基于组合模型的玛纳斯河流域农田土壤盐分反演

doi:10.13304/j.nykjdb.2021.0872

摘要/Abstract

摘要：

盐渍化是影响土壤质量和作物生长的重要因素之一，利用遥感技术大面积获取土壤盐分信息具有重大意义。以新疆玛纳斯河流域农田为研究对象，将偏最小二乘回归模型（PLSR）和BP神经网络模型（BPNN）相结合，构建组合模型来反演土壤盐渍化状况。结果表明，与土壤盐分相关性较高且具有代表性的遥感指数为归-化植被指数（NDVI）、比值植被指数（RVI）和土壤调整植被指数（SAVI），其相关性系数分别为-0.746、-0.663和-0.733。单项预测模型中偏最小二乘回归模型的预测精度最高，其决定系数（R²）为0.759，均方根误差（RMSE）为3.159。组合模型R²为0.797，RMSE为3.611，其验证精度较单项预测模型有所提高，较PLSR模型提高了0.038，较BPNN提高了0.094。组合模型可更准确地预测出玛纳斯河流域农田土壤盐分空间分布状况。玛纳斯河流域农田土壤盐渍化以轻度和中度盐渍化为主，所占比例达到35.34%和25.66%，与实测结果一致。组合模型较单项模型可以获得更准确的土壤盐分空间分布状况，为新疆玛纳斯河流域农田土壤盐渍化治理和土地资源可持续利用提供理论依据。

关键词: 土壤盐分, 组合模型, 遥感反演, 空间分布

Abstract:

Salinization is one of the important factors affecting soil quality and crop growth. It is of great significance to use remote sensing technology to obtain soil salinity information on a large scale. This study took farmland in the Manas River Basin in Xinjiang as the research object， and combined the partial least squares regression model （PLSR） and the BP neural network model （BPNN） to construct a combined model to invert the soil salinization status. The results showed that the representative remote sensing indexes with high correlation with soil salinity were normalized differential vegetation index （NDVI）， ratio vegetation index （RVI） and soil-adjusted vegetation index （SAVI）， and their correlation coefficients were -0.746， -0.663 and -0.733， respectively. Among the single prediction models， the partial least squares regression model had the highest prediction accuracy， the coefficient of determination （R²） was 0.759， and the root mean square error （RMSE） was 3.159. The R² combined model was 0.797， and RMSE was 3.611. Its verification accuracy was higher than that of the single prediction model， which was 0.038 higher than that of PLSR model， and 0.094 higher than that of BPNN. Therefore， the combined model could more accurately predict the spatial distribution of soil salinity in the farmland of the Manas River Basin. The soil salinization of farmland in the Manas River Basin was dominated by mild and moderate salinization， accounting for 35.34% and 25.66%， which were consistent with the results of the field survey. Predicting the soil salt content in the study area based on the combined model could obtain a more accurate spatial distribution of soil salt， and provided theoretical basis for the management of farmland soil salinization and the sustainable use of land resources in the Manas River Basin in Xinjiang.

Key words: soil salinity, combination model, remote sensing inversion, spatial distribution

中图分类号:

S153

杨小虎, 张曼玉, 杨海昌, 张凤华, 江宜霖, 易小兰. 基于组合模型的玛纳斯河流域农田土壤盐分反演[J]. 中国农业科技导报, 2023, 25(1): 134-141.

Xiaohu YANG, Manyu ZHANG, Haichang YANG, Fenghua ZHANG, Yilin JIANG, Xiaolan YI. Inversion of Soil Salinity in Farmland of Manas River Basin Based on Combined Model[J]. Journal of Agricultural Science and Technology, 2023, 25(1): 134-141.

