Kogan1990.pdf

International Journal Of Remote Sensing

ISSN: 0143-1161 (Print) 1366-5901 (Online) Journal homepage: http://www.tandfonline.com/loi/tres20

Remote sensing of weather impacts on vegetation in non-homogeneous areas F. N. KOGAN To cite this article: F. N. KOGAN (1990) Remote sensing of weather impacts on vegetation in non-homogeneous areas, International Journal Of Remote Sensing, 11:8, 1405-1419, DOI: 10.1080/01431169008955102 To link to this article: https://doi.org/10.1080/01431169008955102

Published online: 27 Apr 2007.

Submit your article to this journal

Article views: 278

Citing articles: 248 View citing articles

Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=tres20

INT.

J.

REMOTE SENSING, 1990, VOL II, No.8, 1405-1419

Remote sensing of weather impacts on vegetation in non-homogeneous areas F. N. KOGAN National Environmental Satellite Data and Information Service, NOAA, Washington, OC 20233, U.S.A.

(Received 24 June 1988; infinalform 23 October 1989) Abstract. Successful application of the normalized difference vegetation index (NOVI) for estimating weather impacts on vegetation is currently hindered in nonhomogeneous areas. The problem is that the differences between the level of vegetation in these areas can be related, in addition to weather impacts, to the differences in geographic resources (climate, soil, vegetation type and topography). These differences should be eliminated when weather impacts on vegetation are estimated from NOVI data. This paper discusses a concept and a technique for eliminating that portion of the NOVI which is related to the contribution of geographic resources to the amount of vegetation. The Advanced Very High Resolution Radiometer (AVHRR) data of the Global Vegetation Index format were used for the 1984--1987 seasons in Sudan. The procedure suggests normalization of NOVI values relative to the absolute maximum and the absolute minimum of NOV\. These two criteria were shown to be an appropriate characteristic of geographic resources of an area. The modified NOVI was named the Vegetation Condition Index (VCI). Comparison between VCI, NOVI and precipitation dynamics showed that the VCI estimates better portray precipitation dynamics as compared to the NOV\. The VCI permits not only the desciption of vegetation but also estimation of spatial and temporal vegetation changes and weather impacts on vegetation.

Introduction Major progress in using NOAA Advanced Very High Resolution Radiometer (AVHRR) satellite data for environmental resource monitoring has been achieved since the early 1980s. Specifically, vegetation indices have been developed to quantify different environmental and ecological events. The Normalized Difference Vegetation Index (NDVI) has become the primary tool for description of vegetation changes and interpretation of the impacts of environmental phenomena. Extensive research has shown that the NDVI can be used not only for accurate description of continental land cover, vegetation classification and vegetation phenology (Tucker et al. 1982, Tarpley et al. 1984, Justice et al. 1985) but it is also effective for monitoring rainfall and drought, estimating net primary production of vegetation and crop yields, detecting weather impacts and other events important for agriculture, ecology and economics (Tucker et al. 1985, Prince and Tucker 1986, Hielkema et al. 1986, Malingreau 1986, Justice et al. 1985, 1986, van Dijk 1986, Kogan 1987a). All of this successful developmental work has inspired national and international institutions, relief organizations and private firms to apply accumulated knowledge to real-time monitoring of adverse environmental events, vegetation conditions and even crop and pasture yields (Philipson and Teng 1988, Walsh 1988, Le Comte et al. 1988). Several years in this endeavour have shown that the NDVI has excellent 1.

0143-1161/90 $3.00

© 1990 Taylor & Francis Ltd

1406

F. N. Kogan

potential for determining the impact of environment on vegetation and can be successfully used in early identification of weather-related food shortages. However, this extensive operational and developmental experience has revealed some limitations of the NDVJ. It has been used successfully to identify stressed and damaged crops and pastures but interpretive problems can arise when these results are extrapolated over non-homogeneous areas. In these areas differences between the level of vegetation can be related to differences in environmental resources (climate, soil, vegetation, relief). For example, under similar weather conditions, a region with abundant resources had NDvr values twice as large as those in adjacent regions with insufficient resources (Kogan 1987a). Thus, the application of the NDVI for spatial vegetation analysis, especially for the assessment of weather impacts on vegetation in non-homogeneous areas, requires stratifying the NDVI values to eliminate the differences in vegetation related to specific environmental and economic conditions. This paper discusses the concept of NDVI stratification and some applications of the modified NDVI for more accurate monitoring of vegetation in non-homogeneous areas. The case study was Sudan. The available NDVI data for this country were adjusted to take into account mostly geographic differences among locations. The modified NDVI values were compared with the non-modified values, as well as with the rainfall data. The results were analysed in relation to estimation of vegetation conditions, assessment of weather impacts and detection of large-scale weather patterns controlling short and medium-term changes in phytocenoses. 2.

