Statistical and machine learning methods for crop yield prediction in the context of precision agriculture

Precision Agriculture - It is of critical importance to understand the relationships between crop yield, soil properties and topographic characteristics for agricultural management. This...     

我国不同区域粮食作物产量对有机肥施用的响应差异

探明全国不同气候、土壤和作物类型条件下,化肥配施有机肥对产量的影响及其增产效果的主控因素,可为粮食增产和有机肥的合理施用提供重要的理论依据。本研究通过收集109篇已公开发表的文献,建立了包含不施肥（CK）、单施化肥（NPK）和化肥配施有机肥（NPKM）处理的402组作物产量的数据库。采用整合分析（Meta-analysis）方法分析了不同施肥处理对小麦、玉米和水稻产量的影响,以及不同土壤性质和气候条件下有机肥的增产差异;采用随机森林（Random forest）方法量化土壤和气候因素对有机肥增产效果影响的重要度。全国来看,与NPK相比,NPKM处理下作物的产量平均增幅4.7%,其中小麦、玉米和水稻的增产率分别为5.6%、7.6%和4.5%;作物的平均增产率在西北地区最高,东北和华北地区次之,南方和华东地区较低。配施有机肥对产量的提升作用在我国温带大陆性气候区和温带季风性气候区显著高于亚热带季风性气候区,其中年降雨量是影响小麦和水稻产量对有机肥响应的主要因素,年均温和无霜期是影响玉米产量对有机肥响应的主要因素。有机质和全氮含量是土壤性状中影响有机肥增产效果的主要因素,其中土壤有机质和全氮含量越低,配施有机肥后产量的增幅越高。总的来说,化肥配施有机肥可显著提高作物的产量,尤其在低温少雨、土壤养分含量较低的地区,可通过化肥配施有机肥来进行增产促产。     

Rice yield forecast based on satellite and meteorological data

A methodological approach to forecasting rice yield based on using MODIS satellite data in combination with meteorological information is given. Underlying the approach is modeling the accumulation of rice biomass during growth and development. Rice yield is predicted in the middle of the growing season.     

冬小麦谷粒产量,谷粒蛋白产量,谷粒蛋白含量和稳定性分析

基因型与环境的互作对基因型筛选具有重要作用,并且使品种筛选变得复杂.本文对法国27个冬小麦品种在27个生长环境中获得的试验数据进行了分析比较,同时对高谷粒产量(GY),谷粒蛋白产量(GPY),谷粒蛋白含量(GPC)和稳定性的几种方法进行了筛选:Kang秩和方法(指标1:相同权重的3种谷粒性状和Shukla稳定性方差δ2)以及5种衍生的秩和指标(指标2～6:3种谷粒性状高于δ2 2～6倍的权重),稳定性方差(s2)和回归系数(b),研究它们之间的秩相关性以及它们与3种谷粒性状的相关性.结果表明:3种稳定性统计δ2,s2和b对同时筛选3种谷粒性状指标和稳定性都是非常有用的方法.对于GPY的筛选,指标1秩和方法比稳定性方差s2略保守.指标2比指标1和稳定方差s2更好地筛选到高GY和高GPY.指标3,指标4,指标5和指标6更适合于初始筛选3种谷粒性状.所有的稳定性统计方法都能很好与3种谷粒性状秩相关.稳定性方差s2和回归系数b之间在除了GY之外的3种谷粒性状中秩相关性不高,稳定性方差δ2和稳定性方差s2之间在3种谷粒性状中都具有很强的秩相关性.3种稳定性统计在高产和低产的环境中重复性不高,同样,在2个年份之间的重复性也可以忽略.但是,在随机抽取的4个环境之间的重复性却非常高.     

An integrated framework for software to provide yield data cleaning and estimation of an opportunity index for site-specific crop management

This paper proposes an integrated framework for software that provides yield data cleaning and yield opportunity index (Y _i ) calculation for site-specific crop management (SSCM). The artifacts in many yield data sets, which inevitably occur, can pose a significant effect on the validity of Yi. Automated and standardised yield correction procedures were designed to improve the data quality by removing: (1) unreasonable outliers; (2) distribution outliers (globally and locally); and (3) position errors. The calculation of Yi uses two aspects of crop yield assessment, the magnitude of yield variation and the spatial structure of the variation. The cleaning algorithms were applied to four yield data sets with known integrity issues to demonstrate effectiveness. Approximately 13–20 % of the original yield data were removed, and this resulted in an increased mean yield of 0.13 t/ha (average). The semivariograms of cleaned data were shown to possess smaller nugget values compared with the original data. The opportunity index calculation algorithm was demonstrated on a field with nine seasons of yield data. The results demonstrated that using a ranking of Yi provides a rational, agronomic assessment of the opportunity for SSCM based on the quantity and pattern of production variability displayed in yield data sets. This provides farm managers with a rapid way to assess whether the observed variability deserves further investigation and eventual investment in SSCM operations.     

