中圖分類(lèi)號(hào):S667.1 文獻(xiàn)標(biāo)志碼:A 文章編號(hào):1009-9980(2025)05-1045-12
Abstract: 【Objective】 China is the origin country of litchi (Litchi chinensis Sonn.)and the largest producer in the world. The low or unstable yield caused by unstable flowering is a prominent problem in litchi production,and the flowering time afects not only the maturity offruit,but also the flowering rate and yieldof litchi.The meteorological factors including air temperature,relative air humidity,rainfall, and wind level,and other factors including variety and tree age affect flower diferentiation of litchi. However, there is a lack of systematic research on how the development stage of litchi flowers is affected by the meteorological factors. Accurately predicting the development of the inflorescence and the process of flowering duration,as wellas correctly understanding the quantitative relationship between the flowering phenology and the meteorological factors, is very important for the high-yield and quality production of litchi.The machine learning algorithms can handle high-dimensional nonlinear data with complex interactions,outperform traditional statistical models in ecology,and have been effectively used for plant classification, phenology detection, crop growth detection,and yield prediction. The objective of this study was to develop regression models for litchi inflorescence development duration and flowering duration using machine learning algorithms including RF and STR and to analyze and assess the importance and relevance of selected features on the flowering duration according to RF algorithm in order to provide a theoretical basis for the prediction of litchi flowering period and realizing precise regulation.【Methods】Firstly, the litchi phenological period data were obtained from the National Litchi and Longan Industry Technology System (CARS), with a total of 2204 records. It covered 201 demonstration litchi orchards distributed in 53 cities and counties in Hainan Province, Guangdong Province, Guangxi Zhuang Autonomous Region, Fujian Province and Sichuan Province.The time span was 2009—2018, including 47 varieties such as Guiwei,Nuomici,Huaizhi,F(xiàn)eizixiao,etcs.The meteorological data were downloaded from the website“https:/tianqi.911cha.com/”and recorded at a frequency of one hour, with meteorological factors including atmospheric temperature,atmospheric relative humidity, wind scale and rainfall.Feature engineering of the data,which involved removing irelevant or redundant features and ensuring that there was no high correlation between the retained features,was used to improve the performance and generalization of the model. The six classical machine algorithms including Classified Regression Tree (CART), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Random Forest (RF), Stepwise Regression (STR) and Gradient Boosting Machine (GBM) were used for training. The algorithms (RF and STR) with the smallest Mean Absolute Error (MAE)and the highest residual error (RMSE) and the highest correlation coefficient (RP2) )were selected for further parameter optimization and evaluation.A 5-fold cross-validation with 999 repetitions was performed on alltrained machine learning models. The random seeds are set during resampling, parameter tuning and model training to ensure model reproducibility.The models were applied to be constructed in R-project (version 3.5.2) and the‘caret’package was applied to tune the machine learning algorithm parameters. 【Results】 The residual error of the model were 3.6-3.7 days,and the correlation coefficient were 0.97, so the models had high reliability; The model was further verified with blind test data setof two-year’s phenological ecological characteristics,and the correlation coefficient was between 0.98-0.99.It was indicated that the series of prediction models could be applicable to accurately predict the development of inflorescence. Similarly, the residual error of the model predicted the shedding period were (204號(hào) days,and the correlation coefficient were 0.88-0.97 , so the model had high reliability;The model was further verified with blind test data set of two year’s phenological ecological characteristics,and the correlation coefficient was between 0.96-0.98 , Indicating that the series of prediction models could be applicable to accurately predict the flowering duration. The daily accumulated temperature above 5°C daily average temperature, wind level and rainfall were found to played an important role in the whole process of the florescence period of litchi. In addition, a daily accumulated temperature above 24°C had great impact on the development of inflorescence, while daily a cumulated temperature above 18°C (204號(hào) had significant efect on the flowering duration.【Conclusion】 The robustness and predictive fit of the regression model established in this study were high.After the verification of two year's data,the accuracy and stability of the prediction were ideal. These models would be important to judge and regulate the maturity period and market volume of litchi.And the characteristic features screened out were helpful to understand the complex influence of external meteorological factors on the flowering process.
