Introduction
It is known that the terrestrial biosphere exhibits variation in its properties over a range of both spatial and temporal scales (Van Gardingen et al., 1997; Slaymaker and Spencer, 1998; Trudgill, 2001) and this variability is both naturally and anthro-pogenically driven. This variability has direct implications for human society, as well as the Earth system, since vegetation, the principal biospheric component, is tightly coupled to the radiative, meteorological, hydrological and biogeochemical processes and functions that operate within that system (Bonan, 1997; NASA, 2001). Thus, the quest to fully understand the causes of biosphere variability and to measure and predict terrestrial biospheric responses to natural and anthropogenic influences has assumed great scientific importance (Walker and Steffen, 1997; Rotenberg et al., 1998). However, despite making huge strides in this regard, uncertainties in our knowledge of how the terrestrial biosphere is changing and our ability to estimate future changes are readily apparent and need to be overcome (Harvey, 2000; IGBP, 2001; IPCC, 2001).
In short, there is considerable uncertainty attached to our current understanding of terrestrial biospheric change. Three levels of uncertainty are evident: uncertainty as a result of data shortage, uncertainty as a result of model deficiencies and uncertainty as a result of environmental processes that are indefinable and indeterminate (O’Rior-dan, 2000). Realistically, only the first two of these uncertainty levels can be addressed and one approach to achieving this involves the routine collection of information on the terrestrial biosphere, particularly the spatial distribution, extent and temporal dynamics (intra-annual and inter-annual) of ecosystems. It has been suggested that much of this information may be provided by remote sensing (Wickland, 1989; Stoms and Estes, 1993; NASA, 2001) and the arguments for this suggestion are compelling,particularly with the launch of new Earth remote sensing instruments and sensors designed for the task (e.g. EOS Terra). Unfortunately, despite the potential offered by remote sensing, there are several uncertainties involved in using remotely sensed data. Thus, the uncertainties associated with our knowledge of terrestrial biospheric change are compounded by uncertainties in the use of the remote sensing approach. It is imperative, therefore, that all known uncertainties in the use of remotely sensed data are characterized and accounted for, if at all possible.
This topic focuses on one example. It presents a case study in which an identified uncertainty in the use of remotely sensed data for the provision of information about the terrestrial biosphere is evaluated. Specifically, the uncertainty associated with the use of National Oceanographic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) middle infrared (MIR) reflectance data to monitor the impact of El Nino/Southern Oscillation (ENSO)-induced drought on tropical rainforests is explored. The derivation of useful MIR reflectance data, from total radiant energy measured by the AVHRR sensor in channel 3, requires inversion of a basic physical model. The components of this model are specified in established laws of physics but include a parameter, emissivity, which is rarely known with complete certainty. The impact of using different values for the emissivity within the derivation of MIR reflectance is investigated. Here, therefore, the focus is on uncertainty that arises as a result of measurement error (i.e. not knowing for certain what the emissivity value should be) and ignorance (i.e. using a constant emissivity despite spatial and temporal variability). This topic aims to ascertain whether uncertainty in the magnitude of the emissivity parameter in the derivation of MIR reflectance compromises its use for the monitoring of EN SO-related drought stress on a tropical rainforest ecosystem. Since MIR reflectance offers greater potential for this application than conventional approaches (e.g. based on Normalized Difference Vegetation Index (NDVI)), these results have significant implications for the furthering of our knowledge on change in tropical rainforests caused by drought events and our ability to use this knowledge to aid forest management.
The ENSO
TheENSO phenomenon is a coupled atmospheric and oceanic mechanism responsible for worldwide climatic anomalies (Diaz and Margraf, 1992). These climatic anomalies ariseviateleconnectionsto the anomalous warming of oceanic waters in the central and eastern Pacific and changes in the Walker cell circulation system. TheENSO phenomenon occurs with a periodicity of typically 2 to 7 years, having profound impacts on the terrestrial biosphere (Rasmusson, 1985; Glantz, 1996), and is now widely acknowledged as the largest known climate variability signal on inter-annual timescales (Houghton et al., 1990; Li and Kafatos, 2001). The period from June 1997 to May 1998 saw the global climate system perturbed by the largest ENSO event of the twentieth century (McPhadden, 1999). The resultant deranged weather patterns around the world killed an estimated 2100 people and caused at least 33 billion US dollars in property damage (Suplee, 1999). These sorts of impacts may be more commonplace with the news from some of the latest climatic model predictions that suggest ENSO events may intensify further in their severity and frequency (Houghton et al., 1996; Hulmeand Viner, 1998). The need to understand the nature of impacts of ENSO events on the terrestrial biosphere, and in turn how these impacts affect the Earth system, and subsequently to forecast future impacts is obvious. However, the precise nature of ENSO phenomenon impacts on the terrestrial biosphere is highly differentiated in space (Plisnier et al., 2000). There is a need, therefore, for detailed investigations on the EN SO phenomenon that bring together climatic and ecosystem datasets.