机器学习算法在森林生长收获预估中的应用

森林生长收获预估是森林经理学的一个重要方向,采用模型技术进行森林生长收获估计是森林经营决策的重要前提。传统的统计模型如线性及非线性回归模型、混合效应模型、分位数回归、度量误差模型等统计方法已被广泛应用于研究林木生长,但这些统计方法在应用时常常需满足一定的统计假设前提,诸如数据独立、正态分布和等方差等。由于森林生长数据的连续观测和层次性,上述假设通常难以满足。近年来随着人工智能技术的发展,机器学习算法为森林生长收获预估提供了一种新的手段,它具有对输入数据的分布形式没有假设前提、能够揭示数据中的隐含结构、预测结果好等优点,但在森林生长收获预估中的应用仍十分有限。文章对分类和回归树、多元自适应样条、bagging回归、增强回归树、随机森林、人工神经网络、支持向量机、K最近邻等方法在森林生长收获预估中的应用、软件及调参等进行了综述,讨论了机器学习方法的优势和挑战,认为机器学习方法在森林生长收获预估方面有很大的潜力,必将得到广泛应用,并和传统统计模型相结合成为生长收获模型发展的一种趋势。      

Strategies for monitoring within-field soybean yield using Sentinel-2 Vis-NIR-SWIR spectral bands and machine learning regression methods

Soybean crop plays an important role in world food production and food security, and agricultural production should be increased accordingly to meet the global food demand. Satellite remote sensing data is considered a promising proxy for monitoring and predicting yield. This research aimed to evaluate strategies for monitoring within-field soybean yield using Sentinel-2 visible, near-infrared and shortwave infrared (Vis/NIR/SWIR) spectral bands and partial least squares regression (PLSR) and support vector regression (SVR) methods. Soybean yield maps (over 500 ha) were recorded by a combine harvester with a yield monitor in 15 fields (3 farms) in Paraná State, southern Brazil. Sentinel-2 images (spectral bands and 8 vegetation indices) across a cropping season were correlated to soybean yield. Information pooled across the cropping season presented better results compared to single images, with best performance of Vis/NIR/SWIR spectral bands under PLSR and SVR. At the grain filling stage, field-, farm- and global-based models were evaluated and presented similar trends compared to leaf-based hyperspectral reflectance collected at the Brazilian National Soybean Research Center. SVR outperformed PLSR, with a strong correlation between observed and predicted yield. For within-field soybean yield mapping, field-based SVR models (developed individually for each field) presented the highest accuracies. The results obtained demonstrate the possibility of developing within-field yield prediction models using Sentinel-2 Vis/NIR/SWIR bands through machine learning methods.

     

An artificial neural network model for estimating Mentha crop biomass yield using Landsat 8 OLI

Yield forecasting is essential for management of the food and agriculture economic growth of a country. Artificial Neural Network (ANN) based models have been used widely to make precise and realistic forecasts, especially for the nonlinear and complicated problems like crop yield prediction, biomass change detection and crop evapo-transpiration examination. In the present study, various parameters viz. spectral bands of Landsat 8 OLI (Operational Land Imager) satellite data and derived spectral indices along with field inventory data were evaluated for Mentha crop biomass estimation using ANN technique of Multilayer Perceptron. The estimated biomass showed a good relationship (R²?=?0.762 and root mean square error (RMSE)?=?2.74 t/ha) with field-measured biomass.