Standard procedures for AVHRR data preparation This study used AVHRR data in the Global Vegetation Index (GVI) format for the period January 1984 to December 1987. The GVI is a standard NOAA product resulting from spatial and temporal sampling and processing of the Global Area Coverage (GAC) satellite data (Tarpley et al. 1984, NOAA 1986). The NDVI was calculated on a daily basis from 8-bit Ch I (visible) and Ch2 (near infrared) values using the formula: (Ch2-Chl)j(Chl +Ch2). The temporal sampling consisted of retaining the maximum NDVI value for each pixel and 7 day period. This maximum value has been least contaminated by atmospheric interference and distortions of viewing and illumination geometry (Holben 1986). In producing the GVI data, spatial sampling reduces the resolution of GAC data to approximately 15 km 2 at the equator. Nevertheless, these data contain useful information about vegetation change in relation to environmental impact (Malingreau 1986, Townshend and Justice 1986). Additional spatial sampling consisted of aggregating GVI pixels to larger size areas. For this purpose, Sudan was latitude (figure I). The area divided into rectangles (grid cells) of I ° longitude by of each grid cell was approximately 6100 km '. Average NDVI values were calculated from the 22 to 25 GVI pixels contained in each grid cell. These techniques considerably reduce noise in the surface reflection signal but data can be still contaminated. The principal source of noise is persistent cloudiness, especially over tropical and even sub-tropical areas during the rainy season (Justice et al. 1985, Malingreau 1986, van Dijk 1986). The other sorces of noise are related to specific situations determined by the location of Sun and sensor, surface directional reflectance, type of satellite and random disturbances (Duggin et al. 1984, Gutman 1987, Gallo and Eidenshink 1988). These factors normally reduce the NDVI, although in some circumstances they may increase it (Gutman 1987). In the

to

Remote sensing of weather impacts on vegetation

1407

~, • 2

1016~~5 1 ~

.

3= 1=~7

:0;8 -;::::

4"

• ...kba

•

..

20

32

3.

Figure I. Map of Sudan showing grid cells (1 longitude by to latitude), location of weather stations (e) and studied grids(number). 0

latter case the increased values would be retained as required by the maximum value composite technique. Both reduced and increased NOVI values can be erroneously interpreted. To minimize the errors, additional smoothing of data is usually performed. This includes temporal aggregation of data to represent longer time intervals and smoothing by application of the running average technique (Townshend and Justice 1986, Malingreau 1986, Justice et al. 1986, van Oijk et al. 1987). Compositing weekly NOVI over longer time intervals (2 to 4 weeks) may be undesirable because the phenology of vegetation and short-term changes in vegetation conditions and environment are unlikely to be identified. The running average technique has proved to be more appropriate because this technique not only smooths the data, filters it and identifies time trends but also retains weekly NOVI values that are important for efficient vegetation monitoring. Of all the approaches, median filtering proved to be superior to others because it entirely ignores a single outlier in the time series but does preserve the general pattern of vegetation dynamics (van Oijk 1986, van Oijk et al. 1987). Over the period from the first week in January through to the last week in December, the weekly NOV) data were smoothed with a median filter for each year. The median smoothing removed any single outlier but a seasonal trend in the weekly NOVI time-series and fluctuations related to weather variability became apparent. The resulting product has proven to be effective for vegetation monitoring (van Oijk 1986). However, NOVI values over geographically non-homogeneous areas with different vegetation levels can be interpreted erroneously (Kogan 1987a). This is especially important in analyses which require comparison of vegetation dynamics and weather impact for two or more locations. Application of the NOVI for vegetation monitoring in this case requires geographic filtering of the data.

1408

3.

