基于机器学习算法的天祝藏族自治县草地地上生物量反演

doi:10.11686/cyxb2021072

摘要/Abstract

摘要：

使用机器学习算法快速、准确、大范围监测草地地上生物量（AGB）是目前研究热点，但不同机器学习算法因训练样本、超参数设置不同而存在较大差异。基于实测草地AGB和同期遥感数据、气象数据、地形数据，选择与草地AGB相关性较强的13个因子作为深度神经网络（DNN）、随机森林算法（RF）、梯度提升回归树（GBRT）、支持向量机（SVR）、人工神经网络（ANN）和高斯过程回归（GPR）算法的输入变量，建立草地AGB预测模型并从模型预测精度、稳定性、样本敏感性等方面综合评价6种模型应用潜力，分析2020年天祝藏族自治县生长季（4-9月）内草地AGB时空变化特征及其对气候的响应。结果表明：1）DNN估算草地AGB的综合性能最佳，但稳定性较差，对样本敏感性较高；GPR综合性能次于DNN，稳定性和精度均较好；GBRT、RF模拟精度较高，稳定性差；SVR和ANN精度相对其他模型较差，SVR稳定性较高，ANN稳定性较差。2）天祝藏族自治县草地AGB集中在50~250 g·m^-2，不同月份草地AGB空间异质性较大，整体表现为从西北向东南呈下降趋势。3）山地草甸、高寒草甸和温性草原中的AGB变化与气温表现出较为明显的正相关关系。降水量对高寒草甸、温性草原和山地草甸的影响不明显，但对温性荒漠草原类的影响较大，AGB随降水量减少呈现减少态势。以上研究结果可为监测草地生物量的方法选择和参数设置提供一定技术支持和参考依据。

关键词: 草地生物量, 机器学习, 模型性能, 天祝藏族自治县

Abstract:

Effective， accurate， and large-scale monitoring of grassland aboveground biomass （AGB） using machine learning algorithms is currently a very active field of research， but different machine learning algorithms vary greatly in performance depending on training samples and hyper-parameter settings. The research utilized grassland AGB data collected field， combined with remote sensing data， meteorological data， terrain data for the same period. Thirteen indicators with strong correlation with grassland AGB were selected as input variables for analysis using deep neural network （DNN）， random forest （RF）， gradient boosting regression tree （GBRT）， vector support machine （SVR）， artificial neural network （ANN） and Gaussian process regression （GPR） algorithms for AGB inversion. At the same time， the application potential of 6 models were evaluated from the aspects of model prediction accuracy， stability， sample sensitivity， and characteristics of AGB space-time changes in grasslands during the 2020 Tianzhu County growth season （April-September） and their response to climate. It was found that： 1） The multivariate performance of DNN in grassland AGB inversion was the best， but the stability was poor， and the sensitivity to samples was high； the comprehensive performance of GPR was inferior to DNN， and its stability and accuracy were good； the simulation accuracy of GBRT and RF was high， and its stability was poor； the accuracy of SVR and ANN was relatively poor and the stability of SVR was high， and the stability of ANN was poor. 2） The grassland AGB ranged from 50 to 250 g·m^-2. The spatial heterogeneity of AGB was large and variable over time. In general AGB showed a downward trend from northwest to southeast. 3） Except for temperate desert steppe， the AGB in mountain meadow， alpine meadow and temperate steppe species showed a significant positive correlation with air temperature. Precipitation had no obvious effect on AGB of alpine meadow， temperate grassland and mountain meadow， but had a great effect on AGB of temperate desert grassland. With decrease in precipitation， AGB tended to decrease. The above results provide technical information to support decisions on choice of method and parameter setting when remotely monitoring grassland biomass.

Key words: grassland biomass, machine learning, model performance, Tianzhu Zangzu Autonomous County

秦格霞, 吴静, 李纯斌, 吉珍霞, 邱政超, 李颖. 基于机器学习算法的天祝藏族自治县草地地上生物量反演[J]. 草业学报, 2022, 31(4): 177-188.

Ge-xia QIN, Jing WU, Chun-bin LI, Zhen-xia JI, Zheng-chao QIU, Ying LI. Inversion of grassland aboveground biomass in Tianzhu Zangzu Autonomous County based on a machine learning algorithm[J]. Acta Prataculturae Sinica, 2022, 31(4): 177-188.