     

A flexible approach to managing variability in grain yield and nitrate leaching at within-field to farm scales

We up-scaled the APSIM simulation model of crop growth, water and nitrogen dynamics to interpret and respond to spatial and temporal variations in soil, season and crop performance and improve yield and decrease nitrate leaching. Grain yields, drainage below the maximum root depth and nitrate leaching are strongly governed by interaction of plant available soil water storage capacity (PAWC), seasonal rainfall and nitrogen supply in the water-limited Mediterranean-type environment of Western Australia (WA). APSIM simulates the interaction of these key system parameters and the robustness of its simulations has been rigorously tested with the results of several field experiments covering a range of soil types and seasonal conditions in WA. We used yield maps, soil and weather data for farms at two locations in WA to determine spatial and temporal patterns of grain yield, drainage below the maximum root depth and nitrate leaching under a range of weather, soil and nitrogen management scenarios. On one farm, we up-scaled APSIM simulations across the whole farm using local weather and fertiliser use data and the average PAWC values of soil type polygons. On a 70 ha field on another farm, we used a linear regression of apparent soil electrical conductivity (EC_a) measured by EM38 against PAWC to transform an EC_a map of the field into a high resolution (5 m grid) PAWC map. We then used regressions of simulated yields, drainage below the maximum root depth and nitrate leaching on PAWC to upscale the APSIM simulations for a range of weather and fertiliser management scenarios. This continuous mapping approach overcame the weakness of the soil polygons approach, which assumed uniformity in soil properties and processes within soil type polygons. It identified areas at greatest financial and environmental risks across the field, which required focused management and simulated their response to management interventions. Splitting nitrogen applications increased simulated wheat yields at all sites across the field and decreased nitrate leaching particularly where the water storage capacity of the soil was small. Low water storage capacity resulted in both low wheat yields and large leaching loss. Another management option to decrease leaching may be to grow perennial vegetation that uses more water and loses less by drainage.Paper from the 5th European Conference on Precision Agriculture (5ECPA), Uppsala, Sweden, 2005     

Non-leguminous winter cover crop and nitrogen rate in relation to double rice grain yield and nitrogen uptake in Dongting Lake Plain,Hunan Province,China

Annual ryegrass(Lolium multiflorum Lam.), a non-leguminous winter cover crop, has been adopted to absorb soil native N to minimize N loss from an intensive double rice cropping system in southern China, but a little is known about its effects on rice grain yield and rice N use efficiency. In this study, effects of ryegrass on double rice yield, N uptake and use efficiency were measured under different fertilizer N rates. A 3-year(2009–2011) field experiment arranged in a split-plot design was undertaken. Main plots were ryegrass(RG) as a winter cover crop and winter fallow(WF) without weed. Subplots were three N treatments for each rice season: 0(N_0), 100(N_(100)) and 200 kg N ha–1(N_(200)). In the 3-year experiment, RG reduced grain yield and plant N uptake for early rice(0.4–1.7 t ha–1 for grain yield and 4.6–20.3 kg ha–1 for N uptake) and double rice(0.6–2.0 t ha–1 for grain yield and 6.3–27.0 kg ha–1 for N uptake) when compared with WF among different N rates. Yield and N uptake decrease due to RG was smaller in N_(100) and N_(200) plots than in N_0 plots. The reduction in early rice grain yield in RG plots was associated with decrease number of panicles. Agronomic N use efficiency and fertilizer N recovery efficiency were higher in RG plots than winter fallow for early rice and double rice among different N rates and experimental years. RG tended to have little effect on grain yield, N uptake, agronomic N use efficiency, and fertilizer N recovery efficiency in the late rice season. These results suggest that ryegrass may reduce grain yield while it improves rice N use efficiency in a double rice cropping system.     

Spatio-temporal analysis of yield and weather data for defining site-specific crop management zones

Precision Agriculture - Understanding yield potential and yield-limiting factors is essential for improving profitability and grain yields while avoiding adverse environmental effects. In the USA,...     

Neural network modeling of greenhouse tomato yield, growth and water use from automated crop monitoring data

The recent development of tools to automatically monitor important crop attributes in situ such as yield, growth and water use offers an opportunity to relate real-time crop status to current environmental conditions. In this study, continuous minute-by-minute measurements of crop yield, growth and water use averaged over weekly, daily, or hourly intervals throughout the growing season were used to determine crop response to changes in the greenhouse environment. The data were obtained from crop monitoring stations established in both commercial and research greenhouses. Crop yield measurements obtained from the monitoring system were generally in very close agreement with yields recorded over a much larger area in the commercial greenhouse. Yield was more closely related (R² = 0.65) to radiation from the previous week than to radiation in the current week (R² = 0.56). In addition, a neural network (NN) model of yield which included radiation as an input was better at predicting yield in the following week (R² = 0.70) than yield in the current week (R² = 0.57). These results indicate a lag effect of radiation on yield. Similarly, yield was more positively related to growth from the previous week (R² = 0.32) than to growth from the current week (R² = 0.17). Neural network models of daily growth at both sites (R² = 0.74, 0.69) included day of the year, temperature and CO₂ as inputs. A negative relationship between day of the year and daily growth indicates a decline in crop vigor through the measurement period. Neural network models of daily crop water use for the two sites were stronger (R² = 0.91, 0.85) than those for growth, highlighting the difference in physiological complexity between the two. A model of canopy water status as affected by environmental conditions was generated using hourly measures of tomato canopy mass change. Although the rate of canopy mass gain through the day was often constant, there were days when the plant experienced periods of reduced mass gain mid-day. On those days, the amount of deviation from a constant rate was positively related to radiation, day temperature and water use, suggesting periods of water stress. With subsequent recovery of mass gain rates late afternoon, these deviations did not affect canopy growth for the day. Overall, automated monitoring provides new information on the crop which may readily be incorporated into models of crop performance.     