图/表 8

参考文献 20

1	LI J G, PU L J, HAN M F, et al.. Soil salinization research in China: advance and prospects [J]. J. Geogr. Sci., 2014, 24(5): 943-960.
2	METTERNICHT G I, ZINCK J A. Remote sensing of soil salinity: potentials and constraints [J]. Remote Sens. Environ., 2003, 85(1): 1-20.
3	邱元霖,陈策,韩佳,等.植被覆盖条件下的解放闸灌域土壤盐分卫星遥感估算模型[J].节水灌溉,2019(10): 108-112.
	QIU Y L, CHENG C, HAN J, et al.. Satellite remote sensing estimation model of soil salinity in Jiefangzha irrigation under vegetation coverage [J]. Water Saving Irrig., 2019(10): 108-112.
4	张同瑞,赵庚星,高明秀,等.基于近地多光谱和OLI影像的黄河三角洲冬小麦种植区盐分估算及遥感反演——以山东省垦利县和无棣县为例[J].自然资源学报, 2016, 31(6): 1051-1060.
	ZHANG T R, ZHAO G X, GAO M X, et al.. Soil salinity estimation and remote sensing inversion based on near-ground multispectral and TM imagery in winter wheat growing area in the Yellow River Delta—case study in Kenli County and Wudi county, Shandong province [J]. J. Nat. Resour., 2016, 31(6): 1051-1060.
5	姜红,玉素甫江·如素力,热伊莱·卡得尔,等.基于神经网络模型的干旱区绿洲土壤盐渍化评价分析[J].地球信息科学学报, 2017, 19(7): 983-993.
	JIANG H, Yusufujiang R S L, Reyilai K D E, et al.. Evaluation and analysis of soil salinization in the arid zones based on neural network model [J]. J. Geo-Information Sci., 2017, 19(7): 983-993.
6	BATES J M, GRANGER C W J. The combination of forecasts [J]. Oper. Res. Quarterly, 1969, 20(4): 451-468.
7	冯海宽,杨福芹,李振海,等.最优权重组合模型和高光谱估算苹果叶片全磷含量[J].农业工程学报,2016,32(7):173-180.
	FENG H K, YANG F Q, LI Z H, et al.. Hyperspectral estimation of leaf total phosphorus content in apple tree based on optimal weights combination model [J]. Trans. Chin. Soc. Agric. Eng., 2016, 32(7): 173-180.
8	殷礼胜,魏帅康,孙双晨,等.基于FEEMD-SAPSO-BiLSTM组合模型的短时交通流预测[J].电子测量与仪器学报,2021,35(10):72-81.
	YIN L S, WEI S K, SUN S C, et al.. Short-term traffic flow forecast based on FEEMD-SAPSO-BiLSTM combined model [J]. J. Electron. Measure Instrument, 2021, 35(10): 72-81.
9	唐湘玲,吕新.近50年来玛纳斯河流域气候变化对可利用降水量的影响[J].湖北农业科学, 2011, 50(22): 4582-4585.
	TANG X L, YU X. Influence of climate change on available precipitation in Manas River Basin over last 50 years [J]. Hubei Agric. Sci., 2011, 50(22): 4582-4585.
10	杨海昌,张凤华,王东方,等. 近60年玛河流域绿洲蒸散量变化趋势及其影响因素分析[J].干旱区资源与环境, 2014, 28(7):18-23.
	YANG H C, ZHYANG F H, WANG D F, et al.. The analysis of the variation trend and influencing factors of evapotranspiration in oasis along Manas river valley in the last 60 years [J]. J. Arid Land Resour. Environ., 2014, 28(7):18-23.
11	王海江,石建初,张花玲,等.不同改良措施下新疆重度盐渍土壤盐分变化与脱盐效果[J].农业工程学报,2014, 30(22): 102-111.
	WANG H J, SHI J C, ZHANG H L, et al.. Changes of salinity and desalination effect of severe saline soil in Xinjiang under different improvement measures [J]. Chin. J. Agric. Eng., 2014, 30(22): 102-111.
12	新疆维吾尔自治区农业厅, 新疆维吾尔自治区土壤普查办公室.新疆土壤[M].北京:科学出版社,1996:304-336.
13	WOLD H. Soft modelling by latent Variables: The Non-linear Iterative Partial Least Squares (NIPALS) Approach [J]. J. Appl. Prob., 1975,12(S1): 117-142.
14	SHIRAZI H, NASARIAN S, ABDALLAH I. Implementation of artificial neural networks to automate SASW inversion [C]// Proceedings of Geocongress 2006 Implementation of Artificial Neural Networks to Automate SASW Inversion. Atlanta, Georgia, United States, 2006:1-6.
15	张超.基于组合模型的我国能源消费预测研究[D].长沙:湖南大学,2017.
	ZHANG C. Forecast of China’s energy consumption based on combination model [D]. Changsha: Hunan University, 2017.
16	贾伟,高小红,杨扬,等.基于Landsat 8 OLI影像的三江源区表层土壤全氮空间格局反演[J].干旱区研究,2015, 32(5):890-896.
	JIA W, GAO X H, YANG Y, et al.. Inversion of spatial distribution pattern of topsoil total nitrogen contents in Sanjiangyuan regions based on OLI images [J]. Arid Zone Res., 2015, 32(5): 890-896.
17	张雅莉,塔西甫拉提·特依拜,阿尔达克·克里木,等.基于Landsat8 OLI影像光谱的土壤盐分估算模型研究[J].国土资源遥感,2018,30(1):87-94.
	ZHANG Y L, Tashpolat T Y B, RDAK K L M, et al.. Estimation model of soil salinization based on Landsat8 OLI image spectrum [J]. Remote Sens. Land Resour., 2018, 30(1):87-94.
18	向速林.地下水水质评价的多元线性回归分析模型研究[J].新疆环境保护,2005(4):21-23.
	XIANG S L. A study of multivariate linear regression analysis model for groundwater quality evaluation [J]. Xinjiang Environ. Protection, 2005(4):21-23.
19	李洪涛.灌区降雨预测的组合模型研究[D].哈尔滨:东北农业大学, 2014.
	LI H T. Research on combined model of rainfall forecast in irrigation district [D]. Harbin: Northeast Agricultural University, 2014.
20	王雪梅,周晓红.渭干河-库车河三角洲绿洲棉田土壤盐分估算及遥感反演[J].干旱地区农业研究,2018,36(6): 250-254, 262.
	WANG X M, ZHOU X H. Estimation and inversion modeling of salinity of cotton field soil using remote sensing in the Delta Oasis of Weigan and Kuqa Rivers [J]. Agric. Res. Arid Areas, 2018, 36(6): 250-254, 262.

