 20.941 1.28 lg'R_DWT_H8 0.795 17.979 1.46 0.616 20.211 1.30 lg'(1/R)_DWT_H8 0.799 17.111 1.73 0.644 19.657 1.51 特征波段
Characteristic bandR 0.544 28.087 0.44 0.505 25.327 0.49 (1/R)' 0.793 18.553 1.36 0.573 21.258 1.19 lg'R 0.736 20.404 1.15 0.521 22.550 1.04 lg'(1/R) 0.743 20.966 1.11 0.487 23.315 1.00 R_CWT_27 0.732 19.565 1.54 0.598 21.144 1.42 (1/R)'_CWT_28 0.713 20.243 1.50 0.705 17.917 1.70 lg'R_CWT_28 0.707 20.843 1.65 0.689 19.593 1.75 lg'(1/R)_CWT_28 0.771 18.150 1.60 0.664 19.106 1.51 R_DWT_H9 0.682 21.287 1.41 0.507 23.790 1.26 (1/R)'_DWT_H8 0.744 19.566 1.60 0.623 20.666 1.51 lg'R_DWT_H8 0.785 17.798 1.52 0.680 18.400 1.47 lg'(1/R)_DWT_H8 0.755 19.001 1.52 0.634 19.878 1.45 -
[1] MA S L, HE B Z, XIE B Q, et al. Investigation of the spatial and temporal variation of soil salinity using Google Earth Engine: a case study at Werigan–Kuqa Oasis, West China[J]. Scientific Reports, 2023, 13(1): 2754. doi: 10.1038/s41598-023-27760-8
[2] 阿也提古丽·斯迪克. 宁夏平罗土壤含盐量遥感估算方法研究[D]. 南京:南京大学,2011. SIDIKE Ayetiguli. Estimation Model Of Salt Content In Soils By Remote Sensing: A Case Study Of Pingluo County, Ningxia Provice, China[D]. Nanjing: Nanjing University, 2011. (in Chinese with English abstract)
[3] 顾晨,梁建,刘旭颖,等. 高光谱遥感在草原植物多样性监测中的应用与展望[J]. 应用生态学报,2024,35(5):1397-1407. GU Chen, LIANG Jian, LIU Xuying, et al. Application and prospects of hyperspectral remote sensing in monitoring plant diversity in grassland[J]. Chinese Journal of Applied Ecology, 2024, 35(5): 1397-1407. (in Chinese with English abstract)
[4] 江远东,李新国,杨涵,等. 博斯腾湖湖滨绿洲土壤表层含盐量高光谱估算模型[J]. 中国土壤与肥料,2022(1):1-8. JIANG Yuandong, LI Xinguo, YANG Han, et al. Model of hyperspectral estimation of surface soil salinity in the Bosten Lake lakeside oasis in Xinjiang, China[J]. Soil and Fertilizer Sciences, 2022(1): 1-8. (in Chinese with English abstract)
[5] 孙亚楠,李仙岳,史海滨,等. 基于高光谱数据的盐荒地和耕地土壤盐分遥感反演优化[J]. 农业工程学报,2022,38(23):101-111. SUN Yanan, LI Xianyue, SHI Haibin, et al. Optimizing the inversion of soil salt in salinized wasteland using hyperspectral data from remote sensing[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2022, 38(23): 101-111. (in Chinese with English abstract)
[6] 管铖,刘明月,满卫东,等. 连续小波结合随机森林算法估算互花米草叶片叶绿素含量[J]. 光谱学与光谱分析,2024,44(10):2993-3000. GUAN Cheng, LIU Mingyue, MAN Weidong, et al. Estimation of chlorophyll content in spartina alterniflora leaves based on continous wavelet transformation and random forest algorithm[J]. Spectroscopy and Spectral Analysis, 2024, 44(10): 2993-3000. (in Chinese with English abstract)
