Spatial epidemiology of Highly Pathogenic Avian Influenza (HPAI) in Thailand

Transcription

1 Spatial epidemiology of Highly Pathogenic Avian Influenza (HPAI) in Thailand Exploring the use of remote sensing for risk assessment. Final report of a Letter of Agreement between the Food and Agriculture Organization of the United Nations and the Université Libre de Bruxelles. November 2005 Dr M. Gilbert Biological Control and Spatial Ecology CP160/12 Université Libre de Bruxelles Av F.D. Roosevelt 50 B-1050 Brussels, Belgium mgilbert@ulb.ac.be

2 2 Acknowledgements The author wish to thank Dr Prasit Chaitaweesub, and Dr Sith Premashthira from the Department of Livestock Development (DLD) as well as Dr Wantanee Kalpravidh and Dr Hans Wagner from the FAO Regional Office in Bangkok for their critical inputs, technical assistance, and useful comments and discussions. We would also like to thank all DLD staff who attended the workshop in Bangkok, and the staff of GIS Thai for their technical assistance. The contributions of participants of the workshop held in Rome in October 2005 are gratefully acknowledged. I am particularly grateful to Prof. Xiangming Xiao for his help and support with the processing of the remote sensing data.

3 3 Table of content I INTRODUCTION AND OBJECTIVES 4 The situation in Domestic ducks as HPAI reservoir? 5 Remote sensing to predict duck distributions 6 II WORKSHOPS ON HPAI SPATIAL EPIDEMIOLOGY 7 II. 1 Bangkok 9-12 th May 2005 (Chulalongkorn University) 7 II. 2 Rome 3-5 th October 2005 (FAO Headquarters) 8 III POTENTIAL OF REMOTE SENSING AS A TOOL FOR HPAI RISK ASSESSMENT 10 III.1. Statistics on rice distribution 10 III.2. Association between duck and rice distributions 13 III.3. Modelling the distribution of rice paddy fields using remote sensing 16 State of the art 16 Data processing 17 Results 19 III.4. Conclusions and recommendations 23 IV REFERENCES 24 V APPENDIX 25

4 4 I Introduction and objectives Starting early 2004, Highly Pathogenic Avian Influenza (HPAI) or bird flu swept across east and southeast Asia and left a trail of over one hundred million chickens, ducks and other poultry dead, removed or culled. Influenza type A viruses are known to cross species barriers and move into other host populations, including humans and as of 10 Oct. 2005, there have been 117 human cases of avian influenza A (H5N1) in Vietnam (91), Thailand (17), Indonesia (5) and Cambodia (4) resulting in 60 deaths (WHO 2005). To date there is no strong evidence of adjustment of the H5N1 virus to the human host, which would result in human-to-human transmission, and a possible pandemic of unknown magnitude. The 2004 HPAI H5N1 virus probably first evolved in China, where it has been reported, in time, to progressively circulate in domestic duck populations (Li et al. 2004). A broad variety of H5N1 viruses became widespread in the domestic duck populations in coastal and southeastern parts of China. H5N1 progressively encroached terrestrial poultry as well and finally triggered subcontinental scale spread of the more aggressive H5N1 genotype Z. The spread of this last genotype was particularly fast, possibly aided over long distances by migratory birds, and with secondary spread facilitated by live poultry trade movements. The evolution of the crisis during 2004 showed a retraction of the disease in some countries, mainly the established economies of Japan, South Korea, Malaysia, Hong Kong and the Taiwan Province of China (H5N2). On the other hand, HPAI H5N1 virus may have turned endemic in mainland China, Thailand, Viet Nam and in Indonesia. In the latter three countries and also in Cambodia the virus turned out to be a zoonotic infection. The situation in Lao PDR remains less clear but the country report suggests low incidence. The situation in 2005 In 2005, despite significant control efforts, HPAI continued to be reported as a persistent endemic problem in Thailand, Vietnam, Indonesia and probably also in China. This paved the way for geographical expansion elsewhere as the same H5N1 virus is now reported in domestic and wild birds in Central Asia and Eastern European countries (ProMed-Mail 2005). Still, the situation in southeast Asia appears relatively quiet. For example, Thailand reported much less outbreaks during the summer and early autumn of this year than it did for the same period of 2004 (Fig. 1) Daily cases 2004 Daily cases 2005 Weekly average 2004 Weekly average 2005 Reported cases /1/04 7/8/04 7/15/04 7/22/04 7/29/04 8/5/04 8/12/04 8/19/04 8/26/04 9/2/04 9/9/04 9/16/04 9/23/04 9/30/04 10/7/04 10/14/04 10/21/04 10/28/04 Fig. I.1 Distribution of lab-confirmed HPAI cases in Thailand between the 1 st of July 2004 and the 31 st October 2004 and 2005 respectively.

