The study on biomass estimation in Mongolian grassland using satellite data and field measurement data

Hirokazu Yamamoto, Koji Kajiwara, and Yoshiaki Honda
Center for Environmental Remote Sensing, Chiba University.
1-33, Yayoi-cho, Inageku, Chiba, 263-8522, Japan
Tel : +81-43-290-3835 Fax : +81-43-290-3835
E-mail : [email protected]

Abstract:
It is said that earth environment is affected by changing vegetation biomass. So, it is needed the analysis of physical parameter of vegetation in the wide area by using satellite data, which has middle spatial resolution (approx. 1km2). In this study, biomass is estimated on a large scale using NOAA AVHRR LAC data received in Mongolia and field measurement data.

As the result, we found that it is able to estimate vegetation biomass without influence of cloud contamination and vegetation growth in case of more than 5 days composite of NOAA AVHRR LAC data. And we could estimate biomass in the wide area until less than 10% error.

1. Introduction
Carbon is one of the most important element on the earth, and it can become with key word of the mechanism of earth fluctuation. Also, it is said that vegetation has the important role of the carbon circulation of biosphere-lithosphere-atmosphere. Therefore, it is needed for environment monitoring to understand plant productivity globally.. Many research have done to the estimation of vegetation amount using satellite data which has characteristics of high frequency and middle spatial resolution such as NOAA AVHRR in recent years.

EOS-AM1(Terra) satellite has launched successfully, and MODIS is expected to the understanding of global dynamics. Japan also has plan of launch satellite, ADEOS-II GLI, ADEOS -III SGLI, but there are no methodology of calibration and validation for the above mentioned middle resolution satellite. We already tried the acquisition of various kinds of ground based data in Mongolian grassland in comparison with the satellite data scale, and we succeeded in estimating the relation between vegetation index and vegetation cover ratio (VCR) obtained by the mobile measurement system (Yamamoto et al.,1997). However, actually, such these data have not been applied to satellite data using the model obtained from field measurement data.

2. Objective
The aim of this study is the following items.

• To construct of biomass estimation model which can be calibrated from the relationship between vegetation index – vegetation cover ratio – biomass using the field experiment data
• To apply the field data to satellite data.
• To evaluate the derived satellite biomass estimation by using meteorological station data in Mongolia.

3. Method

3.1 The Study Area and Data
The study area of field measurement and satellite data is Mongolian grassland. Figure.1 shows the range of this study area of field measurement. Field measurement data we used in this study was acquired in summer from 1995 to 1998(Table.1). NOAA AVHRR LAC data received at Ulaanbaatar in Mongolia was used as satellite data. Table.2 shows acquisition date and time of AVHRR data used in this study.

3.2 Analysis of Field Measurement Data
The mobile measurement system is quick measurement system for acquisition of the various kinds of ground information over a large area such as the scale of satellite data. Sensor height from the ground is approximately 2 m. Optical axis of this system is adjusted to become parallel and being fixed to always measure nadir. Spectrometer measures in the visible and near-infrared region (350nm- 1050nm) with 512 channels. Observation item is spectral reflectance and ground digital image for vegetation cover ratio. In this study, vegetation cover ration is defined as the proportion of the green plant that is viewed in nadir direction from the sensor to ground surface.

The biomass measurement is cutting all the grass (green leaf) inside of 1m2 area with the hand, and wet grass weight and dry grass weight by using drying machine for 8.5 hours is measured. And this system is also able to acquire nadir ground digital image, spectral information, and grass height. System is designed the same as mobile measurement system. In this study, 58 good biomass measurement points obtained from August 3 to August 9, 1998 in Site.1 were analyzed. And 5 points data had been obtained in August 10, 1998 in Site.2 and 3 points data had been obtained in August 11 in Site.3. Observation items are spectral reflectance, ground images, and biomass(wet and dry grass weight, and grass height). In this study, biomass is defined as the weight of sufficient dried up grass which is green and fresh matter on the ground.

Calculation of vegetation coverage ratio is derived from digital images using RGB pattern method and hue and intensity method (Yamamoto et al.,1997).

Figure.1 The study Area

Table.1 Acquisition date and time of field measurement data in this study

Table.2 NOAA AVHRR LAC in this study

3.3 Satellite Data Processing
In this study, NOAA-14 AVHRR LAC data received at Ulaanbaatar in Mongolia was used. It is able to calculate reflectance from NOAA AVHRR LAC data ( Mather,1998). The accuracy of the reflectance is about 0.1%. NOAA/NESDIS (NOAA’s National Environmental Satellite Data and Information Service) is opened for acquisition of the latest information of gain and offset for calibration. It is said that calibration coefficient is changeable for sensor deterioration, and so on ( Brest et al., 1992; Rao et al.,1994). In this study, the latest gain and offset was used every date of acquisition.

Geometric correction was conducted using automatic acquisition of GCPs. This method can generate GCP data automatically by matching the coastline data ( Shimoda et al.,1998, Hashimoto et al..,1993). In this study, residuals on image is less than 0.4 pixels, and map projection is Latitude/Longitude Grid system (WGS84) and resampling method is nearest neighbor.

