This action might not be possible to undo. Are you sure you want to continue?
1 1 1
Author affiliation: (1) School of Earth and Environmental Sciences, The University of Adelaide (2) Department of Environment and Natural Resources, South Australia email@example.com Abstract Protecting soil from soil erosion is an issue of global importance, and the magnitude and duration of soil exposure are two of the main determinants of soil erosion risk. However, there are currently very few methods for measuring soil exposure at adequate spatial and temporal frequencies to inform the management of extensive Mediterranean-climate cereal cropping regions. This paper describes progress on a collaborative research program developing remote sensing tools to address this need, and focuses on the cropping regions of South Australia. We have identified MODIS imagery as an ideal spectral, temporal and spatial source of information for monitoring soil erosion. This paper presents a new approach to measuring the relative change in soil, photosynthetic vegetation (PV) and non-photosynthetic vegetation (NPV) cover fractions over time: relative spectral mixture analysis (RSMA). Whereas previous studies have used image indices which typically are restricted to two selected bands, RSMA employs all available spectral information. We employ RSMA to examine change over time of soil, PV and NPV cover fractions in the South Australian cropping regions. As a proof of concept we demonstrate that temporal plots of cover fractions produced by RSMA correspond well with field data and expected seasonal trends. RSMA of MODIS imagery has several advantages over current field cover assessment methods: it provides cover assessments (including soil exposure) every two weeks over the South Australian cropping region. This allows for more accurate measurement of the severity and duration of soil exposure, two of the most important factors for determining soil erosion risk.
While we apply the RSMA method in a South Australian context the method is applicable globally, and would seem appropriate any where broadscale soil, PV or NPV cover measurement is required. Introduction The effective management of crops, pastures and crop residues as soil cover is essential to the continuing profitability of agriculture. Maintaining adequate levels of soil cover minimises water and wind erosion, and has the potential to conserve or increase soil organic carbon levels and soil fertility. The importance of protecting soil resources from erosion is acknowledged as an issue of global importance. As Lal (2003) notes, this importance is illustrated by the strong focus on sustainable management of soil in Agenda 21 from the 1992 Rio summit (UNCED 1992), the UN Framework Convention on Climate Change (UNFCCC 1992), the 1994 UN Framework Convention to Combat Desertification (UNFCD 1996), and Articles 3.3 and 3.4 of the Kyoto Protocol (UNFCCC 1997). Within this context, this study focuses on soil cover management across the extensive agricultural belt of the state of South Australia in central-southern Australia. Current soil cover monitoring throughout Australia is field-based, taking the form of either assessment sites or roadside surveys, dubbed ‘windscreen surveys’ (see Leys et al. 2009 for a summary of current monitoring methods). To take a specific example relevant to this study, in South Australia the State Government Department of Water, Land and Biodiversity Conservation (DWLBC) initiated the Land Condition Monitoring Project (LCMP) in 1998. The aim of the LCMP is to measure changes in soil erosion risk, including soil cover, throughout time and to use this information to continually improve land-management practices to reduce soil erosion risk in South Australian cropping regions. The LCMP conducts a windscreen survey (henceforth referred to as ‘the WSS’) four times a year to monitor the severity and duration of soil cover and disturbance (McCord 2007). These visual estimates, in conjunction with sitebased records of topography and soil type, are used to calculate soil erosion risk for reporting regions. However, these field assessments are neither spatially extensive nor timely: only areas along transects are assessed, and the results are extrapolated to reporting regions; likewise the WSS is run four times a year and takes a number of employees several days to conduct. Furthermore, oblique visual estimates of soil cover, the basis of the WSS, have been demonstrated to be prone to operator bias (Corak et al. 1993; Morrison et al. 1993). There is a need for a more objective, spatially extensive method for monitoring soil cover across the dry cropping districts of southern Australia. We feel that the MODIS sensor captures data at spectral, spatial and temporal scales that make it ideally suited to monitoring soil exposure. However, there has been little research using MODIS for soil cover monitoring. A recent notable exception is the work of Guerschman et al. (2009) which developed an unmixing approach based on MODIS NDVI versus band 6/band 7 space. While 2
