A methodology to support environmental degradation monitoring and analysis using AVHRR data

Save this PDF as:
 WORD  PNG  TXT  JPG

Size: px
Start display at page:

Download "A methodology to support environmental degradation monitoring and analysis using AVHRR data"

Transcription

1 A methodology to suppot envionmental degadation monitoing and analysis using AVHRR data Gilson Alexande Ostwald Pedo da Costa 1,5 Magaeth Simões Penello Meielles 1 Dhamenda Singh 4 Isabelle Helin 3 Jean-Paul Beoi 3 Enio Faga da Silva 2 1 Univesidade do Estado do Rio de Janeio UERJ Rua São Fancisco Xavie, 524-5º Anda, Bloco D, sala Rio de Janeio - RJ, Basil 2 Empesa Basileia de Pesquisa Agopecuáia Embapa Solos Rua Jadim Botânico, Rio de Janeio - RJ, Basil 3 Institut National de Recheche en Infomatique et Automatique INRIA Rocquencout B.P Le Chesnay Cedex - Fance (isabelle.helin, 4 Indian Institute of Technology - Depatment of Electonics & Compute Engineeing Rookee UA, India 5 K2 Sistemas Rua Maques de São Vicente 189, casa Rio de Janeio - RJ, Basil Abstact. This wok poposes a paticula appoach to assess infomation about soil degadation fom NOAA/AVHRR data. As eosive pocesses change physical and chemical popeties of the soil, alteing, consequently, the supeficial colo, monitoing the change in colo ove time can help to identify and analyze those pocesses. A methodology fo the detemination of soil colo fom NOAA/AVHRR data was devised, based on a theoetical model that establishes the elationship among the soil colo, descibed in the Munsell colo system, vegetation indices, suface tempeatue and emissivity. The test aea of the methodology was the Uppe Taquai Basin, in the cental egion of Bazil, whee the lack of land use planning and soil consevation pactices have been causing sevee eosion and siltation of the wate bodies, inceasing the spatial and tempoal significance of flood events ove the Bazilian Pantanal egion. The tests showed that the methodology was efficient in detemining soil colo using the NDVI, MSAVI and PAVI vegetation indices. Best esults wee obtained fo the hue colo component. To futhe test the methodology, the calculated digital colo models wee compaed with the chaacteistic colo of soil classes of that egion. Key wods: emote sensing; soil colo; eosion; vegetation index; suface tempeatue; emissivity; ecology; sensoeamento emoto; co do solo; eosão; índice de vegetação; tempeatua da supefície; emissividade; ecologia. 2941

2 1. Intoduction The Taquai Rive is a majo contibuto to hydological system of the Bazilian Pantanal, a Biosphee Reseve (UNESCO). Its wateshed coves appoximately 80,000 km2, being appoximately one thid located in the highlands of the Paaguay Rive basin. The topogaphy and chaacteistics of most of the soils in the egion make them highly o vey highly susceptible to eosion, potential eosion ates may vay fom 600 to 950 mt/ha/yea. The occupation of the highlands in the cente-west egion of Bazil in the last 30 yeas, chaacteized by the lack of planning of land use, lack of soil consevation pactices, and destuction of ive bank vegetation, has amplified degadation of soil and wate esouces, to a point that it is consideed the pincipal menace to the Pantanal biome integity. Cuently the Taquai ive caies a sediment load of 491 mg/l. Deposition in the Pantanal lowlands occu at ates highe than 100mt/ha/yea, esulting in flood events that, evey yea, incease thei spatial and tempoal significance. The pesent wok descibes a methodology conceived to assist land use management of the Uppe Taquai Basin. The methodology was developed as a component of the ECOAIR Poject, a Scientific Intenational Coopeation Poject fo the development of models and automated outines to identify envionmental paametes associated to eosive pocesses and degadation using satellite images (automatic monitoing of land/use, land cove change). In ode to unavel the complexity of the Taquai issue we adopted a poblem-oiented appoach in which, we have consideed to pedict eosion pocess by satellite data, whee soil colo is consideed as a key facto indicato fo monitoing eosion pocesses. 2. Soil Colo Soil is a complex matix containing mineal, wate, ai and oganic matte at vaious levels. The easiest soil popeties to assess ae the mophological ones, expessions of the appeaance of soil accoding to macoscopic chaacteistics pomptly peceptible, such as colo, textue and stuctue. Often the mophological chaacteistics of soils ae deteminant fo its classification, what indicates its stong coelation with the physical, minealogical and chemical popeties. In fact it is possible to extact a lot of infomation about a paticula soil based on its mophology. Accoding the U.S. Soil Suvey Staff (1981) colo is one of the most useful popeties fo soil identification and appaisal. Qualitative and quantitative infomation can be gained on paametes such as oganic matte content (Schultz, 1993), minealogy (Schulze, 1993), moistue and dainage (Richadson, 1993), ph-eh (Fanning, 1993), and soil hoizon delineation (Soil Suvey Staff, 1999). Dak soils ae usually iche in oganic matte. The ed colo can indicate high amount ion oxides. Cabonate and calcium sulphate give soil a lighte colo, wheeas moistue lowes the intensity of soil colo. Geneally, eoded soils have highe colo values, as a consequence of the emoval of the top layes of soil and the subsequent decease in oganic matte. When the top layes ae totally emoved, soil mateial with colo significantly diffeent fom that of the non-eoded soils will be exposed. The standad method fo specifying the colo of soil is based on a compaison of soil samples of colo chips contained in a Munsell colo chat (Munsel, 1994). The Munsell colo designation makes use of a chaacteization scheme that descibes colo in tems of thee 2942

