Maize Yield Remote Sensing Estimation Model By Quantitative Monitoring Of Key Factors Of Cultivated Land Quality

As a valuable natural resource,the quality of cultivated land is closely related to crop productivity.The quality of cultivated land directly affects food security and the healthy development of agriculture.Crop yield is the ultimate goal of farmland cultivation and a direct indicator of farmland productivity and income.Maize is the largest food crop in China,and timely and accurate monitoring of maize growth and estimating maize yield information can provide decision support for maize planting management.This is of great significance to food production.Remote sensing technology has been widely used in large area,large area of growth monitoring and crop yield estimation and prediction,but there are more and more problems.At present,the yield estimation model lacks systematic research on the selection of influencing factors and the timing of yield estimation.The mining and expression of agricultural knowledge of multi-source data is shallow.Most of the research is limited to the use of fixed phase or growth period characteristic parameters combined with meteorological characteristics to estimate yield.However,the remote sensing data is affected by cloud,light,satellite attitude and so on,which may lead to the lack of key time-phase data in the model,and the robustness of the model is poor.The final yield of crops is directly related to the quality of cultivated land.Therefore,it is necessary to consider the factors affecting the yield from various aspects and angles from the perspective of cultivated land quality monitoring,so as to construct a stable and universal maize yield remote sensing estimation model.In this study,Youyi Farm was taken as the research area,and the long-term sequence NDVI maximum value in 1986 and 2022,NDVI remote sensing image data in the early stage of crop growth,Sentinel-2 NDVI image data,Planet image data in the critical period,meteorological data,terrain data with different resolutions,crop planting structure,soil type,soil organic matter and soil texture and other basic data were obtained.The topographic factors and spatial distribution characteristics of soil and texture extracted from different resolutions in the study area were analyzed,and the soil erosion rate was calculated by RUSLE soil erosion equation.Based on the analysis of the spatial and temporal changes of NDVI in long time series within and between years,from the new perspective of three-dimensional monitoring of cultivated land quality,the key elements of three-dimensional monitoring of cultivated land quality were determined from the perspectives of soil,production capacity and ecology.Finally,a corn remote sensing yield estimation model based on single-phase NDVI,single-phase NDVI and terrain factors,and key elements of cultivated land quality was constructed.The main research contents and results are as follows:(1)The study analyzed the effects of different resolutions of digital elevation model(DEM)data on terrain factors and their relationship with crop yield.By comparing and analyzing key terrain factors extracted from DEMs of varying resolutions,the study found that DEM resolution influenced terrain factor extraction.Lower DEM resolutions resulted in a smaller proportion of high slope areas,with slope areas concentrating in the range of 0-6°at resolutions of 12.5 m and 30 m.Aspect extraction was relatively stable across different resolutions.As DEM resolution improved,topographic relief became more detailed,leading to increased variation within the field.Surface roughness decreased with decreasing DEM resolution,resulting in a smoother surface in higher slope areas.Analyzing the relationship between terrain factors and crop yield revealed that elevation was significantly negatively correlated with yield across all four DEM resolutions.However,slope direction and surface roughness were not correlated with yield.On a regional scale,the topographic factors extracted from rough terrains better represented the large-scale topographic location characteristics compared to higher resolution DEMs.In conclusion,the study demonstrated the varying effects of DEM resolution on terrain factor extraction and its relationship with crop yield.It highlighted the importance of considering DEM resolution and topographic factors when analyzing large-scale topographic characteristics and their impact on agricultural productivity.(2)The study analyzed the spatial characteristics of soil organic matter and soil texture in the research area and calculated the soil erosion modulus to assess the degree of soil erosion over a long period.The results showed that the organic matter content in the soil was unstable,and the content of organic matter varied significantly across different locations.Similarly,the stability of silt and clay content in soil texture was low,while the stability of sand content was very low.As a result,the content of silt and clay in soil varied significantly across the study area,as did the content of sand.The study also found that soil erosion had distinct spatial distribution characteristics in the research area,with relatively low erosion levels in the central and eastern parts of Youyi Farm and a higher level of erosion in the western region.From a temporal perspective,the trend of soil erosion in Youyi Farm from 2000 to 2020 remained consistent,but the study identified an increasing trend of soil erosion since 2000 due to the acceleration of the scale process of Youyi Farm and the use of unreasonable farming methods and fertilization.(3)Spatial autocorrelation analysis was conducted to examine the spatial distribution characteristics of maximum NDVI and initial NDVI of cultivated and dry land in Youyi Farm.Based on SPEI results,the meteorological year scene was determined to divide the period from 1986 to 2020.Results indicated that there were differences in inter-annual variability of NDVI maximum in long time series of different slope positions,paddy fields,soil erosion degrees,and soil types.Variability of slope bottom was more pronounced than slope top in some years.In dry land,changes in NDVI maximum value of time series were more obvious than in paddy fields.The higher the degree of soil erosion,the greater the inter-annual time series NDVI maximum curve changes,and the inter-annual variation of NDVI maximum value was also greater for lower organic matter contents.Sand,silt,and clay content in soil texture exhibited similar time series curves between years,with high and low contents showing greater inter-annual fluctuations than medium content.The coefficient of variation of NDVI was significantly higher than the maximum NDVI at the beginning of the long time series,likely due to farming time,field progress,and terrain-induced differences in emergence time.Results of spatial autocorrelation analysis indicated that the early-stage average value of NDVI was significantly different from the maximum value.Ultimately,it was found that multi-year average value of maximum NDVI in long time series could better reflect the stability of cultivated land quality.Additionally,the maximum NDVI value decreased to varying degrees in different meteorological years.(4)This study analyzed the time series curve of maize yield during the year and used different machine learning algorithms to fit and reconstruct the curve to determine the optimal phase of yield estimation.The correlation analysis method was used to calculate the correlation coefficient between the measured yield and influencing factors to determine the input of the yield estimation model.Linear and machine learning regression algorithms were used to establish the remote sensing model of maize yield in the study area.The results showed that using Whitteaker Henderson to reconstruct the NDVI time series with a single-day step yielded the highest accuracy among the four machine learning algorithms,considering R~2 and RMSE as evaluation indexes.The correlation analysis showed that mid-July NDVI was most correlated with yield,indicating that it was feasible to estimate maize yield using NDVI around mid-July.The prediction model of the spatial distribution map of maize yield in Youyi Farm,established by a random forest model using 9 indexes of cultivated land quality capacity(topography,climate,vegetation growth and stability),soil(soil sand content),and ecology(soil erosion modulus),had the highest accuracy.R~2 was 0.891,and RMSE was35.347 kg/acre.The scatter plot R~2 of the measured and predicted yield of the random forest model reached 0.563,RMSE was 65.278 kg/acre,and MAPE reached 6.5%.The spatial distribution map of maize yield in Youyi Farm showed a large spatial difference,with higher yields in the flat area of the middle,lower yields in the undulating southwest with serious soil erosion,and the lowest yield in the eastern region.