图/表 10

图1 天祝藏族自治县草地类型及野外实测点分布

Fig.1 Spatial distribution of grassland types and field sites in Tianzhu Zangzu Autonomous County

图2 样方分布及野外实测场景照片

Fig.2 Photos of plot distribution and field measurement

图3 实测草地AGB数据分布绿色代表高寒草甸；金黄色代表温性草原；蓝色代表山地草甸；紫色代表温性荒漠草原。Green represents alpine meadow； golden yellow represents warm steppe； blue represents slope meadow； purple represents temperate desert steppe.

Fig.3 Map of the measured data

图4 AGB与待选自变量相关系数*： P<0.05； **： P<0.01；绿色、黄色、粉色、蓝色分别代表极显著正相关、极显著负相关、显著正相关、显著负相关The green， yellow， pink and blue denotes the significant positive effects， significant negative effects， positive effects， negative effects.

Fig.4 Correlation coefficients between the AGB and the explanatory variables

图5 30次重复试验中模型精度箱型图

Fig.5 The boxplot of model accuracy in the 30 repeated experiments

表1 重复30次的R2、RMSE、MAE的统计

Table 1 Statistical table of R2， RMSE and MAErepeated 30 times

机器学习模型 Machine learning models	决定系数R²		均方根误差 RMSE		平均绝对误差 MAE
机器学习模型 Machine learning models	范围Range	平均Mean	范围Range	平均Mean	范围Range	平均Mean
DNN	0.80~0.88	0.85	25.43~32.30	32.30	4.23~6.14	5.63
GBRT	0.74~0.88	0.84	21.51~49.63	32.31	4.55~7.29	5.95
RF	0.75~0.88	0.84	30.41~43.48	37.21	4.55~6.63	5.58
SVR	0.77~0.84	0.80	39.55~49.24	46.01	6.33~7.92	7.46
ANN	0.74~0.87	0.81	38.04~54.19	45.99	4.95~9.36	6.54
GPR	0.76~0.88	0.83	32.60~51.03	39.72	4.94~7.14	6.02

图6 机器学习模型对于训练样本大小的敏感性

Fig.6 Sensitivity of the machine learning models to the training sample size

图7 使用雷达图从不同角度综合对比6种机器学习模型的表现

Fig.7 Comprehensive comparison of six machine learning models with different metrics using a radar chart

图8 草地AGB空间分布

Fig.8 Spatial distribution of AGB in grassland

图9 各种草地类型AGB对气温和降水的响应机制Ⅰ：天祝藏族自治县4种草地类型平均值Average of grassland types in Tianzhu Zangzu Autonomous County； Ⅱ：高寒草甸Alpine meadow； Ⅲ：山地草甸Slope meadow； Ⅳ：温性草原Warm steppe； Ⅴ：温性荒漠草原Temperate desert steppe； PRE：降水Precipitation； TEM：气温Temperature.

Fig.9 Response mechanism of AGB in different grassland types to temperature and precipitation

