Estimation of Maize Aboveground Biomass Based on CNN-LSTM-SA

doi:10.12133/j.smartag.SA202412004

Abstract

Abstract:

[Objective] Maize is one of the most widely cultivated staple crops worldwide, and its aboveground biomass (AGB) serves as a crucial indicator for evaluating crop growth status. Accurate estimation of maize AGB is vital for ensuring food security and enhancing agricultural productivity. However, maize AGB is influenced by a multitude of dynamic factors, exhibiting complex spatial and temporal variations that pose significant challenges to precise estimation. At present, most studies on maize AGB estimation rely primarily on single-source remote sensing data and conventional machine learning algorithms, which limits the accuracy and generalizability of the models. To overcome these limitations, a model architecture that integrates convolutional neural networks (CNN), long short-term memory networks (LSTM), and a self-attention (SA) mechanism was developed in this research to estimate maize AGB at the field scale. [Methods] The research utilized vegetation indices, crop parameters, and meteorological data that were collected under varying gradient water treatments in the experimental area. First, an optimized CNN-LSTM-SA model was constructed. The model employed two-dimensional convolutional layers to extract both spatial and temporal features, while utilizing max-pooling and dropout techniques to mitigate overfitting. The LSTM module was used to capture temporal dependencies in the data. The SA mechanism was introduced to compute global attention weights, enhancing the representation of critical time steps. Nonlinear activation functions were applied to mitigate multicollinearity among features. A fully connected layer was used to output the estimated AGB values. Second, the Pearson correlation coefficients between influencing factors and maize AGB were analyzed, and the importance of multi-source data was validated. recursive feature elimination (RFE) was used to select the optimal input features. The local interpretable model-agnostic explanations (LIME) method was employed to interpret individual samples. Finally, ablation experiments were conducted to assess the effects of incorporating CNN and SA into the model, with performance comparisons made against random forest (RF) and support vector machine (SVM) models. [Results and Discussions] The correlation analysis revealed that crop parameters exhibited strong correlations with AGB. Among the vegetation indices, the improved normalized difference red edge index (NDREI) demonstrated the highest correlation (r = 0.63). To address multicollinearity issues, the visible atmospherically resistant index (VARI), soil adjusted vegetation index (SAVI), and normalized difference red edge index (NDRE) were excluded from the analysis. The CNN-LSTM-SA model integrated crop parameters, vegetation indices, and meteorological data and initially achieved a coefficient of determination (R²) of 0.89, a root mean square error (RMSE) of 129.38 g/m², and a mean absolute error (MAE) of 65.99 g/m². When only vegetation indices and meteorological data were included, the model yielded an R² of 0.83, an RMSE of 161.36 g/m², and an MAE of 89.37 g/m². Using a single vegetation index further reduced model accuracy. Based on multi-source data integration, RFE removed redundant features. After excluding the 2-meter average wind speed, the model reached its best performance with R² of 0.92, RMSE of 107.53 g/m², and MAE of 55.19 g/m². Using the LIME method to interpret feature contributions for individual maize samples, the analysis revealed that during the rapid growth stage, the model was primarily influenced by the current growth status and vegetation indices. For samples in the mid-growth stage, multi-day crop physiological characteristics had a substantial impact on model predictions. In the late growth stage, higher vegetation index values showed a clear suppressive effect on the model outputs. During the mid-growth stage of maize under varying moisture conditions, the model consistently demonstrated heightened sensitivity to low temperatures, moderate humidity levels, and optimal vegetation indices. The CNN-LSTM-SA model demonstrated more consistent fitting performance and accuracy across different growth stages and water conditions compared to the LSTM, LSTM-SA, and CNN-LSTM models. Additionally, it also exceeded the performance of the RF model and the SVM model in all evaluation metrics. [Conclusions] This study leveraged the feature extraction capabilities of CNN, the temporal modeling strength of LSTM, and the dynamic attention mechanism of the SA to enhance the accuracy of maize AGB estimation from a spatiotemporal perspective. The approach not only reduced estimation errors but also improved model interpretability. This research could provide valuable insights and references for the dynamic modeling of crop AGB.

Key words: maize, aboveground biomass, convolutional neural network, long short-term memory network, self-attention mechanism

CLC Number:

TP399
S513

WANG Yi, XUE Rong, HAN Wenting, SHAO Guomin, HOU Yanqiao, CUI Xitong. Estimation of Maize Aboveground Biomass Based on CNN-LSTM-SA[J]. Smart Agriculture, 2025, 7(4): 159-173.