this method showed promise, evaluation was hampered by a lack of extensive field validation data. Grass curing measurements were obtained for ten sites, and the index performed well at six sites and poorly at four. This paper presents the first use in Australia of Relative Spectral Mixture Analysis (RSMA), a new approach to measuring cover from MODIS first developed by Okin (2007). Whereas the method employed by Guerschman et al. (2009) only utilised four MODIS bands, RSMA utilises all eight 500 m bands. The aim of this research is to determine whether RSMA of MODIS imagery is capable of measuring changes in soil, PV and NPV fractions in the South Australian cropping regions. We evaluate RSMA cover fractions against LCMP field measures of soil erosion risk. Methods Study Area Our study is based in the rain-fed cropping regions of South Australia, an extensive belt of approximately 11.3 million ha across the southern portion of the State (Figure 1). The climate here is Mediterranean, with hot dry summers (November – April) and mild wet winters (May-October); winter rainfall predominates, ranging from 600 mm in the south to 250 mm pa in the north. Agriculture in the region is dominated by annual rotations of cereal crops, legumes, pasture and fallow. Crop and pasture growth and senescence in the study area follows a predictable annual pattern. Through the summer (December-March), the landscape is largely dry, although summer weeds and pastures can thrive and produce significant green vegetation growth in some areas. This is followed by rainfall in late March through to May and subsequent weed and pasture growth, until chemical spraying of weeds and seeding, or direct-drill seeding, which reduce cover to a minimum in May-June. Following seeding, annual crops germinate and growth peaks in September. Finally crops ripen, senesce and are harvested in November and December. Stubble remaining after harvest is commonly grazed by stock through summer. Retention of stubble and pasture cover is encouraged to minimise soil exposure, but it is understood that the risk of erosion is usually greatest in late autumn when plant residues and pastures have been grazed down and early cultivation takes place.
Figure 1. Project reporting regions and windscreen survey transect locations, and general study area location within Australia. Field Data The LCMP windscreen survey methodology has been described in detail by McCord et al. (2000). In summary, the cropping regions were divided into 45 zones of similar climatic, soil and land-use characteristics. These zones were grouped into regions for reporting purposes. Road transects for a ‘windscreen survey’ (WSS) were designed to cover as much variability within zones as possible, except where precluded by a lack of roads or by native vegetation obscuring views of agricultural land. Both the reporting regions and WSS transects are presented in Error! Reference source not found.. The Windscreen Survey is conducted four times a year (March, May, June and October), with each survey date timed to coincide with critical phases in the annual cropping cycle. The same team drives the same transect each year, except where sickness or employee-turnover requires a change of surveyors. Sites are intended to be representative of the region, hence anomalies are avoided. The data collected by the WSS allows LCMP staff to estimate the proportion of land vulnerable to erosion in each of the reporting regions. It is this estimate, proportion of land vulnerable to erosion, that is compared to RSMA fractions in this study.
Relative Spectral Mixture Analysis The RSMA technique was pioneered by Okin (2007), and is thoroughly described therein. In summary, RSMA is based on spectral mixture analysis (SMA) and provides relative indices of PV, NPV, snow and soil cover. With reference PV, NPV and snow spectra RSMA uses SMA to infer relative changes in these cover fractions from a reference time, or baseline. An advantage of RSMA over SMA of single images is that it is not necessary to know the exact soil spectrum for every image pixel. The soil spectrum for each pixel is contained in the baseline image. Changes in the soil fraction are implicitly measured as the remaining variation after changes in PV, NPV and snow are accounted for. In the example presented in this paper the RSMA was run on 138 MODIS NBAR 16-day composite images covering the period from the beginning of 2002 to the end of 2007, with a baseline image date of 26 May 2003. Results Preliminary results for one cropping region, the Eyre Peninsula, for the period beginning of 2002 to the end of 2007 are presented in Figure 2. This figure compares the LCMP estimate of percent of land at risk of erosion (largely influenced by soil exposure) to our MODIS RSMA relative soil, PV and NPV cover fractions. Within Figure 2 there is good agreement in the timing of peaks and troughs in both the LCMP land at risk of erosion and the RSMA soil cover fraction, as expected. Furthermore the temporal pattern of soil, PV and NPV cover fractions corresponds with the known temporal variation of these cover fractions determined by climate and phenology previously detailed in the Study Area description. However, it is worth noting that the RSMA soil fraction records a longer period of high soil exposure than the LCMP data. As an example, the RSMA soil fraction image for 30 August 2003, during the southern Australian winter when