[7] 郭斌,白昊睿,张波,等. 基于RF和连续小波变换的露天煤矿土壤锌含量高光谱遥感反演[J]. 农业工程学报,2022,38(10):138-147. GUO Bin, BAI Haorui, ZHANG Bo, et al. Inversion of soil zinc contents using hyperspectral remote sensing based on random forest and continuous wavelet transform in an opencast coal mine[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE), 2022, 38(10): 138-147. (in Chinese with English abstract)
[8] 王凡,陈龙跃,段丹丹,等. 小波分析的茶鲜叶全氮含量高光谱监测[J]. 光谱学与光谱分析,2022,42(10):3235-3242. WANG Fan, CHEN Longyue, DUAN Dandan, et al. Estimation of total nitrogen content in fresh tea leaves based on wavelet analysis[J]. Spectroscopy and Spectral Analysis, 2022, 42(10): 3235-3242. (in Chinese with English abstract)
[9] 张娟娟,牛圳,马新明,等. 基于离散小波的土壤全氮高光谱特征提取与反演[J]. 光谱学与光谱分析,2023,43(10):3223-3229. ZHANG Juanjuan, NIU Zhen, MA Xinming, et al. Hyperspectral feature extraction and estimation of soil total nitrogen based on discrete wavelet transform[J]. Spectroscopy and Spectral Analysis, 2023, 43(10): 3223-3229. (in Chinese with English abstract)
[10] 安柏耸,王雪梅,黄晓宇,等. 基于连续小波变换的土壤重金属镉含量的高光谱估测[J]. 地球与环境,2023,51(2):246-253. AN Baisong, WANG Xuemei, HUANG Xiaoyu, et al. Hyperspectral estimation of heavy metal cadmium content in soil based on continuous wavelet transform[J]. Earth and Environment, 2023, 51(2): 246-253. (in Chinese with English abstract)
[11] 王延仓,李笑芳,李莉婕,等. 基于离散小波-微分变换算法定量反演火龙果茎枝叶绿素含量的研究[J]. 光谱学与光谱分析,2023,43(2):549-556. WANG Yancang, LI Xiaofang, LI Lijie, et al. Quantitative inversion of chlorophyll content in stem and branch of pitaya based on discrete wavelet differential transform algorithm[J]. Spectroscopy and Spectral Analysis, 2023, 43(2): 549-556. (in Chinese with English abstract)
[12] LIN D, LI G Z, ZHU Y D, et al. Predicting copper content in chicory leaves using hyperspectral data with continuous wavelet transforms and partial least squares[J]. Comput Electron Agric, 2021, 187: 106293. doi: 10.1016/j.compag.2021.106293
[13] 叶淼,朱琳,刘旭东,等. 基于连续小波变换、SHAP和XGBoost的土壤有机质含量高光谱反演[J]. 环境科学,2024,45(4):2280-2291. YE Miao, ZHU Lin, LIU Xudong, et al. Hyperspectral inversion of soil organic matter content based on continuous wavelet transform, SHAP, and XGBoost[J]. Environmental Science, 2024, 45(4): 2280-2291. (in Chinese with English abstract)
[14] 郭玉雪,于洪丽,么航,等. 基于离散小波变换的卷积自编码运动想象脑电信号的分类[J]. 电子测量技术,2023,46(19):188-196. GUO Yuxue, YU Hongli, YAO Hang, et al. Classification of convolutional autoencoder motor imagery EEG signals based on discrete wavelet transform[J]. Electronic Measurement Technology, 2023, 46(19): 188-196. (in Chinese with English abstract)
[15] 周贝贝,李恒凯,龙北平. 离子吸附型稀土矿区复垦植被光谱特征变异提取与分析[J]. 光谱学与光谱分析,2023,43(12):3946-3954. ZHOU Beibei, LI Hengkai, LONG Beiping, et al. Variation analysis of spectral characteristics of reclaimed vegetation in an lonic rare earth mining area[J]. Spectroscopy and Spectral Analysis, 2023, 43(12): 3946-3954. (in Chinese with English abstract)
