- Количество слайдов: 43
Applications of Geostationary Satellite Data for Drought and Fires: Indian Experience C. P. Singh S. Panigrahy, J. S. Parihar Space Applications Centre (ISRO) 2 nd SALGEE Workshop TSMS – EUMETSAT “MSG Land Surface Applications: Drought & Fires” (4 – 7 April 2011, Antalya, Turkey)
Introduction Wildfire is a major natural disturbance. Forest fires: critical in the context of climate change. Deciduous forests || summer seasons || increasing trend. Geostationary satellite: real time detection of fire, rate, and fire spread and progression Important to do: validation of the binary active fire detects characterization of fire detects fire temperature modeling. ATSR/AATSR based fire analysis Maharashtra > M. P. > Chhattisgarh > Orissa > Gujarat > Jharkhand > Karnataka > Mizoram > Uttaranchal > U. P. > A. P.
India’s fire scenario The main contributors to the biomass burning are forests and Agricultural lands 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Tree Cover, broadleaved, evergreen Tree Cover, broadleaved, deciduous, closed Tree Cover, broadleaved, deciduous, open Tree Cover, needle-leaved, evergreen Tree Cover Regularly flooded: Mangrove Shrub Cover, closed-open, deciduous Shrub Cover, open, deciduous Herbaceous cover, closed-open Herbaceous Cover, closed-open with sparse trees Sparse Herbaceous or Sparse Shrub Cover Cultivated and Managed Areas Mosaic Types- Tree Cover/ Other Natural Vegetation Bare Areas Water Bodies (Artificial and Natural) Snow and Ice (Natural and Artificial) Land cover data source: http: //www. gvm. jrc. it/GLC 2000/ Artificial Surfaces and Associated Areas Input Data : Spot Vegetation and DMSP data ; Period : 1 -11 -1999 to 31 -12 -2000 This has been used as an input to RCM.
INSAT-3 D Algorithm & Status of Algorithm for Fire detection using INSAT-3 D Imager data. S. N. Components Status 1 ATBD preparation Completed 2 Algorithm /Software prototype Developed 3 Software integration at IMDPS Done 4 Software Testing at IMDPS Testing using MODIS data INSAT-3 D Imager based testing to be done 5 Sensitivity of algorithm parameters Done with MODIS data To be done with Imager 6 Validation Plan Document Prepared CAs finalization The hotter portion of the pixel will contribute relatively more radiance in the shorter wavelengths than in the longer wavelengths To Do
Remote sensing of fire : üAny object above absolute zero (0 o. K; -273. 16 o. C; -459. 69 o. F emits EMR. üAs temp. increases the maxima shifts towards shorter W. L. üλmax=2897/T ; µm (Wien’s displacement law) can give max. W. L. at which the exitance is maximum & is related to Temp. The hotter portion of the pixel will contribute relatively more radiance in the shorter wavelengths than in the longer wavelengths The band at 3. 9µm has a strong thermal response even if only a small portion of the pixel is covered by fire and this characteristic is vital for fire detection. Spectral Exitance Distribution Curve and the location of INSAT-3 D Channel 3, 5 and 6
Testing fire algorithm
Each condition in the algorithm brings down the number of candidate pixels qualifying to be flagged as fire pixels.