Atmospheric correction was conducted by 6S code, which is focused on visible and near infra-red region ( Vermote et al., 1997 ). In this study, LUT method (Lei et al.,1998) was used. At first, LUT is calculated for each Sun-Target-Sensor geometry condition. GTOPO30 which has approx. 1km resolution is used as height above sea level . And high reliable data (AeroNet) is used as optical thickness (Optical thickness @550nm was re-calculated.), which was measured at Mandalgobi in Aug. of 1998. These data are obtained every day, and it is varied from 0.036 to 0.071. Therefore, in this study, 0.071 is used for 6Scode. Table.3 shows input parameters for 6S code.

Table.3 Input Parameters for 6S

3.4 Analysis of Meteorological Station data
15 meteorological stations are located at flat and homogeneous grassland in eastern part of Mongolia (Figure.2). Acquisition date is from the beginning of June to the end of September every 10 days. Observation items are ground image by photographs, wet and dry grass weight, and grass height. Cutting area is 1m ×1m, and cutting method is almost same as field biomass measurement. Dry weight is the grass weight dried by sun for a few days. In this study, meteorological station data was used as evaluation of estimated biomass in the wide area and construction of VCR-Biomass model.

In model construction, using 4 corners of the clipping area as control point, affine transformation method was conducted to the non-vertical digitized photograph. After this process, the RGB Pattern method and Hue and Intensity method were conducted to the image. 270 good points of 420 measurement points were used as VCR data. And then VCR-Biomass model was constructed by simple linear regression between this VCR data and biomass data of meteorological station.

In evaluation of estimated biomass, as above-mentioned, corrected reflectance of satellite sensor is calculated each channels, therefore NDVI is derived from these reflectance data. Moreover, biomass map in the wide area can be calculated by using biomass estimation model derived from field experiment data. The following equation was used as accuracy of the estimated biomass:

e = (T-M)/T×100
e: Relative Error
M: Estimated Biomass
T: Meteorological Station Data (15 Stations) … (1)

Moreover, the estimation biomass was evaluated every composite period. The influence of the plant growth for composite period was also evaluated.

Figure.2 The Distribution of 15 Meteorological Stations in Mongolia

4. Result

4.1 Biomass Estimation Model Using Field data and Meteorological Station Data
Figure.3 depicts example of averaged spectral reflectance obtained by mobile measurement. It is clearly that enough amount of data was collected in the wide area, and it can be said the representative reflectance on a satellite scale. NDVI was calculated from convolved reflectance to NOAA AVHRR spectral resolution. The left side of Figure.4 shows relationship between NDVI and vegetation cover ratio (VCR). It is good correlation between VCR and NDVI. Therefore they are able to be constructed NDVI-VCR model on a satellite scale. This model can be obtained by simple regression analysis, and it expressed as the following formula:

VCR = 0.9375 NDVI-0.0830 (R = 0.847)
VCR : Vegetation Cover Ratio …(2)

Figure.3 Example of Averaged Spectral Reflectance obtained by Mobile Measurement

Figure.4 Relation between NDVI and VCR obtained by Mobile Measurement(Up)
Relation between VCR and Biomass obtained by Field Measurement and Meteorological Station(Down)
Moreover, the following simple linear regression equation is expressed with vegetation cover ratio (VCR) and biomass :
DW = 135.98 VCR+3.1703 (R = 0.723)
DW : Biomass (g/m2)
VCR : Vegetation Cover Ratio (VCR) … (3)

In case of using meteorological station data:
DW = 115.47 VCR+5.964 (R = 0.723)
DW : Biomass (g/m2)
VCR : Vegetation Cover Ratio (VCR) … (4)

Using formula (2) and (3), biomass estimation model by simple linear regression of field measurement data is expressed as:
DW = 127.481 NDVI-8.116
DW: Biomass(g/m2) … (5)

Similarly, biomass estimation model using meteorological station data is expressed derived from formula (2) and (4):
DW = 108.25 NDVI-3.620
DW : Biomass(g/m2) … (6)

Figure.5(i) Biomass estimation using 5days composite NOAA AVHRR LAC(1996)

Figure.5(ii)Biomass estimation using 8days composite NOAA AVHRR LAC(1998)
4.2 Evaluation of Estimated Biomass for Composite Period
Therefore, estimation biomass in the whole Mongolian grassland is made by equation (5) or (6). Figure.5 depicts estimated biomass derived from model of field measurement data.

Figure.5(i) is the result of 5 days composite of 1996 and figure.5(ii) shows 8 days composite of 1998.Figure.6(i) shows the relative errors of the average of all 15 meteorological stations for 5 days composite of 1996. The influence of plant growth is not appearing with the 5th days when composite is completed.Figure.6(ii) is the relative errors of the average of all 15 meteorological station composite for 1998 year 8 of days. Relative error becomes 0% to the 4th days ,and after the 5th days, it becomes to negative. Therefore, it is clear that composite data can be reflected the vegetation quantity without the influence of the plant growth, if cloud cover removal is completed with 5 days composite. This result shows that 10 days composite data and monthly composite data which are used generally may not reflect actual vegetation quantity.

