Remote sensing retrieval of nature grassland biomass in Menyuan County， Qinghai Province experimental area based on Sentinel-2 data

doi:10.11686/cyxb2022147

Abstract

Abstract:

Above-ground biomass （AGB） is an important indicator for evaluating grassland productivity. Accurate inversion of AGB of natural grassland is of great significance for monitoring grassland growth and evaluating the feed balance of forage-livestock. As commonly used remote sensing data （such as Landsat and MODIS） suffer from by many problems caused by low temporal and spatial resolution， it is extremely important to explore Sentinel-2 satellite data with higher temporal and spatial resolution and more spectral bands in monitoring grassland vegetation at county scale. In this study， we used Sentinel-2 satellite remote sensing imagery and the AGB data of Menyuan County， Qinghai Province to construct a grassland biological monitoring system based on random forest （RF）， support vector machine （SVM） and artificial neural network （ANN） methods， to study the spatial and temporal distribution characteristics of natural grassland biomass in Menyuan County from 2019 to 2021. It was found that： 1） The three original bands（B2，B6，B11）and two vegetation indices， inverted red edge chlorophyll index（IRECI），and pigment specific simple ratio chlorophyll index（PSSRa），were the important variables for AGB quantification in natural grassland. Among these，red-edge bands（B5，B6，B7）play an important role in remote sensing inversion of natural grassland AGB. 2） The AGB estimation model based on the RF algorithm was the optimal model（validation set R² 0.72， RMSE 622.616 kg·ha^-1）for natural grassland biomass estimation in Menyuan County， which was superior to the SVM model（validation set R² 0.66，RMSE 698.271 kg·ha^-1）and the ANN model（validation set R² 0.63，RMSE 730.676 kg·ha^-1）. 3） The average value of AGB of natural grassland in Menyuan County from 2019 to 2021 ranged from 3360.26-3544.00 kg·ha^-1. In general， the AGB of grassland in Menyuan County from 2019 to 2021 showed a trend of initially increasing and then decreasing， with a spatial distribution pattern of gradual decrease from the periphery to the middle.

Key words: Sentinel-2, aboveground biomass, machine learning, nature grassland, inversion mode

Rui GUO, Shuai FU, Meng-jing HOU, Jie LIU, Chun-li MIAO, Xin-yue MENG, Qi-sheng FENG, Jin-sheng HE, Da-wen QIAN, Tian-gang LIANG. Remote sensing retrieval of nature grassland biomass in Menyuan County， Qinghai Province experimental area based on Sentinel-2 data[J]. Acta Prataculturae Sinica, 2023, 32(4): 15-29.

Figures/Tables 12

Fig. 1 Spatial distribution of sampling sites and random points of the study area from 2019 to 2021

Table 1 Parameters of Sentinel-2 satellite images of the study area from 2019 to 2021

年份 Year	卫星成像时间 Satellite ingestion period	地面观测时间 Ground observation time	成像卫星 Imaging satellite	轨道号 Track number	云量百分比 Cloud cover percentage （%）	云影百分比 Cloud shadow percentage （%）
2019	05-28	05-29	2B	T47SQB	6.27	0.30
	06-30	06-27-06-28	2B	T47SQB	48.71	1.01
	07-25	07-28-07-29	2A	T47SQB	17.36	0.64
	08-29	08-27-08-28	2B	T47SQB	37.02	0.75
	09-20	09-20-09-22	2A	T47SQB	5.90	1.09
2020	07-06	07-07-07-08	2A	T47SQB	27.12	0.30
2020	09-04	09-02-09-03	2A	T47SQB	2.95	0.48
2021	07-06	07-07	2B	T47SQB	42.49	0.44
	07-31	08-04	2A	T47SQB	12.34	1.89
	09-09	09-03	2A	T47SQB	18.14	1.99
	09-29	09-28-09-29	2A	T47SQB	6.41	0.12