Figures/Tables 18

Fig. 1

Table 1

Table 2

Table 3

Table 4

Formulas of major vegetation indices applied in maize AGB estimation

植被指数	公式
归一化差值植被指数（Normalized Difference Vegetation Index， NDVI ）^［27］	$N D V I = (R N I R - R R) / (R N I R + R R)$ （1）
土壤调整植被指数（Soil Adjusted Vegetation Index， SAVI）^［28］	$S A V I = (1 + L) × (R N I R - R R) / (R N I R + R R + L)$ （2）
增强型植被指数（Enhanced Vegetation Index， EVI）^［29］	$E V I = 2.5 × (R N I R - R R) / (R N I R + 6 × R R - 7.5 × R B + 1)$ （3）
改进型叶绿素吸收植被指数（Transformed Chlorophyll Absorption In Reflectance Index，TCARI）^［30］	$T C A R I = 3 × (R R E - R R) - 0.2 × (R R E - R G) × (R R E / R R)$ （4）
绿度归一化植被指数（Green Normalized Difference Vegetation Index， GNDVI）^［31］	$G N D V I = (R G - R R) / (R G + R R)$ （5）
抗大气指数（Visualized Atmospheric Resistance Index， VARI）^［32］	$V A R I = (R G - R R) / (R G + R R - R B)$ （6）
比值植被指数（Simple Ratio， SR）^［33］	$S R = R N I R / R R$ （7）
归一化差异红边指数（Normalized Difference Red Edge Index， NDRE）^［34］	$N D R E = (R N I R - R R E) / (R N I R + R R E)$ （8）
标准化差值红边指数（Improved Normalized Difference Red Edge Index， NDREI）^［35］	$N D R E I = (R R E - R G) / (R R E + R G)$ （9）
改进的叶绿素吸收反射指数（Modified Chlorophyll Absorption in Reflectance Index， MCARI）^［36］	$M C A R I = (R R E - R R) - 0.2 × (R R E - R G) × (R R E / R R)$ （10）