F. N. Kogan

Geographic filtering of NDVI The geographic filtering procedure suggested in this paper eliminates that portion of the NOVIs spatial variability which is related to the contribution of geographic resources to the amount of vegetation. This contribution fluctuates considerably, depending mainly on climate, soils, vegetation type and topography of an area. For example, in tropical areas high NOVI values result from the lush tropical forest vegetation, whereas in deserts low NOVI values are to be expected. Obviously these difTernces are not due to the impact of the weather. For purposes of estimating weather impact, they must be removed. A simple approach is to estimate NOVI values relative to the historical averages for each location. Unfortunately, this technique cannot be applied because AVHRR records are insufficient for determining reliable average NOVI values. To overcome this limitation, it was suggested that NOVI values should be estimated relative to their historical maximum. This maximum can be determined from limited historical records which contain at least 1 year of data with favourable weather impacts on vegetation. It is possible to use two types of maximum: absolute historical maximum which is determined as the highest seasonal NOVI value for several seasons (Kogan 1987 a) and weekly maximum composited from the highest weekly value of several years of NDV] data (Kogan 1987 b). The latter is more appropriate for assessment of weather impacts, although lack of satellite data limits currently the use of this criterion. In this research the absolute historical maximum was used. This historical maximum was determined from the 1985 NOVI records and, partially, 1986 and 1987 records. Examination of this maximum showed that its spatial distribution imitates fairly well spatial distribution of climatic, soil and vegetation zones (Kogan 1987 a). In addition, as was recently discovered, the absolute maximum NOV I values correlate strongly with May-September precipitation normals. In view of these facts the established historical maximum in Sudan can be interpreted as an integrated indicator of geographic resources. To remove the contribution of geographic resources from the NOVI, this index is normalized relative to its absolute historical maximum. This principle is illustrated schematically in figure 2. Consider two fairly similar seasonal NOVI curves CI and C2 recorded in neighbouring regions I and 2. Assume that for a particular week the NOVI value for each of these regions is equal to A, i.e. assume the curves intersect at value A. The lines MAXI and MAX2 in figure 2 represent the absolute historical maximum NOVI values for regions I and 2 respectively. In the context of NOVI interpretation, the equality at A does not necessarily indicate similarity of vegetation conditions or equality of weather impacts. The observed NOVI value A is closer to the absolute historical maximum MAX2 (AB 2 ) for region 2 than to the absolute historical maximum MAXI (AB!) for region!. This suggests that vegetation condition in region 2 was better than in region I due to more favourable cumulative weather impacts. Normalization of the NOVI value relative to the regional absolute historical maximum would highlight this fact. Our operational work with interpretation of NOVI data also suggests that in areas with clearly defined seasonal dynamics of vegetation it might be useful to take into account the minimum vegetation level as well. This minimum can be considered as an initial threshold from which the vegetation starts to grow. As with the maximum concept the difference between the observed NOVI and this threshold value can characterize the intensity of vegetation growth. In this context, the larger the difference, the more intensive the growth is likely to be. As seen in figure 2, this

Remote sensing of weather impacts on vegetation

1409

maxi

8 1 •, I

, ,

, : B2,r-

-

-

max2

_.- -

-

-

- --

, ,..-- .....

so z

';J'

"

-

-

-

~ - -

"

'C2

,/ ~

CI

'" v2

- or .,.,...

mine:

Week Figure 2. Diagram explaining a principle of geographic filtering of NOV! data.

difference was slightly greater in region I (AC ,) than in region 2 (AC 2 ) . Accordingly, the vegetation growth in region I is estimated to be more intensive than in region 2. The absolute minimum NOVI for Sudan is less than 0·05, which is normally considered the threshold 'soil line' (Hiernaux and Justice 1986, Malingreau 1986). However, it is important to emphasize that in Sudan the minimum value of the NOVI, between 0·05 and 0'025, represents vegetation. The 0·025 threshold was selected as a 'stratification' value for bare soil, based on NOAA-7 satellite data (Holben 1986). In Sudan the 0·025 isoline, plotted from the complete set of historical data, mimics the topography (figure 3). It effectively outlines the areas above 500 metres in altitude, which do exhibit at least a minimal amount of vegetation early in the season. In addition to the large-scale pattern discussed in the preceding paragraph a closer analysis reveals a number of small-scale patterns in the distribution of absolute minimum NOV!. For example, increased minimum values in south-central Sudan (0'02 to 0,05) are produced by the swampland prevailing over this area; in far southwestern and south-eastern Sudan high (0,04-0'06) absolute minimum values similarly coincide with savannah woodland vegetation and enhanced precipitation levels (above 1000 mm). The following expression determines the required adjustment of the NOVI values for a given set of geographically determined criteria. This adjustment provides additional filtering, thereby enhancing the weather-related signal in NDVI values. IOO(NDVli j - NOVI- MIN;) VCI;j=NOVI _ MAX-NDVI-MIN. J

J

where VCI is the modified NOVI, or Vegetation Condition Index (expressed in per cent); NOVI is the smoothed weekly composite Normalized Difference Vege-

F. N. Kogan

1410

MINI:~

NOVI

....

.. ...

•

..

'.

.

:

•

-

.... ..

.'

• r- r-. ...'

..

-(

..

.'

•

:

-

..

..

.

••

.. '"

: ;

.. ....

..

:

"B" ..

"

,', ','

'

3.

(0)

A • below 0.02 B • 0.02-0.05 C • above 0.05

•

• •

'"

••

\1

;

,

>;

•

RELIEF in metres

•

AJ

.. .. ..

.... '

,..r

3.

..

.

.. ... '

••

3'

3'

(b)

A .0-500 B ·501-1000

Figure 3. Comparison of Sudan maps showing (0) relief (in metres), (van Chi-Bonnerdel 1973) and (b) the absolute minimum ofNDVl for 1984-1987growing seasons,

tation Index; NOVI_ MTN and NOV I _ MAX are absolute minimum and maximum, respectively, of the smoothed weekly composite NOVT defined from historical data; i and j define week and location, respectively. It should be noted that the absolute minimum contributes considerably less to NOV] modification than does the absolute maximum. Setting the absolute minimum to zero, simplifies the expression above (Kogan 1997 a). However, recent work with well-filtered NOVI time-series has demonstrated that, in areas with clearly identified seasonal changes in vegetation, the absolute minimum criterion facilitates more accurate estimation of environmental impacts. This can be demonstrated by assigning numbers to the parameters in the above equation. 4.