Key words: Machine learning; Prediction model; Phenological phase; Inflorence development duration;Blooming duration
荔枝(LitchichinensisSonn.)是原產(chǎn)于中國(guó)南部和東南亞的重要亞熱帶常綠果樹(shù),在世界范圍內(nèi)主要分布在南北緯 17°~32° 。成花不穩(wěn)定而導(dǎo)致的低產(chǎn)或不穩(wěn)產(chǎn)是荔枝生產(chǎn)中的突出問(wèn)題[1-2],而開(kāi)花的早晚既影響果實(shí)成熟期,也影響荔枝成花率和產(chǎn)量3]。影響荔枝花芽分化的主要因子包括枝梢狀態(tài)和氣象因子如氣溫、水分等4,但氣象因子與花期的定量關(guān)系尚未得到系統(tǒng)研究。準(zhǔn)確預(yù)測(cè)花穗發(fā)育以及開(kāi)花持續(xù)物候進(jìn)程,正確理解開(kāi)花物候與氣象因子之間的定量關(guān)系,對(duì)荔枝高產(chǎn)優(yōu)質(zhì)種植有著重要意義。
氣象因子對(duì)荔枝花發(fā)育階段的影響研究主要局限于氣溫4與土壤相對(duì)濕度。氣溫、空氣相對(duì)濕度、風(fēng)力、降雨量、天氣等如何綜合影響荔枝花發(fā)育持續(xù)期,對(duì)其重要影響因子的挖掘還未見(jiàn)報(bào)道。物候模型[1-是研究生態(tài)系統(tǒng)對(duì)氣候變化響應(yīng)的重要工具。但物候與環(huán)境因子之間的非線(xiàn)性關(guān)系使得現(xiàn)有物候模型往往難以獲得較高的模擬精度。
機(jī)器學(xué)習(xí)算法可以處理具有復(fù)雜交互作用的高維非線(xiàn)性數(shù)據(jù),在生態(tài)學(xué)方面的表現(xiàn)優(yōu)于傳統(tǒng)的統(tǒng)計(jì)模型[12],已經(jīng)有效應(yīng)用于物候檢測(cè)[13]、作物生長(zhǎng)檢測(cè)[14]和產(chǎn)量預(yù)測(cè)[15-i]。隨機(jī)森林算法(RandomFor-est,RF)[能夠模擬輸入特征之間復(fù)雜的相互作用和非線(xiàn)性關(guān)系,逐步回歸算法(StepwiseRegression,STR)[8可以自動(dòng)選擇回歸建立的重要特征,并評(píng)估參數(shù)和獨(dú)立特征之間的相關(guān)系數(shù),并且由于其魯棒性和易用性而被廣泛應(yīng)用。
針對(duì)荔枝開(kāi)花期預(yù)測(cè)的空白,筆者在本研究中從荔枝現(xiàn)“白點(diǎn)”日、初花日至謝花日物候、大氣溫濕度變化、降雨量、風(fēng)力、一定溫度以下或以上的冷熱積累量等建立特征集,并首次利用隨機(jī)森林(RF)和逐步回歸(STR)算法構(gòu)建荔枝花穗發(fā)育和開(kāi)花持續(xù)期雙階段預(yù)測(cè)模型體系,通過(guò)機(jī)器學(xué)習(xí)模型量化氣象因子(如積溫、風(fēng)力)對(duì)荔枝花期的線(xiàn)性和非線(xiàn)性影響,為荔枝花期物候期預(yù)測(cè)、生理機(jī)制研究和精確調(diào)控提供數(shù)據(jù)支持。
1 材料和方法
1.1數(shù)據(jù)收集
荔枝物候期數(shù)據(jù)來(lái)自國(guó)家荔枝龍眼產(chǎn)業(yè)技術(shù)體系,共2204條記錄。覆蓋201個(gè)示范荔枝園,分布于海南?。?02條)、廣東?。?52條)廣西壯族自治區(qū)(562條)、福建?。?72條)、四川省(116條)等的53個(gè)縣市區(qū)。時(shí)間跨度2009一2018年,包括桂味(380條)、糯米糍(306條)懷枝(110條)、妃子笑(798條)等47個(gè)主栽或地方特色品種。示范園采取常規(guī)的栽培管理方式,成花與坐果正常。
開(kāi)花物候包括現(xiàn)“白點(diǎn)”(即“花序出現(xiàn)”日、初花日和謝花日。記載標(biāo)準(zhǔn)為,“白點(diǎn)”日:指全株 25% 或以上的枝梢頂芽鱗片完全打開(kāi)后露出內(nèi)部白色茸毛體的日期;初花日:全樹(shù) 25% 花序上的小花開(kāi)放的日期;謝花日:全樹(shù) 95% 花序上的小花開(kāi)放后凋謝的日期。
示范荔枝園所在縣區(qū)開(kāi)花期間的氣象數(shù)據(jù)從\"https://tianqi.911cha.com/\"網(wǎng)站下載,記錄頻率為 1h .氣象因子包括大氣溫度、大氣相對(duì)濕度、風(fēng)力和降雨量。
1.2 植株生長(zhǎng)發(fā)育特征建立