5 5 Whilst encouraging, these results need to be interpreted with caution because the Thailand epidemic only started to rise in September last year, and the subsequent epidemic that peaked in end October was rather unexpected. The future will tell if a new peak of cases will again be recorded in December, but what remains is that the virus circulation appears at a much lower level this year than in The few reported cases over the summer suggest that the virus has been forced back into its reservoir, with sporadic introduction into terrestrial poultry, mainly in quails and native chickens. The relatively low level of virus circulation suggests that more focus on this reservoir could possibly lead to HPAI extinction in Thailand. Hence, the ecological identification and delineation of this reservoir is therefore of very high importance. Domestic ducks as HPAI reservoir? Two studies highlight the critical role of domestic ducks as possible reservoir in the persistence and spread of HPAI in Asia. Hulse-Post et al. (2005) demonstrated that domestic ducks could remain relatively healthy whilst excreting sufficient amount of virus to sustain transmission, and may thus form an important reservoir. An analysis by FAO, LUBIES and the Thailand Department for Livestock Development (DLD) revealed that i) on a large geographical scale, at country level, there was no association between the distribution of chicken outbreaks and the abundance of chickens, suggesting a low level of chicken-to-chicken transmission, ii) the spatial distribution of HPAI outbreaks in both chicken and ducks in Thailand was strongly associated with the distribution of free grazing ducks, which corroborates earlier results pointing to domestic ducks as main HPAI reservoir (Gilbert et al. 2005). In 2005, DLD implemented movement control regulations, which probably have substantially contributed to the observed reduction in virus circulation, even if the number of those free grazing ducks in 2005 is similar to that of 2004 (Chaitaweesub 2005). Duck husbandry in most Asian countries is characteristically associated with rice production (feed) and presence of lowland areas (water). Ducklings are allowed on the paddies at the age of one month and herded to forage on the paddy fields until they are ready to lay. Scavenging ducks benefit feed utilization efficiency and take in large amounts of insects, snails and even grasshoppers. Thus, ducks also form an important aid in plant pest management. With duck husbandry involving frequent field movements of flocks, brought together in night shelters which are often located within the villages, and with marketing of live birds and eggs extending to beyond the village, apparently healthy ducks may indeed play an important role in virus transmission and thus explain the observed spatial pattern of HPAI. In addition, free grazing ducks are probably the domestic poultry species group most exposed to transmission by water that may become contaminated by wild birds that frequently forage in the same paddy fields. Analysis of rice statistics in Thailand, and in particular in the province of Suphanburi, showed that the duck production cycle is closely intertwined with rice cropping because it provides the duck feed (Gilbert et al. 2004). Most rice fields in the eastern part of Thailand produce one crop per year whereas crops located in the central plains permit the production of two or even three crops per year. Single rice crop areas do associate with duck farming but count less ducks and a production cycle so as to match the period of rice harvest. In contrast, in double rice crop areas there is year round availability of postharvest rice paddy fields, sustaining the low-input low-output free grazing duck farming system, and representing a very large proportion of the total duck numbers. With the duck production associated with rice fields, the hypothesis emerged that the timing of the rice production and harvest in fact drives the availability of duck feed, and with it the distribution of the free ranging ducks. Available national rice statistics suggest two basic crop seasons. The main season account for 86% (1988/ /93 average) of the annual rice harvest and thus constitutes the most important feed resource for the ducks. The planting season is from May to August during the wet season, the southwest monsoon, running from May to September. The harvest starts early October and runs through to end January. The second season is quite different in that planting is done during January/February, with the harvest during May/June. A most remarkable feature is the fact