Comparison between VCT and NDVI The smoothed NOVI data were normalized using the equation above. To examine how effectively the modified VCT characterizes the impact of weather, the resulting VCI values were compared with NOVI values for neighbouring grid cells which have similar seasonal dynamics of vegetation. Figure 4 shows the principal differences between the two indices. Figure 4 (0) compares the state of vegetation in 1986 for neighbouring grid cells (I and 2) in the west-central part of South Darfur province (west of Nyala). This area has a humid tropical climate with annual precipitation between 500 and 650mm falling entirely during the growing season. More importantly, dryness of this area increases towards the lowlands in the east, resulting in a lower absolute maximum of the NOV I (0'21) in grid 2 as compared with grid 1 (0'27). Figure 4(b) shows the 1986

Remote sensing of weather impacts on vegetation

1411

(0)

> o

z

0(b)

o >

.32 -

-

.16

> 0

z

o WEEK

Figure 4.

Dynamics of 1986 weekly Vel and NOV! for grid cells (0) 1-2 and (b) 3-4 (see figure 1 for grid location).

numerical estimate of vegetation vigour in the south-western part of South Kordofan, with annual precipitation between 700 and 800 mm. The second pair of grids is located in the tropical steppe zone but grid 3 with a drier climate (annual precipitation 700mm) has a lower absolute maximum of NOV I value (0'26) than grid 4 (0'30). As figure 4 indicates, in both locations the areas with a drier climate (grids 2 and 3) had lower NOV 1 values throughout the growing season. However, this did not necessarily indicate unfavourable vegetation conditions that were due to adverse weather impacts on vegetation. It is important to emphasize again that NOV I values reflect the contribution of both weather and geographical resources. But the VCI values reflect the contribution of only weather. When VCI values were analysed, they showed that only early in the season (weeks 26 to 30) were conditions in the areas with a dryer climate worse than in the areas with a wetter climate. Late in the season (after

1412

F. N. Kogan

week 30) the VC[ indicated similar vegetation conditions for grids I and 2 but superior conditions for grid 3 versus grid 4. These were opposite to NDV[ estimates. The diagram in figure 4 illustrates only some of the differences between the VCI and the NDVI for estimation of vegetation conditions. Depending on the characteristics of an area these differences can be opposite to those shown in figure 4. But in any of these cases the advantages of the VC[ over the NDV[ for estimation of weather impacts can be demonstrated. Moreover, a comparison of precipitation dynamics with VC[ patterns provides further support for use of the VC[ over the NDV[ as a tool for estimating weather impact. 5.

Rainfall regime during 1984-1987 Rainfall records for the 18 meteorological stations used in this study were obtained from the Sudan Meteorological Service as cumulative monthly amounts for the rainy season, May to October 1984-1987. The locations of these stations are shown in figure I. It is evident that the network with available records is very limited for the purpose of detailed spatial comparison. It is widely accepted that rainfall is the principal parameter of the environmental regime which governs life in Africa. The vegetation cycle strictly follows the arrival and retreat of rainfall. The rainfall regime in Africa is neither stable over time nor space. A number of researchers have concluded that the annual amount of rainfall in sub-Saharan Africa has been gradually declining since the late 1960s (Nicholson 1983, Todorov 1985, Lamb et al. 1986). Several devastating droughts (1972-1973, 1982-1984), coupled with over-utilization of land, are thought to have caused a selfperpetuating process of increasing dryness. Long climatic records in Sudan have also indicated that rainfall is generally decreasing. For 75 per cent of the years in the 1968-1984 period, the total annual precipitation was below the long-term mean (Eldredge et al. 1988). The 1984 drought was considered one of the worst in the last decade, although abundant rains covered much of Sudan in 1985 and part of its territory in 1986 and 1987 (table I). Rainfall records are incomplete but a spatial analysis of seasonal rainfall indicates that in 1985 the area south of 120 N, southern Kassala and the northern Bille Nile provinces received above-normal precipitation. In central and western Sudan seasonal rainfall was less plentiful but, considering timeliness of rains (excessive rains in June, July, and partly August and September) and beneficial post-drought impact on vegetation growth, it is likely that the 1985 vegetation even in these regions was very well developed. In some areas well-developed vegetation was observed in 1986 or 1987. In summary, it is important to stress the diversity of the rainfall regime over the period of this study. This diversity resulted in a wide variety of conditions, from extremely dry in 1984, to fairly wet in 1985, and in some areas in 1986 and 1987. Following this pattern, extremely low biomass accumulation was observed all over Sudan in 1984, high accumulation in 1985 and also some areas in 1986 and [987. The absolute maximum of the NDV[ was determined mostly from 1985 records, with 1986 data being determinative for some of the areas between 11 0 and 140 N, and 1987 data for the area south of l l N, The absolute minimum of the NDV! was defined mostly from 1984 satellite data. r

6.

