Updated February 1998
Prepared by the Food and Agriculture Organization of the United Nations
of behalf of the United Nations International Drug Control Programme
The Expert Group Meeting on the Detection of Illicit Cultivation of Narcotic Plants by Remote Sensing (Vienna, 23 - 27 October 1989) concluded that remote sensing (RS) techniques could conceivably be used to detect illicit crops under optimal conditions. Since then, several new orbital sensors have been launched, and new image processing techniques have been developed to aid in the detection of illicit crops. In addition, a large number of new, high-resolution civilian RS satellites are planned to be launched within the next few years. Many of these new satellite sensors will be of use in illicit-crop monitoring programmes.
In addition to the orbital sensors which are central to the conceptual development of this study, new airborne sensors are also being developed that could be of use in illicit-crops monitoring, including hyperspectral scanners containing more than one hundred narrow spectral bands; cartographically accurate airborne multispectral scanners using differential GPS (DGPS) and inertial navigation system (INS) data for geocoding images without the need for ground control; and interferometric synthetic aperture radar (SAR) sensors that will produce single-pass digital elevation models (DEMs) and orthorectified radar intensity images. Each of these new sensors provides additional tools for use in the identification of illicit crops.
Newly-developed techniques include the use of geographical information systems (GIS) to integrate remotely sensed image data with Global Positioning System (GPS) data. Desktop and workstation computer processing power has more than doubled since 1989, and continues to increase. Desktop personal computers (PCs) now contain more processing power than large mainframe or mini-computer systems did only a few years ago. New computer operating systems (e.g. Windows 95 using Pentium chip technology), can operate at true 32-bit processing level. New illicit-crop detection algorithms can be designed to take advantage of this increased processing power, including such highly computer-intensive activities as distance modelling, slope and aspect calculation, and sophisticated image processing, such as data fusion and orthorectification resampling.
The results of UNDCP-funded FAO-executed projects in Afghanistan and Lebanon for monitoring the opium poppy crop provided the stimulus for UNDCP's evaluation of RS data: it was felt that a fuller evaluation was warranted of RS and related technologies. This evaluation was to be carried out within the framework of a desk study. The result is this report, which involved the input of various parties, consolidated into a coherent study and amplified as necessary with information from publications and from internal reports and studies available through the offices of FAO, UNDCP, the United States Drug Enforcement Agency (DEA) and other agencies.
The objective of UNDCP was to obtain an assessment of the potential and limitations of current and expected RS technology for the monitoring of illicit crops worldwide.
The study starts with an assessment of the problem, and then considers the parameters that influence the ability to identify illicit-crop production and associated activities. The technology available is then covered, concentrating on particular advantages and disadvantages.
A number of case studies are presented where the use of RS has featured strongly, and some project proposals that have relied greatly on RS for their implementation are also considered.
The final technical section looks at the use of geographical information systems, as the means of choice for bringing together all the inputs, manipulating them to achieve correspondence, and providing outputs in a form that is meaningful and that can be used as the basis for assessments of illicit-crop production, inventory, monitoring and planning eradication efforts.
The current state of the art in RS is considered in Annexes. Annex 1 describes the technical characteristics of the RS sensors currently available and planned, while Annex 2 discusses the need for accurate elevation data, and the use of GPS technology for the necessary precise ground positioning.
In particular, in the technical Annexes, the effect is considered of the rapid developments in desktop and workstation computer processing power, which continues to increase. New illicit-crop detection algorithms exploiting RS technology can be designed to take advantage of this increased processing power, including such highly computer-intensive activities as distance modelling, slope and aspect calculation, and sophisticated image processing, such as data fusion and orthorectification resampling. The ongoing developments in this sphere will have to be borne in mind when assessing the potential of RS for the purpose of illicit-crop monitoring.
This study presents an overview of the problems associated with monitoring of illicit crops, and focuses on the potential for use of remote sensing (RS) technology data in illicit-crop production monitoring. The relative utilities of existing and planned sensors are assessed. The utility of various data processing environments and data processing algorithms is also analysed, together with the relative value and role of other types of information, including field measurements and ancillary georeferenced data.
The relative merits are analysed both of custom and off-the-shelf software and of the various RS data categories and resolutions available.
As a result of these analyses, it is suggested that the most effective monitoring system will probably include both coarse and fine resolution RS data, coupled with field data and ancillary data. The specific role each of these elements of information should have in a monitoring system is also discussed. The characteristics of the most cost-effective monitoring system will depend on the acceptable level of cost, the desired accuracy of the survey and other factors.