6 6 that some 80% of the total production during the second season is confined to the central wetlands where free grazing ducks are found in the highest number. More detailed data were however required to firm up (or otherwise) on the statistical association between free grazing duck distribution and the distribution of second crop rice harvest areas. Remote sensing to predict duck distributions With the strong association between free grazing ducks and HPAI outbreaks, and also between rice and duck farming systems, an option that presented itself was to explore the application of remote sensing as a risk assessment decision support tool for HPAI. Building such utility for Thailand approach required several steps which were explored in this study: i) understanding better the free grazing duck husbandry, ii) test the association between free grazing duck distribution and rice statistics, iii) test how rice distribution can be predicted using multi-temporal satellite imagery. The duck movement control measures effectively implemented by DLD in 2005 may in effected have lowered the importance of free grazing ducks as a risk factor in 2005, as compared to its importance during the 2 nd wave in However, the study of the interrelationship between HPAI, free range duck and rice husbandry remains valid and important given that i) a duck movement is likely to continue to take place even in Thailand, escaping control; ii) the availability of leftover grains in rice paddy fields remains attractive to both migratory and resident wild birds that may harbour the virus; iii) other countries in the region such as Vietnam have to some extent similar rice and duck production practices, and duck movement schemes could also be considered here, subject to similar risk assessment scheme, and iv) this study provides the first step of a risk model that may incorporate relevant variables, such as the spatio-temporal distribution of wetlands (as main migratory birds habitat) and temperature (driving HPAI virus survival). This current assignment has two main components. The first part reports on interaction with DLD staff, technical support in spatial epidemiology, feedback in terms of expert knowledge on the Thai field conditions of rice and duck production in relation to HPAI. This part was achieved in a workshop in Bangkok in May The second part was carried out in the Université Libre of Brussels (ULB), and concerned the analysis of duck and rice statistics, and the use of satellite imagery to predict their distribution.

7 7 II Workshops on HPAI spatial epidemiology II. 1 Bangkok 9-12 th May 2005 (Chulalongkorn University) The workshop was organized in Bangkok by the FAO regional office, the Department of Livestock Development, the GISTHAI group from Chulalongkorn University, and ULB/FAO Rome and took place from the 9 th to the 12 th May 2005 in Chulalongkorn University. The workshop programme, introductory session, and technical manual are provided in appendix. Approximately 30 participants, mainly veterinary staff, attended the workshop, and the GISTHAI group provided technical assistance. As detailed in the workshop agenda (Appendix I), most of the time was spent providing technical training in the management, processing, and analysis of epidemiological GIS data. This was complemented by introduction to spatial epidemiology. Most of the final day was dedicated to discussions on the association between wetlands, rice production, free grazing duck populations and HPAI distribution. The results of these discussions are summarised below. Consumption. Duck is usually encountered in markets in the form of ready-made cooked meat, and rarely as fresh meat. The highest consumption periods go from October to February, in association with the Chinese new year, and April (Cheng Meng day). Free grazing ducks. The free grazing ducks are introduced as soon as possible after the harvest, at densities up to 4000 ducks / ha to feed on leftover rice grain as main source, but also cited were alternative feed resources such as snails, eggs of snails, and in some places also little fishes or crabs, and vegetables. The large majority of the free grazing ducks are raised for their eggs, and their typical production cycle is as follows: hatchery (mainly in Chonburi & Nakong Pathon; 28 days) ducklings in nursery (1 month mainly from Jul. to Aug.) grazing (14-15 months ) -slaughterhouses (in both areas). An alternative cycle, apparently more frequent in the southernmost plain area, allows the free grazing ducks to be brought back to farms after 4-5 months of free grazing for egg production. Free grazing ducks raised for meat have a similar cycle, except that they spend only 2-3 months free grazing, and are then brought back to farm for a month before being sent to slaughterhouses. The free grazing ducks are becoming less popular among consumers and producers because of the increased awareness of their possible role in HPAI transmission. Possible transmission to other poultry. Hatching and duckling nursing occur within villages (in contact with native chickens), and importantly the grazing flocks pass through villages, where there is possible contacts with other poultry. cocks. Fighting cocks. No seasonal or geographical pattern was reported for fighting Migratory Birds. Migratory birds are frequently observed from Nov. to Jan. in the same fields as free grazing ducks. These are identified as waterfowls, small wild ducks, egrets & herons, with some resident bird populations.