Precipitation and indices dynamics Unfortunately, the network of complete and reliable precipitation records in Sudan is sparse. Of these reliable stations only two were located in the neighbouring

Remote sensing of weather impacts on vegetation Table I.

1413

Precipitation (per cent of normal) during May through October. Station

Province

Gezira White Nile

Geneina EI Fasher EIObeid En Nahud Khartum Kassala Gedaref Wad Mcdani Ed Dueim

Blue Nile

Abu Na'arna

North Darfur North Kardofan Khartum Kassala

Kosti Darnazinc

Senar Nayala Kadugli Renk Malakal Wau Juba

South Darfur South Kardofan Upper Nile Bahr EI Ghazel East Equatorial

1984

1985

1986

23 24 24 37 2 35 57 38 22 25 61 84 40 46 71 49 88 96 91

90 92 89 130 24 51 122 III 121 89 94 91 94 78 96 105 97 112 105

65 76 100 76

1987

58

27

92 110 66 98 75 73 86 96 55 105 79 88

58 74 67 81 115 58 67

99

grids. Additionally, the requirement of similarity of vegetation dynamics places a further constraint on selection of stations and grids for comparative analysis. Therefore, only two stations, Wad Medani and Ed Daeim, located in grids 5 and 6, respectively (figure I), were found to satisfy entirely the required conditions. These stations were selected for analysis. Comparative analysis was also performed for two other stations, Renk (grid 7) and Damazine (grid 8), although they were separated by one grid (220 km). Figure 5 shows the seasonal trend in weekly vcr, NDvr, and total monthly

1984

1985

1986

96

1987

vel

I.ft fL

100

~g[~,

[/

L~ .

~

ND\i1

02[

fdC?

o.ll.~

26 30 34 38

·96

160~ 80

oJun

Jul

Aug

Sep

26 30 1

34 38

f.~·

26 30 34 38

[/~ l ...

~

26 30 34 38 Week

Precipita jion

~ t-.J • JlIl Jul Aug Sep Jun Jul Aug

~:"---::"-~

Sep

Jun Jul

Aug

Sop

Figure 5. Dynamics of 1984-1987: (a) weekly Vel, (b) weekly NDVI for grid cells 5-6 (see figure 1 for grid location) and (e) monthly precipitation (per cent of normal) for weather stations Wad Mcdani (_) and Ed Dacim (~).

1414

F. N. Kogan

rainfall for the growing seasons of 1984-1987. Grid 5 represents EI Gezira and grid 6 the White Nile provinces. Both grids have a dry climate but the Wad Medani weather station (grid 5), being under greater influence from the neighbouring highlands, receives more rainfall during the growing season (350 mm) than the Ed Daeim station (250 mm). Accordingly, the absolute maximum of the NDvr for grid 5 is slightly higher (0'13) than for grid 6 (0'10). Analysis of figure 5 indicated that there was no difference between vcr and NDVr estimates for 1984 and 1985. This is expected because grid 5 (Ed Daeim) received a greater amount of rainfall (early in the season during 1985 and in the mid-season during 1984) and had wetter conditions than grid 6 (Wad Medani). But the differences are noted for 1986 and 1987. Contrary to their climatic resources, grid 6 received larger amounts of rainfall early in the season (June, July) and this resulted in better vegetation conditions and higher vcr values through mid-August. Characteristically, the NOVI values for the same period did not show a difference between the grids. At the end of the 1987 season the higher vcr values suggested better conditions in grid 5 versus 6, in accord with the two-fold difference in August rainfall. There was a noted inconsistency between NDvr and precipitation dynamics for the first three weeks of September 1987-larger NDVr values were observed for the grid with the smaller amount of rainfall. At the end of the 1986 season, vcr values for both grids were equal, in accord with near-normal precipitation for both locations from the end of July through August. The NDVr values again did not show such agreement with precipitation dynamics. Another example shows the seasonal dynamics of the indices for grids 7 and 8 and precipitation for stations Renk and Damazine, respectively (figure 6). The first grid is located in the far northern Upper Nile and the second in the southern Blue Nile provinces. These places have a tropical climate with a clearly defined dry period. Grid 8, being under the influence of the Ethiopian High Plateau, normally receives around 700 mm rainfall during May-October (Damazine), while grid 7 receives only 500 mm (Renk). Accordingly, the absolute maximum of the NDvr decreases from 0·34 in grid 8 to 0·20 in grid 7. Analysis of figure 6 showed that the NDvr for grid 7 was considerably larger than

1984

1985

1986

1987

vel

~ __ " ~ ---' ~/.

gLC °26 30 34 38

I~U ~

L

~ _~.""""