Performance of the algorithm • The RMSE values varied from 1. 972 to 3. 444 pixels for 11 x 11 kernal with 2. 5σ and 3 x 3 kernel with 4σ in the fire detection algorithm, respectively. • In general the RMSE value decreased with increase in kernel size. Kernel Size σ Value Range 13 11 9 7 5 3 Range 4 2. 214 2. 338 2. 478 2. 678 3. 072 3. 444 1. 230 3. 5 2. 080 2. 120 2. 242 2. 458 2. 826 3. 315 1. 235 3 2. 140 2. 004 2. 110 2. 313 2. 740 3. 200 1. 196 2. 5 2. 616 1. 972 2. 255 2. 204 2. 646 3. 120 1. 148 2 3. 131 2. 324 2. 622 2. 426 2. 606 3. 108 0. 807 1. 051 0. 366 0. 512 0. 473 0. 466 0. 335 The MAE varied from 1. 704 to 2. 905 pixels using 11 x 11 kernal with 3σ and 9 x 9 kernel with 2σ, respectively. The least MAE of 1. 704 pixels was obtained using 11 x 11 kernel size and 3σ. σ Value Kernel Size 13 11 9 7 5 3 4 1. 831 1. 887 2. 000 2. 127 2. 451 2. 817 3. 5 1. 820 1. 845 1. 958 2. 268 2. 732 3 1. 845 1. 704 1. 775 1. 915 2. 239 2. 662 2. 5 2. 141 2. 379 1. 845 1. 817 2. 183 2. 577 2 2. 451 2. 905 2. 085 2. 028 2. 169 2. 592
Performance of the algorithm… Index of agreement (d) indicates relative error of the estimate. ‘d’ varied from 0. 443 to 0. 802 using 3 x 3 kernal size with 2σ and 11 x 11 kernal with 3σ, respectively. The max. ‘d’ was obtained using the 11 x 11 kernel size and 3σ. Increasing the kernel size from 9 to 13, ‘d’ value approached saturation and that the variation then onwards was observed due to the change in the sigma (σ) σ Value value. Kernel Size Range 13 11 9 7 5 3 Range 4 0. 771 0. 762 0. 738 0. 681 0. 578 0. 447 0. 127 3. 5 0. 796 0. 786 0. 767 0. 708 0. 604 0. 464 0. 131 3 0. 788 0. 802 0. 780 0. 711 0. 602 0. 466 0. 133 2. 5 0. 740 0. 748 0. 743 0. 722 0. 609 0. 464 0. 114 2 0. 671 0. 674 0. 684 0. 661 0. 576 0. 443 0. 094 0. 051 0. 050 0. 037 0. 025 0. 016 0. 011 To know which of the parameter is more sensitive, an analysis was carried out and it was found that more variation was expressed by changing the kernel size than σ value
• The results reveal that kernel sizes were sensitive in identifying potential fire pixels as compared to different σ values. • The reduced kernel size will not include the large area around the fire pixels as background as a result, the immediate pixels surrounding potential fire pixels with a relatively higher temperature values will render the potential fire pixel as non-fire pixel. Thus, reducing the number of pixels detected as fire. • These results are in accordance to the findings of Giglio et al. , (2003). • Considering the next lowest RMSE of 2. 004 (the lowest being 1. 972), lowest MAE of 1. 704 and best index of agreement of 0. 802, all of which were obtained using 11 x 11 kernel size and 3σ in fire detection algorithm, it can be concluded that, the 11 x 11 kernel size and 3σ is the best setting for agricultural fire detection in day time MODIS data.
About 60% fire detects matches with the fire locations in MOD 14. (a). Classified agriculture fire events for 16 -Oct-2005 in Punjab state, detected using contextual algorithm in comparison to (b). fire events by MOD 14 for 16 -Oct-2005 in Punjab state.
Conclusion • Kernel size were more sensitive as compared to sigma (σ) value. • 11 x 11 kernel size and 3σ is the best setting for agricultural fire detection.
Residue burning from agricultural system in India
Introduction Residue - leaves, straw and husks left in the field after harvest. Controlled fires but AAQ issues. IGP Rice (Oryza sativa)–wheat (Triticum aestivum) cropping system (RWS) is rice and wheat crop rotation which is predominant in IGP. The region is named Indo-Gangetic Plains after the Indus and the Ganges, the twin rivers that drain it. Indian states, namely, Punjab, Haryana, Uttar Pradesh, Bihar, and West Bengal falling in vast Indo-Gangetic Plain (IGP). Geographical area of 5, 71, 490 sq. km. Most of the area is a gently undulating plain. The total area under RWS in India is roughly around 20 m ha. Almost 90– 95% of the rice area in Punjab, Haryana and western UP is used under intensive RWS.
Data used Data Product Satellites/Senso rs used Spatial resolution Time period MOD 14* (Terra/Aqua)/ MODIS 1 km SPOT-VGT 1 km Daily fire of 3 years (August, 2006 to July, 2009) Year 2002 -03 Croprotation** [Source: * MODIS fire information system, Geoinformatics center, Asian Institute of Technology, Thailand ** Panigrahy et al, 2009]
Crop-rotation map of the study area.
