A primary study on crop
production prediction using global vegetation index
Xuemei Bai and Shurji Murai
Institute of Industrial Science University of Tokyo 7-22
Roppongi, Minato-ku, Tokyo 106 Japan
Abstract NOAA GVI (Global Vegetation Index)
data have been considered as the index of the amount of chlorophyll of
green biomass of land cover. In this study, a primary model for crop
production prediction was developed by using the NOAA GVI data, weather
data and other geographical data. In this model, the GVI volume, which is
the product of the summation of GVI value of a certain time duration and
the area, was used as the index of green biomass. Study area was divided
into several subareas, and the relationship between GVI volume and
production of these subareas every year was approximated by a straight
line. The coefficients of the linear function was determined by the
weather data. Once the linear function is determined, the crop production
of the study are can be predicted by adding up the estimated crop
production of every subarea. The GVI data sand weather data needed were
both only from June to August. Huagn-Huai River catchment area were chosen
as study area including six provinces (Hebei, Shanxi, Shanxi, Gansu,
Shandong, Henan) and relatively satisfactory results were obtained.
Introduction Recently, research on crop production
prediction using satellite remote sensing data has become very important.
Land sat data is most widely used for this purpose. However, for large
scale prediction, NOAA GVI data have better characteristics compared with
other satellite data. Previous studies by the authors showed that there
exist strong relationships between NOAA GVI data, weather data to predict
crop production. The study area includes six provinces like Hebei, Henal
Shanxi1, Shanxi2, Gansu, Shandong, which is located in Huang-Guai River
catchments area, in the central part of china and inside the study area
the combination of cultivating crop is correlation between GVI and crop
production with all the correlation coefficient over 90% were obtained.
The weather affections on the annual change of this linear relationship
were studied and a preliminary prediction model was developed.
Brief description of data
- Monthly maximum value of GVI (global Vegetation
Index):
The original data are hemispheric pairs of Polar
-stereographic arrays where each hemispheric arrya is 1024 by 1024
pixels in size with weekly maximum value of GVI all over the world (1).
From original images, monthly maximum value component of GVI were made
firstly, and resampled to new images which cover the areas from
longitude 70 to 140 degree East and from latitude 10 to 70 degree north
by the size of 512 by 480 pixels. Longitude latitude projection were
used for the convenience of are calculation.
- Monthly average value of temperature, total rainfall of every
month:
The weather data were provided by the Meteorological
Agency of Japan. From all 2 thousand weather observation stations all
over the world, those stations that located in study area ere chosen and
the rainfall and temperature data were studies. The average temperature
and rainfall were calculated for every province by taking the average
value of the data of weather stations that located in that province.
This calculation was carried out for different time durations like April
to August, June to August and January to December.
- Other Geographical Data
- Crop production data of every province in study area: The crop
production data in terms of the total crop production of every
province of the study are from 1983 to 1987 were used.
- Cultivated area data of every province: the ctual cultivated area
data from983 to 1987 were used.
- Map of river system in China: Political boundary map of china was
scanned geometrically corrected, resampled to the same size as new
images and overlaid with them to decide the study area boundary on
images.
Prediction Model
Derivation Mathematically, if these is an arrays [x1, y1], [x2,
y2],
[xn. Yn], the regression line is y= ax+b, then the original y
and regressive value y' have the relationship: y1 + y2+
..+yn = y' 1+2'
2+
.=y' n. Using this theory, the study area was divided into several
subareas, and the crop production of study area could be estimated by
predicting the crop production of every subarea and add them up if only
the regression line for cop production can be decided.