~I~

R~IM L<~. 26 30 34 38

26 30 34 38

t

~-

~

L 26 30 34 38 Week

LWrt:rti lui

Jun Jul Aug Sep Jill Jul Aug Sep Jun Jul Aug S'l' Jun Jul Aug Sep

Figure 6. Dynamics of 1984-1987: (a) weekly Vel, (b) weekly NDVI for grid cells 7-8 (see figure I for grid location) and (e) monthly precipitation (per cent of normal) for weather stations Damazine (.) and Renk (1?1l).

Remote sensing of weather impacts on vegetation

1415

for grid 8 regardless of whether rainfall was abundant (1985) or deficient (1984). However, the VCI values indicated that 1985 and 1986 conditions in grid 8 were similar to or even better than in grid 7. These estimates agree with the precipitation dynamics. As figure 6 shows, both stations had almost equal amounts and similar distribution of precipitation during the 1986 season. This resulted in roughly equal VCI values for both grids, while the NOVT for grid 8 was much higher than for grid 7. In 1985 (figure 6 ) the principal difference in vegetation conditions between these two sites developed early in the season. Above normal (120 per cent) June rains fell in grid 8 (Damazine), while unusually dry weather (25 per cent of normal rainfall) persisted through early July in grid 7 (Renk). This resulted in lower values for both VCI and NOVT in grid 7 versus 8. The difference between VCT and NOV I estimates appeared in July. July rains much above normal (160 per cent) equalized the conditions for both locations as estimated by VCI, while NOVI values continued to show considerable differences between these two sites. August-September VCI estimates continued to follow precipitation dynamics for both locations. However, NOVI values did not show adequate agreement with precipitation dynamics. This discussion clearly indicates that VCT estimates portray precipitation dynamics fairly well, and with higher accuracy as compared with NOV!. Both conclusions were supported by correlation analysis of all available satellite and precipitation records for this study. Figure 7 illustrates this correlation between May-September rainfall and end-of-September (week 39) VCT for 1984-1987 for grids containing weather stations. As seen, there is a well defined correlation for all three selected intervals of precipitation normals. An increase in precipitation leads to an increase in the VCI. It is characteristic that the rate of VCT increase is slightly reduced for higher precipitation normals. This is because the same amount of precipitation has a greater impact on vegetation in a drier climate. The relationship in figure 7 can be used to estimate VCI based on precipitation data. For this purpose the equation was approximated in a form where VCT was the dependent variable and precipitation was the independent variable.

VCT =0·19p-0·llP+46·00

Preclpltatk>n normal (mm)

• -2Slr340 lit - -

350-600

o q,

o - - ···550-650

c

e

o

o

","'0

J> ' • , , 0 00.

o

e

•

,......

•

• o

•

40

60

80

100

VCI<%)

Figure 7. Correlation between May-September precipitation and VCI.

1416

F. N. Kogan

Table 2. Determination coefficients between vegetation index and precipitation. Precipitation normal

vel

NDVI

250-340 350-500 550-650

0·88 0·74 0·57

0·82 0·64 0·49

where VCI is end-of-September VCI and p and P are May-September actual and normal precipitation respectively. The correlation between VCI and precipitation was compared with the corresponding correlation between NOV I and precipitation for the same selected intervals of precipitation normals. The results, shown in table 2, indicate that the determination coefficients are persistently larger for precipitation-VCI than precipitationNOVI relationship. To sum up the discussion so far, it is imperative to stress again the advantages of the VCI over the NOVI for estimating vegetation conditions and weather impacts on vegetation in geographically non-homogeneous areas. 7.

Analysis of vegetation conditions The advantages of the VCI for estimating the impact of weather on vegetation become more pronounced when patterns of the index are compared with those for precipitation. Both parameters were plotted on the map of Sudan and a visual interpolation and a contour analysis of their values were carried out. From figure I it can be seen that most (85 per cent) of the rainfall reporting stations represent the area between II ° and 16° N, although the eastern part had only three stations. Southern Sudan, with one station (Juba) south of 7° N, was the least covered. The 1987 rainfall data were even less representative than this would indicate because only nine station reports from central and east-central Sudan were available (EWSB 1987). Figure 8 shows the distribution of the end-of-month cumulative VCI, along with total monthly precipitation (per cent of normal) from May through September 1985. In addition to the VCI-rainfall comparison, the NOVI distribution was also analysed versus the VCI and precipitation patterns. The 1985 season was selected for display because rainfall was plentiful and evenly distributed across the country in comparison with other years. The patterns shown on the maps represent three major categories of environmental conditions: above normal, normal and below normal. Index values corresponding to the upper 30 percentile of their change were selected to represent above normal conditions. Similarly, the index values corresponding to the lower 30 percentile represented below normal conditions. The values between these percentiles were considered to approximate normal conditions. Accordingly, the vegetation condition index for the indicated categories ranged as follows: above 70, 70 to 30 and below 30 per cent of the absolute maximum. The NOVI values for the same categories were selected in the range above 0,25, 0·25 to 0·15 and below O' I 5. In view of the recent increase in dryness in central Africa, the following categories of precipitation were selected: above 80, 80 to 60 and below 60 per cent of the 1951-1980 rainfall mean. In terms of precipitation impacts on vegetation, these categories can be currently interpreted as favourable, normal and unfavourable, respectively. As data in figure 8 indicate, very good rains fell during the 1985 season. After starting in southern Sudan in May, timely rains moved to the north, bringing plentiful