(i)

(ii)

Figure.6 Averaged Relative Error versus Composite Period of NOAA AVHRR LAC data((i) is 1996’s, (ii) is 1998’s)
4.3 Biomass Estimation Using Field Data and Satellite Data
The evaluation of the relative error was carried out only by using 5 days composite data (1996) in this research, because it was confirmed that 1998 composite data has the influence of plant growth.

Figure.7(i) shows the error of the estimated biomass by field measurement model on the map. The relative error is 0.45%, and extremely high accuracy with Mandalgobi. Mandalgobi meteorological station may be periphery close to field measurement site. However, a relative error is growing up in the place which is distant from Mandalgobi.

(i)

(ii)

Figure. 7 Relative error of estimated biomass using 5 days composite data of 1996 (Filed Measurement Model, figure (i))
Relative error of estimated biomass using 5 days composite data of 1996 (Meteorological Station data Model, figure (ii))
4.4 Biomass Estimation Using Meteorological Station Data and Satellite Data
Figure.7(ii) shows the relative error of estimated biomass of 5 days composite data of 1996 using meteorological station data model on the map. The accuracy is -1.92% in Mandalgobi. The accuracy is less than biomass estimated by field measurement model, but Ulaanbaatar are -3.32%. Relative error at the place among Ulaanbaatar and Mandalgobi. is within about 10 %.

5. Conclusions
In this study, we estimated biomass in the wide area using field measurement data and satellite data, NOAA AVHRR LAC data., and evaluated the estimated biomass using meteorological station data. As the result, it is able to conclude the following items.

It is developed the methodology of field experiment strategy (Mobile Measurement, Biomass Measurement) for footprint of satellite sensor scale, and NDVI-Biomass model is constructed with vegetation cover.

Both of field measurement data and meteorological station data can be applied to NOAA AVHRR LAC data, and it can estimate biomass on a large scale.

It is able to estimate biomass in the wide area by using NOAA AVHRR LAC data and field data with high accuracy. The error ratio of biomass estimation is less than approx. 10% between Ulaanbataar and Mandalgobi. Moreover, LAC data should be less than 5 days composite, because more than 5 days composite data has capability of influence of plant growth in the Mongolian grassland.

Acknowledgments
This work has been supported by CREST (Core Research for Evolutional Science and Technology) of Japan Science and Technology Corporation (JST). And we are very grateful to many staffs of Dundgobi local government and The Information and Computer Center in Mongolia.

References

• Hashimoto, T. and Murai., S., Geometric Correction of NOAA AVHRR Imagery in accordance with the number of GCPs., Journal of the Japan Society of Photogrammetry and Remote Sensing, Vol.36, No., 1993.(in Japanese)
• Honda, Y., Kajiwara, K., Yamamoto, H., Hashimoto, T., Takagi, M., and Takamura, T., Project for Establishment of Plant Production Estimation Using Remote Sensing, Proceeding of The International Symposium on Global Change in East Asia – Vegetation Monitoring -, p.113-127, 1998.
• Lei, L., Koide, M., Iikura, Y., and Yokoyama, R., Correction of atmospheric effects on AVHRR imagery by 6S code, The Remote Sensing Society of Japan ,Journal of The Remote Sensing Society of Japan,Vol. 18, No. 2, p.51-64, 1998.
• Mather, P. M., Computer Processing of Remotely-Sensed Images, NOAA Polar Orbiter Data (POD) User’s Guide, November 1998 revision Section 3.3, p.202-206, 1998.
• Shimoda, H., Fukue, K., Cho, K., Matsuoka, R., Hashimoto, T., Nemote, T., Tokunaga, M., Tanba, S., and Takagi, M., Development of a Software Package for ADEOS and NOAA Data Analysis, Proceedings of the IGARSS’98, Vol.2, Int03, ISBN, 0-7803-4406-5, 1998.
• Tucker, C. J., Red and Photographic Infrared Linear Combinations for Monitoring Vegetation, Remote Sensing of Environment., Vol. 8, p.127-150, 1979.
• Vermote, E. F., Tanre, D., Herman, M., and Morcrette, J. J., Second Simulation of the Satellite Signal in the Solar Spectrum 6S An over view, IEEE Trans. Geosci. Remote Sensing, Vol. 35, No. 3, p.657-686, 1997.
• Yamamoto, H.,Kajiwara, K., Honda, Y.:The Estimation of Vegetation Coverage Ratio by Car Survey — The Experience in Mongolian Grassland —, Proceeding of the Symposium of the Japan Society of Photogrammetry and Remote Sensing,241-244, 1996.(in Japanese)
• Yamamoto, H., Honda, Y., Kajiwara, K., Okano, C., Huang, S. and Matsuoka, .:The Analysis of Field Data Measured in Grassland for Biomass Monitoring Using Satellite Data, Journal of the Japan Society of Photogrammetry and Remote Sensing, Vol.36, No.4, 1997.(in Japanese)