Several recent studies have applied remote sensing techniques to estimate the extent of oil palm plantations in Southeast Asia. Studies aimed at quantifying plantation extent on a national scale or larger have deployed visual, expert-based interpretation methods 2 , 3 , 4 , 5 , 14 or semi-automatic approaches 15 , blended with extensive field information 16 , These approaches are, however, very labour intensive.
The availability of radar imagery has given rise to a new generation of remote sensing studies benefiting from penetration of clouds, improved resolution and high revisiting frequency 18 , 19 , Sentinel data have proven particularly useful in detecting smallholder and industrial plantations 22 , 23 , 24 and plantation age 25 , which, in turn, is a good predictor of oil palm yields 10 , Productive oil palm plantations comprise both industrial and smallholder plantations that have an age of at least three years.
We have excluded roads, mill infrastructure and other land uses occurring in oil palm-dominated landscapes. The methodology for determining the extent of plantations builds on Sentinel-1 microwave data whereas the Landsat archive has allowed us to build a time series reaching back to the s to determine when the plantation was established. The methodology identifies oil palm plantations of all types independent of weather and daylight conditions or ground-sourced information amidst other tree crops in the landscape.
We determined the accuracy of the resulting oil palm map against an independently collected and interpreted sample of very high-resolution satellite images. The main input data used in the detection of oil palm include: 1 the Copernicus Sentinel-1 microwave backscattering time series; 2 the Landsat multispectral time series; and 3 additional auxiliary data sets. Oil palm area detection requires preparation of Sentinel-1 data and the application of an unsupervised oil palm detection routine based on the Normalized Difference Vegetation Index NDVI and texture features.
The combination of NDVI and texture is important to reliably distinguish oil palm from other woody vegetation. To this end, an annual mosaic of backscatter coefficients for was produced by processing all available Sentinel-1 data 1A and 1B with single polarizations VV and VH; V is vertical, H is horizontal. The main objectives in producing this annual mosaic were: i to build a baseline C-band backscattering coefficient mosaic of Southeast Asia; and ii to assess radar backscatter changes between different years in land cover change studies.
To calculate the SAVG, all input data were scaled to bytes. We deployed a stratified unsupervised classification algorithm to detect oil palm without training data; a conceptual overview of the workflow is presented in Fig.
First, to account for regional differences, we divided Southeast Asia into 12 equally sized grids with a side length of five degrees and randomly allocated 50, training points. For each grid, the detection algorithm returns the optimal number of clusters between 10 and 16 to fit the sample; each cluster represents one land cover class or what the algorithm recognizes as having a similar pattern.
Subsequently, we identified those clusters that correspond to oil palm areas by comparing them with very high-resolution imagery. We then applied the oil palm classification algorithm from each grid to the entire study area, resulting in twelve, at times contradicting, oil palm maps. We reconciled these competing classifications by applying a majority filter, i. Post-processing was used to apply rule-based corrections to the intermediate oil palm map resulting from the unsupervised classification.
These corrections were necessary in settlements, mangroves and sloping areas, as well as for other objects like roads running through the plantations. From a reflectance and backscattering viewpoint, settlements are very heterogeneous, such that the backscattered values in some parts of settlements are similar to those observed from oil palm fields.
Setting a minimum NDVI value of 0. Similar confusion exists with mangroves. This was resolved by deploying a mangrove mask, derived from three regional and global mangrove maps 27 , 28 , 29 , and removing areas falsely classified as oil palm from within the mangrove area. On Kalimantan, coverage by Sentinel satellites is in descending orbit only, i. This results in gaps, notably in hilly areas.
This process introduced a de facto minimum mapping unit of m 2 for the oil palm map. For a refinement of the final product, we removed pixels with NDVI less than 0. This was required to correct the Sentinel 1 signal with respect to hillshades As an oil palm canopy develops over time, the area fraction of bare soil decreases, which can be detected with optical remote sensing through the bare soil index BSI.
To estimate plantation stand age, we moved backwards in time from oil palm plantations detected in until bare soil exceeded a certain value in young stands. For this we used surface reflectance values from the Landsat 5 and Landsat 7 image collections. We then generated a time series of BSI development starting from To smooth the time series and to remove noisy observations, we calculated the median BSI from a moving window of 12 months of the time series and omitted any month time period in which we had less than three observations.
To determine a reference BSI for oil palm, we calculated a histogram of BSI values from established plantations in and used the 95 th percentile as a threshold value. The 95 th percentile coincides with the presence of very young plantations two to three years of age with an open canopy and a high BSI , as confirmed by the visual interpretation of high-resolution imagery.