8 8 II. 2 Rome 3-5 th October 2005 (FAO Headquarters) A second workshop was held in FAO headquarters in Rome from the 3 rd to the 5 th October 2005 within the framework of this LOA. The aim of the workshop was to present and summarize the most recent advances in the understanding of HPAI agro-ecology in southeast Asia, and to suggest recommendations for future work. The workshop agenda and the list of participants are detailed below. The presentations of the workshop are available upon request to Dr Jan Slingenbergh (FAO AGA, jan.slingenbergh@fao.org). Agenda: Workshop on Remote Sensing and HPAI Risk Assessment (3-5 October 2005, Rome HQ) Monday, 3 rd October Introductory remarks/fao's response to the Avian Influenza crisis (Juan Lubroth, FAO AGA) Introduction of participants to each other/purpose of the workshop (Jan Slingenbergh, FAO AGA) Migratory birds and HPAI spread/proposal for epidemiological framework (Soares Magalhaes, Royal Veterinary College, UK) Remote Sensing (RS) and HPAI risk assessment (RA) in Thailand (Marius Gilbert, ULB Belgium) Rice and duck calendar in the central plains of Thailand (Prasit Chaitaweesub and Wantanee Kalpravidh, DLD Thailand) Thailand s priorities in the control and prevention of HPAI (Prasit Chaitaweesub: DLD and other government institutions; Wantanee Kalpravidh: private sector) What are the main questions/issues to be addressed in Thailand? (Jan Slingenbergh: general issues; Marius Gilbert: The scope for RS driven HPAI RA) Discussion groups/drafting of proposals (i) How to follow-up in Thailand? (ii) How can the international scientific/technical community and FAO assist? Plenary Session to harvest the results of the discussions. Tuesday, 4 th October Satellite imagery application in the clarification of land cover/land use and cropping patterns in Eastern Asia and China (Xiangming Xiao, CSRS, USA) Contrasting the Mekong and SE China agroecology (Robert Brinkman, FAO) SE Asia livestock-environment aspects; nutrient balances (Pierre Gerber, FAO AGL) EMPRES activities in China and Viet Nam (Vincent Martin, FAO AGA) Institutions in Hanoi that could be involved in RS in HPAI RA (Michael Epprecht, GIS Cons, Vietnam) Discussion groups/drafting proposals: How to build a platform for RS in HPAI RA, in (i) China and (ii) Viet Nam? Elaboration of concrete plans concerning Thailand/Viet Nam and China (Defining direct objectives, work programmes, timetables, actors and resources) Diverse satellite meetings Wednesday, 5 th October Environmentally suitable areas for pig production in China (Pierre Gerber) Physical planning, agroecology and traffic/trade in Asia; early recognition/mitigation of biological risks (Peter Kenmore, FAO AGPP) Asia: rice ecology and IPM (Peter Kenmore and William Settle) Asia: old and new forms of agriculture; do biologically safe intermediate forms exist? (Jan Slingenbergh) Discussion session moderated by Peter Kenmore on Asian agriculture, risks and solutions Workshop wrap-up

9 9 List of Participants: Workshop on Remote Sensing and HPAI Risk Assessment (3-5 October 2005, Rome HQ) Name Affiliation External experts Robert Brinkman Prasit Chaitaweesub Michael Epprecht Marius Gilbert Ward Hagemeijer Wantanee Kalpravidh Ricardo Soares Magalhaes Xiangming Xiao ex-director, Land and Water Development, FAO Dept. Livestock Development, Thailand GIS consultant Biological control and spatial ecology, University of Brussels Wetland Species Conservation Programme, Wetland International Regional Project Coordinator, FAORAP Bangkok Royal Veterinary College Inst. of the study of earth, oceans and space, Univ. of New Hampshire AGAH - Animal Health Service Parasitic Diseases Group - PDG Jan Slingenbergh Senior Officer Raffaele Mattioli Animal Health Officer Giuliano Cecchi RS-GIS consultant Alienor Fromont Volunteer student EMPRES Juan Lubroth Senior Officer Peter Roeder Animal Health Officer Vincent Martin Animal Health Officer Akiko Kamata Animal Health Officer Sophie von Dobschuetz Associate Professional Officer Francesca Pozzi GIS consultant William Karesh Wildlife consultant, Wildlife Conservation Society Craig Lewis Volunteer student AGAL - Livestock Information, Sector Analysis and Policy Branch Henning Steinfeld Pierre Gerber Chief Livestock Policy Officer Livestock, Environment and Development (LEAD) Initiative Tom Wassenaar Livestock & Environment Officer Pro-Poor Livestock Policy Facility (PPLPF) Joachim Otte Coordinator Katinka de Balogh Animal Health Officer AGAP - Animal Production Service Emmanuelle Guerne Bleich Barbara Rischkowsky Animal Production Officer Animal Production Officer AGPP - Plant Protection Service Peter Kenmore William Settle Senior Officer Technical Officer SDRN - Environment and Natural Resources Service Rene Gommes Senior Officer