参考文献 42

1	Scurlock J， Hall D. The global carbon sink： A grassland perspective. Global Change Biology， 1998， 4： 229-233.
2	Jin Y X， Yang X C， Qiu J J， et al. Remote sensing-based biomass estimation and its spatiotemporal variations in temperate grassland， Northern China. Remote Sensing， 2014， 6： 1496-1513.
3	Party Group of the Ministry of Environmental Protection of the Communist Party of China. Building a new modernization construction pattern of the human and nature harmonious development-based on the theory and practice since the eighteenth congress of the ecological civilization construction. Environmental Protection， 2016， 44（13）： 11-13.
	中共环境保护部党组. 构建人与自然和谐发展的现代化建设新格局——党的十八大以来生态文明建设的理论与实践. 环境保护， 2016， 44（13）： 11-13.
4	Gao J B， Jiao K W， Wu S H， et al. Past and future effects of climate change on spatially heterogeneous vegetation activity in China. Earth’s Future， 2017， 5（7）： 679-692.
5	Parmesan C， Yohe G. A globally coherent fingerprint of climate change impacts across natural systems. Nature， 2003， 421： 37-42.
6	Huang D Q， Yu L， Zhang Y S， et al. Aboveground numerical characteristics of five natural grassland and their relationships to environmental factors in the northern slopes of the Mountains Qilian. Acta Agriculturae Boreali-occidentalis Sinica， 2011， 20（6）： 174-180.
	黄德青，于兰，张耀生，等. 祁连山北坡5类天然草地地上部数量特征及其与环境因子的关系. 西北农业学报， 2011， 20（6）： 174-180.
7	Zhang Q R， Zhang J J. Effects of grassland resources on the development of white yak industrialization in Tianzhu County. China Cattle Science， 2012， 38（2）： 53-56.
	张起荣，张金菊. 天祝藏族自治县草地资源对白牦牛产业化发展的影响. 中国牛业科学， 2012， 38（2）： 53-56.
8	Akiyama T， Kawamura K. Grassland degradation in China： Methods of monitoring， management and restoration. Grassland Science， 2007， 53： 1-17.
9	Flombaum P， Sala O E. A non-destructive and rapid method to estimate biomass and aboveground net primary production in arid environments. Journal of Arid Environments， 2007， 69： 352-358.
10	Xu K X， Su Y J， Liu J， et al. Estimation of degraded grassland aboveground biomass using machine learning methods from terrestrial laser scanning data. Ecological Indicators， 2020， 108： https：//doi.org/10.1016/j.ecolind.2019.105747.
11	Xia J Z， Ma M N， Liang T G， et al. Estimates of grassland biomass and turnover time on the Tibetan Plateau. Environmental Research Letters， 2018， 13（16）： 14-20.
12	Wu C F， Shen H H， Shen A H， et al. Comparison of machine-learning methods for above-ground biomass estimation based on landsat imagery. Journal of Applied Remote Sensing， 2016， 10（3）： 10-35.
13	Chen X H. Analysis on the prospect of grass and livestock industry in Tianzhu County. Developing Strategic Surplus Production and Management， 2012， 32（1）： 57-59.
	陈新辉. 天祝藏族自治县草畜产业前景浅析. 发展战略与生产经营， 2012， 32（1）： 57-59.
14	Li Y Y， Dong S K， Li X Y， et al. Effect of enclosure on vegetation photosynthesis and biomass of degraded grasslands in headwater area of Qinghai-Tibetan plateau. Acta Agrestia Sinica， 2012， 20（4）： 621-625.
	李媛媛，董世魁，李小艳，等. 围栏封育对三江源区退化高寒草地植物光合作用及生物量的影响. 草地学报， 2012， 20（4）： 621-625.
15	Zhang Y， Zhang C B， Wang Z Q， et al. Quantitative assessment of relative roles of climate change and human activities on grassland net primary productivity in the Three-River Source region， China. Acta Prataculturae Sinica， 2017， 26（5）： 1-14.
	张颖，章超斌，王钊齐，等. 气候变化与人为活动对三江源草地生产力影响的定量研究. 草业学报， 2017， 26（5）： 1-14.
16	Zhang F P， Wang H W， Zhu Y W， et al. Study on the aboveground biomass of natural grassland and balance between forage and livestock in Qilian county. Journal of Natural Resources， 2017， 32（7）： 1183-1192.
	张福平，王虎威，朱艺文，等. 祁连县天然草地地上生物量及草畜平衡研究. 自然资源学报， 2017， 32（7）： 1183-1192.
17	McEniry J， Crosson P， Finneran E， et al. How much grassland biomass is available in Ireland in excess of livestock requirements？ Irish Journal of Agricultural and Food Ressearch， 2013， 52（1）： 67-80.
18	Gao X X， Dong S K， Li S， et al. Using the random forest model and validated MODIS with the field spectrometer measurement promote the accuracy of estimating aboveground biomass and coverage of alpine grasslands on the Qinghai-Tibetan Plateau. Ecological Indicators， 2020， 112： 106-114.
19	Haydock K P， Shaw N H. The comparative yield method for estimating dry matter yield of pasture. Australian Journal of Experimental Agriculture and Animal Husbandry， 1975， 15： 663-670.