For all pixels above the threshold in , we analysed the BSI time series backwards in time where canopy closure is defined as the point when the BSI index drops below the cut-off value. This means that the resulting oil palm age map will register the first observation of oil palm plantations at an age of two to three years.
At this point, the plantation is 2 to 3 years of age. The data values range from 0 to 37 where 0 is the No Data value.
Values 1 to 3 are not present and a value of 4 corresponds to the year , the first year oil palm was detected, and each consecutive number represents the next year, i.
Table 1 provides a summary of the extent of oil palm plantations in by country, with the percentage that was planted before and the percentages planted between and and between and , showing the acceleration in planting over time. Figure 2 shows the location of the remotely sensed oil palm plantations in Indonesia, Malaysia and Thailand in , classified by year of planting.
An overview of the extent and age of detection of oil palm plantations in Indonesia, Malaysia and Thailand. Figure 3 shows zoomed in locations of remotely sensed oil palm plantations in Indonesia, Malaysia and Thailand, classified by year of planting. We validated the oil palm plantation map using visual interpretation of very high-resolution satellite imagery with a random stratified sample. This is a conservative approach to accuracy assessment as the focus is on areas that are difficult-to-map, which will most likely reduce the accuracy numbers compared to other approaches such as that adopted by Descals et al.
For each randomly located sample, we downloaded an image from Bing Maps or Google Maps using an automated workaround implemented in R. After removing those samples where very-high resolution images were not available, 44 independent citizen scientists performed visual interpretation of the 8, sites with the available very-high resolution imagery.
The Picture Pile application for collection of validation data. Each image was assessed 5 to 8 times by different users, allowing for cross-user comparison of the same image. The results are mapped in Fig. Previous data collection campaigns with Picture Pile have shown high accuracies in identifying cropland 38 and building damage assessment Although the majority of the imagery used in the validation was dated between and based on a sample of images selected, some of the imagery used in the validation was older.
We acknowledge the limitation of this approach due to the temporal availability of very high-resolution satellite imagery, which may contribute to some omission errors. A map of the results of the validation procedure across South-East Asia.
Colours of the points indicate agreement of remotely sensed and visually interpreted classifications. A further comparison was undertaken between the map produced here and the one from Descals et al. We randomly distributed , points across regions in common between both maps. Hence there is good general agreement between these two products although we additionally provide age of detection.
We do, however, acknowledge that the age has not been validated in the same way as oil palm extent using visual interpretation, as this was not possible due to the lack of sufficient very high-resolution imagery to carry out a rigorous accuracy assessment.
Very high-resolution satellite imagery is also provided to show forest before and oil palm in subsequent years after Similar manual verification was carried out at other locations to demonstrate that the method worked. We acknowledge that there is some confusion between oil palm and coconut, which will lead to possible commission errors. These patterns are visible when we overlaid polygons of coconut plantations 39 onto our map and the map of Descals et al.
Complementing national inventories with the map produced here could support the calculation of spatially-explicit estimates of greenhouse gas emissions and removals 40 , while also allowing for an independent check of the official statistics. The oil palm map in combination with spatial information about estate boundaries would allow specific actors and their adherence to environmental legislation and compliance with sustainability standards to be identified.
The oil palm map could also be used in analyses related to determining the economic trade-offs in different types of land use. Plantation stand age is an important predictor of oil yield as palm age influences the quality and quantity of the fresh fruit bunches In that sense, renewing oil palm stands at an age of around 25 years is one potential contribution towards reducing oil yield gaps, notably in the oldest production areas such as parts of Malaysia.
Woittiez et al. Furthermore, interlinkages exist between overaged plantations and these causes: high-yielding tenera varieties have been becoming more widely used over time, meaning that old plantations still contain higher fractions of dura. Many diseases tend to strike harder on old plantations and harvesting of older and hence higher palms requires more labour.
The remote sensing product presented here can be used to estimate productivity and assist high-level planning of the oil palm sector to adapt its strategies regarding plantation management, e.
Finally, the generation of such a data set in near real-time could provide timely, independent, transparent and consistent monitoring of palm oil production across large geographical areas, bridging the gap between technology and land policy. Earth, Planets and Space volume 68 , Article number: 98 Cite this article. Metrics details. A probabilistic seismic hazard analysis PSHA for Thailand was performed and compared to those of previous works.
This PSHA was based upon 1 the most up-to-date paleoseismological data slip rates , 2 the seismic source zones, 3 the seismicity parameters a and b values , and 4 the strong ground-motion attenuation models suggested as being suitable models for Thailand.