生長(zhǎng)發(fā)育特征用來(lái)描述植物的生長(zhǎng)狀況,包括樹(shù)齡(age)、現(xiàn)“白點(diǎn)”日(initialdate,ID)、初花日(earlybloomingdate,EBD)、謝花日(terminalbloom-ingdate,TBD)、花穗發(fā)育持續(xù)期(inflorencedevelopmentduration,IDD)、開(kāi)花持續(xù)期(bloomingdura-tion,BD)。
1.3 氣象特征建立
針對(duì)每個(gè)果園觀察物候期植株分別提取花穗發(fā)育持續(xù)期和開(kāi)花持續(xù)期的氣象數(shù)據(jù),以5d為時(shí)間尺度計(jì)算每天氣象數(shù)據(jù)的滑動(dòng)平均大氣溫度(meantemperature,MT)、平均空氣相對(duì)濕度(meanrelativehumidity,MRH)、平均風(fēng)力(meanwindscale,MWS)和平均降雨量(mean precipitation,MP)。以此基礎(chǔ)上,參考前期工作使用了一系列候選基準(zhǔn)溫度,從5~35°C 區(qū)間,以 1°C 為間隔設(shè)置一系列候選基礎(chǔ)溫度,計(jì)算成花期間每天高于基礎(chǔ)溫度[meanthermalaccumulation(5~35),MTA(5~35)]的熱量總和作為衡量開(kāi)花有效積溫的特征。熱量積累量由公式( 1~ 2)給出:
其中, Athrmi 是5d時(shí)間尺度下的冷量。 Tbi 是導(dǎo)期的基礎(chǔ)溫度。 T(i) 是某個(gè)時(shí)間的測(cè)量溫度。
Athrmt 是某個(gè)時(shí)間尺度成花誘導(dǎo)期熱量的總和。
以上數(shù)據(jù)合并后,獲得針對(duì)因變量花穗發(fā)育持續(xù)期的39個(gè)特征和1102條記錄、針對(duì)因變量開(kāi)花持續(xù)期的41個(gè)特征和1102條記錄。
1.4 特征篩選
首先刪除恒定值、超過(guò) 50% 等于0或方差 ?0.05 的噪聲特征。對(duì)高相關(guān)性特征 (|r|gt;0.95 保留方差較大者,以降低數(shù)據(jù)過(guò)擬合風(fēng)險(xiǎn)。
計(jì)算以皮爾遜相關(guān)系數(shù)Pearson( CORRp ,公式3)和斯皮爾曼相關(guān)系數(shù)Spearman( CORRsp ,公式4)表示的開(kāi)花物候持續(xù)時(shí)間和氣象數(shù)據(jù)的相關(guān)矩陣:
VARrEx 和 VARrgy 是秩變量的標(biāo)準(zhǔn)差,而cov( rgx-rgy )是秩變量的協(xié)方差。在這項(xiàng)工作中, Lsig 被設(shè)置為0.05。
最終針對(duì)因變量花穗發(fā)育持續(xù)期和開(kāi)花持續(xù)期的特征分別為8個(gè)和10個(gè)(表1)。
1.5 預(yù)測(cè)模型建立
首先綜合考慮常用經(jīng)典機(jī)器算法(及其特點(diǎn))進(jìn)行預(yù)訓(xùn)練:ClassifiedRegression Tree,CART(可解釋性)、K-NearestNeighbor,KNN(局部特征敏感)、SupportVectorMachineSVM(高維數(shù)據(jù))、RandomForest,RF(抗過(guò)擬合)、StepwiseRegression,STR(線(xiàn)性關(guān)系)、GradientBoostingMachine,GBM(預(yù)測(cè)性能優(yōu)化),篩選表現(xiàn)較優(yōu)的模型做進(jìn)一步參數(shù)優(yōu)化和評(píng)估。選擇平均絕對(duì)誤差(預(yù)測(cè)值與實(shí)際值之間絕對(duì)差值的平均值,meanabsoluteerror,MAE)和均方根誤差(預(yù)測(cè)值與實(shí)際值之間差值的平方的平均值的平方根,root mean squared error,RMSE)最小而相關(guān)系數(shù) (RP2) 最大的算法(RF和STR)建立回歸預(yù)測(cè)模型。在重采樣、參數(shù)整定和模型訓(xùn)練時(shí)設(shè)置隨機(jī)種子,以保證模型的重現(xiàn)性。對(duì)所訓(xùn)練的機(jī)器學(xué)習(xí)模型都進(jìn)行了5-fold交叉驗(yàn)證,999次重復(fù)。模型應(yīng)用R-project(3.5.2版本)構(gòu)建,應(yīng)用‘caret'包(Kuhn,2008)調(diào)整機(jī)器學(xué)習(xí)算法參數(shù)。
2 結(jié)果與分析
2.1 開(kāi)花物候和氣候分析