Table 2 Parameters of 29 common vegetation index

变量属性Variable attributes	计算公式Calculation formula
比值植被指数Ratio vegetation index （RVI）	$R V I = B 8 ∕ B 4$
差值植被指数Difference vegetation index （DVI）	$D V I = B 8 - B 4$
权重差值植被指数Weighted difference vegetation index （WDVI）	$W D V I = B 8 - 0.5 × B 4$
红外植被指数Infrared vegetation index （IPVI）	$I P V I = B 8 ∕ (B 8 + B 4)$
红边位置指数Red edge position index （REP）	$R E P = B 5 + 40 × B 4 + B 7 / 2 - B 5 B 6 - B 5$
垂直植被指数Perpendicular vegetation index （PVI）	$P V I = s i n ? 45 ° × B 8 - c o s ? 45 ° ? × B 4$
增强植被指数Enhanced vegetation index （EVI）	$E V I = 2.5 × B 8 - B 4 B 8 + 6 × B 4 - 7 × B 2 + 1$
增强植被指数-2 Enhanced vegetation index-2 （EVI2）	$E V I 2 = 2.5 × B 8 - B 4 B 8 + 2.4 × B 4 + 1$
多时相植被指数-2 Multitemporal vegetation index-2 （MTVI2）	$M T V I 2 = 1.5 × 1.2 × (B 8 - B 3) - 2.5 × (B 4 - B 3) (2 × B 8 + 1) 2 - 6 × (B 8 - 5 B 4) - 0.5$
归一化差值植被指数Normalized difference vegetation index （NDVI）	$N D V I = B 8 - B 4 B 8 ? + ? B 4$
再归一化差植被指数Renormalization the vegetation index （RDVI）	$R D V I = D V I × N D V I$
B4和B5归一化差值植被指数NDVI with band4 and band5 （NDVI45）	$N D V I 45 = B 5 - B 4 B 5 + B 4$
绿波归一化差值植被指数NDVI of green band （GNDVI）	$G N D V I = B 7 - B 3 B 7 + B 3$
反红边叶绿素指数Inverted red edge chlorophyll index （IRECI）	$I R E C I = B 6 × B 7 - B 4 B 5$
三角形植被指数Triangle vegetation index （TVI）	$T V I = 0.5 × [120 × B 6 - B 3 - 200 × B 4 - B 3]$
土壤调节植被指数Soil adjusted vegetation index （SAVI）	$S A V I = 1.5 × (B 8 - B 4) ∕ (B 8 + B 4 + 0.5)$
优化型土壤调节植被指数Optimized soil regulates vegetation index （OSAVI）	$O S A V I = (1 + 0.16) × (B 8 - B 4) B 8 + B 4 + 0.16$
转化土壤调节植被指数Transformed soil adjusted vegetation index （TSAVI）	$T S A V I = 0.5 × B 8 - 0.5 × B 4 - 0.2 0.5 × B 8 + B 4 + 0.1$
修正型土壤调节植被指数Modified soil adjusted vegetation index （MSAVI）	$M S A V I = (2 - N D V I × W D V I) × (B 8 - B 4) ∕ 8 × (B 8 + B 4 + 1 - N D V I × W D V I)$
二次修正型土壤调节植被指数Secondly modified soil adjusted vegetation index （MASVI2）	$M A S V I 2 = 0.5 × 2 × B 8 + 1 - 0.5 × 2 × B 8 + 1 2 - 8 × B 8 - B 4$
大气阻抗植被指数Atmospherically resistant vegetation index （ARVI）	$A R V I = B 8 - (2 × B 4 - B 2) ∕ B 8 + (2 × B 4 - B 2)$
叶绿素吸收比值指数Chlorophyll absorption ratio index （CARI）	CARI =（ $B 5 - B 4) - 0.2 × (B 5 - B 3)$
改进红边比值植被指数Modified red-edge simple ratio index （MSR）	$M S R = B 8 ∕ B 4 - 1 B 8 ∕ B 4 + 1$
特定色素简单比值植被指数Pigment specific simple ratio chlorophyll index （PSSRa）	$P S S R a = B 7 ∕ B 4$
Meris陆地叶绿素指数Meris terrestrial chlorophyll index （MTCI）	$M T C I = B 6 - B 5 B 5 - B 4$
修正型叶绿素吸收比值指数Modified chlorophyll absorption ratio index （MCARI）	$M C A R I = (B 5 - B 4) - 0.2 × B 5 - B 3 × (B 5 ∕ B 4)$
“哨兵2号”红边位置指数Sentinel-2 red edge position index （S2REP）	$S 2 R E P = 705 + 35 × [B 4 + B 7 / 2 - B 5] × (B 6 - B 5)$
红边感染点指数Red edge infection point index （REIP）	$R E I P = 700 + 40 × (B 4 + B 7) ∕ 2 - B 5 B 6 - B 5$
全球环境监测指数Global environmental monitoring index （GEMI）	$G E M I = e t a × 1 - 0.25 × e t a - (B 4 - 0.125) 1 - B 4$ ， $e t a = 2 × (B 8 A 2 - B 42) + 1.5 × B 8 A + 0.5 × B 4 B 8 A + B 4 + 0.5$