3 vaiables: hue, value and choma. The hue notation of a colo indicates its elation to ed, yellow, geen, blue and puple, accoding to Munsell (1907) hue is the quality by which we distinguish one colo fom anothe. Value is a neutal axis that efes to the gay level of the colo, it is the quality by which we distinguish a light colo fom a dak one. Choma is the quality that distinguishes the diffeence fom a pue hue to a gay shade. 3. Theoetical Model The poposed appoach aims at assessing soil colo fom NOAA/AVHRR data. It stats fom establishing the coelation models between soil colo, collected in situ by pedologists, and Vegetation Indices and Emissivity, calculated fom the NOAA images. The next step is the invesion of the models so that soil colo can be detemined diectly fom the NOAA data. It is a semi-empiical appoach. The detemination of the coelation models between vegetation indices and emissivity and soil colo/moistue stats fom the definition of the physical significance of the vegetation indices and emissivity. Vegetation Indices have been used extensively fo the deivation of the biophysical popeties of vegetation and soil. In this wok a few types of vegetation indices wee used, in ode to detemine witch is moe useful fo the assessment of colo: Nomalized Adjusted Vegetation Index (NDVI); NDVI = (ρ ni ρ ed ) / (ρ ni - ρ ed ) (1) Modified Soil Adjusted Vegetation Index (MSAVI) (Qi, 1994); MSAVI = ((2ρ ni + 1) - ((2ρ ni + 1) 2-8(ρ ni - ρ ed )) 0.5 ) / 2 (2) Global Envionment Monitoing Index (GEMI) (Pinty, 1992); GEMI = ξ(1-0.25ξ) - (ρ ed ) / (1 - ρ ed ) (3) ξ = (2(ρ ni - ρ ed 2 ) + 1.5ρ ni + 0.5ρ ed ) / (ρ ni + ρ ed + 0.5) Puified Adjusted Vegetation Index (PAVI) (Singh, 2004). PAVI = (ρ ni - ρ ed 2 ) / (ρ ni - ρ ed 2 ) (4) Suface Emissivity is a measue of the inheent efficiency of the suface in conveting heat enegy into adiant enegy above the suface. It depends upon the composition, oughness and moistue content of the suface and on the obsevation conditions (i.e. wavelength, pixel esolution and obsevation angle). Suface emissivity vaiation, consequently, have a diect elationship with suface composition change (Sobino, 2000). The channel emissivity diffeence and mean channel emissivity can be calculated diectly fom AVHRR/NOAA data using NDVI Theshold Method - NDVI THM (Sobino, 2000). Vegetation indices and suface emissivity can be consideed as a function of the ecosystem investigated, climate, teain, soil and hydology vaiables. Conceptually the vegetation indices and emissivity can be modeled using those envionmental vaiables: VI / Emissivity = f(cl, Ve, Ph, S) + K (5) The sub-models may in tun be epesented as a function of thei majo components: climate (Cl), Vegetation/Ecosystem (Ve), Physiogaphy (Ph), Soil/Hydology (S). Whee K, is the modeling eos caused by envionmental vaiables and potential inaccuate measuements. The model could evidently be moe complex, howeve, not all envionmental 2943