Mapping sub-field maize yields in Nebraska,USA by combining remote sensing imagery,crop simulation models,and machine learning

Precision Agriculture - Crop yield maps are valuable for many applications in precision agriculture, but are often inaccessible to growers and researchers wishing to better understand yield...     

哈尔滨市主要农作物籽实、秸秆、根茬产量及其养分含量的分析

通过两年的多点试验表明 ,哈市玉米、水稻、大豆作物的籽实、秸秆、根茬在产量上有一定的比例关系 ,其中平均比例为玉米 1:1.2 4 :0 .2 8,水稻 1:0 .87:0 .38;大豆 1:0 .2 6 :0 .33。养分分配也有差异 ,以玉米为例 ,氮、磷、钾养分在籽实、秸秆、根茬中的含量分别为 1.32 % ,0 .85 % ,0 .71% ;0 .4 0 % ,0 .16 % ,0 .2 4 %和 0 .5 3% ,0 .90 % ,0 .6 0 %。3种养分摄取的总量排序为籽实 >秸杆 >根茬     

An approach for assessing the effects of site-specific fertilization on crop growth and yield of durum wheat in organic agriculture

Precision agriculture (PA) technologies allow us to assess field variability and support site-specific (SSP) application of inputs. The joint application of PA and organic farming practices might be synergetic. The objective of this 3-year study was to propose a multivariate statistical and geostatistical approach, to evaluate the effects of SSP nitrogen (N) fertilization on durum wheat in transition to organic farming. Soil parameters were measured to assess soil fertility level before the SSP fertilization on wheat, which was carried out by management zones in the third year. Radiometric measurements were performed with a hyperspectral spectroradiometer and N-uptake at anthesis and grain yield were determined. The expected values and 95 % confidence intervals of the soil parameters, N-uptake and yield data were estimated with polygon kriging for each management zone. Reflectance data were reduced through principal component analysis and the retained principal components were submitted to factorial co-kriging analysis to estimate orthogonal scale-dependent factors. Comparisons between N-uptake and yield and between the retained regionalized factors (F1) and yield were performed. The spatial pattern of F1 at shorter scales was mostly reproduced in the N-uptake map, suggesting the predictive capacity of hyperspectral data for crop N-status. Within-cluster variance for yield was reduced, quite probably as a combined effect of meteorological pattern and management. The preliminary results seem to be promising in the perspective of PA. Moreover, an inverse relationship between grain yield and crop N-status was observed.     

基于机器学习方法的母猪高低产分类模型研究

为帮助猪场管理者更好地对母猪进行繁殖管理、预测母猪的高低产、及时淘汰低产母猪,收集和整理包含出生场地、分娩栏位、品种和不同胎次、初生窝重信息的3个母猪群体的生产数据集,制定母猪高低产的分类标准,使用R软件中的Boruta包筛选出影响母猪高低产的重要特征,使用4种不同的机器学习方法——逻辑回归(logistic regression, LOG)、决策树(decision tree, DT)、随机森林(random forest, RF)和支持向量机(support vector machine, SVM)构建母猪高低产的分类模型,并进行决策树视图分析探究影响母猪最高产的相关因素。结果显示:4种机器学习方法构建母猪高产分类模型的分类准确率均在71%左右,最高可达84%,并且发现SVM作为最佳建模方法在所有数据集和不同分类标准下出现的频率最高,其次是LOG和DT。决策树视图显示出生场地、品种和初生窝重是划分最高产母猪的重要叶节点,利用这些特征预测最高产母猪准确率可达73%～82%。以上结果表明在未来的养猪生产中,利用机器学习方法实现母猪高低产的早期预测将会是一个不错的选择。     