crop and pasture cover is at its highest, is presented in Figure 3. In this image, lighter pixels indicates an area where soil exposure is higher than in the baseline image, medium grey indicates no change, and black indicates less soil exposure. For reference, the baseline image date was 26 May 2003. Areas of native vegetation (e.g. large Conservation Parks in the central-west of Eyre Peninsula, and near the SAVictoria border) show little change in soil exposure between the two dates, whereas the extensive cropping regions show marked decrease in soil exposure, as crops as pastures grow during the winter. Discussion and Conclusions We have demonstrated that RSMA of MODIS imagery is a promising tool for the measurement of variation in soil, PV and NPV cover fractions within and between seasons. We show good agreement of RSMA soil, PV and NPV cover fractions with a field measure of soil exposure, and expected temporal variation in these cover fractions due to climate and phenology in the study area, the Eyre Peninsula of South Australia. 5
Figure 2. Comparison of (1) Land Condition Monitoring Project (LCMP) Erosion Hazard Index (EHI) estimate of percent of land in the Eyre Peninsula cropping region at risk of erosion to (2) relative soil, photosynthetic vegetation (PV) and non-photosynthetic vegetation (NPV) cover fractions derived from relative spectral mixture analysis (RSMA) for the same region. While there was generally good agreement between the RSMA fractions, field data and expected temporal pattern, there was one exception worth noting. The RSMA soil cover fraction records a longer period of high soil exposure than the LCMP data. This is significant, because both severity and duration are essential determinants of soil erosion risk. If this result is accurate MODIS RSMA could improve the accuracy of soil erosion risk estimates significantly. The increased temporal and spatial resolution of MODIS RSMA offer significant advantages over current field methods. These characteristics should allow RSMA based cover measures to provide information on temporal dynamics of soil, PV and NPV cover that were previously unavailable. The increased spatial resolution will allow monitoring of soil exposure, stubble retention and crop growth at a sub-regional level, while the increased temporal resolution will improve understanding of the temporal dynamics of these cover types. 6
Additionally, MODIS RSMA could well assist us in understanding the impact of climate change on soil exposure and soil carbon inputs by measuring the amount of green vegetation production and the rate of dead vegetation decay. This is still research in progress. In the future we will perform more direct comparisons of MODIS RSMA to spatially-located LCMP field data and other field-based fractional cover measures. Additionally, we aim to develop a method for conversion of RSMA to an absolute cover measure.
Figure 3. RSMA soil fraction image for 30 August 2003. Change in soil exposure is measured relative to the baseline image date, 26 May 2003. Acknowledgements Research conducted with funding from an Australian Research Council Linkage Grant (LP0990019) with Partner Organisation support from SA Department of Water, Land and Biodiversity Conservation (now Dept. Environment and Natural Resources). References Corak, S. J., T. C. Kaspar, et al. (1993). "Evaluating methods for measuring residue cover." Journal of Soil and Water Conservation 48: 700-704. Guerschman, J. P., M. J. Hill, et al. (2009). "Estimating fractional cover of photosynthetic vegetation, non-photosynthetic vegetation and bare soil in the
Australian tropical savanna region upscaling the EO-1 Hyperion and MODIS sensors." Remote Sensing of Environment 113(5): 928-945. Lal, R. (2003). "Soil erosion and the global carbon budget." Environment International 29(4): 437-450. Leys, J., J. Smith, et al. (2009). Improving the capacity to monitor wind and water erosion: a review, Australian Government Department of Agriculture, Fisheries and Forestry. McCord, A., M. Thomas, et al. (2000). A Description of Key Land Condition Monitoring Methodology Used in South Australia from 1999-2000. Adelaide, South Australia, Primary Industries and Resources. McCord, A. K. (2007). State survey monitoring manual: cropping districts. Adelaide, Department of Water, Land and Biodiversity Conservation. Morrison, J. E., C.-C. Huang, et al. (1993). "Residue measurement techniques." Journal of Soil and Water Conservation v48(n6): p479(476). Okin, G. S. (2007). "Relative spectral mixture analysis - a multitemporal index of total vegetation cover." Remote Sensing of Environment 106: 467-479. UNCED (1992). Agenda 21: programme of action for sustainable development, rio declaration on environment and development, statement of principles. Final text of agreement negotiated by governments at the United Nations Conference on Environment and Development (UNCED). Rio de Janeiro, Brazil, UNDP, New York: 3-14. UNFCCC (1992). United Nations framework convention on climate change. Bonn, Germany, UNFCC. UNFCCC (1997). Kyoto Protocol to the United Nations Framework Convention on Climate Change. Bonn, Germany, UNFCC. UNFCD (1996). United Nations framework convention to combat desertification. Bonn, Germany.