荔枝現(xiàn)“白點(diǎn)”日(圖1-A)、初花日(圖1-B)和謝花日(圖1-C)的年積日均接近正態(tài)分布,現(xiàn)“白點(diǎn)”日從年前第71天開(kāi)始至第102天結(jié)束,中位數(shù)為19天;初花日自第15天至第135天,中位數(shù)為85天;謝花日自第37天持續(xù)至第155天,中位數(shù)為103天?,F(xiàn)“白點(diǎn)”日至初花日為花穗發(fā)育持續(xù)時(shí)間,主要集中在50\~75d,平均值為63天(圖1-D);初花日至謝花日為開(kāi)花持續(xù)時(shí)間,主要集中在 10~20d ,平均值為16天(圖1-E),指標(biāo)在年間差異比較小。
將開(kāi)花的完整過(guò)程分為自現(xiàn)“白點(diǎn)\"至初花日和初花日至謝花日。從圖2-A可見(jiàn),自現(xiàn)“白點(diǎn)\"至初花日平均氣溫為 14.72°C ,在 13°C 附近分布最集中;初花日至謝花日平均氣溫為 18.13°C ,密集分布于上下四分位。從圖2-B可見(jiàn),自現(xiàn)“白點(diǎn)\"至初花日平均大氣濕度為 56.61% ,在 47% 附近分布最為集中;初花日至謝花日平均大氣濕度為 69.67% ,接近正態(tài)分布。自現(xiàn)“白點(diǎn)\"至初花日平均降雨量為0.06mm ,在初花日前大部分低于 0.06mm ;初花日至謝花日平均降雨量為 0.21mm ,比自現(xiàn)“白點(diǎn)\"至初花日多1倍,主要集中分布在下四分位 0.30mm 附近(圖2-C)。自現(xiàn)“白點(diǎn)\"至初花日平均風(fēng)力為6.14級(jí);初花日至謝花日平均風(fēng)力6.56級(jí),密集分布于上下四分位(圖2-D)。
2.2 特征相關(guān)性評(píng)價(jià)
每個(gè)特征與因變量的 CORRp 和 CORRsp 較為一致?;ㄋ氚l(fā)育持續(xù)期和開(kāi)花持續(xù)期均與大于 5°C 積溫呈強(qiáng)線(xiàn)性正相關(guān),與平均降雨量呈中強(qiáng)單調(diào)正相關(guān),與平均大氣溫度、平均空氣相對(duì)濕度和平均風(fēng)力相關(guān)性較低;與樹(shù)齡幾乎沒(méi)有線(xiàn)性或單調(diào)相關(guān)性(圖3)。
此外,花穗發(fā)育持續(xù)期與現(xiàn)“白點(diǎn)”日呈中強(qiáng)線(xiàn)性負(fù)相關(guān)(圖3-A),而開(kāi)花持續(xù)期與現(xiàn)“白點(diǎn)”日相關(guān)性較低,與大于 18°C 積溫呈中強(qiáng)單調(diào)正相關(guān)(圖3-B)。可見(jiàn)影響花穗發(fā)育持續(xù)期和開(kāi)花持續(xù)期的因素較多,大于 5°C 積溫越多,持續(xù)期越長(zhǎng),但單個(gè)氣象因子平均值、上一個(gè)物候出現(xiàn)的時(shí)間等特征與因變量線(xiàn)性相關(guān)度較低,說(shuō)明以上特征的簡(jiǎn)單相關(guān)分析并不能很好地解釋因變量的特征,機(jī)器學(xué)習(xí)算法則有望為這一復(fù)雜問(wèn)題提供解決方案。
2.3顯式和隱式預(yù)測(cè)模型的建立與評(píng)價(jià)
從圖4可見(jiàn),對(duì)于花穗發(fā)育持續(xù)期與開(kāi)花持續(xù)期數(shù)據(jù)集,MAE和RMSE最小的是RF、GMS、SVM和STR模型,其 R2 相應(yīng)較高,前四位模型中SVM模型的RMSE和 R2 波動(dòng)較大,說(shuō)明其準(zhǔn)確性和穩(wěn)定性較差。因此,分別選擇RF和STR進(jìn)一步建立顯式和隱式模型,并優(yōu)化模型參數(shù)。以2009—2016年數(shù)據(jù)建立模型,2009—2016年數(shù)據(jù)按7:3比例隨機(jī)分為訓(xùn)練集(trainingdataset)和測(cè)試集(validationdataset),分別建立RF和STR模型。以2017年以及2018年數(shù)據(jù)為盲測(cè)集(blindtestdataset)。
對(duì)于花穗發(fā)育持續(xù)期,最終確定RF模型參數(shù)為mtry -8 ,ntree =4000 ,STR模型最終確定參數(shù)nvmax 3;對(duì)于開(kāi)花持續(xù)期,最終確定RF模型參數(shù)為mtry
ntree=4000,STR模型最終確定參數(shù)nvmax :=3 。以上參數(shù)下預(yù)測(cè)模型RMSE最小,錯(cuò)誤率較低而且較穩(wěn)定,降低了數(shù)據(jù)依賴(lài)性,確定為模型的最佳參數(shù)。
