RELATIONSHIP OF AIRCRAFT RADIOMETRIC MEASUREMENTS TO ...

6 downloads 0 Views 1MB Size Report
B.D. Foran and G. Pickup. CSIRO Division of' IVildlifc and Rarzgelands Researclz, Central Australiarz I,aborator~~, Y.O. Box 211 1,. Alice Springs, N. T. 5 750.
Aust. Range]. J. 6(2) 1984, 59-68.

RELATIONSHIP OF AIRCRAFT RADIOMETRIC MEASUREMENTS TO BARE GROUND ON SEMI-DESERT LANDSCAPES IN CENTRAL AUSTRALIA B.D. Foran and G. Pickup CSIRO Division of' IVildlifc and Rarzgelands Researclz, Central Australiarz I,aborator~~, Y.O. Box 211 1, Alice Springs, N. T. 5 750. Australia.

Abstract Radiance values in Landsat MSS bands were measured over five landscapes typical of central Australia: a calcareous plain, an eroded calcareous plain, a gibber plain, a floodplain and a savannah woodland. using an aeroplane-mounted i s o t e c h four band radiometer flown at a height which gave Landsat pixe!-sixed sampling areas. Forty-five radiometric variables derived from these radiance data were correlated with the amount of bare soil interpreted frotn aerial photos. On the calcareous plains and the floodplain, the highest correlations were obtained with the visible bands, Band 4 and Band 5, and the difference between them. On the gibber plains and savannah woodlands the best correlations were obtained with the ratio of Band 5 to Band 7 and the subtraction ratio of B4-B5/ B4-B7. A series of linear relationships were developed linking bare ground \vith radiometric variables which expl'ained between 705%and 85% of the variation. Multiple regression analysis on the radiometer indiccs most highly correlated with total bare ground showed that inclusion of other cover variables such as litter, shadow, soil colour and herbage covcr increased the variation explained by only 5-10%. A test of the effect of differences in solar elevation angle on the 45 radiometric variables showed that the radiance values in each band and differences between them \$cue significantly affected. Eight indices were affected to a lesser degree and may bc useful for multitemporal work. Some considerations of the use of the selected indices for practical landscape monitoring are also discussed.

Introduction

Pastoral properties in inland Australia are large and subject t o an erraiic and unpredictable climate. Government agencies which are responsible for pasture lands must monitor Australia's rangelands at a level commensurate with the large areas involved, low productivity per unit area and within the limits of available personnel. Landsat technology with its repetitive coverage of large areas is an appropriate tool t o apply in the initial stages of landscape monitoring. Landsat MSS radiance data o n rangelands are a function mainly of the soil reflectance and the degree t o which the soil surface is masked by vegetation cover, particularly if that vegetation is green. The vegetation cover is a function of the preceding rainfall and grazing levels and is frequently sparse. If Landsat data are used t o aid monitoring of rangelands, they must be able t o separate the effects of vegetation from those of soil characteristics. This may be difficult when only a t vegetation is present. limited a n ~ o u n of Vegetation cover on semi-arid landscapes comprises three structural units: trees and sl~rubs,standing herbaceous, and litter cover. Of these, the herbage layer is the most important for rangeland use, but detection of change within it is confounded by the influence of the other units. Furthermore, it has been suggested that threshold levels of plant cover exist below which plant plus soil reflectance cannot be distinguished frotn soil reflectance (Satterwllite ct al. 1982). Plant cover may often be below threshold levels

and so we need t o determine these levels. Because plant covers are small and transitory in nature. we have chosen the amount of 'bare ground' as the most appropriate landscape characteristic for monitoring in these regions. Rationale

There are three main p r o b l e m in developing a relationship between radiance and vegetation cover in rangelands. These arise from the diversity of soil and vegetation colours, changes in illumination during the year, and the amount of green vegetation present. Central Australia contains a wide range of soil and vegetation types and colours. This makes it difficult t o predict vegetation cover from radiance because of interaction with soil colour. Uniform soil-vegetation systems allow more accurate measurenlent of vegetation characteristics (hlcDaniel and Haas 1982) than heterogeneous sytelns (Deering et al. 1977; Westin and Lenime 1978). Thus in exploring relationships between bare gro~ind and radiance, a selective approach was taken. Five con1111o11 landscape types were chosen (Table I ) , which are broadly representative of the best and most intensively used grazing country in central Australia. Differences in the angle of solar elevation and. hence the degree of scene illumination, are important variables in multi-temporal work with Landsat. Our radiometer data were obtained for solar elevation angles between 38" and 58' for the landscapes sampled

Paper

-

Relationship of aircraft radiometric measurements to bare ground

Table 1. Landscape types used in the study Code

Type

Description"

CAL

Calcareous

Gently sloping limestone country on calcareous (Gcl. 12) and texture contrast (DR 1.13) soils with scattered trees, bluebush Mairearla astrotriclia) and short grasses and forbs.