4 vaiables ae completely independent, what makes it possible to obtain theoetical VI/Emissivity with a limited numbe of envionmental vaiables. Vegetation indexes ae influenced by vaiations of vegetation and soil. They can be consideed as the sum of two components: The same can be said about suface emissivity: VI = VIsoil + Vivegetation (6) ε = εsoil + εvegetation (7) In this wok we segmented the images and investigated only the locations whee the influence of soil in the indices ae geate than that of the vegetation. So, we esticted the application of the model to the space of vegetation indices whee the influence of the component VI vegetation is small (NDVI between 0 e 0.2). Futhemoe, fo a specific geogaphic location, the vegetation/ecosystem and phisiogaphy sub models become elatively less time vaiant. Theefoe, NDVI fo a specific time (t) at a specific geogaphic location becomes pimaily a function of climate vaiables and soil moistue/colo. To simplify the models: VI(t) = f [soil colo/moistue] + f [climate vaiables] + K1 (8) VI = f [soil colo] + f [tempeatue)] + K2 (9) Emissivity = f [soil colo] + K3 (10) Suface tempeatue was calculated though one of the split window algoithms (Uliviei, 1992); Ts = T (T4 - T5) + 48(1 - ε) - 75 ε (11) Whee Ts is the suface tempeatue, T4 is the bightness tempeatue fom band 4 of AVHRR and T5 is the bightness tempeatue fom band Mateials and Methods The gound tuth data consisted of 60 colo pofiles (in Munsell notation) of the soil top laye collected in the yeas of 1995, 1996 and 1999, fom diffeent sites in the Uppe Taquai basin. Local Aea Coveage (LAC) AVHRR/NOAA images wee acquied fom the Satellite Active Achive fom the NOAA Administation (< fo the dates the soil samples wee collected. Afte geometic coection and atmospheic calibation, the Vegetation Indices (NDVI, GEMI, MSAVI and PAVI) and Emissivity (mean and diffeence) wee calculated. Then, VIs and Emissivities wee calculated fo the gound tuth sites Diffeent kinds of egession analysis wee tested to detemine the best coelation models among the VIs and Emissivities and the soil colo. 2944

5 Data wee sepaated into two sets, data fom 1995 and 1996 (22 samples), and data fom 1999 (37 samples). Regession coefficients wee calculated using data fom 1999 and the egession model was tested against 1995 data. 5. Results The calculations made taken all data into consideation shown that the best coelations wee obtained by linea egession fo all the vegetation indices and suface emissivity. The best coelation coefficients wee obtained by linea/multiple egession: VI (H) (H) (V) (V) (C) (C) NDVI 0,69 0,47 0,46 0,21 0,36 0,13 MSAVI 0,69 0,48 0,48 0,23 0,36 0,13 GEMI 0,32 0,10 0,27 0,07 0,31 0,10 PAVI 0,68 0,46 0,46 0,21 0,37 0,14 Table 1. Values of coelation coefficient () and coefficient of deteminations ( ) fo Hue, Value and Choma espectively (fom left to ight) with vaious vegetation indices by equation Hue/Value/Choma = b0 + b1 (VI) + b2 (Suface Tempeatue) E (H) (H) (V) (V) (C) (C) ε 0,45 0,20 0,18 0,03-0,25 0,06 ε 0,49 0,24 0,15 0,02-0,29 0,08 Table 2. Values of and fo Hue, Value and Choma espectively (fom left to ight) with channel and diffeence emissivity (E) by equation Hue/Value/Choma = b0 + b1 (E) When testing the data against the colo (hue, value and choma) gids calculated fom the NOAA images, using the egession models detemined fom the 1999 data, the following esults we obtained: VI/E (H) (H) (V) (V) (C) (C) NDVI 0,86 0,75 0,71 0,51 0,48 0,23 MSAVI 0,86 0,74 0,73 0,54 0,47 0,22 GEMI -0,52 0,27 0,60 0,36 0,38 0,14 PAVI 0,86 0,74 0,70 0,49 0,49 0,24 ε -0,62 0,39-0,42 0,18 0,39 0,15 ε -0,71 0,51-0,43 0,18 0,44 0,19 Table 3. Values of coelation coefficient () and coefficient of deteminations ( ) fo Hue, Value and Choma espectively (fom left to ight) with obseved and calculated values of Hue, Value and Choma using vaious vegetation indices, channel and diffeence emissivity. 2945