假設(shè)兩因變量與最重要的性狀呈線(xiàn)性關(guān)系,從而獲得顯式模式STR預(yù)測(cè)的線(xiàn)性方程:
BD=8.73-17.67×MT+2.09 MWS-23.14× MAT18+78.29×MAT5. 0 (6)
從以上方程可見(jiàn),平均空氣相對(duì)濕度、大于 5°C 積溫與花穗發(fā)育持續(xù)期呈正相關(guān),而平均大氣溫度與花穗發(fā)育持續(xù)期呈負(fù)相關(guān)。平均風(fēng)力、大于 5°C 積溫與開(kāi)花持續(xù)期呈正相關(guān),而平均大氣溫度、大于18°C 積溫與開(kāi)花持續(xù)期呈負(fù)相關(guān)。
2.4特征重要性評(píng)價(jià)
用RF算法?;诓患兌冉档蛯?duì)特征重要性進(jìn)行排序,結(jié)合單個(gè)特征與 CORRp 和 CORRsp 表示的花穗發(fā)育持續(xù)期和開(kāi)花持續(xù)期的相關(guān)性來(lái)評(píng)估特征的重要性。對(duì)于每個(gè)特征攜帶的信息,RF算法的重要性提供非線(xiàn)性評(píng)估,而兩個(gè)相關(guān)系數(shù)分別給出線(xiàn)性和單調(diào)評(píng)估。
從圖7-A可以看出,大于 5°C 積溫對(duì)花穗發(fā)育持續(xù)期模型變量重要性、線(xiàn)性和單調(diào)相關(guān)性都最高;平均大氣溫度對(duì)花穗發(fā)育持續(xù)期的非線(xiàn)性相關(guān)性要高于線(xiàn)性和單調(diào)相關(guān);平均風(fēng)力與花穗發(fā)育持續(xù)期呈負(fù)線(xiàn)性相關(guān),變量重要性較高,高于平均降雨量和平均空氣相對(duì)濕度,現(xiàn)“白點(diǎn)”日對(duì)花穗發(fā)育持續(xù)期模型的重要性則較低。從圖7-B可以看出,對(duì)開(kāi)花持續(xù)期模型變量重要性前4位的特征,它們與因變量的線(xiàn)性相關(guān)性和單調(diào)相關(guān)性表現(xiàn)與花穗發(fā)育持續(xù)期模型相同。而初花日、現(xiàn)“白點(diǎn)”日和花穗發(fā)育持續(xù)期對(duì)開(kāi)花持續(xù)期模型的重要性則較低。另外,大于 18°C 積溫對(duì)開(kāi)花持續(xù)期模型的影響要高于大于24°C 積溫對(duì)花穗發(fā)育持續(xù)期模型的影響,可見(jiàn),花發(fā)育后期對(duì)高溫的敏感性要強(qiáng)于花發(fā)育早期。
3討論
3.1建立物候與氣象數(shù)據(jù)庫(kù)的意義
地面觀測(cè)是一種傳統(tǒng)的物候?qū)W研究方法[19-20],,可以準(zhǔn)確地記錄特定地點(diǎn)和物種的物候時(shí)間,提供物候變化的第一手直接證據(jù)。近年來(lái)衛(wèi)星遙感通過(guò)檢測(cè)與綠色相關(guān)的植被指數(shù)2、葉綠素?zé)晒鈁22或高頻采集圖像[23]等手段拓展了傳統(tǒng)植物物候觀測(cè)的視野,但這些方法仍存在空間分辨率和圖像解釋準(zhǔn)確率低等缺陷,因此地面觀測(cè)獲得物候期數(shù)據(jù)仍然是常綠果樹(shù)特別是荔枝物候研究最可靠的手段。國(guó)家荔枝龍眼產(chǎn)業(yè)體系自2009年成立以來(lái),在廣東、廣西、海南、福建、云南和四川6個(gè)省區(qū)的荔枝產(chǎn)區(qū)設(shè)立荔枝物候期觀察點(diǎn),統(tǒng)一制定物候期指標(biāo)和判斷標(biāo)準(zhǔn)。自觀測(cè)體系運(yùn)行以來(lái),已系統(tǒng)記錄超過(guò)1500條數(shù)據(jù),為挖掘數(shù)據(jù)關(guān)系和研究荔枝物候打下了重要基礎(chǔ)。
3.2模型評(píng)價(jià)及特征的意義分析
物候模型是研究植物物候?qū)ξ磥?lái)氣候變化響應(yīng)的重要工具[0-],多數(shù)物候模型基于“度-日\(chéng)"概念,只關(guān)注特定時(shí)期內(nèi)溫度總和,忽略了溫度的時(shí)間變化,此外空氣濕度、降雨量等可能對(duì)植物物候產(chǎn)生重要影響的氣象因子尚未很好地嵌入到現(xiàn)有的物候模型中,使得極端氣候條件或全球變暖趨勢(shì)下統(tǒng)計(jì)模型可能會(huì)導(dǎo)致相當(dāng)大的偏差[24。而且物候期與荔枝的成花率以及產(chǎn)量關(guān)系密切。筆者課題組在廣州地區(qū)對(duì)荔枝的4個(gè)品種一一妃子笑、懷枝、桂味和糯米糍進(jìn)行了系統(tǒng)的農(nóng)業(yè)氣象研究,發(fā)現(xiàn)調(diào)控這些品種的末次秋稍成熟期在10月16日至10月31日之間,以及確保在1月15日前出現(xiàn)花芽分化“白點(diǎn)”,可以在多變的農(nóng)業(yè)氣象條件下實(shí)現(xiàn)高成花率和高產(chǎn)出]。