One ecosystem particularly under threat from ENSO events is the ever-wet tropics (Whitmore, 1988; Walsh and Newbery, 1999). Such ecosystems face continuing anthropogenic pressures as well as a predicted increase in the frequency and severity of ENSO-related climatic events (e.g. drought, floods) (Harwell, 2000; Harrison, 2001; Couturier et al., 2001). This, coupled with the significant role that the ever-wet tropics play within the Earth system (Philips, 1998; Artaxo, 1998; Houghton et al., 2000) and the fact that these processes may change under ENSO conditions (Potter et al., 2001), mean that they deserve specific attention.
The Role for Remote Sensing
Over the past decade remote sensing has emerged as the only realistic approach to measure much of the necessary data for studies of the terrestrial biosphere (Town-shend et al., 1994; Nemani and Running, 1997). It affords the accurate and systematic measurement of long-term instrumental and proxy observations of key environmental indicators of terrestrial biospheric change (Houghton et al., 1996; Nemani and Running, 1997; Franklin and Wulder, 2002). Remote sensing provides the only means to obtain a view of the ecosystems of the terrestrial biosphere at appropriate spatial and temporal resolutions.Remotely sensed data can be linked to measurements acquired in situ, allowing detailed analysis of key processes and controlling factors of terrestrial biospheric change. Furthermore, since remotely sensed data are acquired under fully traceable conditions they can be used to establish a baseline from which any divergence from the global environmental norm can be quantified.
Since the study of impacts of ENSO events on the terrestrial biosphere requires information on vegetation properties at fine temporal resolutions across large areas, there is an obvious contribution from remote sensing. The general approach has been to couple remotely sensed data with indicators of ENSO strength and phase. ENSO indicators used include those measured in the Pacific Ocean: Pacific sea surface temperature (SST), the southern oscillation index (SOI) and outgoing long-wave radiation (OLR) (Philander, 1990; Trenberth and Tepaniak, 2001). Other indicators include those that are more localized in scale, for example, rainfall totals (Nicholson and Entekhabi, 1986; Nicholson, 1996) or air temperature and pressure statistics (Ropelewski and Halpert, 1986; Trenberth and Shea, 1987). The remotely sensed data used have been mainly those acquired in channels 1 (visible reflected radiation; 0.58-0.68 ^m) and 2 (near infrared reflected radiation (NIR); 0.75-1.10 ^m) of the NOAA AVHRR sensor, combined in the NDVI. The AVHRR NDVI is believed to provide an effective measure of photosynthetically active biomass (Goward et al., 1985; Myneni and Williams, 1994) and has been highly correlated with vegetation biophysical properties and hence can be of considerable value for the study of ecosystem properties. As well as using the AVHRR NDVI over space,it has been realized that intra-annual time-series of AVHRR NDVI data can be used to resolve pheno-logical canopy properties (e.g. Spanner et al., 1990; Achard and Estreguil, 1995; Duchemin et al., 1999; Moody and Johnson, 2001). Moreover, inter-annual time-series NDVI data have been correlated with several climatic variables in a wide range of environments (Liu et al., 1994; Shultz and Halpert, 1995; Wellens et al., 1997; Yang et al., 1998), as well as to SST, SOI and OLR (Myneni et al., 1996; Anyamba and Eastman, 1996; Plisnier et al., 2000; Anyamba et al., 2001). As a result of its successful use, it has been tentatively postulated that NDVI data be used as a supplementary index of ENSO activity, whereby the variability of time-series NDVI may serve as a proxy for climate variations, provide insights into understanding regional climate impacts of ENSO teleconnections on ecosystems and forecast their occurrence (Hess et al., 1996; Verdin et al., 1999; Plisnier et al., 2000; Li and Kaftos, 2001).