MODIS based agriculture fire data
Steps • Daily MODIS fire products (MOD 14) downloaded for the study period from both the Terra and Aqua platforms for day and night acquisitions in ASCII format. • The data is having Time, Day / Night tag, Satellite platform tag, Latitude, Longitude, Reflectance of Band 2, Brightness Temperature of Band 21 & Band 31, Fire power and Fire confidence. • The ASCII data converted to CSV and based on the mapping of the latitude and longitude fields points were converted to GIS ready vector format (. SHP files). • It was found that night time data from Terra platform pertaining to September 5, 2008 having reflectance value of band 2 as -1 was not reliable and therefore omitted from the data analysis. • this drop in reflectance may be due to shadows, cast by clouds and surface relief and so exhibit similar spectral changes as those caused by fire.
• The remaining point vector files were extracted for the IGP area using a mask of the IGP state’s administrative boundary. The fire point data was overlaid on the croprotation image and the raster values were transferred to the point coverage for further analysis.
Results… Cumulative monthly fire counts in IGP states [CR: Crop Rotation; MW: Maize-Wheat; PW: Pearlmillet-Wheat; RFF: Rice-Fallow; RPW: Rice-Potato-Wheat; RW: Rice-Wheat; SB: Sugarcane-based; CW: Cotton-Wheat; OW: Other-Wheat; PM: Pmillet-Mustard; POW: Pmillet-Other. Wheat; RP: Rice-Pulse; RVW: Rice-Veg. Wheat; RFR: Rice-Fallow-Rice; RMJ: Rice. Mustard-Jute]
Results… Aggregate monthly, crop-rotation wise agriculture residue burning for Indo. Gangetic Plain states and percentage contribution of individual states.
Results • Total fire detects (8, 726), ~69% comes from agricultural areas and remaining (31%) comes from non-agricultural (forest and bush) fire. • Out of total agriculture based burnings, 84% were in the rice-wheat and riceother crop-wheat based cropping systems or RWS. • Fire also detected in the Cotton-Wheat(2%), Maize-Wheat(3%), Pearlmillet. Wheat(1%), Rice-Fallow(5%), and Sugarcane-based cropping (4%). • RWS is known for residue burning, but other cropping systems showing small fires may also be due to the saturation due to small fraction of the pixels where rice is being cultivated. • Fire events were mainly confined to the months of March – May (36%) and October- December (55%). • In the 1 st part, April shows the highest fire events with 18. 25% and the in the 2 nd part, October shows the highest fire events with 45. 59% of total agricultural fire. • This is due to the burning of leftover crop residues of rice crop before planting of rabi season crop.
Conclusion Rice is grown during warm, humid season between June and October and wheat in cool, dry season between November and March. Due to the use of combine harvesters, there has been a sharp increase in the share of open field burning as it leaves behind large quantities of agricultural residues. The agricultural residue burning is more serious in the state of Punjab. The fire events in Punjab are detected in the Months of April – May and Oct- Nov only. April and May Wheat residues being burned (8%) Oct and Nov Rice residues being burned (92%) This is because of the fact that the time-gap available for planting Rice is quite high, therefore farmer may wait for rainfall to get the residues naturally leaves its nutrients to soil. The time gap available between rice to wheat cropping is not sufficient for nutrient enrichment, the reason for higher rice residues burning. Residue burning also causes nutrient and resource loss, and reduces total N and C in the topsoil layer, thus calling for improvement in harvesting technologies and sustainable management of the RWS.
Conclusion… Agricultural fire detection in the near real time is very essential. The uncontrolled and badly chosen time of burning of residues generally causes pollution in the ambient air and deteriorates the quality of air we breathe. The associated health hazards are well known. Therefore, the meteorological parameters should be considered and the authorities should inform farmers about a proper time window for controlled burning, so that the gases goes straight to the upper atmosphere and should not cause health hazard by concentrating in ambient air. Burning of crop residue results in emission of trace gases and particulate matters, loss of plant nutrients and thus adversely affects the pedology. The future Indian mission, INSAT-3 D having capability to detect fire at every 15 to 30 minutes will enhance the capability of such active fire detection from space and will be able to provide more realistic picture of short lived fire events in RWS.