10 1 0 III Potential of Remote Sensing as a tool for HPAI risk assessment As detailed in the introduction, the potential for remote sensing application in the risk assessment of HPAI in Thailand was explored mostly through modelling the distribution of the rice distribution, given that rice paddy fields had shown to be probable drivers of free grazing duck distribution and abundance, and as meeting point between wild birds and domestic poultry. The first step of this analysis involved the compilation of official rice statistics on the spatial and temporal distribution of rice planting and subsequent harvest throughout Thailand. The second steps involved matching the distribution of rice distribution against the distribution of free grazing ducks. The last steps focused on the use of remote sensing to model the distribution of rice paddy fields in space and time. III.1. Statistics on rice distribution Rice statistics were provided by DLD in the form of one text file by province and crop cycle (first and second crop). Each file included acreage measurements of newly planted area, planted area, and harvested area month by month, as well as the reported yield. Some province data were broken down by district. There were a total of 74 files and 66 files for the first crop and second crop respectively. These data were processed and integrated into a single Excel database that could be easily queried through PivotTable to generate monthly statistics by the available statistical unit (province or district level). The data on the first crop were patchy, with some information at the province level, and some information at the district level. In contrast, all second crop statistics were available at the district level. These data are summarized in figures III-1 to III First Crop Second Crop Newly planted area (Km 2 ) Apr May 2003 Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May 2004 Jun Jul Aug Sep Fig. III.1 Monthly distribution of the total newly planted rice areas in Thailand per statistical records (the actual data run from Apr to Apr 2004 for the first crop, and Sep to Sep for the second crop; remaining data have been reproduced for display purpose)

11 First crop Second crop Planted area (Km 2 ) Apr May 2003 Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May 2004 Jun Jul Aug Sep Fig. III.2 Monthly distribution of the total rice planted area under crop in Thailand as per statistical records (the actual data run from Apr to Apr 2004 for the first crop, and Sep to Sep for the second crop; remaining data have been reproduced for display purpose) First crop Second crop Harvested area (Km 2 ) Apr May 2003 Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May 2004 Jun Jul Aug Sep Fig. III.3 Monthly distribution of the total harvested rice crop area in Thailand as per statistical records (the actual data run from Apr to Apr 2004 for the first crop, and Sep to Sep for the second crop; remaining data have been reproduced for display purpose).

12 1 2 New First Crop Second Crop Harvested area (Km 2 ) Planted Apr May 2003 Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May 2004 Jun Jul Aug Sep Apr May 2003 Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May 2004 Jun Jul Aug Sep Apr May 2003 Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May 2004 Jun Jul Aug Sep Fig. III.4 Seasonality of newly planted (top), undercrop (middle), and harvested (bottom) rice crop area in Thailand as per statistical records (the actual data run from Apr to Apr 2004 for the first crop, and Sep to Sep for the second crop; remaining data have been reproduced for display purpose). First crop planting takes place mainly May to Oct. for the first crop, and from Nov. to Apr. for the second crop. The planting peak is in July. The period showing the largest planted area runs from July to Dec. and Jan to June for the first and second crop respectively. The harvest is mainly taking place in Nov. and Dec. for the first crop, much more so than during the previous or following months. The second crop harvest is much more scattered over time, with some harvest taking place from Feb. to July. As can be noted, the surface area planted for the first crop is much larger than for the second crop, except for the situation in the central plains where the first and second planting areas are equivalent (see Fig III) (see Gilbert et al. 2004, Appendix A). The geographical distribution of those plantings is presented in Fig. III-5. When we consider the geographic distribution of rice paddy fields (black and white insert), the statistic of the first crop rice statistics appear patchy at best. The information was not available for several provinces, and province-level statistics are probably a too coarse resolution to assess the match with observed patterns of cultivation. The second crop statistics are more comprehensive with all data available at the district level and few provinces with no data available, all distributed in areas unsuitable for second crop production anyway. The display of the second crop production shows the concentration of the production in the central plains of Thailand, with one localized area of two crop rice production in the eastern part of the country.

13 1 3 Fig. III.5 Proportion of land made up of rice paddy fields in October 2003 (left; peak of the 1 st crop period), and March 2003 (right; peak of the 2 nd crop period). Areas in white represent areas for which there are no data. The little black and white insert shows the distribution of rice paddy fields according to the land use database of Thailand. III.2. Association between duck and rice distributions The distribution of farm ducks, free grazing ducks and rice statistics are shown in Fig. III-6. The match between the geographical distribution of free grazing ducks and second rice crop areas is outstanding, especially if one recalls that the two data sets originate from entirely different sources: the X-Ray door-to-door survey and the official agriculture statistics for the duck and rice distribution, respectively. The results are very similar when it comes to the quantitative association between the availability of harvested rice paddy fields during the second crop (expressed in number of month with second crop harvested paddy fields) and the free grazing duck population (Fig. III.7). In contrast, the same plot with the number of ducks raised in farm shows no real association, with large number of ducks raised in areas with no second crop. Data on the seasonality of the first crop rice harvest were only available at the province level and the analysis of year-round data was relevant on province level data.