20	Gartner R J W， Laws L， O'Rourke P K. Millet compared with wheat and sorghum in high-grain diets for cattle. Animal Feed Resources Information System， 1975， 15（75）： 446-450.
21	Omer R M， Hester A J， Gordon I J， et al. Seasonal changes in pasture biomass production and offtake under the transhumance system in northern Pakistan. Journal of Arid Environments， 2006， 67： 641-660.
22	Zhang X H， Chen X L， Tian M R， et al. An evaluation model for aboveground biomass based on hyperspectral data from field and TM8 in Khorchin grassland， China. PloS One， 2020， 15（2）： 1-7.
23	Schaefer M T， Lamb D W. A combination of plant NDVI and LiDAR measurements improve the estimation of pasture biomass in tall fescue （Festuca arundinacea var. fletcher）. Remote Sensing， 2016， 8（2）： https：//doi.org/10.3390/rs8020109.
24	Marildo G F， Tatiana M K， Quadros F L F D. Estimating natural grassland biomass by vegetation indices using sentinel 2 remote sensing data. International Journal of Remote Sensing， 2020， 41（8）： 2861-2876.
25	Zeng N， Ren X， He H L， et al. Aboveground biomass of grassland in the three-river headwaters region based on neural network. Research of Environmental Sciences， 2017， 30（1）： 59-66.
26	Zeng N， Ren X L， He H L， et al. Estimating grassland aboveground biomass on the Tibetan Plateau using a random forest algorithm. Ecological Indicators， 2019， 102： 479-487.
27	Xie Y C， Sha Z Y， Yu M， et al. A comparison of two models with landsat data for estimating above ground grassland biomass in Inner Mongolia， China. Ecological Modelling， 2009， 220（15）： 1810-1818.
28	Mobyen U A， Peter A， Tim A， et al. A machine learning approach for biomass characterization. Energy Procedia， 2019， 158： 1279-1287.
29	Pablito M. López S， Carlos A， et al. A comparison of machine learning techniques applied to Landsat-5 TM spectral data for biomass estimation. Canadian Journal of Remote Sensing， 2019， 42（6）： 690-705.
30	Amir S， Hormoz S， Scott P， et al. A comparative assessment of multi-temporal landsat 8 and machine learning algorithms for estimating aboveground carbon stock in coppice oak forests. International Journal of Remote Sensing， 2019， 38（22）： 6407-6432.
31	He L， Li A， Yin G， et al. Retrieval of grassland aboveground biomass through inversion of the PROSAIL model with MODIS imagery. Remote Sensing， 2019， 11（13）： http：//ir.imde.ac.cn/handle/131551/26624.
32	Powell S L， Cohen W B， Healey S P， et al. Quantification of live aboveground forest biomass dynamics with landsat time-series and field inventory data： A comparison of empirical modeling approaches. Remote Sensing of Environment， 2010， 114（15）： 1053-1068.
33	Karlson M， Ostwald M， Reese H， et al. Mapping tree canopy cover and aboveground biomass in Sudano-Sahelian woodlands using landsat 8 and random forest. Remote Sensing， 2015， 7： 10017-10041.
34	Aggarwal C C. An Introduction to neural networks// In： Neural networks and deep learning. Cham： Springer， 2018.
35	Su D X. The compilation and study of the grassland resource map of China on the scale of 1∶1000000. Journal of Natural Resources， 1996， 11（1）： 75-83.
	苏大学. 1∶1000000中国草地资源图的编制与研究. 自然资源学报， 1996， 11（1）： 75-83.
36	Zhang Z L. Investigation report on grassland industrialization in Tianzhu County. Tianzhu： Annual Academic Meeting of Gansu Province， 2019.
	张子廉. 天祝藏族自治县草业产业化调研报告. 天祝：甘肃省学术年会， 2019.
37	Van G M. Computational foundations of natural intelligence. Frontiers in Computational Neuroscience， 2017， DOI： 10.3389/fncom.2017.00112.
38	Scornet E. Random forests and kernel methods. IEEE Transactions on Information Theory， 2016， 62（3）： 1485-1500.
39	Natekin A， Knoll A. Gradient boosting machines， a tutorial. Frontiers in Neurorobotics， 2013， 7（21）： 1-21.
40	Vapnik V， Golowich S E， Smola A J. Support vector method for function approximation， regression estimation and signal processing//Mozer M C， Jordan M， Petsche T， edit. Advances in Neural Information Processing Systems， 1996： 281-287.
41	Rasmussen C E， Williams C K I. Gaussian processes for machine learning. Cambridge： the MIT Press， 2006.
42	Xu J H. Quantitative geography （2nd Edition）. Beijing： Higher Education Press， 2014.
	徐建华. 计量地理学（第二版）. 北京：高等教育出版社， 2014.