溫度通常被認(rèn)為是植物物候的主要控制因素[10,25]。一般認(rèn)為春季晝間 20°C 以上、夜間低至10°C 以下有利于荔枝開(kāi)花,但物候事件對(duì)溫度變化的反應(yīng)在很大程度上是非線(xiàn)性的2。降水已被認(rèn)為是干旱半干旱地區(qū)調(diào)節(jié)植物物候的主要因子,荔枝在土壤保持干旱的情況下并不能開(kāi)花??諝鉂穸瓤赡苡绊懼参锏拇杭疚锖?,因物種不同有所差異[。本研究結(jié)果表明,風(fēng)力超過(guò)5級(jí)時(shí)會(huì)對(duì)果樹(shù)開(kāi)花坐果造成危害,微風(fēng)濕潤(rùn)天氣可能促進(jìn)授粉,有待結(jié)合田間試驗(yàn)驗(yàn)證。大于 5°C 積溫以及開(kāi)花過(guò)程中的平均氣溫對(duì)花穗發(fā)育持續(xù)時(shí)間影響最大。大于 24°C 積溫加速花芽分化,但持續(xù)高溫可能降低成花率,與前人的報(bào)道一致4。此外,平均溫度、平均風(fēng)力和平均降雨量對(duì)花發(fā)育早期模型與花發(fā)育后期模型都有重要影響,而在花發(fā)育后期對(duì)高溫的敏感性要強(qiáng)于花發(fā)育早期。
筆者在本研究中聚焦于主效應(yīng),但隨機(jī)森林算法通過(guò)節(jié)點(diǎn)分裂亦能自動(dòng)捕捉特征交互作用,特征重要性排序也間接反映了交互效應(yīng)。通過(guò)隨機(jī)森林評(píng)估了特征重要性,發(fā)現(xiàn)非線(xiàn)性效應(yīng)(如平均大氣溫度對(duì)花穗發(fā)育持續(xù)期的非線(xiàn)性影響)、積溫(大于5°C 積溫)與溫度(平均大氣溫度)的耦合通過(guò)RF模型的不純度降低體現(xiàn)。STR模型的關(guān)系式也反映了各特征之間的數(shù)量關(guān)系。后續(xù)可結(jié)合因果推斷方法(如貝葉斯網(wǎng)絡(luò))和結(jié)合SHAP值等解釋性方法進(jìn)行深入分析。
3.3研究結(jié)果的意義、作用與應(yīng)用
自2014年以來(lái)中國(guó)荔枝種植面積超過(guò)3.6萬(wàn) hm2 年產(chǎn)量均超過(guò)200萬(wàn)t,種植環(huán)節(jié)直接產(chǎn)值達(dá)280多億元[28],但接近一半荔枝集中在6月份成熟上市[29],產(chǎn)期重疊給各產(chǎn)區(qū)帶來(lái)很大的鮮銷(xiāo)壓力。對(duì)各荔枝主產(chǎn)區(qū)和品種成熟時(shí)間進(jìn)行預(yù)測(cè)判斷,可以為各地提早制定銷(xiāo)售預(yù)案,減少市場(chǎng)風(fēng)險(xiǎn),對(duì)產(chǎn)業(yè)穩(wěn)定發(fā)展有著重要意義?;ㄋ氚l(fā)育至開(kāi)花最終到謝花持續(xù)時(shí)間的長(zhǎng)短影響果實(shí)成熟的早晚和產(chǎn)量。建立穩(wěn)定可靠的花發(fā)育預(yù)測(cè)模型,可以加深對(duì)荔枝開(kāi)花進(jìn)程的認(rèn)識(shí),在預(yù)測(cè)天氣變化的基礎(chǔ)上形成有針對(duì)性地調(diào)控花期和果實(shí)成熟期的農(nóng)藝措施,有較高的應(yīng)用價(jià)值。
4結(jié)論
建立了針對(duì)花穗發(fā)育持續(xù)期和開(kāi)花持續(xù)期的RF和STR回歸模型?;?倍交叉驗(yàn)證和999次魯棒性測(cè)試,對(duì)花穗發(fā)育持續(xù)期預(yù)測(cè)均方根誤差為3.6~3.7d(R2≥0.97) ,盲測(cè) R2?0.96 ;開(kāi)花持續(xù)期均方根誤差 1.2~2.6d(R2≥0.88) ,盲測(cè) R2?0.96 ,說(shuō)明該系列模型有較高的準(zhǔn)確性和穩(wěn)健性;大于 5°C 日積溫、日平均溫度、風(fēng)級(jí)以及降雨量對(duì)荔枝花的整個(gè)開(kāi)放過(guò)程起著重要作用,大于 24°C 積溫對(duì)花穗發(fā)育影響顯著,可為花期調(diào)控提供決策依據(jù)。
致謝:研究工作得到國(guó)家荔枝龍眼產(chǎn)業(yè)技術(shù)體系海口、儋州、湛江、茂名、深圳、玉林、欽州、北海、漳州、寧德、瀘州、保山等綜合試驗(yàn)站及荔枝示范園園主幫助,特此致謝!