14 Spatial Epidemiology of HPAI in Thailand: exploring the use of RS for risk assessment 1 4 Fig. III-6 Distribution of all duck (top-left) and free grazing duck (top right) densities, and of the proportion of land used for second crop rice (bottom left)from available statistics, and of rice paddy fields (bottom-right; from land use maps).

15 1 5 Free grazing duck population (Log(x+1)) Farm duck population (Log(x+1)) Number of months with second crop rice harvest Number of months with second crop rice harvest Fig. III-7 Box-plot of the log-transformed number of free grazing ducks (left) and ducks raised in farms (right) by district, and the number of months with some degree of second crop rice harvest at the district level (The bold line is the median, the box limits are the 75% percentiles, the end of the vertical bars, the whiskers the min and max values, and the points outside those lines are those considered as suspected outliers). The first shows a much stronger statistical association with an F 1,294 statistics of and for free grazing ducks and farm ducks respectively. Free grazing duck population (Log(x+1)) Farm duck population (Log(x+1)) Number of months with rice crop harvest Number of months with rice crop harvest Fig. III-8 Log-transformed free-grazing (left) and farm(right) duck population as a function of the number of month with rice harvest all along the year at the province level for the full year period (Free grazing ducks: y = x 0.284, R 2 = 0.527; F 1,74 = 83, p < 0.001; Farm ducks: y = 0.067x , R 2 = 0.174, F 1,74 = 15.75, p < 0.001). Here again, the year-round number of months during which some rice crop harvest takes place constitutes a very strong driver of the size of the free grazing duck population, and much less so or not at all for farm ducks. (Fig. III.8). These results support the hypothesis that the distribution of rice forms the main driver of the distribution of free grazing ducks in Thailand. Ducks can be raised everywhere rice is found (as shown by the geographical match between all duck distribution and rice paddy fields distribution, and also by the significant association between ducks numbers and rice production in Fig. III-9), but year-round feed availability sustains the free ranging duck husbandry, representing nearly 13 million birds distributed in the central plain (whilst ducks in farms represent a total of 20 million birds).

16 1 6 Duck population (Log(x+1)) Duck population (Log(x+1)) Rice production (Tons) Rice production (Log(Tons)) Fig. III-9 Duck population as a function of rice production at the province level in Thailand in normal (left) and log (right) scales. The above relationships between free grazing duck and rice distributions indicate that the prediction of rice paddy field distribution using earth observation data may have interesting potential for the mapping of free grazing duck population in Thailand. III.3. Modelling the distribution of rice paddy fields using remote sensing State of the art Given the importance of rice as one of the world major crops, and food resource, the potential of earth observation satellite data for predicting rice distribution and production is not new and has been explored in previous studies using several approaches at different spatial and temporal resolutions. In particular, the use of optical sensor data is reviewed in Xiao et al.(2002) and more recently in Xiao et al. (2005). One of the challenges of predicting rice crop distribution using optical sensor data is that a significant part of production period takes place during the monsoon period, at a time when clouds cover large parts of the country. This can be resolved by applying a multi-temporal approach to generate cloud-free imagery, but only if the temporal frequency is high enough, which precludes the use of high spatial resolution images such as those generated by SPOT and Landsat. The production of rice is characterised by a flooding stage that corresponds to the time of planting. Xiao et al. (2005) identified the spectral fingerprint corresponding to this particular stage as temporary inversion between vegetation indices (high before harvest, and after the first growth, but low at the time of flooding), and a moisture index (showing a peak at the time of flooding), followed in a relatively short period of time by an increase in the vegetation index. These indices require the use of a middle infra-red band, which is part of the channels available in the MODIS and SPOT VEGETATION imagery. These two satellites produce images of a spatial resolution of 500 m and 1 km, and are available as 8-days and 10 days composite, respectively. As the association between free grazing ducks and rice distribution was not considered to be very strong at fine resolution but more likely to show up at the landscape scale, we chose to apply SPOT VEGETATION data because this also significantly reduced the processing time.