ECAL

Eroded Calcareous

As above but with about 100 years of intermittent heavy grazing resulting in tree death, soil erosion, and dominance by unpalatable wire grass Aristido strigma.

GIBI, G1B2

Gibber

Undulating plains or texture contrast soils (Dr1.43) with a strong surface pavement (gibber) consisting of siliceous material and a plant cover of Mitchell grass Astrehla pectirmta, annual grasses and forbs, and occasional small trees and perennial shrubs Atrip1e.u spp.

FPL (DRY) FPL (GREEN)

Floodplain

A floodplain and channel conlplex on scalded texture contrast soils (DR 2.33) with mainly ephemeral grasses and forbs and some juvenile river red gum Eucalj'pf~iscannzlditler~sisin floodout areas. The same area was examined under both dry and green conditions.

SAV (DRY)

Savannah

A savanna woodland o n earthy sand (Uc5.21) soils with trees and shrubs such as whitewood Atalaya hcrniglauca, supplejack Vcr?tilagoi~iiriirrulis,corkwood Hakea siiberra and witchetty bush Acacia kernpcarm over a mixture of annual and perennial grasses and forbs. The area was examined under both dry and green conditions.

*Soil classifications are from the Northcote (1971) system.

Table 2. Radiometric counts (mean & range) and site cover data (mean & range) for six landscapes in central Australia Site

CAL

CCAL

GIBl

GIR2

Band 4

mean range

37 27 48

48 19-80

57 44-77

50 45- 5 6

77 46-103

62 46-82

78 63-99

Band 5

mean range

64 37-83

88 25-138

101 67-135

81 66 9 3

140 8 0 180

103 64-144

136 100 175

97 5 8 149

Band 6

mean range

59 36-73

72 24-113

82 66-103

67 58-72

104 65-125

96 81 108

102 85- 125

94 8 2 - 108

Band 7

mean range

94 57-113

103 30--139

li3 85-135

96 80110

148 95- 170

141 1 2 0 I60

147 130-170

140 125 165

Total Bare

mean lange

33 2-68

70 28-100

72 29-97

48 12-69

87 42100

59 26 87

81 5 0 100

52 10 97

Soil red-orange

mem range

33 4-68

46

38 5- 9 2

13 1 28

62 7-97

52 26-77

62 0- 100

44 9 -73

Item

Sod yellow-red

O 97

mean rmgc

0

0

-

-

Soil white

mean range

1 0-12

Gjbber purple

mean range

0

mean range

0

Stone hght

22 0-99 0

-

-

1 0 36

0 6 0-51 27 0-74

1

0

7 0 29

28 0-55 0

Total Plant Cover mean range

62 32-93

24 0-62

28 3-71

52 3188

Tree and qhrub

mean range

II 0-44

12 0-53

4 0-20

0 0-3

Herbage layer

mean range

48 9 79

11 0- 76

23 3-63

52 28-88

Litter

mean range

1 0-10

1 0-13

0 0-4

Shadow

mean range

5 0-17

6 0-24

0 0-5

Solar elevation angle (degrees)