- GVI and Crop Production
Using the average value of GVI
from July to September, the study area were classified into 4 categories
such as forest, agricultural area, grass land, water or desert by using
the threshold classification method. For those agricultural areas of
every province, the GVI volume of different time duration were
calculated, which can be expressed as follows : GVIVLMK: SIGMA (AREA)
(SIGMA (TIME) PIXGVI) * PIXAREA
.(1) Where, GVIGLM: GVI volume of
agricultural area of certain time duration. PIXGVI: the monthly maximum
value of GVI of a certain agricultural pixel. PIXAREA; the area of the
same pixels. GVI data were usually considered as to represent the
greenness of ground vegetation cover, therefore the GVI volume should be
able to represent the amount go green biomass of ground vegetation
cover, and the crop production should be related to the amount of green
biomass in agricultural area. Since the original data had low resolution
and the classification were carried out without ground truth data, the
result of classification can show the potential agricultural area but it
can't represent the accurate cultivated area. In order to get better
result, the GVI volume of agricultural are calculated former was
calibrated using the area ratio between classified agricultural area and
real cultivated area. GVIVLM (MIDIF): Calibrated GVI Volume AREAR:
Cultivated area AREAC: Classified agricultural area The relationship
between the calibrated GVI volume of different time duration and crop
production of six province in the study area were studies and it was
found that the crop production was highly related to the GVI volume of
different time duration with all the correlation coefficient more than
90%. Fig. 1 shoes the relationships between the GVI volume from April to
August and the crop production of that year.
- Weather Affection
From Fig. 1 it can be seen that the
crop production has direct ratio with the calibrated GVI volume but the
slop and segment of the linear equation for every year are different.
Fig. 2 and Fig. 3 show the annual change of these tow values. If the
straight line of GVI and crop production for certain year can be
decided, the crop production of that year can be obtained. The weather
affection was considered as the most important factors to decide the
linear equation.
Average temperature, rainfall and aridity of
different time duration were calculatied on province level. In order to
reflect the range of affection of weather data, weight method was used.
The weight of a province was decided by dividing the cultivating area of
all six provinces. Table 1 shows the weight values can be used for
prediction even without the cultivating area data of that year. The
average temperature, rainfull and aridity of every province were
multiplied by the weight of that province and added up to be the average
rainfall and temperature of the study area of that year, and the
relationship between these weighted weather factors and the coefficient
of the crop-GVI linear function were studied. As the results, it has
been fount that the weighted average temperature from June to August
highly related both to the slope and segment with correlation
coefficient 94% and 80% ( see Fig. 4 and Fig. 5 ). Therefore, It's
possible to decide the crop-GVI equation by observing temperature from
June to August.
- Prediction Processing
Using the results obtained up to
now, the crop production prediction of the study ara can be carried out
as follows:
Step 1: Collecting weather data from June to
August and using the weight value in Table1, calculated the average
temperature of this time duration, and decide the crop- GVI equation
according to following equations:
SLOPE = 0.03623 * t -
0.06988------------------------(3)
SEGMENT = -127.6 * t +
2701.6------------------------(4) where,
t : weighted
temperature
The crop-GVI equation is:
Crop = SLOPE * GVIVLM + SEGMENT------------------(5)
Step 2: Calculate the GVI volume using equation (1)
from June to August for every province and using the ratio of
cultivating area and agricultural area, calibrate the calculated GVI
volume using equation. (2)
Step 3: Add up the caliberated
GVI volume of every province and introduce it to equation (5) to get the
crop production prediction value of study area.
Conclusions Fig. 6 shows the real and estimated value
of crop production of study area. The average accuracy is approximately
95%.
The results obtained upto now
show the possibility of using NOAA GVI data and weather data for large
scale crop production prediction. Study area can be divided into several
subareas and by deciding the regression line for crop production to GVI
volume, the crop production of every subarea can be obtained. The
summation of these predicted value forms the prediction value of the study
area. The regression line can be determined by weather factors. The study
should be improved in following aspects 1) the possibility of earlier
prediction should be studies 2) Find out better correlationships which was
used to decide the GVI crop regression line.
References
- Satellite Data Division, USA, Global Vegetarian Index User's
guide
- Xuemei Bai and Shunji Murai, (1989), "Regional Environment
Evaluation of Changjiang and Huang - Huai River Catchment Area in China
using global vegetation index and weather data" the 10th Asian
conference on Remote Sensing, Nov, 23-29 1989, Kualalumpur,
Malaysia.
- Shunji Murai and Xuemei Bai, (1990), Habitability anaysis in China
Using Global Vegetation Index, The 2nd GIS workshop, Beijing, Aug. 6-8
1990, Beijing
China.
|