指数Index	SSC	SAVI	RVI	SI	SI₁	SI₂	SI₃	NDVI	NDWI	DVI
SSC	1.000
SAVI	-0.733^**	1.000
RVI	-0.663^**	0.947^**	1.000
SI	0.640^**	-0.665^**	-0.776^**	1.000
SI₁	0.664^**	-0.680^**	-0.789^**	0.984^**	1.000
SI₂	0.354^**	-0.174^**	-0.368^**	0.817^**	0.834^**	1.000
SI₃	0.663^**	-0.680^**	-0791^**	0.983^**	0.999^**	0.836^**	1.000
NDVI	-0.746^**	0.967^**	0.985^**	-0.802^**	-0.820^**	-0.393^**	-0.820^**	1.000
NDWI	0.742^**	-0.935^**	-0.923^**	0.801^**	0.804^**	0.359^**	0.798^**	-0.951^**	1.000
DVI	-0.680^**	0.975^**	0.864^**	-0.507^**	-0.519^**	0.031	-0.518^**	0.887^**	-0.869^**	1.000

指数Index	SSC	SAVI	RVI	SI	SI₁	SI₂	SI₃	NDVI	NDWI	DVI
SSC	1.000
SAVI	-0.733^**	1.000
RVI	-0.663^**	0.947^**	1.000
SI	0.640^**	-0.665^**	-0.776^**	1.000
SI₁	0.664^**	-0.680^**	-0.789^**	0.984^**	1.000
SI₂	0.354^**	-0.174^**	-0.368^**	0.817^**	0.834^**	1.000
SI₃	0.663^**	-0.680^**	-0791^**	0.983^**	0.999^**	0.836^**	1.000
NDVI	-0.746^**	0.967^**	0.985^**	-0.802^**	-0.820^**	-0.393^**	-0.820^**	1.000
NDWI	0.742^**	-0.935^**	-0.923^**	0.801^**	0.804^**	0.359^**	0.798^**	-0.951^**	1.000
DVI	-0.680^**	0.975^**	0.864^**	-0.507^**	-0.519^**	0.031	-0.518^**	0.887^**	-0.869^**	1.000

建模方法 Modeling method	验证集Validation set
建模方法 Modeling method	决定系数R²	均方根误差RMSE
PLSR	0.759	3.325
BPNN	0.703	4.794
组合模型 Combination model	0.797	3.111

建模方法 Modeling method	验证集Validation set
建模方法 Modeling method	决定系数R²	均方根误差RMSE
PLSR	0.759	3.325
BPNN	0.703	4.794
组合模型 Combination model	0.797	3.111

土壤盐渍化分级 Soil salinization classification	面积 Area/km²	比例Ratio/%
非盐渍土Non saline soil	762.832	8.46
轻度盐渍土Slightly saline soil	3 186.417	35.34
中度盐渍土Moderately saline soil	2 313.325	25.66
重度盐渍土Severe saline soil	2 230.275	24.74
盐土Saline soil	523.626	5.80