Table 2 Parameters of 29 common vegetation index

变量属性Variable attributes	计算公式Calculation formula
比值植被指数Ratio vegetation index （RVI）	$R V I = B 8 ∕ B 4$
差值植被指数Difference vegetation index （DVI）	$D V I = B 8 - B 4$
权重差值植被指数Weighted difference vegetation index （WDVI）	$W D V I = B 8 - 0.5 × B 4$
红外植被指数Infrared vegetation index （IPVI）	$I P V I = B 8 ∕ (B 8 + B 4)$
红边位置指数Red edge position index （REP）	$R E P = B 5 + 40 × B 4 + B 7 / 2 - B 5 B 6 - B 5$
垂直植被指数Perpendicular vegetation index （PVI）	$P V I = s i n ? 45 ° × B 8 - c o s ? 45 ° ? × B 4$
增强植被指数Enhanced vegetation index （EVI）	$E V I = 2.5 × B 8 - B 4 B 8 + 6 × B 4 - 7 × B 2 + 1$
增强植被指数-2 Enhanced vegetation index-2 （EVI2）	$E V I 2 = 2.5 × B 8 - B 4 B 8 + 2.4 × B 4 + 1$
多时相植被指数-2 Multitemporal vegetation index-2 （MTVI2）	$M T V I 2 = 1.5 × 1.2 × (B 8 - B 3) - 2.5 × (B 4 - B 3) (2 × B 8 + 1) 2 - 6 × (B 8 - 5 B 4) - 0.5$
归一化差值植被指数Normalized difference vegetation index （NDVI）	$N D V I = B 8 - B 4 B 8 ? + ? B 4$
再归一化差植被指数Renormalization the vegetation index （RDVI）	$R D V I = D V I × N D V I$
B4和B5归一化差值植被指数NDVI with band4 and band5 （NDVI45）	$N D V I 45 = B 5 - B 4 B 5 + B 4$
绿波归一化差值植被指数NDVI of green band （GNDVI）	$G N D V I = B 7 - B 3 B 7 + B 3$
反红边叶绿素指数Inverted red edge chlorophyll index （IRECI）	$I R E C I = B 6 × B 7 - B 4 B 5$
三角形植被指数Triangle vegetation index （TVI）	$T V I = 0.5 × [120 × B 6 - B 3 - 200 × B 4 - B 3]$
土壤调节植被指数Soil adjusted vegetation index （SAVI）	$S A V I = 1.5 × (B 8 - B 4) ∕ (B 8 + B 4 + 0.5)$
优化型土壤调节植被指数Optimized soil regulates vegetation index （OSAVI）	$O S A V I = (1 + 0.16) × (B 8 - B 4) B 8 + B 4 + 0.16$
转化土壤调节植被指数Transformed soil adjusted vegetation index （TSAVI）	$T S A V I = 0.5 × B 8 - 0.5 × B 4 - 0.2 0.5 × B 8 + B 4 + 0.1$
修正型土壤调节植被指数Modified soil adjusted vegetation index （MSAVI）	$M S A V I = (2 - N D V I × W D V I) × (B 8 - B 4) ∕ 8 × (B 8 + B 4 + 1 - N D V I × W D V I)$
二次修正型土壤调节植被指数Secondly modified soil adjusted vegetation index （MASVI2）	$M A S V I 2 = 0.5 × 2 × B 8 + 1 - 0.5 × 2 × B 8 + 1 2 - 8 × B 8 - B 4$
大气阻抗植被指数Atmospherically resistant vegetation index （ARVI）	$A R V I = B 8 - (2 × B 4 - B 2) ∕ B 8 + (2 × B 4 - B 2)$
叶绿素吸收比值指数Chlorophyll absorption ratio index （CARI）	CARI =（ $B 5 - B 4) - 0.2 × (B 5 - B 3)$
改进红边比值植被指数Modified red-edge simple ratio index （MSR）	$M S R = B 8 ∕ B 4 - 1 B 8 ∕ B 4 + 1$
特定色素简单比值植被指数Pigment specific simple ratio chlorophyll index （PSSRa）	$P S S R a = B 7 ∕ B 4$
Meris陆地叶绿素指数Meris terrestrial chlorophyll index （MTCI）	$M T C I = B 6 - B 5 B 5 - B 4$
修正型叶绿素吸收比值指数Modified chlorophyll absorption ratio index （MCARI）	$M C A R I = (B 5 - B 4) - 0.2 × B 5 - B 3 × (B 5 ∕ B 4)$
“哨兵2号”红边位置指数Sentinel-2 red edge position index （S2REP）	$S 2 R E P = 705 + 35 × [B 4 + B 7 / 2 - B 5] × (B 6 - B 5)$
红边感染点指数Red edge infection point index （REIP）	$R E I P = 700 + 40 × (B 4 + B 7) ∕ 2 - B 5 B 6 - B 5$
全球环境监测指数Global environmental monitoring index （GEMI）	$G E M I = e t a × 1 - 0.25 × e t a - (B 4 - 0.125) 1 - B 4$ ， $e t a = 2 × (B 8 A 2 - B 42) + 1.5 × B 8 A + 0.5 × B 4 B 8 A + B 4 + 0.5$