I'PL-DRY

08 0-5 25 0- 93 0

TPL-CRIEN

SAV-GREEN 63 4 8 85

0

0

0

-

-

-

7 0 -45 0

-

0

SAV-DRY

0

19 0 67

8 0-48

0

0

-

-

0

0

-

-

11 0-55

40 13 73

17 0-48

47 3 90

9 0-53

4 0-2 1

6 0-19

5 0--31

2 0-9

36 13-58

6 0 39

42 2-82

0

1 0 -6

0 0-2

5 0 15

0 0-4

0

2 0- 8

1 0-1 1

2 0-11

1 0-7

B.D. Foran and G. Pickup

(Table 2) making comparison between landscapes difficult. Several data collection runs were carried out t o examine the effect of changes in solar elevation and also t o determine the extent t o which the data for different sun elevations could be combined. These runs involved the flying of fixed transects o n the same day over floodplain and savannah landscapes at sun elevations of 31°, 45" and 57". Frame by frame conlparison was not possible because of shifts in frame position along the flight path. Instead the sample means were compared using a simple t test after an F test which confirmed that sample variances were the same. This provided an indication of the extent t o which changes in scene illumination affected leflected radiance in the measured wave-length bands and derived indices. The herbage layer of rangeland vegetation is usually sparse and dry although the treelshrub layer may be green. After significant rainfall, the green flush which develops over the landscape affects the a ~ n o u n t of reflected radiation and may reduce the accuracy of predictive relationships derived for dry conditions. This effect was measured o n the floodplain and savannah transects a few weeks after the first period of data collection and following 120-160 m m of rain. Methodology

The method of data collection used was similar t o that o f Hick and Tapley (1981). Radiometric data in the MSS spectral bands and colour photographs were obtained from an aeroplane mounted Exotech lOOA radiometer and Hasselblad 7 0 mm camera. The radiometer was connected t o a data logger and synchronized so that a photo was taken of each area sensed b y the radiometer. A flying height of 3 0 0 nl with a view angle of 15" was used so that the area sampled was approximately equivalent in size t o a Landsat pixel. Several types of data processing were carried o u t . Cover characteristics for each pixel-sized area were assessed b y projecting the 70 m m colour transparencies onto a gridded screen. Preliminary studies showed that 100 points per transparency were the best compron~ise between precision and time taken for interpretation. At this intensity of sampling, error levels were high for minor cover components but less than 5% for the major component, total bare ground. At each of the grid intersections cover characteristics were classified into soil, rock, gibber (stone pavement), tree or shrub, herbage layer, litter or shadow. The soil, rock and gibber categories were further subdivided according t o colour. Total bare ground was computed by summing the soil, gibber and rock categories, while total cover was determined b y summing the tree/shrub, herbage and litter categories. Data in nlillivolts from the radionleter were converted t o counts using the standard calibration range

for the instrument and were expressed o n a scale of 0-255. The radiometer data may be expressed as counts or in the form of derived indices. Most indices in the literature have been derived for crops and/or well vegetated areas rather than poorly vegetated rangeland areas. Counts in MSS bands are sensitive t o sun angle because of the effect of illumination. while commonly used band ratios are sensitive t o the effects of greening. Because of these known deficiencies and because simple indices had not been evaluated for Australian rangelands to date, it was decided t o use a full range of indices in all combinations of MSS bands. In this study, 4 5 variables were employed. These were: (i) Counts in MSS bands B4, B5, B6, B7 (ii) Ratio of band counts

RAT46, RAT47, RAT.56, RAT57, RAT67

(m) Dlffe~encesbetween band counts e g DlFF45 = B4-B5 DIFF46. DIFF47, DIFF56, DIFF57, DIFF67 (iv) Dlffelence r a t ~ o s e.g. DRAT45 = (B4 -B5)/(B4+B.5) DRAT46, DRAT47, DRAT56. DRAT57, DRAT67 (v) Ratlos of r a t ~ o s e.g B4557 = (B4/B5)/(B5/B7). B4657, B5675. B5674. B5774, B6774. B4567. B4667, B5667, B4564, B4574, B4556, B4674, B4665, B5776 (VI) Subtiaction ratios e g SRAT456 = (B4 B5)/(B4 B6) SRAT457, SRAT457, SRAT.567, SRAT564, SRAT574. SRAT675. SRAT674 -

The suitability of the derived indices was determined b y their correlations with the percentage of bare ground pt esent. Although good correlations may exist between the amount of bare ground and a number of different radiometric variables, other ground cover parameters also affect reflectance. T o determine the effect of these parameters on the landscapes examined, a stepwise multiple regression analysis was used. Quadratic regressions were also employed t o allow for situations in which the ielationship between ground characteristics and reflectance may be non-linear. Results

Relationship of Radia~zceVariables to Bare C r o w d Radionleter counts for all sites and times varied from 19 t o 1 0 3 (Band 4), 25 t o 1 8 0 (Band 5). 2 4 to 1 2 5 (Band 6) and 3 0 t o 170 (Band 7) (Table 2). These

Paper

-

Relationship of aircraft radiometric measurements to bare ground

Tabie 3. Correlation coefficient 'r' between % bare ground and radiometric variables for various landscapes.* Radiometer Index

CAL ECAL

* Correlation coefficients