Chang Dong , Xiangtian Meng , Weimin Ruan , Jian Cui , Xinle Zhang , Huanjun Liu
{"title":"An innoval hyperspectral prediction model for soil organic matter in croplands of the Northeast China Mollisols Region","authors":"Chang Dong , Xiangtian Meng , Weimin Ruan , Jian Cui , Xinle Zhang , Huanjun Liu","doi":"10.1016/j.still.2025.106666","DOIUrl":null,"url":null,"abstract":"<div><div>Soil organic matter (SOM) is a primary source of soil nutrients, and accurately estimating SOM content is crucial for boosting agricultural productivity and enhancing soil fertility. Addressing the challenges of high spatial heterogeneity of SOM across large regions and complex soil environments, this study presents a new paradigm for SOM prediction using Lab-measured spectral data (VIS-NIR-SWIR) integrated with advanced deep learning techniques. In this paradigm, hyperspectral reflectance data from 1087 surface soil samples were combined with results from China's second national soil census. Three grouping strategies were tested: no-grouping (NG), traditional grouping (TG), and spectral feature difference grouping (SG). Four modeling algorithms—random forests (RF), convolutional neural networks (CNN), long short-term memory networks (LSTM), and the improved A-LSTM model—along with four spectral preprocessing methods (raw spectra, continuum removal, fractional-order differential, and spectral characteristic parameters (SCPs)-extracted spectra) were evaluated for quantifying SOM content. The following conclusions were drawn: (1) The new paradigm integrates various grouping strategies, models, and inputs. The highest prediction accuracy (R² = 0.89, RMSE = 0.55 %) was achieved by combining the A-LSTM model with SCPs as input variables and SG as the grouping strategy. (2) The A-LSTM increased R² by 0.22, 0.15, and 0.04, and reduced RMSE by 0.36 %, 0.45 %, and 0.06 %, respectively, compared to other models. Additionally, the new paradigm demonstrates strong performance in SOM prediction by optimizing the combination of grouping strategies, models, and inputs, establishing it as the optimal approach. Notably, errors in the low (0–2 %) and high (8–10 %) SOM content intervals were significantly reduced. (3) Among the grouping strategies, SG was the most effective, followed by TG, with SG increasing R² by 0.31 compared to NG. (4) Among the input variables, the SCPs-based prediction model performed the best, improving R² by 0.35 compared to the original spectra. The proposed A-LSTM model successfully captured the nonlinear relationship between spectra and organic matter, offering strong technical support for future large-scale SOM monitoring.</div></div>","PeriodicalId":49503,"journal":{"name":"Soil & Tillage Research","volume":"253 ","pages":"Article 106666"},"PeriodicalIF":6.1000,"publicationDate":"2025-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Soil & Tillage Research","FirstCategoryId":"97","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S016719872500220X","RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"SOIL SCIENCE","Score":null,"Total":0}
引用次数: 0
Abstract
Soil organic matter (SOM) is a primary source of soil nutrients, and accurately estimating SOM content is crucial for boosting agricultural productivity and enhancing soil fertility. Addressing the challenges of high spatial heterogeneity of SOM across large regions and complex soil environments, this study presents a new paradigm for SOM prediction using Lab-measured spectral data (VIS-NIR-SWIR) integrated with advanced deep learning techniques. In this paradigm, hyperspectral reflectance data from 1087 surface soil samples were combined with results from China's second national soil census. Three grouping strategies were tested: no-grouping (NG), traditional grouping (TG), and spectral feature difference grouping (SG). Four modeling algorithms—random forests (RF), convolutional neural networks (CNN), long short-term memory networks (LSTM), and the improved A-LSTM model—along with four spectral preprocessing methods (raw spectra, continuum removal, fractional-order differential, and spectral characteristic parameters (SCPs)-extracted spectra) were evaluated for quantifying SOM content. The following conclusions were drawn: (1) The new paradigm integrates various grouping strategies, models, and inputs. The highest prediction accuracy (R² = 0.89, RMSE = 0.55 %) was achieved by combining the A-LSTM model with SCPs as input variables and SG as the grouping strategy. (2) The A-LSTM increased R² by 0.22, 0.15, and 0.04, and reduced RMSE by 0.36 %, 0.45 %, and 0.06 %, respectively, compared to other models. Additionally, the new paradigm demonstrates strong performance in SOM prediction by optimizing the combination of grouping strategies, models, and inputs, establishing it as the optimal approach. Notably, errors in the low (0–2 %) and high (8–10 %) SOM content intervals were significantly reduced. (3) Among the grouping strategies, SG was the most effective, followed by TG, with SG increasing R² by 0.31 compared to NG. (4) Among the input variables, the SCPs-based prediction model performed the best, improving R² by 0.35 compared to the original spectra. The proposed A-LSTM model successfully captured the nonlinear relationship between spectra and organic matter, offering strong technical support for future large-scale SOM monitoring.
期刊介绍:
Soil & Tillage Research examines the physical, chemical and biological changes in the soil caused by tillage and field traffic. Manuscripts will be considered on aspects of soil science, physics, technology, mechanization and applied engineering for a sustainable balance among productivity, environmental quality and profitability. The following are examples of suitable topics within the scope of the journal of Soil and Tillage Research:
The agricultural and biosystems engineering associated with tillage (including no-tillage, reduced-tillage and direct drilling), irrigation and drainage, crops and crop rotations, fertilization, rehabilitation of mine spoils and processes used to modify soils. Soil change effects on establishment and yield of crops, growth of plants and roots, structure and erosion of soil, cycling of carbon and nutrients, greenhouse gas emissions, leaching, runoff and other processes that affect environmental quality. Characterization or modeling of tillage and field traffic responses, soil, climate, or topographic effects, soil deformation processes, tillage tools, traction devices, energy requirements, economics, surface and subsurface water quality effects, tillage effects on weed, pest and disease control, and their interactions.