[16] 姜志伟,李云飞,姜涛,等. 基于离散小波分解与重构的多源土壤含水量数据融合方法与评估[J]. 农业工程学报,2023,39(7):145-156. JIANG Zhiwei, LI Yunfei, JIANG Tao, et al. Multi-source soil moisture data fusion method and its evaluation using decomposition and reconstruction of discrete wavelet transform[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE), 2023, 39(7): 145-156. (in Chinese with English abstract)
[17] 毛继华,赵恒谦,金倩,等. 河北铅锌尾矿库区土壤重金属含量高光谱反演方法对比[J]. 农业工程学报,2023,39(22):144-156. MAO Jihua, ZHAO Hengqian, JIN Qian, et al. Comparative study on the hyperspectral inversion methods for soil heavy metal contents in Hebei lead-zinc tailings reservoir areas[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE), 2023, 39(22): 144-156. (in Chinese with English abstract)
[18] 杨丽萍,任杰,王宇,等. 基于多源遥感数据的居延泽地区土壤盐分估算模型[J]. 农业机械学报,2022,53(11):226-235. YANG Liping, REN Jie, WANG Yu, et al. Soil salinity estimation model in Juyanze based on multi-source remote sensing data[J]. Transactions of the Chinese Society for Agricultural Machinery, 2022, 53(11): 226-235. (in Chinese with English abstract)
[19] 乔木,周生斌,卢磊,等. 新疆渭干河流域土壤盐渍化时空变化及成因分析[J]. 地理科学进展,2012,31(7):904-910. QIAO Mu, ZHOU Shengbin, LU Lei, et al. Causes and spatial-temporal changes of soil salinization in Weigan River basin, Xinjiang[J]. Progress in Geography, 2012, 31(7): 904-910. (in Chinese with English abstract)
[20] 李想,张永彬,刘明月,等. 滨海湿地土壤质地高光谱估测模型对比分析[J]. 光谱学与光谱分析,2024,44(9):2568-2576. LI Xiang, ZHANG Yongbin, LIU Mingyue, et al. Comparative analysis of hyperspectral estimation models for soil texture in coastal wetlands[J]. Spectroscopy and Spectral Analysis, 2024, 44(9): 2568-2576. (in Chinese with English abstract)
[21] 刘潭,王雯琦,李子默,等. 基于DWT-DE变换和AHA-ELM算法的水稻叶片氮含量预测[J/OL]. 农业机械学报,1-11[2024-12-19]. http://kns.cnki.net/kcms/detail/11.1964.S.20240904.1720.008.html. LIU Tan, WANG Wenqi, LI Zimo, et al. Prediction of nitrogen content in rice leaves based on DWT-DE transformation and AHA-ELM algorithm[J/OL]. Transactions of the Chinese Society for Agricultural Machinery, 1-11[2024-12-19]. http://kns.cnki.net/kcms/detail/11.1964.S.20240904.1720.008.html. (in Chinese with English abstract)
[22] 钟亮,钱家炜,储学远,等. 利用高光谱遥感技术监测小麦土壤重金属污染[J]. 农业工程学报,2023,39(5):265-270. ZHONG Liang, QIAN Jiawei, CHU Xueyuan, et al. Monitoring heavy metal contamination of wheat soil using hyperspectral remote sensing technology[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE), 2023, 39(5): 265-270. (in Chinese with English abstract)
[23] CSILLAG F, Pásztor L, BIEHL L L. Spectral band selection for the characterization of salinity status of soils[J]. Remote Sensing of Environment, 1993, 43(3): 231-242.