For the PSHA mapping, both the ground shaking and probability of exceedance POE were analyzed and mapped using various methods of presentation. In a comparison between the ten selected specific provinces within Thailand, the Kanchanaburi and Tak provinces had comparatively high seismic hazards, and therefore, effective mitigation plans for these areas should be made.
Although Bangkok was defined as being within a low seismic hazard in this PSHA, a further study of seismic wave amplification due to the soft soil beneath Bangkok is required.
At present, much evidence supports the idea that Thailand is an earthquake-prone area. Paleoseismological investigations have indicated that Thailand is dominated by active fault zones Charusiri et al. During the past century — , seven published isoseismal maps Pailoplee have depicted that Thailand and, in particular, the northern and western parts have been subjected to earthquakes of an intensity range of II—VII on the modified Mercalli intensity MMI scale according to both local-moderate M w of 5.
Based mainly on the present-day instrumental seismicity data, Pailoplee and Choowong investigated and revealed that most of the seismic source zones in mainland Southeast Asia area are seismically active. In addition, according to the region—time—length algorithm Huang et al. This evidence indicates that Thailand is not shielded from earthquake hazards. As a result, the probabilistic seismic hazard analysis PSHA Cornell ; Kramer in Thailand has been progressively modified over the last three decades Table 1.
The detailed location and earthquake source parameters of each fault are expressed in Additional file 1. The pink dots are the earthquake data recorded from to present.
Triangles denote the locations of the ten significant provinces recognized in this PSHA. The black squares are the new sites of paleoseismological investigations used in this study with more details shown in Fig. After the devastation following the M w 9. Based on 10 seismic source zones SSZs , 18 active faults in Thailand, and various weighting schemes in attenuation models Youngs et al.
In addition, Pailoplee et al. Compared with the nine existing strong ground motions that had been recorded in northern Thailand, the attenuation models of Kobayashi et al.
Finally, Ornthammarath et al. After weighting some attenuation models Zhao et al. During the last 5 years, some data, assumptions, and models have been improved and altered significantly. In addition, based on the latest hazardous earthquakes of 6. The results obtained should help in the understanding of the severity of earthquake hazards and allow the necessary action to be taken to sustain the development of new, as well as the ongoing, engineering works, including serving as a resource for the further development of effective earthquake mitigation plans for Thailand.
Maps of different provinces in Thailand showing the locations of the new paleoseismological investigations used in this study. The index of these maps is illustrated in Fig. The number of each site is equivalent to the column no. With the present-day tectonic activities of the Indian-Eurasian plate collision, a number of seismogenic faults have originated within and nearby Thailand.
However, due to the limitations of the investigated fault data, most previous PSHA has roughly applied the SSZs as the earthquake sources Shrestha ; Warnitchai and Lisantono ; Pailoplee et al. Although Petersen et al.
The geometry and strike of each fault do not exactly conform to the details compared with the geomorphological evidence, e. In addition, some utilized seismogenic faults are ambiguous, e. Quantitatively, Pailoplee et al. Theoretically, the paleoseismological parameters of 1 the maximum credible earthquake MCE , 2 the rupture area, and 3 the rate of fault slip should be defined in each fault segment.
Nevertheless, according to the limitation of paleoseismological data, Pailoplee et al. For example, Pailoplee et al. In addition, according to the strong ground-motion attenuation model of Kobayashi et al. Up to the present, at least 55 sites of paleoseismological investigations in Thailand have been reported in addition to 13 technical reports of paleoseismological investigations Table 2. For example, there are 31 locations no. The fault slip rates cover a range of 0.
In some fault segments, more than one site for each active fault has been investigated. This gave fault slip rates from 0. In western Thailand Fig. In addition, according to the projects by the RID , 11 sites no. Among these, three sites no. For the other eight sites, all in the Klong Marui fault no. It is widely recognized that paleoseismological data are significant characteristics in deriving a reliable PSHA Andreou et al. When more paleoseismological evidence is used, the PSHA is likely to be more accurate.
In this study, the location, the geometry, and the strike of each fault were, therefore, recognized according to Pailoplee et al. According to the 55 additional paleoseismological investigations, i. As mentioned above, where fault segments had active fault data at more than one site, the highest fault slip rate was utilized.
The other paleoseismological data from outside Thailand also required for the PSHA were obtained from publications and technical reports Pailoplee et al. The MCE, the rupture areas, and the fault slip rates were obtained from the investigation of the active faults at each specific individual site. In Fig. Therefore, in addition to the active faults recognized in this PSHA, the SSZs were also applied in this study as the background seismicity.