參考文獻(xiàn)References:
[1] 李建國(guó).荔枝學(xué)[M].北京:中國(guó)農(nóng)業(yè)出版社,2008:44. LIJianguo.The litchi[M].Beijing:China Agriculture Press, 2008:44.
[2] MENZEL CM.Thecontrolof floral initiationin lychee:Areview[J].ScientiaHorticulturae,1983,21(3):201-215.
[3] 蘇鉆賢,楊勝男,黃悅,萬(wàn)志遠(yuǎn),申濟(jì)源,陳厚彬.荔枝成花、坐 果與現(xiàn)\"白點(diǎn)”期和末次秋梢期成熟期的關(guān)系研究[J].果樹(shù)學(xué) 報(bào),2023,40(8):1628-1639. SU Zuanxian,YANG Shengnan,HUANG Yue,WAN Zhiyuan, SHEN Jiyuan,CHEN Houbin. Relationship between flowering rate and fruit set,and the dates of“white millete”appearance and last autumn shoot flush maturation in litchi[J]. Journal of Fruit Science,2023,40(8):1628-1639.
[4] 陳厚彬,蘇鉆賢,張榮,張紅娜,丁峰,周碧燕.荔枝花芽分化研 究進(jìn)展[J].中國(guó)農(nóng)業(yè)科學(xué),2014,47(9):1774-1783. CHEN Houbin,SU Zuanxian,ZHANG Rong,ZHANG Hongna, DING Feng,ZHOU Biyan.Progresses in research of litchi floral differentiation[J]. Scientia Agricultura Sinica,2014,47(9):1774- 1783.
[5] SUZX,LIULY,LIYQ,CHENHB.Predicting flowerinductionof litchi (Litchi chinensis Sonn.) with machine learning techniques[J].Computers and Electronics in Agriculture,2023,205; 107572.
[6] 李志強(qiáng),袁沛元,邱燕萍,李建光,凡超.冬季灌溉與桂味荔 枝成花率的關(guān)系初探[J].熱帶作物學(xué)報(bào),2012,33(3):402- 407. LI Zhiqiang,YUAN Peiyuan,QIU Yanping,LI Jianguang,F(xiàn)AN Chao.Therelationshipbetween winter irigation and spring floweringofGuiwei litchi trees[J].Chinese Journal of Tropical Crops,2012,33(3):402-407.
[7] SPARKS T H,MENZEL A. Observed changes in seasons: An overview[J].International Journal of Climatology,20o2,22(14): 1715-1725.
[8] 張倩.影響庫(kù)爾勒香梨開(kāi)花與果實(shí)生長(zhǎng)的氣象條件分析[D]. 烏魯木齊:新疆師范大學(xué),2013. ZHANG Qian.Analysis ofthe effect of meteorological conditions on the flowering and fruit growth of Korla Fragrant Pear[D]. Urumqi: Xinjiang Normal University,2013.
[9] 范曉明,袁德義,唐靜,田曉明,張旭輝,王碧芳,譚曉風(fēng).錐栗 開(kāi)花授粉生物學(xué)特性[J].林業(yè)科學(xué),2014,50(10):42-48. FAN Xiaoming,YUAN Deyi, TANG Jing,TIAN Xiaoming, ZHANG Xuhui,WANG Bifang,TAN Xiaofeng. Biological characteristicsof flowering and pollinationin Castanea henryi[J]. Scientia Silvae Sinicae,2014,50(10):42-48.
[10] CLELAND E E,CHUINE I,MENZEL A,MOONEY H A, SCHWARTZ MD.Shifting plant phenology in response to global change[J].Trends in Ecologyamp;Evolution,20o7,22(7):357- 365.
[11]LIU Q,PIAO S L,F(xiàn)U Y H,GAO M D,PENUELAS J,JANSSENSI A. Climatic warming increases spatial synchrony in spring vegetation phenology across the Northern Hemisphere[J]. Geophysical Research Letters,2019,46(3):1641-1650.
[12] REICHSTEIN M,CAMPS-VALLS G,STEVENS B,JUNG M, DENZLER J, CARVALHAIS N,PRABHAT. Deep learning and processunderstanding for data-driven earth system science[J]. Nature,2019,566(7743):195-204.