[24] 齐文文. 基于高光谱数据的民勤绿洲土壤含盐量预测[D]. 兰州:兰州大学,2011. QI Wenwen. Estimating Soil Salt Content Based On Hyper-Spectral Data In Minqin Oasis[D]. Lanzhou: Lanzhou University, 2011. (in Chinese with English abstract)
[25] 勾宇轩,赵云泽,李勇,等. 基于CWT-sCARS的东北旱作农田土壤有机质高光谱反演[J]. 农业机械学报,2022,53(3):331-337. GOU Yuxuan, ZHAO Yunze, LI Yong, et al. Soil organic matter content in dryland farmland in northeast China with hyperspectral reflectance based on CWT-sCARS[J]. Transactions of the Chinese Society for Agricultural Machinery, 2022, 53(3): 331-337. (in Chinese with English abstract)
[26] 李铠,常庆瑞,陈倩,等. 基于连续小波变换耦合CARS算法的冬小麦冠层叶片含水量估算[J]. 麦类作物学报,2023,43(2):251-258. LI Kai, CHANG Qingrui, CHEN Qian, et al. Estimation of water content in canopy leaf of winter wheat based on continuous wavelet transform coupled CARS algorithm[J]. Journal of Triticeae Crops, 2023, 43(2): 251-258. (in Chinese with English abstract)
[27] 王延仓,金永涛,王晓宁,等. 传统光谱变换与连续小波耦合定量反演潮土有机质含量[J]. 光谱学与光谱分析,2018,38(8):2571-2577. WANG Yancang, JIN Yongtao, WANG Xiaoning, et al. Quantitative inversion of organic matter content besed on interconnection trasitional spectral transform and continuous wavelet transform[J]. Spectroscopy and Spectral Analysis, 2018, 38(8): 2571-2577. (in Chinese with English abstract)
[28] 玉米提·买明,王雪梅. 连续小波变换的土壤有机质含量高光谱估测[J]. 光谱学与光谱分析,2022,42(4):1278-1284. MAIMING Yumiti, WANG Xuemei. Hyperspectral estimation of soil organic matter content based on continuous wavelet transformation[J]. Spectroscopy and Spectral Analysis, 2022, 42(4): 1278-1284. (in Chinese with English abstract)
[29] 江远东,李新国,杨涵. 基于连续小波变换的表层土壤有机碳含量的高光谱估算[J]. 江苏农业学报,2023,39(1):118-125. JIANG Yuandong, LI Xinguo, YANG Han, et al. Hyperspectral estimation of organic carbon content in surface soils based on continuous wavelet transform[J]. Jiangsu Journal of Agricultural Sciences, 2023, 39(1): 118-125. (in Chinese with English abstract)
[30] 王延仓,杨贵军,朱金山,等. 基于小波变换与偏最小二乘耦合模型估测北方潮土有机质含量[J]. 光谱学与光谱分析,2014,34(7):1922-1926. WANG Yancang, YANG Guijun, ZHU Jinshan, et al. Estimation of organic matter content of north fluvo-aquic soil based on the coupling model of wavelet transform and partial least squares[J]. Spectroscopy and Spectral Analysis, 2014, 34(7): 1922-1926. (in Chinese with English abstract)
-
期刊类型引用(6)
1. 高婧,李茂春,毛荣,张青,井立军. 1991―2022年新疆塔城地区植棉区气候变化对棉花生育期的影响. 中国棉花. 2024(02): 16-23 . 百度学术
2. 李云霞,王国栋,刘瑜,吕宁,梁飞,范军亮,尹飞虎. 新疆典型绿洲灌区土壤理化性状与盐分离子分布特征. 农业机械学报. 2024(07): 357-364+414 . 百度学术
3. 刘雲祥,张礼,咸文荣. 青海地区花生适宜播期研究. 青海农林科技. 2024(03): 39-43 . 百度学术
4. 熊坤,马美娟,金趁意,余卫东. 播期对花生‘豫花65’生长及产量影响. 中国农学通报. 2024(32): 16-22 . 百度学术
5. 陈春波,李均力,赵炎,夏江,田伟涛,李超锋. 新疆草地时空动态及其对气候变化的响应——以昌吉回族自治州为例. 干旱区研究. 2023(09): 1484-1497 . 百度学术
6. 李欢,翟孟如,江伟,韦钢,刘苏,吕晓霞,钟昀平. 新疆冷凉区域花生高产栽培技术. 新疆农业科技. 2023(06): 24-27 . 百度学术
其他类型引用(3)