| The sustainable use and protection of the croplands in the black soil region of Northeastern China is of paramount importance for maintaining the country’s food security.Since the founding of the People’s Republic of China,tillage practices in the black soil region has been dominated by conventional methods such as rotary tillage,straw removal and burning.The continuous pressure of high-intensity farming on the croplands has caused serious land degradation in this area.In recent years,with the implementation of the National Action Plan for Conservation Tillage in the black soil region,the area under conservation tillage has been increasing.However,due to the lack of a mature spatiotemporal monitoring system for conservation tillage,little is known about the the magnitude and extent of this increasein conservation tillage.In this thesis,Gongzhuling,Lishu County and Yushu City,which are located in the black soil area of central Jilin Province,were selected as the study area,and field surveys and surface straw cover RGB image collection were carried out to facilitate the development of a straw cover image recognition method based on the entangled random forest model,in order to overcome the current difficulty in acquiring such data;furthermore,building on the established spatial database of the cropland straw cover percentage,we built a remote sensing method for identifying and spatially characterizing the extent of conservation tillage based on Sentinel-2 imagery,and quantified the spatial and temporal change of the extent of conservation tillage in the study area from 2018 to 2021.The main conclusions are summarized in the following:(1)The overall ratio of no-tillage to conventional tillage in the 473 surveyed plots in the study area is close to 1:1,among which no-tillage plots in Lishu County account for72.86%,benefiting from the highly effective promotion of the“Lishu model”,while no-tillage plots in Gongzhuling City and Yushu City account for only 36.03%and 39.09%,respectively,.This suggests that the promotion of conservation tillage in these two areas is still in the initial stage,and still has a lot of room for improvement.(2)An entangled random forest model was developed to achieve automatic image recognition of four types of land elements(bare soil,straw,green vegetation,and gravel).Using this method,an extensive data set of straw coverage percentage was established for the study area,offering a promising solution to the limitation that such type of data is difficult to obtain due to hight costs of time and labor.The image recognition results show that the implementation of no-tillage methods brought an substantial increase in surface straw cover percentage in all three regions,with the average straw cover on no-tillage plots(8.05%)being more than twice as high as that on conventional tillage plots(3.45%).In the mean time though,there were still 30%-35%of no-tillage plots with straw cover<4%.It indicates that during the large-scale promotion of no-tillage,mininal tillage and other conservation tillage measures,attention should be paid to the simultaneous increase in straw return rate so as to improve the straw cover percentage in no-tillage plots.(3)A remote sensing inversion method was built to detect the extent conservation tillage.A total of 29 indicators consisting of Sentinel-2 surface reflectance,various spectral indices,texture features and topographic indexes were used as explanatory variables for modelling development.Random forest,support vector machine and logistic regression were used as modelling algorithms,and their modelling performances were cross-compared.Results show that the latter two of the three tested methods achieved satisfactory classification accuracy(>68%)in classifying tillage practices,among which the logistic regression model was found to be the optimal solution in terms of overall identification accuracy and generalization.Both the user accuracy and producer accuracy reached over70%for logistic regreesion,which enabled effective mapping of conservation tillage extent at regional scale.(4)The logistic regression classification model was applied to the Sentinel-2 images acquired in2018 and 2021,in order to produce the spatial distribution maps of conservation tillage in the study area.From the temporal perspective,both the absolute area and relative proportion of conservation tillage in the study area increased significantly during the three-year period,with the highest coverage of conservation tillage found in Lishou County,where the area and proportion of conservation tillage to total arable land were 2.93×10~6 mu and77.79%,respectively;the largest increase during the three years was in Yushu City,where the area under conservation tillage increased from 9.85×10~5 mu in 2018 to 2.96×10~6 mu in 2021;the smallest area covered by conservation tillage was in Gongzhuling City,which also had an increase of 18.78%,from 11.23%(5.71×10~5 mu)to 30.01%(1.53×10~6 mu).From the spatial perspective,Lishu County achieved a widespread coverage of conservation tillage across the entire cropland extent,while in Gongzhuling City and Yushu City,conservation tillage plots were mostly concentrated in areas close to urban area and/or with flatter terrain,possibly due to the limited availability of no-till seeders at the beginning of conservation tillage promotion.In this study,an automatic straw cover image recognititon method was combined with Sentinel-2 remote sensing to enable systematic and spatial quantification and monitoring of conservation tillage at regional scale.This methodology can provide new insights for the investigation of conservation tillage promotion in the black soil region. |
PDF Full Text Request