Substituting nitrogen and phosphorus fertilizer with optimal amount of crop straw improves rice grain yield,nutrient use efficiency and soil carbon sequestration

Crop straw return after harvest is considered an important way to achieve both agronomic and environmental benefits.  However, the appropriate amount of straw to substitute for fertilizer remains unclear.  A field experiment was performed from 2016 to 2018 to explore the effect of different amounts of straw to substitute for fertilizer on soil properties, soil organic carbon (SOC) storage, grain yield, yield components, nitrogen (N) use efficiency, phosphorus (P) use efficiency, N surplus, and P surplus after rice harvesting.  Relative to mineral fertilization alone, straw substitution at 5 t ha^–1 improved the number of spikelets per panicle, effective panicle, seed setting rate, 1 000-grain weight, and grain yield, and also increased the aboveground N and P uptake in rice.  Straw substitution exceeding 2.5 t ha^–1 increased the soil available N, P, and K concentrations as compared with mineral fertilization, and different amounts of straw substitution improved SOC storage compared with mineral fertilization.  Furthermore, straw substitution at 5 t ha^–1 decreased the N surplus and P surplus by up to 68.3 and 28.9%, respectively, compared to mineral fertilization.  Rice aboveground N and P uptake and soil properties together contributed 19.3% to the variation in rice grain yield and yield components.  Straw substitution at 5 t ha^–1, an optimal fertilization regime, improved soil properties, SOC storage, grain yield, yield components, N use efficiency (NUE), and P use efficiency (PUE) while simultaneously decreasing the risk of environmental contamination.     

Continuous applications of biochar to rice: Effects on grain yield and yield attributes

Biochar is considered as a beneficial soil amendment for crop production. However, limited information is available on the effects of continuous applications of biochar on rice. In this study, a fixed field experiment was conducted in the early and late rice-growing seasons from 2015 to 2017. Grain yield and yield attributes with a widely-grown rice cultivar Zhongzao 39 were compared, with and without applications of biochar in each season. The results showed that grain yield initially decreased with biochar applications in the first three seasons due to decreases in grain weight and harvest index. Although there were further relative decreases in grain weight and harvest index for rice that was supplied with biochar in the fourth to sixth seasons, grain yield was increased(by 4–10%) because of increases in sink size(spikelets per m2) and total biomass. The increased sink size in rice whose soil had been supplied with biochar in the fourth to sixth seasons was achieved by increasing panicle size(spikelets per panicle) or number of panicles, or both. Our study suggests that the positive effects of biochar application on rice yield and yield attributes depend on the duration of biochar application. Further investigations are needed to determine what are the soil and physiological processes for producing yield responses associated with ongoing applications of biochar. Also, it should be evaluated the performance of biochar application combined with other management practices, especially those can increase the grain weight and harvest index in rice production.     

开荒作物对后茬作物产量、杂草群落及土壤理化性质的影响

【目的】探讨传统农业中利用芝麻作为开荒作物的作用机理。【方法】采用大田种植方式,在南方红壤新垦荒地以芝麻作为先锋作物,大豆和玉米作为对照,调查不同作物对后茬作物玉豆和萝卜产量、土壤理化性质以及田间杂草的影响。【结果】新垦荒地种植芝麻对后茬玉豆和萝卜单株产量增产率分别为44.25%和148.48%;对土壤p H、有机碳和速效磷含量有增加效应,使土壤碱解氮和速效钾分别减少了89.93%和47.95%;种植芝麻后未发现原生境中的杂草粗耳草,但出现优势杂草凹头苋和三叶鬼针草;杂草种类和生物量受后茬作物种类的影响差异较大,其中芝麻-自然长草处理杂草种类为5.67种,与芝麻-萝卜处理11.67种和芝麻-玉豆处理10种相比差异达显著水平,杂草的地上部净质量为萝卜(1 639.13 g·m~(-2))玉豆(976.89 g·m~(-2))自然长草(944.13 g·m~(-2));芝麻-长草与长草-长草处理相比,杂草种类由4种增加到5.67种,但杂草生物量由2 698.6 g·m~(-2)减少到944.13 g·m~(-2)。【结论】芝麻可以提高后茬作物产量,对杂草的生长有抑制作用,适宜作为开荒作物。芝麻凋落物和残茬易腐解,导致土壤有机质和速效磷增加,这可能是后茬作物增产的主要原因。