17 1 7 Data processing We used SPOT VEGETATION 10-days BiDirectional Composite syntheses that can be downloaded free from the VEGETATION Programme 1. The 10-days composite radiometric data were downloaded for Southeast Asia for the period going from March 2003 to March 2005, so that the data series could be tested against the rice statistics, and the 2 nd wave of HPAI outbreaks in late 2004 / early The data series therefore consisted of 75 compressed files (one for each 10-days period), for a total of approximately 32 GB. Preliminary treatments of those compressed data included the extraction of the 4 spectral bands from each composite file [B0 ( nm, blue), B2 ( , red), B3 ( , near-infrared), MIR ( , mid infrared)] and for the area of interest. This step was carried out using the software Crop_VGT written by Silvio Griguolo 2. Three indices were estimated for each 10-days period of the time series using the available bands (the characteristics of these three indices are detailed in Boles et al. 2004). Normalised Difference Vegetation Index NDVI: (B3-B2)/(B3+B2) Enhanced Vegetation Index EVI: 2.5 x ((B3 B2) / (B3 + 6.B2 7.5B0 + 1)) Land Surface Water Index LSWI: (B3 MIR)/(B3 + MIR) The rice detection or demarcation process itself was carried out in two main steps. The first step involved masking out areas that can be identified as clouds, permanent water or evergreen vegetation, and to leave out those pixels for the rest of the analysis (details of the procedure can be found in Xiao et al. 2005). The second step involved classifying as rice all pixels where LSWI plus a constant value (equal to 0.05 in Xiao et al. 2005) is higher than the NDVI, or the EVI, and where this inversion is followed by a rapid increase in EVI (EVI 40 days later is higher than half the maximum EVI in the ten following periods). The two steps were implemented in R-package (R Development Core Team 2004) at a resolution of 1 km by pixel

18 1 8 Fig. III-10 Predicted distribution of number of rice crops (left) between May 2003 and April 2004 as compared to the land use database of rice paddy fields distribution Observed Predicted Newly planted area (Km 2 ) Mar Apr May 2003 Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Fig. III-11 Observed and satellite-derived predictions of newly planted areas in Thailand.

19 1 9 Results As can be seen from Fig. III.10 there is a good overall match between the predicted distribution of rice planting and the distribution of rice paddy fields obtained from landuse data. The method also captures the seasonality of rice planting, with both predicted and observed rice planting curves running in parallel, with a month lag between the peak of the statistics on planting and their detection by the earth observations techniques (Fig. III-11). The predictions distribution of first crop and second crop are compared in Fig III- 12, which shows important deviations between satellite-derived predictions and rice statistics during the second crop season. As second crop rice statistics and the distribution of free grazing ducks are known to match, it is believed that the observed discrepancies observed here stem from the satellite-derived prediction areas. There are apparently some problems with the identification of the second crop rice planting areas in the northern parts of the central plains, and tendency to overestimate second rice crop areas in southern Thailand. Fig. III-12 Predicted distribution of first rice crop planting (left) from May to Oct. 2003, and of second rice crop plantings (right) from Nov to Apr. 2004, with the distribution derived from statistical records in small inserts.

20 2 0 The method developed by Xiao et al. (2005) and the different thresholds used aimed to map the year-round area of planted rice paddy fields. The thresholds, in particular the value added to the LSWI to detect inversion, require to be adjusted however, if one wishes to map all pixels where rice fields may have been present. For example, Fig.III-13 shows the results of the first and second rice crop distributions when this value is set as This clearly overestimates rice acreage because not all pixels can be assumed to be covered by 100% rice. A more realistic distribution of second crop predictions can be produced with an adjusted threshold fitting the distribution of second crop statistics and free grazing duck populations. Fig. III-13 Predicted distribution of first rice crop planting (left) from May to Oct. 2003, and of second rice crop plantings (right) from Nov to Apr. 2004, using a higher threshold for LSWI and NDVI or EVI inversions. When considering the association between second rice crop and the distribution of 2 nd wave HPAI outbreaks (Fig. III.14) the association appears very strong, in particular along the northern part of the central plain rice paddy fields, which coincides the area where outbreaks were reported in ducks. Of course, many of the small patches distributed across the country outside the central plains are probably false positives, and do not necessarily correspond to areas with free grazing ducks, blurring the statistical association between HPAI distribution and predicted second rice crop distribution.

21 2 1 Fig. III-14 Predicted distribution of second rice crop planting with 2 nd wave outbreaks in chicken (left) and ducks (right). Balancing the discrepancies between rice planting, rice statistics, and free grazing duck distribution is critical to the understanding the predictions one can derive from the satellite-derived rice distribution maps. For example (Fig. III-15), the province of Phra Nakhon Si Ayutthaya reported hardly any free grazing ducks whereas both second crop rice predictions and statistics indicated plenty of rice for free grazing duck raising. This calls for verification of the Xray survey data on free ranging ducks. Conversely, there was no second rice crop planting figuring in the statistics concerning southern Thailand, despite the presence of a local subpopulation of free grazing ducks was reported in the X-ray census, also the satellite-derived predictions indicated a significant amount of second rice crop.

22 2 2 Fig. III-14 Grazing duck distribution (top left) with the province of Phra Nakhon Si Ayutthaya marked, distribution of second rice crop plantings as predicted from the satellite imagery (top right) and extracted from statistical records (bottom left).