Table 4

Table 5

Table 6

Fig. 2

Fig.3

Fig. 4

Table 7

Fig. 5

Fig. 6

Table 8

Fig. 7

Table 9

Fig.8

Fig. 9

References 52

[1]	李娜. 玉米秸秆青贮饲料关键技术分析[J]. 农业科技与发展, 2022, 1(3): 137-139.
	LI N. Key technologies for silage feed production from maize straw[J]. Agricultural science and technology and development, 2022, 1(3): 137-139.
[2]	FARIAS G D, BREMM C, BREDEMEIER C, et al. Normalized Difference Vegetation Index (NDVI) for soybean biomass and nutrient uptake estimation in response to production systems and fertilization strategies[J]. Frontiers in sustainable food systems, 2023, 6: ID 959681.
[3]	MALTESE N E, MADDONNI G A, MELCHIORI R J M, et al. The allometric relationships between biomass and nitrogen of vegetative organs affect crop N status in maize at silking stage[J]. Field crops research, 2023, 294: ID 108861.
[4]	LI Z P, WAN S M, CHEN G D, et al. Effects of irrigation regime on soil hydrothermal microenvironment, cotton biomass, and yield under non-film drip irrigation system in cotton fields in southern Xinjiang, China[J]. Industrial crops and products, 2023, 198: ID 116738.
[5]	LI H T, SHAO L W, LIU X W, et al. What matters more, biomass accumulation or allocation, in yield and water productivity improvement for winter wheat during the past two decades [J]. European journal of agronomy, 2023, 149: ID 126910.
[6]	ABRAHA T, BASIR FAL, OBSU L L, et al. Pest control using farming awareness: impact of time delays and optimal use of biopesticides[J]. Chaos, solitons & fractals, 2021, 146: ID 110869.
[7]	HAN P, LAVOIR A V, RODRIGUEZ-SAONA C, et al. Bottom-up forces in agroecosystems and their potential impact on arthropod pest management[J]. Annual review of entomology, 2022, 67: 239-259.
[8]	HYNDMAN R J. Forecasting: principles and practice[M]. Melbourne: OTexts, 2018.
[9]	OSHAN T M. The spatial structure debate in spatial interaction modeling: 50 years on[J]. Progress in human geography, 2021, 45(5): 925-950.
[10]	TA X X, LIU Z H, HU X, et al. Adaptive spatio-temporal graph neural network for traffic forecasting[J]. Knowledge-based systems, 2022, 242: ID 108199.
[11]	CHEN G, CHEN S, LI D, et al. A hybrid deep learning air pollution prediction approach based on neighborhood selection and spatio-temporal attention[J]. Scientific reports, 2025, 15(1): ID 3685.
[12]	PADMAVATHI B, BHAGYALAKSHMI A, VISHNUPRIYA G, et al. IoT-based prediction and classification framework for smart farming using adaptive multi-scale deep networks[J]. Expert systems with applications, 2024, 254: ID 124318.
[13]	SHI X J, CHEN Z R, WANG H, et al. Convolutional LSTM network: a machine learning approach for precipitation nowcasting[J]. Advances in neural information processing systems, 2015, 802-810.
[14]	WANG L, ZHANG H C, BIAN L M, et al. Poplar seedling varieties and drought stress classification based on multi-source, time-series data and deep learning[J]. Industrial crops and products, 2024, 218: ID 118905.
[15]	ZHOU H Y, ZHANG S H, PENG J Q, et al. Informer: beyond efficient transformer for long sequence time-series forecasting[EB/OL]. arXiv: 2012.07436, 2020.
[16]	ANDRYCHOWICZ M, DENIL M, COLMENAREJO S G, et al. Learning to learn by gradient descent by gradient descent[J]. Advances in neural information processing systems, 2016: 3988-3996.
[17]	VASWANI A. Attention is all you need[C]// 31st Conference on Neural Information Processing Systems (NIPS 2017). Red Hook, New York, USA: Curran Associates Inc, 2017.
[18]	WU B F, ZHANG M, ZENG H W, et al. Challenges and opportunities in remote sensing-based crop monitoring: A review[J]. National science review, 2023, 10(4): ID nwac290.
[19]	SARKER I H. Deep learning: A comprehensive overview on techniques, taxonomy, applications and research directions[J]. SN computer science, 2021, 2(6): ID 420.
[20]	肖武, 陈佳乐, 笪宏志, 等. 基于无人机影像的采煤沉陷区玉米生物量反演与分析[J]. 农业机械学报, 2018, 49(8): 169-180.
	XIAO W, CHEN J L, DA H Z, et al. Inversion and analysis of maize biomass in coal mining subsidence area based on UAV images[J]. Transactions of the Chinese society for agricultural machinery, 2018, 49(8): 169-180.
[21]	RAYA-SERENO M D, ORTIZ-MONASTERIO J I, ALONSO-AYUSO M, et al. High-resolution airborne hyperspectral imagery for assessing yield, biomass, grain N concentration, and N output in spring wheat[J]. Remote sensing, 2021, 13(7): ID 1373.
[22]	WANG T J, CRAWFORD M M, TUINSTRA M R. A novel transfer learning framework for sorghum biomass prediction using UAV-based remote sensing data and genetic markers[J]. Frontiers in plant science, 2023, 14: ID 1138479.
[23]	WEI L L, YANG H S, NIU Y X, et al. Wheat biomass, yield, and straw-grain ratio estimation from multi-temporal UAV-based RGB and multispectral images[J]. Biosystems engineering, 2023, 234: 187-205.
[24]	ZHANG L Y, HAN W T, NIU Y X, et al. Evaluating the sensitivity of water stressed maize chlorophyll and structure based on UAV derived vegetation indices[J]. Computers and electronics in agriculture, 2021, 185: ID 106174.
[25]	SHAO G M, HAN W T, ZHANG H H, et al. Mapping maize crop coefficient Kc using random forest algorithm based on leaf area index and UAV-based multispectral vegetation indices[J]. Agricultural water management, 2021, 252: ID 106906.