Table 3 Descriptive statistical analysis of grassland above ground biomass from 2019 to 2021 （n=325）

指标 Indicators	月份 Month
指标 Indicators	5	6	7	8	9	10
平均值 Mean value （kg·hm^-2）	716.411	1999.937	2701.109	3161.746	2926.367	2750.452
最大值 Maximum value （kg·hm^-2）	1220.000	4666.700	6160.600	5930.000	5509.200	4647.333
最小值 Minimum value （kg·hm^-2）	187.200	744.700	1368.700	1398.933	716.800	904.667
标准差 Standard deviation （kg·hm^-2）	240.516	909.195	732.878	998.728	891.322	1139.032
变异系数 Coefficient variable	0.336	0.455	0.271	0.316	0.305	0.414

Table 4 Spectral reflectance values of nature grassland in different months

月份 Month	项目 Item	波段Band
月份 Month	项目 Item	B1	B2	B3	B4	B5	B6	B7	B8	B8A	B9	B11	B12
5	Max	0.059	0.086	0.123	0.140	0.195	0.300	0.334	0.358	0.373	0.363	0.390	0.259
	Min	0.051	0.073	0.109	0.123	0.182	0.279	0.309	0.327	0.345	0.342	0.365	0.246
	Mean	0.055	0.080	0.117	0.131	0.187	0.287	0.319	0.340	0.356	0.350	0.377	0.253
	STD	0.00280	0.00286	0.00333	0.00404	0.00321	0.00454	0.00533	0.00671	0.00569	0.00407	0.00462	0.00338
6	Max	0.019	0.032	0.070	0.038	0.116	0.364	0.438	0.462	0.466	0.445	0.220	0.101
	Min	0.016	0.025	0.060	0.031	0.104	0.329	0.392	0.394	0.418	0.418	0.202	0.091
	Mean	0.018	0.029	0.066	0.034	0.109	0.346	0.411	0.425	0.438	0.429	0.211	0.096
	STD	0.00074	0.00175	0.00252	0.00151	0.00299	0.00939	0.01061	0.01247	0.01050	0.00730	0.00427	0.00245
7	Max	0.016	0.026	0.066	0.036	0.120	0.381	0.463	0.482	0.496	0.490	0.199	0.088
	Min	0.013	0.019	0.052	0.029	0.100	0.337	0.406	0.432	0.445	0.449	0.174	0.073
	Mean	0.014	0.022	0.059	0.030	0.109	0.364	0.441	0.458	0.474	0.466	0.184	0.078
	STD	0.00075	0.00191	0.00329	0.00242	0.00521	0.01266	0.01403	0.01331	0.01332	0.01069	0.00581	0.00381
8	Max	0.024	0.051	0.086	0.062	0.141	0.349	0.418	0.440	0.461	0.443	0.229	0.111
	Min	0.016	0.026	0.061	0.040	0.107	0.287	0.344	0.357	0.386	0.383	0.186	0.084
	Mean	0.019	0.036	0.072	0.049	0.124	0.319	0.381	0.397	0.423	0.410	0.208	0.098