23 2 3 III.4. Conclusions and recommendations Two main conclusions can be drawn from the results presented in this study. First, the distribution of free grazing ducks in Thailand is strongly driven by the availability of the feed resource. More in particular, we have shown that areas where some rice paddy fields harvest is taking place throughout the year these are the areas where the highest population of free grazing ducks can be raised. This is an important result, which offers prospects for application in areas where statistics on duck production systems are not as detailed as in Thailand. In addition, these results help to explain the links between the dynamic production of duck meat and eggs to the spatiotemporal feed availability. Second, earth observation data have the potential to be applied to map the distribution of rice fields planting in space and time, even if minor refinements to the algorithm are still required to obtain the best balance between the spatial sensitivity and specificity of the technique. In particular the accuracy of the method in terms of predicting the distribution of rice planting over time remains to be more fully assessed. Further developments should include: Predicting rice harvest. It would be useful to develop algorithms permitting the identification of rice harvest, given that the movements of free grazing duck flocks directly follows the rice harvest; the duck population dynamics is expected to be more closely driven by the dynamics of the harvest than of the planting. Exploring the rice / duck association in other countries with high duck productions. The seasonality in rice production in Asia differs strongly from areas with year-round rice production (e.g. Indonesia, Philippines), to areas where rice can only be grown for a short period of time (e.g. Northern China). These patterns can be explored using RS based techniques. Moreover, duck husbandry certainly will show identical variation, and it is therefore critical to understand how duck production cycles relate to the rice calendars in other countries. Spatio temporal model. Wetlands, rice fields, free grazing duck and water temperature (because virus survival in water contaminated with faeces is a function of temperature) are key variables of HPAI ecology. A next step would therefore be to develop a spatio-temporal model of the association between the rice calendar, free grazing duck production, water temperature, and the distribution of HPAI outbreaks.

24 2 4 IV References Boles, S., Xiao, X., Zhang, Q., Munkhutya, S., Liu, J., and Ojima, D.S., 2004, Land cover characterization of Temperate East Asia, using multi-temporal image data of VEGETATION sensor, Remote Sensing of Environment, 90(4): Chaitaweesub, P. (2005) Thailand s priorities in the control and prevention of HPAI. Workshop on Remote Sensing and HPAI Risk Assessment, 3-5 October 2005, FAO Headquarters. Gilbert, M., Chaitaweesub P., Parakamawongsa, T., Premashthira, S., Kalpravidh, W., Wagner, H. & Slingenbergh, J. (2004) Highly Pathogenic Avian Influenza in Thailand: a two-scale analysis of second wave outbreaks. Consultancy report prepared for FAO Animal Health Division (Rome, Italy), 22 pp. Hulse-Post DJ, Sturm-Ramirez KM, Humberd J, Seiler P, Govorkova EA, Krauss S, Scholtissek C, Puthavathana P, Buranathai C, Nguyen TD, Long HT, Naipospos TSP, Chen H, Ellis TM, Guan Y, Peiris JSM and Webster RG (2005) Role of domestic ducks in the propagation and biological evolution of highly pathogenic H5N1 influenza viruses in Asia. Proc Natl Acad Sci USA, 102: Li, K. S., Y. Guan, J. Wang, G. J. D. Smith, K. M. Xu, L. Duan, A. P. Rahardjo, P. Puthavathana, C. Buranathai, T. D. Nguyen, A. T. S. Estoepangestie, A. Chaisingh, P. Auewarakul, H. T. Long, N. T. H. Hanh, R. J. Webby, L. L. M. Poon, H. Chen, K. F. Shortridge, K. Y. Yuen, R. G. Webster, and J. S. M. Peiris Genesis of a highly pathogenic and potentially pandemic H5N1 influenza virus in eastern Asia. Nature 430: R Development Core Team (2004). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN , URL Xiao, X., Boles, S., Liu, J., Zhuang, D., Frolking, S., Li, C., Salas, W., and Moore, B. III., 2005, Mapping paddy rice agriculture in southern China using multi-temporal MODIS images, Remote Sensing of Environment, 95(4): Xiao, X., S. Boles, S. Frolking, W. Salas, B. Moore, C. Li, L. He, and R. Zhao, 2002, Observation of flooding and rice transplanting of paddy rice fields at the site to landscape scales in China using VEGETATION sensor data. International Journal of Remote Sensing, 23(15):

25 2 5 V Appendix Workshop on HPAI spatial epidemiology (Bangkok 9-12th May 2005) See enclosed document.