Remote sensing of wealher impacts on ueqetation

MAY

..lLY

AUGUST

1417

SEPTEMBER

NDVI

PRECPfTAIDN

Figure 8. Patterns of the 1985 cumulative vel (%), NOVI and monthly precipitation (per cent of normal) in Sudan. moisture during the most important part of the growing season, June to August. Progression of the vegetation followed the general tendency of rain movement as indicated by both indices. However, contour analysis revealed better correlation with precipitation for the VCI. For example, the northern limit of favourable and normal vegetation conditions in 1985 based on vcr estimates coincided with the advancement of increased amounts of rainfall (above 80 per cent of normal). The NDVr estimates did not show such correlation: the northern limit line was positioned 10 (June) to 30 (September) south of either the vcr or precipitation lines. The differences between the VCI and NDVr estimates for favourable conditions reached 4.5 0 (near 450 km) in east-central Sudan. South-eastern Sudan presents another example demonstrating the differences between the vcr and NDvr estimates. As the vcr values (above 70 per cent) indicated, the vegetation conditions in May were mostly

1418

F. N. Kogan

favourable but the NDVI values (below 0'15) suggested unfavourable conditions. These differences remained through the end of the 1985 season. Unfortunately precipitation data were not available for this location to verify these estimates directly. However, indirect verification based on cloud imagery analysis and rainfall in the neighbouring areas suggested that VCI estimates were accurate. Only late in the season of 1985, when rainfall decreased, did VCI values emphasize unfavourable conditions. But this was on a much smaller area compared with the NDVI estimate. Interestingly, low NDVT values (below 0,15) in south-eastern Sudan were also recorded for the 1986 and 1987 vegetation seasons while the VCI again indicated comparatively higher estimates. The noted discrepancies are related to the aforementioned differences in the ability of the two indices to interpret vegetation conditions. In the case described, lower NOVI values were conditioned by deficient climatic resources rather than by the impact of unfavourable weather on vegetation. 8.

Conclusions One can conclude that the VCI promises definite improvement in analysis of vegetation conditions in non-homogeneous areas. This was supported by independent evaluation of conditions based on meteorological reports. The VCI permits not only the description of land cover and spatial and temporal vegetation change but also allows the quantification of the impact of weather on vegetation. This is extremely important for timely determination of problems with agricultural production in areas with marginal climatic resources. The VCI allows one to compare weather impact in neighbouring areas with different ecological and economic resources. This is very important when using this index for drought monitoring. Furthermore, the VCI is more convenient and understandable for users because VCI values indicate how much the vegetation advanced due to weather and how far the vegetation development is from the potential maximum and minimum defined by the geographical resources. It is also important to underline that the selection of the absolute maximum NDVI as a criterion for index modification is justified by the fact that the patterns of this criterion excellently match the established geographical zoning. The technique discussed here minimizes noise in the AVHRR data by means of elimination of areal non-uniformity and increases the vegetation signal related to the impact of weather alone on vegetation. Acknowledgments The author wishes to thank his colleagues Drs L. Lambert and J. Sullivan, and also D. Le Comte and C. Steinborn, for their extremely useful discussions, comments and generous suggestions. References DUGGIN, M., SCHOCH, L., CUNIA, T., and PIWINSKI, D., 1984, Effectsof random and systematic variations in unresolved cloud on recorded radiance and on target discriminability. Applied Optics, 23, 387-395. ELDREDGE, E., KHALIL, S., NICHOLDS, N., and RYDJESKI, D., 1988, Changing rainfall patterns in Western Sudan. Journal of Climatology, 8,45-53. EWSB (EARLY WARNING SYSTEM BULLETIN), 1987, Sudan Early Warning System, Vol. 2, No. 10 (Khartoum, Sudan). GALLO, K. P., and EIDENSHINK, J. c., 1988, Difference in visible and near-IR responses, and derived vegetation indices, for the NOAA-9 and NOAA-IO AVHRRs: a case study. Photoqrametric Engineering and Remote Sensing, 54, 485-490.

