Mapping Key Soil Properties Based On Crop Rotation Information

Due to long-term intensive agriculture,the issues of soil fertility decline and soil acidification in the lateritic red soil region of southern China are increasingly prominent,which seriously affect cropland quality and threaten food and ecological security.There is an urgent need for accurate spatial information of soil properties to provide a basis for improving cropland quality and ensuring food and ecological security.Human agricultural activities including crop rotation constantly change cropland soil properties,which significantly influence the mapping accuracy of cropland soil properties.However,the lack of relevant spatial data makes it hard to represent human agricultural activities in soil mapping to reflect the long-term impacts of human on soil,thus limiting the mapping accuracy.Continuously improved spatial and temporal resolutions of remote sensing data make it possible to identify human agricultural activities quickly,accurately over a large extent and capture their dynamics.Therefore,how to use time-series remote sensing data to accurately identify human agricultural activities and apply human agricultural activity information to soil mapping to improve the mapping accuracy is a key scientific issue for soil mapping.Zengcheng District,Guangzhou City,Guangdong Province,a typical agricultural production region with lateritic red soil,was selected as the study area.In addition,the time-series Sentinel satellite images,crop and soil samples,and agricultural census data were employed to explore the methods for mapping cropland soil properties based on crop rotation information.The main contents and conclusions of this study are as follows:(1)Mapping crop rotation based on remote sensing.First,based on the field survey,farmer interviews,questionnaires and agricultural census data,a regional tailored crop rotation classification system was proposed to fully characterize the complex crop rotation conditions.Three main systems were determined according to the main crop types,i.e.,paddy field,vegetable and orchard system,which were further divided into nine subsystems according to their seasonal dynamics.Then,the time-series Sentinel-1 radar and Sentinel-2 optical images were applied to estimate four key indicators:flooding frequency,cropping intensity,cropping diversity and coefficient of variation.Finally,the three main systems were classified by the flooding frequency and random forest,based on which each subsystem was identified by combining the four key indicators and specific rules.Results show that the overall accuracies of the crop rotation systems in 2020 and 2021 are 81%and 84%,respectively.The orchard system has the highest mapping accuracy,followed by the paddy and vegetable systems.Spatially,the northern part of the study area mainly has paddy and orchard systems,while the southern part is dominated by orchard and vegetable systems.The paddy system is mainly composed of single rice rotated with vegetables and double rice rotated with vegetables,and the vegetable system is mainly composed of low-diversity vegetables,while the orchard system is mainly composed of short-term orchard.(2)Mapping soil organic matter(SOM)in cropland based on crop rotation information.First,the differences in SOM among different crop rotation systems were analyzed.Then,the framework of digital soil mapping(DSM)was adopted to conduct this study.The key environmental factors influencing cropland SOM were determined based on a literature review and collected from multiple data sources.Combining them with the mapped crop rotation systems,five groups of environmental variables were designed.Finally,based on the soil samples,the SOM prediction models were constructed by random forest regression and forward recursive feature selection algorithm,and the model performances were evaluated by a 10-fold cross-validation procedure.Results show significant differences in SOM among the crop rotation subsystems.Single rice rotated with vegetables and high-diversity vegetables have the highest SOM contents,followed by long-term orchard and double rice rotated with vegetables,while short-term orchard and low-diversity vegetables have the lowest contents.For different combinations,the addition of time-series remote sensing data slightly increases the prediction accuracy;adding the crop rotation main-systems in 2021 does not alter the results,while adding the sub-systems in 2021 greatly enhances the predictability of SOM with a decrease in RMSE by7%and an increase in R~2 by 24%.And the crop rotation sub-systems in 2021 appears more important in the SOM predictive models than the soil,topographic,and climatic variables.Besides,the addition of the crop rotation main-and sub-systems in 2020 does not make a big difference in the results.(3)Mapping soil pH in cropland based on crop rotation information.First,the differences in soil pH among different crop rotation systems were analyzed.Then,based on existing environmental variables,the key environmental factors that affect soil pH were incorporated.Combining them with the identified crop rotation systems,six groups of environmental variables were designed.Finally,the soil pH prediction models were constructed by random forest regression and forward recursive feature selection algorithm,and the prediction models were validated.Results show that significant differences in soil pH are observed among the crop rotation main-and sub-systems.The paddy system has the highest soil pH,followed by the vegetable system,and the orchard system has the lowest soil pH.For different combinations,the addition of time-series remote sensing data greatly increases the prediction accuracy,and the addition of the crop rotation main-and sub-systems in 2021 further improves the predictability,with an increase in R~2 from 0.39 to 0.43.And the addition of the crop rotation main-and sub-systems in 2020 does not alter the results.The crop rotation main-and sub-systems in 2021 also present a high importance in the soil pH predictive models.Based on a comprehensive understanding of the crop and soil characteristics in the study area,this thesis constructs a intra-annual crop rotation classification system.This system fully summarizes the complex characteristics of crop rotation and can reflect the differences in SOM and pH.On this basis,this study developes an innovative method for mapping the complex intra-annual crop rotations based on remote sensing.It can directly map crop rotation instead of overlaying crop layers,providing a new perspective for mapping and understanding crop rotation.This method can effectively identify crop rotation in cloudy and rainy areas in southern China,offering a solution for mapping complex crop rotations.This study also demonstrates the effectiveness of including crop rotation in the predictive models for SOM and soil pH,providing a reference for the application of crop rotation in DSM.The research is of great significance in exploring effective ways to improve the mapping accuracy of soil properties in cropland,as well as addressing soil issues and improving cropland quality through optimizing crop rotation distribution and composition.