[26]	PÔÇAS I, CALERA A, CAMPOS I, et al. Remote sensing for estimating and mapping single and basal crop coefficientes: A review on spectral vegetation indices approaches[J]. Agricultural water management, 2020, 233: ID 106081.
[27]	陈雨烨, 王培娟, 张源达, 等. 基于3种遥感指数的东北春玉米干旱识别对比[J]. 应用气象学报, 2022, 33(4): 466-476.
	CHEN Y Y, WANG P J, ZHANG Y D, et al. Comparison of drought recognition of spring maize in Northeast China based on 3 remote sensing indices[J]. Journal of applied meteorological science, 2022, 33(4): 466-476.
[28]	MSADEK J, TLILI A, CHOUIKHI F, et al. Assessing the impacts of climate change scenarios on soil-adjusted vegetation index in north African arid montane rangeland: Case of toujane region[J]. Climate, 2025, 13(3): 59.
[29]	SUN X B, ZHANG P L, WANG Z H, et al. Potential of multi-seasonal vegetation indices to predict rice yield from UAV multispectral observations[J]. Precision agriculture, 2024, 25(3): 1235-1261.
[30]	GAO S, YAN K, LIU J X, et al. Assessment of remote-sensed vegetation indices for estimating forest chlorophyll concentration[J]. Ecological indicators, 2024, 162: ID 112001.
[31]	KUMAR M, BHATTACHARYA B K, PANDYA M R, et al. Machine learning based plot level rice lodging assessment using multi-spectral UAV remote sensing[J]. Computers and electronics in agriculture, 2024, 219: ID 108754.
[32]	COSWOSK G G, GONÇALVES V M L, DE LIMA V J, et al. Utilizing visible band vegetation indices from unmanned aerial vehicle images for maize phenotyping[J]. Remote sensing, 2024, 16(16): ID 3015.
[33]	张慧, 李平衡, 周国模, 等. 植被指数的地形效应研究进展[J]. 应用生态学报, 2018, 29(2): 669-677.
	ZHANG H, LI P H, ZHOU G M, et al. Advances in the studies on topographic effects of vegetation indices[J]. Chinese journal of applied ecology, 2018, 29(2): 669-677.
[34]	束美艳, 顾晓鹤, 孙林, 等. 基于新型植被指数的冬小麦LAI高光谱反演[J]. 中国农业科学, 2018, 51(18): 3486-3496.
	SHU M Y, GU X H, SUN L, et al. High spectral inversion of winter wheat LAI based on new vegetation index[J]. Scientia agricultura sinica, 2018, 51(18): 3486-3496.
[35]	邵亚杰, 李珂, 丁文浩, 等. 基于无人机多光谱影像特征估算棉花生物量[J]. 新疆农业科学, 2024, 61(6): 1328-1335.
	SHAO Y J, LI K, DING W H, et al. Study on cotton biomass estimation based on multi-spectral imaging features of unmanned aerial vehicle[J]. Xinjiang agricultural sciences, 2024, 61(6): 1328-1335.
[36]	QIAO L, TANG W J, GAO D H, et al. UAV-based chlorophyll content estimation by evaluating vegetation index responses under different crop coverages[J]. Computers and electronics in agriculture, 2022, 196: ID 106775.
[37]	VIRTANEN P, GOMMERS R, OLIPHANT T E, et al. SciPy 1.0: Fundamental algorithms for scientific computing in Python[J]. Nature methods, 2020, 17(3): 261-272.
[38]	NIU Z Y, ZHONG G Q, YU H. A review on the attention mechanism of deep learning[J]. Neurocomputing, 2021, 452: 48-62.
[39]	GOODFELLOW I, BENGIO Y, COURVILLE A. Deep learning: Hyperparameter optimization [M]. Cambridge: MIT Press, 2016.
[40]	LIU Z, XU Z, JIN J, et al. Dropout reduces underfitting[C]// Proceedings of the 40th International Conference on Machine Learning. New York, USA: Proceedings of Machine Learning Research, 2023, 202: 22233-22248.
[41]	XIAO F, SHI B Y, GAO J, et al. Enhanced probabilistic prediction of pavement deterioration using Bayesian neural networks and cuckoo search optimization[J]. Scientific reports, 2025, 15(1): ID 8665.
[42]	HODSON T O. Root-mean-square error (RMSE) or mean absolute error (MAE): When to use them or not[J]. Geoscientific model development, 2022, 15(14): 5481-5487.
[43]	LI Z Z, BI S D, HAO S, et al. Aboveground biomass estimation in forests with random forest and Monte Carlo-based uncertainty analysis[J]. Ecological indicators, 2022, 142: ID 109246.
[44]	GUIDOTTI R, MONREALE A, RUGGIERI S, et al. A survey of methods for explaining black box models[J]. ACM computing surveys, 2019, 51(5): 1-42.
[45]	FANG H L, BARET F, PLUMMER S, et al. An overview of global leaf area index (LAI): methods, products, validation, and applications[J]. Reviews of geophysics, 2019, 57(3): 739-799.
[46]	YUAN L M, ZHOU T Y, LI K, et al. Moderate soil drying improves physiological performances and kernel yield of maize[J]. Food and energy security, 2023, 12(3): ID e444.
[47]	SHAO R X, YU K K, LI H W, et al. The effect of elevating temperature on the growth and development of reproductive organs and yield of summer maize[J]. Journal of integrative agriculture, 2021, 20(7): 1783-1795.
[48]	HOU P, LIU Y E, LIU W M, et al. Quantifying maize grain yield losses caused by climate change based on extensive field data across China[J]. Resources, conservation and recycling, 2021, 174: ID 105811.
[49]	YUAN A, KUMAR S D, WANG H T, et al. Dynamic interplay among soil nutrients, rhizosphere metabolites, and microbes shape drought and heat stress responses in summer maize[J]. Soil biology and biochemistry, 2024, 191: ID 109357.
[50]	MOLNAR C. Interpretable machine learning[M]. Saarbrücken: Lulu. com, 2020.
[51]	YANG Q, LIU L C, ZHOU J X, et al. A flexible and efficient knowledge-guided machine learning data assimilation (KGML-DA) framework for agroecosystem prediction in the US Midwest[J]. Remote sensing of environment, 2023, 299: ID 113880.
[52]	ZHANG Y F, WU R, DASCALU S M, et al. A novel extreme adaptive GRU for multivariate time series forecasting[J]. Scientific reports, 2024, 14(1): ID 2991.