	STD	0.00209	0.00559	0.00663	0.00563	0.00893	0.01754	0.02161	0.02279	0.02296	0.02074	0.01259	0.00730
9	Max	0.018	0.033	0.075	0.067	0.137	0.277	0.317	0.346	0.358	0.353	0.268	0.144
	Min	0.012	0.018	0.053	0.042	0.109	0.240	0.278	0.304	0.318	0.333	0.219	0.104
	Mean	0.014	0.024	0.062	0.052	0.121	0.262	0.303	0.331	0.343	0.343	0.239	0.121
	STD	0.00197	0.00475	0.00628	0.00688	0.00896	0.00907	0.00971	0.01111	0.01001	0.00651	0.01513	0.01224

Fig.2 Spectral reflectance curves of nature grassland in different months

Table 5 Results of important characteristic variables were selected by LASSO method

LASSO ID	变量Variables
LASSO 1	B2， B6， B11， IRECI， PSSRa
LASSO 2	B2， B8A， B11， EVI， TVI， REP， CARI， IRECI， PSSRa
LASSO 3	B2， B11， EVI， CARI， IRECI， PSSRa
LASSO 4	B2， B8A， B11， EVI， TVI， B11， IRECI， PSSRa
LASSO 5	B2， B6， B11， EVI， IRECI， PSSRa

Table 6 Model accuracy evaluation of five combinations of LASSO variables

LASSO ID	模型 Models	训练集 Training set		验证集 Validation set
LASSO ID	模型 Models	T-R²	T-RMSE （kg·hm^-2）	V-R²	V-RMSE （kg·hm^-2）
LASSO 1	RF	0.93	316.937	0.72	622.616
	SVM	0.69	656.276	0.66	698.271
	ANN	0.73	601.837	0.63	730.676
LASSO 2	RF	0.95	284.129	0.72	624.594
	SVM	0.71	638.798	0.68	685.008
	ANN	0.83	477.670	0.67	686.136
LASSO 3	RF	0.94	296.973	0.71	625.002
	SVM	0.69	655.849	0.66	691.631
	ANN	0.78	546.190	0.66	686.613
LASSO 4	RF	0.94	295.179	0.71	627.529
	SVM	0.71	642.576	0.67	681.942
	ANN	0.81	501.278	0.67	695.866
LASSO 5	RF	0.94	294.397	0.72	632.793
	SVM	0.69	659.055	0.66	691.451
	ANN	0.76	561.079	0.63	729.430

Fig.3 The validation results by comparing AGB predicted values and observed values based on machine learning model

Fig.4 Importance ranking of variables

Fig.5 Spatial distribution of AGB of nature grassland in 2019， 2020 and 2021 in Menyuan County

Table 7 Nature grassland biomass in Menyuan County from 2019 to 2021

年份

Year

最小值

Minimum value （kg·hm^-2）

最大值

Maximum value （kg·hm^-2）

均值

Mean value （kg·hm^-2）

标准差

Standard deviation （kg·hm^-2）

总和

Sum （ $×$ 10⁶ t）