Assessment of biomass burning and CO emission over India Regional distribution of fire counts (year 2002 – 2006) over India (1 deg. Grid size) Distribution of average CO (1018 Molecules/cm 2) conc. (year 2002 -2006) over India
Atmospheric CO variations from MOPITT Vs ATSR based Fire Count over Indian region for year 2000 – 2007
INFFRAS Established as part of the Disaster Management Support Programme of the Department of Space to facilitate forest fire monitoring and management. INFFRAS is designed to meet the user requirement of the forest department at three levels. 1. Pre fire (preparatory planning for fire control), 2. During fire (near real time active fire detection and monitoring), 3. Post fire (damage and recovery assessment and mitigation planning). INFFRAS provides active forest fire alerts during the fire season (Feb – June). MODIS sensors aboard the Terra and Aqua platforms (for daytime observations), OLS sensor on the DMSP satellites (for night time observations). The fire alerts report only active forest fires observed by these satellites. This is made available to users on a no-cost basis
INFFRAS The daytime products are produced using TERRA / AQUA MODIS data. While the exact overpass times are variable they are approximately between 0930 and 1430 Hrs. Fire locations represents the center of a 1 km “fire” pixel. Nighttime fire detections are produced by using DMSP-OLS data which overpass India between 1900 and 2200 Hrs. “Fire” pixels from this data set have a side of 2. 7 km. IRS P 6 AWi. FS and ASTER are used in validation of the fire products. The products are different in terms of the sensor used, over pass time and methodology of detection. It is observed that the spatial occurrence of fires between the sensors are in general agreement Fire alerts are available through NRSA website and are also e-mailed to decision makers (forest department etc). Size of Fire: MODIS - of the order of 100 m 2. DMSP-OLS visible band instrument - of the order of ~45 m 2. Algorithm: 1. MODIS data is received at NRSC. Fire detection is performed using a contextual algorithm.
INFFRAS Algorithm: 2. DMSP-OLS operates in sun-synchronous orbits with nighttime overpasses ranging from about 7 pm to 10 pm local time with a swath width of 3000 km. The Operational Line Scanner System (OLS) is an oscillating scan radiometer with two spectral bands, basically designed for global observation of cloud cover. At night, the visible band is intensified with a photo-multiplier tube (PMT) to permit detection of clouds illuminated by moonlight. The light intensification enables observation of faint sources of visible- near infrared emissions present at night on the earth's surface including cities, towns, villages, gas flares, heavily lit fishing boats and fires. Fires present at the Earth's surface at the time of the nighttime overpass of the DMSP are readily detected in the visible band data. A forest mask is used to isolate active forest fires. Near real time (3+ hours old) nighttime DMSP-OLS data of the Indian region is provided via an automated subscription service from the NOAA. Archived information on active fires from Jan 01, 2006 for night time detections and from Feb 01 for daytime fire detections are available.
Fire Validation To remove uncertainties from fire product. Two kinds of active fire products 1). “yes/no” binary fire maps 2). Characterization of fires e. g. flaming, smoldering, and unburned areas and assigning temperature values to each of them. Methods: Ground based observations time consuming process limited spatial coverage and sampling capacity Simultaneous airborne imaging costs tend to be prohibitive sampling can be poor if campaings are not well coordinated (multimission campaigns make it even more difficult) High resolution orbital sensors sampling will be restricted to morning hours (~10 am local) Inter-product comparison Smoke plumes Difference in algorithms and approach Controlled fire burning
Probable Validation Areas in MP & Punjab Year 1995 to 2008 Based on a recurrence analysis done for 10 x 10 km grid using ATSR/AATSR fire data, following Taluka in M. P. have better potenitial for fire product validation: Harda, Harsud, Betul, Budhni, Piparia, Singrauli, and Bijaipur. April – May , 2009 • April – May || October- November • April and May Wheat residues burned • October and November Rice residues • Total Agricultural fire Wheat residues (8%) during April-May & Rice residues (92%) during October-November in Punjab.