Remote sensing of weather impacts on vegetation

1419

GUTMAN, G., 1987, The derivation of vegetation indices from AVHRR data. International Journal of Remote Sensing, 8,1235-1243. HIELKEMA, J. U., PRINCE, S. D., and ASTLE, W. L., 1986, Rainfall and vegetation monitoring in the Savanna zone of Democratic Republic Sudan using NOAA Advanced Very High Resolution Radiometer. International Journal of Remote Sensing, 7,1499-1514. HIERNAUX, P. H. Y., and JUSTICE, C. 0.,1986, Suivi du developpement vegetal au cours de l'etc 1984 dans Ie Sahel Malien. International Journal of Remote Sensing, 7, 1515-1532. HOLBEN, B. N., 1986, Characteristics of maximum-value composite images from temporal AVHRR data. International Journal of Remote Sensing, 7,1417-1434. JUSTICE, C. 0., HOLBEN, B. N., and GWYNNE, M. D., 1986, Monitoring East African vegetation. International Journal of Remote Sensing, 7, 1453-1474. JUSTICE, C. 0., TOWNSHEND, J. R. G., HOLBEN, B. N., and TUCKER, C. J., 1985, Analysis of the phenology of global vegetation using meteorological satellite data. International Journal of Remote Sensing, 6,1271-1318. KOGAN, F. N., 1987 a, Vegetation index for areal analysis of crop conditions. Proceedings ofthe

18th Conference on Agricultural and Forest Meteorology, AMS, held in W. Lafayette, Indiana, on 15-18 September 1987, pp. 103-106. KOGAN, F. N., 1987 b, On using smoothed vegetation time-series for identifying near-optimal climate conditions. Proceedings of the 10th Conference on Probability and Statistics, AMS. held in Edmonton, Canada, pp. 81-83. LAMB, P. J .. PEPPLER, R. A., and HASTENRATH, S., 1986, Interannual variability in the tropical Atlantic. Nature, 322, 238-240. LE COMTE, D. M., KOGAN, F. N., STEINBORN, C. A., and LAMBERT, L., 1988, Assessment of crop conditions in Africa. NOAA Technical Memorandum NESDIS AISC 13, Washington, D. c., U.S.A. MALINGREAU, J. P., 1986, Global vegetation dynamics: satellite observations over Asia. International Journal of Remote Sensing, 7,1121-1146. NICHOLSON, S. E., 1983, Sub-Sahara rainfall in years 1976-1980: evidence of continued drought. Monthly Weather Review, Ill, 1646-1654. NOAA, 1986, Global vegetation index user's guide (Washington D. C: NESDlS). PHILIPSON, W. R., and TENG, W. L., 1988, OperationallnIerpretation of AVHRR vegetation indices for world crop information. Photoqrammetric Engineering and Remote Sensing, 54,55-59. PRINCE, S. D., and TUCKER, C. J., 1986, Satellite remote sensing of rangelands in Botswana. International Journal of Remote Sensing, 7,1555-1570. TARPLEY, J. D., SCHNIEDER, S. R., and MONEY, R. L., 1984, Global vegetation indices from NOAA-7 meteorological satellite. Journal of Climate and Applied Meteorology, 23, 491-503. TODOROV, A. V., 1985, Sahel: the changing rainfall regime and the 'normals' used for assessment. Journal of Applied Meteorology, 24, 97-104. TOWNSHEND, J. R. G., and JUSTICE, C. 0., 1986, Analysis of the dynamics of African vegetation using the normalized difference vegetation index. International Journal of Remote Sensing, 7, 1435-1446. TUCKER, C. J., GATLIN, J., SCHNIEDER, S. R., and KUCHINOS, M. A., 1982, Monitoring large scale vegetation dynamics in the Nile delta and river valley from NOAA AVHRR data.

Proceedings of the Conference on Remote Sensing of Arid and Semi-Arid Lands held in Cairo, Egypt, in January 1982 (Ann Arbor: Environmental Research Institute of Michigan), pp. 973-977. TUCKER, C. J., TOWNSHEND, J. R. G., and GOFF, T. E., 1985, African land-cover classification using satellite data. Science, 227, 369-375. VAN CHI-BoNNARDEL (editor), 1973, The Atlas of Africa. Paris. VAN DIJK, A., 1986, A crop condition and crop yield estimation method based on NOAA/AVHRR satellite data. Ph.D. Dissertation, University of Missouri-Columbia, Columbia, Missouri, U.S.A. VAN DIJK, A., CALLIS, S. L., SAKAMOTO, C. M., and DECKER, W. L., 1987, Alternative method for reducing radiometric disturbance in NOAA/AVHRR data. Photogrammetric Engineering and Remote Sensing, 53, 1059-1067. WALSH, J., 1988, Famine early warning system wins its spurs. Science. 239, 249-250.