The characteristics of various RS data types for illicit-crop monitoring and their relative costs are presented in the Table below.
The applicability of the various technologies varies according to the target crop:
|DATA TYPE||Coverage||Stratification||Enumeration||Accuracy||Data cost||Pre-processing costs||Processing and analysis|
|Coarse spatial resolution of multispectral satellite data||Large area coverage, (>150 km2) @ 20 to 30-m resolution||Potential for improved stratification for overall project area||Potential for total enumeration (no sampling error); information on crop location||Low to high accuracy for crop type and area||$US4,400 per Landsat scene; $US2,800 for SPOT-XS||Roughly equal to data costs for registration, mosaicking, and normalizing||Approximately 3 work-days per frame of data (ca $US3,000)|
|High spatial resolution panchromatic or multispectral satellite data||Moderate area coverage (10 - 30 km2) @ 1 to 5 m resolution||Limited potential for stratification (incomplete coverage)||Sampling error due to incomplete coverage||Moderate to high accuracy for crop type and area||Projected data costs $US10 to $US20/km2 for raw data||For 1-m data, pre-processing may be as high as $US50 to $US100/km2||Approximately 3 work-days per frame of data (ca $US3,000)|
|Field data||Small area coverage (0.5 km2 sampling units)||Potential for high sampling error (imprecision)||Sampling error due to incomplete coverage||Potential high for crop identification accuracy; modest to high for area||High cost per unit area||Modest investment per sample unit (develop sampling frame, etc.)||Compile maps and area tables from field data, ca $US1,000 per sample unit|
|Ancillary data (e.g., in a GIS)||Large area, complete coverage likely||Useful for stratification and prioritization||General information available; may be dated; may not include specific requirements||Useful for modelling||Relatively low cost per unit area||Modest pre-processing costs, mostly format conversion||Application dependent|
Some general points made within the context of the report are:
1. The role of RS data will differ between situations where illicit crops are abundant and located in close proximity to each other (e.g., Afghanistan) and where they are sparse and not clumped (e.g., Lebanon). In the former case the primary value of RS may be stratification, leading to optimal and efficient sampling. In the latter case, the primary value of RS may be the ability to examine all elements of the project area, thus reducing sampling error, which might be unacceptably large without this capability.
2. The two elements of potential future RS data that will most enhance the ability of RS to assist in crop monitoring appear to be improved spatial resolution and improved temporal resolution.
3. An operational crop monitoring programme should have access to a facility and staff which can effectively integrate and process both satellite data and other sources of data (field data, ancillary data), and is also flexible enough and knowledgeable enough to design specific monitoring procedures for each different situation, so that the procedures optimally match the desired accuracy, timeliness and acceptable cost.
4. Multi-date approaches are likely to be much more accurate than single-date approaches, because they permit the use of crop profile information. Although single-date approaches may be effective in situations where there is a single, optimal acquisition date, factors such as variable topography, climate, cultural practices and cloud cover make single date approaches less robust in many regions.
5. Data from multiple sensors with different informational content (e.g. optical and radar) is likely to improve crop identification compared to results from single-sensor approaches. The use of two or more similar optical sensors (e.g., SPOT and Landsat TM) to improve temporal resolution may improve crop identification accuracy, but this improvement is dependent on accurate spatial registration of the different data sets, which can be difficult in areas of significant topography and inadequate digital terrain data. Area estimation is most accurately done with high-resolution data, but the most cost-effective area estimation approach may be to use a multiple sensor mix of coarse and fine resolution data, and an effective multi-stage or multi-phase sampling approach.
An optimal crop monitoring system would probably draw on information from a range of data sources, with the relative roles of the various sources of information fairly clear, based on their characteristic advantages and limitations. A qualitative description of such a system is presented.
Sample-based field data will still be required, for use in several ways, including developing the spectral-spatial-temporal classification (categorization) algorithm for coarse-resolution RS data, where appropriate, and + through double sampling or other procedures + for bias-correcting any crop identifications and area estimates made with either coarse- or fine-resolution RS data, or both.
Among the factors expected to affect costs and benefits are:
The pros and cons of centralized versus regional centres is addressed, as are the particular needs for various ancillary information.
The general conclusion is that RS has the potential to provide timely, accurate information on illicit-crop cultivation in wide areas, but that the technology currently available means that considerable effort is required in any one situation to develop a monitoring system that is easily usable by a non-specialist. Once the preliminary work to allow easy orthorectification of images and the superimposition of those images in exact registration in a GIS, subsequent activities should proceed routinely.