水分处理区域	快速生长期（DAP： 30—73 d）			生长中期（DAP： 74—104 d）			生长后期（DAP： 105—125 d）
水分处理区域	灌溉量/mm	降雨量/mm	总量/mm	灌溉量/mm	降雨量/mm	总量/mm	灌溉量/mm	降雨量/mm	总量/mm
T1	56.0	19.6	75.6	68.3	94.2	162.5	12.0	0	12.0
T2	36.4	19.6	56.0	56.3	94.2	150.5	7.8	0	7.8
T3	22.4	19.6	42.0	64.6	94.2	158.8	9.6	0	9.6
T4	36.4	19.6	56.0	46.8	94.2	141.0	7.8	0	7.8
T5	22.4	19.6	42.0	51.3	94.2	145.5	4.8	0	4.8

无人机系统	主要参数	参数值
无人机可见光图像采集系统	重量/g	1 388
	飞行时间/min	30
	通行半径/km	5
	飞行速度/（km/s）	<72
	像素	5 472×3 648
	焦距/mm	8.8/24
	视场角/（°）	84
	传感器	1英寸CMOS
	RGB颜色空间	sRGB
	ISO范围	100~12 800
	快门速度/s	8~1/8 000
	图像格式	JPEG、DNG
无人机多光谱图像采集系统	轴距/mm	900
	起飞重量/kg	6
	载重/kg	2
	飞行时间/min	18
	通信半径/km	3
	飞行速度/（m/s）	5
	搭载相机	MicaSense RedEdge-M
	像素	1 280×960
	数据格式	12位Raw
	光谱波段	蓝、绿、红、近红外、红边
	焦距/mm	9.6
	视场角/（°）	47.2

数据类型	特征参数
作物数据	H、LAI、Tc
植被指数	NDVI、SAVI、EVI、TCARI、GNDVI、VARI、SR、NDRE、NDREI、MCARI
气象数据	RH、Rn、U2

折数	训练集样本范围	测试集样本范围
第一折	第1到第67个样本	第68到第100个样本
第二折	第1到第100个样本	第101到第134个样本
第三折	第1到第134个样本	第135到第167个样本
第四折	第1到第167个样本	第168到第200个样本
第五折	第1到第268个样本	第269到第335个样本

CNN-LSTM-SA模型参数	参数值
LSTM单元数	16
Dense单元数	32
训练轮数	100
批次大小	32
激活函数	ReLU
Dropout率	0.2
损失函数	均方误差函数
优化器	Adam