[13]ALMEIDA J,DOS SANTOS JA,ALBERTONB,TORRES R DS,MORELLATO LPC.Applying machine learning based on multiscale classifiers to detect remote phenology patterns in Cerrado savanna trees[J].Ecological Informatics,2014,23:49-61.
[14]DAI W J,JIN HY,ZHANG Y H,LIU T,ZHOU Z Q. Detecting temporal changes in the temperature sensitivity of spring phenologywith global warming:Application of machine learning in phenological model[J].Agricultural and Forest Meteorology, 2019,279:107702.
[15]THESSEN A.Adoption of machine learning techniques in ecology and earth science[J]. One Ecosystem,2016,1:e8621.
[16] CZERNECKI B,NOWOSAD J,JABLONSKA K. Machine learning modeling of plant phenology based on coupling satelliteand gridded meteorological dataset[J].International Journal of Biometeorology,2018,62(7):1297-1309.
[17]BREIMAN L.Random forests[J].Machine Learning,2001,45 (1):5-32.
[18] ZHAO HS,ZHU XC,LIC,WEIY,ZHAO G X,JIANG Y M. Improving the accuracy of the hyperspectral model for apple canopy water content prediction using the equidistant sampling method[J]. Scientific Reports,2017,7(1):11192.
[19]AONO Y,KAZUI K.Phenological data series of cherry tree floweringinKyoto,Japan,and itsapplicationto reconstruction of springtime temperatures since the 9th century[J]. International Journal of Climatology,2008,28(7):905-914.
[20]SPARKS TH,CAREY PD.The responses of species to climate over two centuries:An analysis of the Marsham phenological record,1736-1947[J]. Journal ofEcology,1995,83(2):321.
[21]LIUQ,PIAO SL,JANSSENSIA,F(xiàn)UYS,PENG SS,LIANX, CIAISP,MYNENIRB,PENUELASJ,WANGT.Extensionof the growing season increases vegetation exposure to frost[J].Nature Communications,2018,9(1):426.
[22]SMITHWK,BIEDERMANJA,SCOTTRL,MOOREDJP, HEM,KIMBALLJS,YAND,HUDSONA,BARNESML, MACBEANN,F(xiàn)OXAM,LITVAKME.Chlorophyll fluorescence better captures seasonal and interannual gross primary productivity dynamics across dryland ecosystems of southwestern North America[J].Geophysical Research Letters,2018,45 (2):748-757.
[23]RICHARDSONAD,HUFKENSK,MILLIMANT,AUBRECHTDM,CHENM,GRAYJM,JOHNSTONMR,KEENANT F,KLOSTERMANST,KOSMALAM,MELAASEK,F(xiàn)RIEDL MA,F(xiàn)ROLKING S. Tracking vegetation phenology across diverse North American biomes using PhenoCam imagery[J]. Scientific Data,2018,5:180028.
[24]LIUQ,F(xiàn)UYH,LIU YW,JANSSENS IA,PIAO SL. Simulatingthe onset of spring vegetation growth across the Northern Hemisphere[J].Global Change Biology,2018,24(3):1342-1356.
[25]CHUINE I. A unified model for budburst of trees[J]. Journal of TheoreticalBiology,2000,207(3):337-347.
[26]FUYH,PIAO SL,VITASSEY,ZHAOHF,DE BOECKHJ, LIUQ,YANGH,WEBERU,HANNINENH,JANSSENSIA. Increased heat requirement for leaf flushing in temperate woody species over 1980-2012:Effects of chilling,precipitation and insolation[J]. Global Change Biology,2015,21(7):2687-2697.
[27]LAUBEJ,SPARKSTH,ESTRELLAN,MENZELA.Does humiditytrigger treephenology?Proposal for an air humidity based framework for bud development in spring[J].New Phytologist,2014,202(2):350-355.
[28]陳厚彬.我國(guó)荔枝產(chǎn)業(yè)發(fā)展情況:在2018年中國(guó)國(guó)際荔枝產(chǎn) 業(yè)大會(huì)開(kāi)摘節(jié)暨新聞發(fā)布會(huì)上的講話(huà)[J].中國(guó)熱帶農(nóng)業(yè),2018 (3):6-7. CHENHoubin.The development ofChina’slitchi industry:In the 2018 China International Litchi Industry Conference opening festival and press conference speech[J]. China Tropical Agriculture,2018(3):6-7.
[29]蘇鉆賢,楊勝男,陳厚彬,申濟(jì)源.2020年我國(guó)荔枝主產(chǎn)區(qū)的 生產(chǎn)形勢(shì)分析[J].南方農(nóng)業(yè)學(xué)報(bào),2020,51(7):1598-1605. SU Zuanxian,YANG Shengnan,CHEN Houbin,SHEN Jiyuan. Analysis of the production situation for litchi in main planting areasof China in 2O2o[J]. Journal of Southern Agriculture, 2020,51(7):1598-1605.