Probability of Occurrence of Droughts A perennial event Tamilnadu, J&K, Telangana, West Rajasthan - every 2. 5 years Gujarat, E. Rajasthan & Western Uttarpradesh – every 3 years Other states have droughts every 4 -5 years Based on the rainfall deficiency for the last 100 years
Severe droughts in India in last 100 years Year % of country area affected % of less rainfall over entire India % of less rainfall over drought reg. 1918 71 -26 -49 1965 41 -17 -36 1972 47 -25 -35 1979 45 -21 -38 1987 50 -18 -45 After 1987, India experienced severe drought in 2002 and 2009 due to large negative anomaly of Indian summermonsoon rainfall and its intra-seasonal distribution. .
NADAMS • NADAMS provides near real-time information on prevalence, severity level and persistence of agr. drought at state/district/sub-district level. • Currently, the project covers 13 states, which are predominantly agriculture based and prone to drought situation. • Monitoring is restricted to Kharif season (June – Oct/Nov. ) because of rainfall dependency. • Agri. drought assessments are made at using daily-observed NOAA AVHRR (1. 1 km) data. • The system consideration are; (1) seasonal NDVI progression from the beginning of the season, (2) comparison of agri. area NDVI profile with previous normal years, (3) weekly rainfall status compared to normal, and (4) weekly progression of sown area compared to normal. • The relative deviation of NDVI from that of normal and the rate of progression of NDVI during the season give indication about the agricultural situation. • The drought bulletins are published regularly.
• By definition, the severity of agricultural drought is determined by rootzone soil water. Thus, root zone soil moisture budgeting is one of the best methods for successful monitoring of crop status in rain fed areas. • With the availability of satellite derived rainfall products, the possibility of using rainfall in operational forecasting models is realized. India has enhanced its effort of deriving rainfall product from KALPANA series of meteorological satellites.
Drought detection using INSAT 3 A CCD • Multi-stage regional drought impact assessment uses time series daily or composite vegetation index (VI) data. • The regional-continental coverage at multispectral bands with constant view direction from geostationary satellite senor is ideal for this purpose. • The fortnightly composites of operational NDVI product from INSAT 3 A CCD observations between September 1 to October 31 in 2008 (normal year) and 2009 (drought year) were used to detect late kharif (June to October agriculture growing period) season drought conditions in India over permanent agricultural area. • The maximum deviation of NDVI within two years was found over Karnataka state and its ranging from 0. 22 to 0. 43 where as minimum deviation was found over Punjab within range of 0. 05 to 0. 16 due to availability of irrigation facilities. • Deviation of district average NDVI of each composite for two consecutive years were compared with deviation of rainfall sum over June to September. • Maximum correlation of 0. 73 was found for central Indian consists of states like Madhya Pradesh, Chhattisgarh, Jharkhand Maharashtra. The poor correlation was observed in case of irrigated districts mainly lies north part of India in Punjab, Haryana and UP states in all fortnightly composite of NDVI.
Late season agricultural drought detection from INSAT 3 A CCD NDVI 30 th. Sept to 15 th. Oct 2008 30 th. Sept to 15 th. Oct 2009 IMD June-Sept 2009 NDVI ranges -1. 00 – 0. 00 0. 35 – 0. 37 0. 01 – 0. 12 0. 38 – 0. 41 0. 13 – 0. 15 0. 42 – 0. 46 0. 16 – 0. 18 0. 47 – 0. 50 0. 19 – 0. 21 0. 51 – 0. 60 0. 22 – 0. 24 0. 61 – 0. 69 0. 25 – 0. 26 0. 70 – 0. 79 0. 27 – 0. 30 0. 80 – 0. 90 0. 31 – 0. 34 16 th. Oct to 31 st. Oct 2008 16 th. Oct to 31 st. Oct 2009
Future Scope • Apart from generating the active fire products and drought indices it is important to investigate its linkages through other parameters like, the normalized vegetation index (NDVI), land surface temperature (LST), surface albedo, surface insolation and time series phenological variables. • Currently, optical and thermal observations from radiometer and multispectral observations from CCD payloads are being used to address some of these. • The use of observations from six-channel ‘Imager’ at 1 -4 km and 19 channel sounder at 10 km spatial resolution from INSAT 3 D would enhance the possibility of operationalization of land surface products.