The successful use of the NDVI for studying terrestrial biospheric responses to climatic anomalies lies partly in the inherent characteristics of the NOAA AVHRR sensor (Cracknell, 1997), theeasewith which theNDVI index can becomputed and the physical principles determining the reflectivity of vegetation in visible and NIR wavelengths. The anomalous climatic characteristics associated with an ENSO event can produce phenological and canopy biophysical property changes that determine the NDVI values that deviate from the norm. Consequently, the NOAA AVHRR NDVI has been touted as offering a long-term data set, starting in July 1981, that is unequalled for monitoring terrestrial land cover and condition (Los et al., 1994; Batista et al., 1997).
There are several uncertainties associated with the use of NOAA AVHRR NDVI data for the above purpose. Problems with aerosol contamination, instrument degradation and orbital drift are common (Eastman and Fulk, 1993; Gutman, 1995; Batista et al., 1997) and thus the NDVI used may be a function of these as well as the ecosystem properties from which they have been acquired. Furthermore, low NDVI-precipitation correlations have been noted in some ecosystems, for example, tropical forest (Schultz and Halpert, 1995; Eklundh, 1998; Richard and Poccard, 1998). These uncertainties may, however, be overcome by modelling their influences, particularly those of a systematic nature (e.g. instrument degradation; Los, 1993), or combining NDVI data with other remotely sensed data to minimize their influence or using an alternative remotely sensed data set which is more suited for use in a particular ecosystem. Indeed, the NDVI is not the only remotely sensed product that can be generated from NOAA AVHRR data. Other products include surface temperature (Ts), generated from radiation acquired in AVHRR channels 4 and 5, and middle infrared (MIR) reflectance, derived from the total MIR radiation acquired in AVHRR channel 3. Ts measurements have been used to study drought impacts on ecosystems, both, alone (e.g. Plisnier et al., 2000), and within vegetation indices such as ND VI/Ts (e.g. McVicor and Bierwirth, 2001) and the Vegetation and Temperature Condition Index (e.g. Unganai and Kogan, 1998). MIR reflectance data have also been shown to have potential for assessing landscape dryness (Boyd and Duane, 2001). The use of Ts and MIR reflectance data for the study of ENSO impacts is in its infancy, even though the data used to generate these products have been measured simultaneously to those used to derive the NDVI. Thus, we still do not wholly understand the properties of all these remotely sensed data being acquired in such great quantities and have, therefore, yet to realize their full potential for the study of ENSO impacts on the terrestrial biosphere.
MIR Reflectance at 3.751 m
The total MIR radiation signal acquired in AVHRR channel 3 (between 3.55 and 3.93 |im) comprises both the radiation reflected from the Earth’s surface and shorter wavelength radiation that has been absorbed at the surface and emitted at MIR wavelengths. This hybrid signal, part reflected and part emitted radiation, introduces a degree of uncertainty into its use and thus the use of MIR radiation for terrestrial biospheric research has been limited (Boyd and Curran, 1998). Nonetheless, the MIR radiation has several attributes that should encourage its wider use in global environmental research. It has been observed that the use of MIR radiation offers favourable atmospheric penetration capabilities over that of visible and NIR radiation and is relatively insensitive to scan angle effects (Kaufman and Remer, 1994; Franca and Setzer, 1998). Moreover, MIR radiation, used in conjunction with and in place of the NDVI or Ts, has been used successfully in the mapping, monitoring and prediction of forest types and their properties.Furthermore, it has been demonstrated that the removal of the emitted radiation portion of the MIR radiation to leave the reflected portion (known as MIR reflectance) increases the accuracy of remotely sensed prediction of ecosystem properties (Boyd et al., 1999; Boyd et al., 2000).
To derive MIR reflectance from MIR radiation there is the assumption that
where
is the total radiation signal measured in AVHRR channel 3,
is the reflected component of the channel and Ech3 is the emitted component of the channel. Based on this assumption, Kaufman and Remer (1994) proposed a model developed from earlier work by Ruff and Gruber (1983) and Gesell (1984) that can be used to derive MIR reflectance from the total radiant energy measured in AVHRR channel 3. The model assumes that calculating and removing the emitted component from the total radiation signal can be used to derive MIR reflectance. It is assumed that the radiation measured in channel 4 (10.50- 11.30 pm) of the AVHRR sensor, which is almost wholly composed of energy emitted by the Earth, is related to the emitted component of the total radiant energy measured in AVHRR channel 3 through Planck’s function. The correction for surface temperature emission is accomplished by calculating the brightness temperature (T) in channel 4 by inverting Planck’s function (equation 2)
where T is brightness temperature calculated from channel 4 (K), v is the central wave number of channel 4 (cm-1), C1 is constant 1.1910659 x 10-5, C2 is constant 1.438833 and E is a radiance value from channel 4 (mW m2 sr cm-1)
The brightness temperature calculated from channel 4 is then used to calculate the emitted component of the total radiant energy measured in AVHRR channel 3. The total radiant energy measured in AVHRR channel 3 (Lch3) can be described (equation 3)
where
is the two-way transmission function in MIR channel of AVHRR sensor in channel 3,
is the reflectance in MIR channel of AVHRR sensor in channel 3,
is the incident solar radiation at top of atmosphere in channel 3,
is the cosine of solar zenith angle,
is a one-way transmission function in channel 3,
is the emissivity of the surface in MIR channel of AVHRR sensor in channel 3,
is the channel 3 Planck function and T is the brightness temperature calculated from channel 4 (equation 2).