Based on the available literature, there are at least three models of SSZs for mainland Southeast Asia Nutalaya et al. According to the updated data and reasonable assumptions, the 13 SSZs of zones A—M proposed by Pailoplee and Choowong were used in this study Fig. The a and b values of the Gutenberg—Richter relationships of each SSZ, including the fault data within each SSZ, were provided by the most up-to-date data provided by Pailoplee and Choowong , as given in Table 3.
The detailed location and earthquake source parameters of each earthquake sources are expressed in Additional file 1. Conceptually, every point within each earthquake source was assumed to have the same probability of being the epicenter of a future earthquake Erdik et al.
But the information is as practical as the drawing is artistic. River systems, mountains, forts, and towns are precisely located and identified. Another special-purpose map of the same part of the interior and from the same period is by Henri Mouhot, the French naturalist who rediscovered Angkor in Cambodia. He drew ten route maps now in the Royal Geographical Society in London for three journeys from Siam to Cambodia and Laos in the years John Arrowsmith, draftsman and map publisher, who had produced earlier maps of South-East Asia, assisted Mouhot in organizing his maps.
Notations were written on the map by Mouhot. Its unique position as the only country in South-East Asia that was never colonialized undoubtedly retarded foreign mapping of Thailand. And, perhaps the Thais did not feel the need for maps since precise borders were not established until foreign intervention in surrounding countries forced the delineation of erritorial boundaries in the last half of the nineteenth century.
Thus, the arrival of modern cartography in Thailand did not begin until when the Royal Survey Department was officially established by royal decree Maps of Bangkok The stamp of the Royal Survey Department appeared on maps thereafter. Foreign advisors and Thais worked together, under the patronage of King Chulalongkorn, for the next twenty-five years on the mapping of Thailand. The second area of focus of this period was town plans of Bangkok and charts of the access.
Richards in chart the river from the mouth of the gulf to Bangkok. It gives detailed sailing instructions for navigating across the bar into the entrance of the river. The title on the map says it was drawn by a native. The place names mix transliterations of Thai and English words. While unimpressive in content, this small plan has an appealing naivete. A town plan by the Royal Survey Department was compiled in Seven symbols represent canals, paddy fields, orchards, tramways, roads, buildings, and houses along both sides of the Menam Chao Phraya.
This detail shows the Grand Palace and fortified wall around it. The growth of the city is visible by the concentration of shops and districts grouped around the outer wall of the palace.
The Port of Bangkok was charted by McCarthy and his team in It was printed in English and showed European private residences, floating houses, and New Road, along the river. Until the last eight years of the nineteenth century, maps were sent to England for making copper plates and for printing. This plan is representative of the last phase of foreign input in the mapping of Thailand. A detail of a town plan of Bangkok of is an example of full-fledged Thai cartography Plate It was printed in four colours which were used to depict symbols.
Toponyms on this map are in English. It identifies wooden buildings, railway tracks, dykes, boundaries, and even foot paths. Aerial photography for maps was introduced later in the twentieth century. The highest degree of sophistication in the mapping of Thailand to date was achieved in when the Royal Thai Survey Department issued a satellite map of Bangkok Plate As reflected in the title, this paper is only an introduction.
More work needs to be done. Many maps of Thailand undoubtedly lie buried in the archives of Burma and India. The records of all nationalities of missionaries in Thailand need to be conscientiously examined. And, the search for indigenous maps must continue.
It is in these directions that I shall turn my future research. Birch, De Gray W. The Commentaries of the Great Afonso Dalboquercfue. London: The Hakluyt Society, vol 2.
Boeles, J. Jan Crawfurd, J. Singapore: Oxford University Press. Crone, G. London: Hutchinson University Library,. Ferrand, G. Gerini, Colonel G. Hakluyt, R. The Principall Navigations. AndDiscoveries of the English Nation. Jumsai, S. In Memorium Phya Amman Rajadhon, pp. Kaempfer, E. A Description of the Kingdom of Siam Bangkok: White Orchid Press. Kennedy, V. Bangkok: Siam Society, pp.
La Loubere, S. The Kingdom of Siam. Extra Series No. Maps of Bangkok: A. Patanonda, Major Ubolwan. Topographic Atlas of Thailand. Promboon, S. The Siamese Maritime Trade, A. Cisarua, Indonesia: Nov. Smithies, M. Oct Living in Thailand. Sternstein, L.
0コメント