In equation (3) the second term now represents the emitted component of the signal. To obtain MIR reflectance the equation can be rearranged, thus (equation 4)
where
is the full radiation signal,
is the channel 3 one-way transmission function,
is the channel 3 two-way transmission function,
is the emissivity of the surface in MIR channel of AVHRR sensor in channel 3,
is the channel 3 Planck function, T is the brightness temperature,
is the incident solar radiation at top of atmosphere and
is the cosine of solar zenith angle.
MIR reflectance from a vegetation canopy is thought to be principally a function of the liquid water content of the canopy. Thus, any change in the leaf biomass of the canopy, such as that induced by an ENSO event, will be accompanied by a change in the amount of canopy water, and thereby the ability of the canopy to absorb or reflect MIR radiation (Kaufman and Remer, 1994; Boyd and Curran, 1998). In addition to change in leaf biomass, an ENSO event that leads to drought conditions in an ecosystem could promote a change in the canopy condition promoting canopy senescence, which would increase the MIR reflectance from a canopy (Elvidge, 1988; Salisbury and D’Aria, 1994). Indeed a study of intra-annual temporal variability of MIR reflectance from a forest ecosystem has been observed and attributed to both leaf area variability and landscape dryness, believed to be driven by climatic characteristics (Boyd and Duane, 2001).
MIR reflectance from an ecosystem has yet to be related to ground collected climatic measurements and so its potential for the task of monitoring ENSO impacts on the terrestrial biosphere is unrealized. This topic presents a study in which the value of NOAA AVHRR MIR reflectance for the monitoring of ENSO-related drought stress of a tropical forest ecosystem is investigated. The possible correlation between precipitation values and remotely sensed response is explored. It is hypothesized that the MIR reflectance would be more sensitive than the NDVI to changes in the forest canopy properties and condition occurring as a result of anomalous rainfall, a product of an ENSO event. However, there are uncertainties with using NOAA AVHRR MIR reflectance; the effect of one, the accuracy of emissivity predictions, will be investigated here.
Study Area
The rainforests of Danum Valley, Sabah, Malaysia (117°E4°N) during the 1997-98 ENSO event were the focus of study (Figure 8.1). Rainfall statistics show that during the 1997-98 ENSO event there was a marked reduction in rainfall, with the year of 1997 having the lowest rainfall total on record (Walsh and Newbery, 1999) (Figure 8.2).
Figure 8.1 Location of the Danum Valley Conservation Area
Figure 8.2 Monthly rainfall (mm) at Danum Valley for the period under study
At this site is a conserved area of rainforest comprising 43 800 ha of primary forest, known as the Danum Valley Conservation Area (DVCA). The DVCA was established in May 1995 by an enactment in the Sabah State Legislature. It is a Class I Forest reserve to remain unlogged and otherwise undisturbed for posterity. The DVCA lies in a key position close to the drought-prone eastern coastlands of Borneo. During ENSO years, high atmospheric pressure persists over the Sabah region leading to more stable atmospheric conditions and a marked reduction in rainfall. Rainfall records of Sabah show that drought intensity and frequency have increased in the last few decades and this tallies with a recent increase in the frequency and magnitude of ENSO events. Furthermore, evidence acquired via climatic modelling suggests a further increase in the intensity of ENSO events (Walsh and Newbery, 1999). This, coupled with human interference of the forests of Sabah, mean that ENSO-related drought impacts on this ecosystem are likely to increase significantly. A monitoring system to analyse and describe forest response to drought stress and subsequently to forecast future impacts of ENSO events would be advantageous for the effective management of this threatened ecosystem.