6 6. Soil Class Matching To evaluate the pecision of colo calculation tough the methodology an application was devised to confim the soil classification of the Uppe Taquai Basin, poduced ealie by EMBRAPA (Basil, 1997). Fo the application, we selected the Neossolos Quatzaênicos Óticos soil class, which coves an aea of km2, coesponding to 47% of the Basin aea. Each pixel of the digital colo model calculated, situated inside the aeas associated to the selected class and in locations whee the NDVI values anged fom 0 and 0.2, was tested. The colo inteval fo the Neossolos Quatzaênicos Óticos soil class was detemined fom a set of mophological desciptions of soil pofiles collected in the High Paaguai Rive Basin. Fom the colo ecods povided by EMBRAPA, the colo component intevals fo the selected class ae: 2.5YR a 10YR fo Hue; 3 a 4 fo Value; and 2 a 5 fo Choma. Fou AVHRR images fom diffeent yeas wee tested. The digital soil colo models wee calculated using both NDVI and MSAVI vegetation indices. Consideing only the hue component, the compaison gave the following esults: Image Date Hit Rate Hit Rate (MSAVI) (NDVI) 19/8/ % % 29/11/ % % 24/11/ % % 15/10/ % % Table 4. Pecentage of pixels fo which the calculated hue lies inside the chaacteistic colo inteval fo the selected class. Consideing all the thee colo components (hue, value and choma) at the same time, the compaison gave the following esults: Image Date Hit Rate (MSAVI) Hit Rate (NDVI) 19/8/ % % 29/11/ % 90.5 % 24/11/ % % 15/10/ % % Table 5. Pecentage of pixels fo which the calculated colo (hue, value and choma) lies inside the chaacteistic colo inteval fo the selected class. It must be noted that the colo inteval consideed fo the selected class is consideed lage, with espect to hue it vaies fom 2.5YR to 10YR. In a futue application we intend to investigate classes chaacteized by smalle colo intevals, such as the Latossolos Vemelhos o Latossolos Vemelho-Amaelo. 2946

7 7. Conclusions The esults show a good coelation between NDVI, PAVI and MSAVI (in that ode) and Hue. They also show that Hue can be pedicted with a good level of accuacy diectly fom the NOAA images The low coelation between the colo components and emissivity indicates that unaccounted chaacteistics of soil have a lage influence on emissivity than colo. As emissivity has been linked with the stuctue of soils, maybe othe factos, such as textue, oughness o chemical composition can be bette coelated to emissivity. A fai coelation has been established between Value and Vegetation Indices (specially MSAVI, NDVI and PAVI, in that ode), and a low coelation between Choma and Vegetation Indices. That can be patially explained by the highe influence moistue has on Value and Choma than on Hue. The investigation of soil pofile ecods obtained fom EMBRAPA (fom the same Cental-West Region of Bazil), shows that a wet soil sample has usually the same Hue, but lowe Value and Choma values than a dy sample. Futhe tests ae cuently being made to evaluate the capacity of pediction of soil degadation pocesses of the poposed appoach. In the futue moistue infomation should be added to the models, what we believe will impove the esults obtained by the appoach. 8. Acknowledgment The pesent wok is a pat of ECOAIR Poject (Digital Image Pocessing Technology fo Change Detection of Envionment Infomation), sponsoed by CNPq Bazilian National Council fo Scientific and Technological Development (< and INRIA Institut National de Recheche en Infomatique et Automatique, Fance. Authos ae vey much thankful fo thei financial suppot. 9. Refeences Bem-Do, E.; Ions, J.R.; Epema, G. Soil Reflectance. Remote Sensing fo the Eath Sciences, v. 3, Wiley, New Yok, p , Basil. Ministéio do Meio Ambiente, Recusos Hídicos e da Amazônia Legal. Pogama Nacional do Meio Ambiente. Pojeto Pantanal. Plano de Consevação da Bacia do Alto Paaguai (PCBAP), v. 2, Basilia, Cihla, J.; St.Lauent, L.; Dye, J. A. Relation between the Nomalized Diffeence Vegetation Index and Ecological Vaiables. Remote Sensing of Envionment, v. 35, Elsevie, New Yok, p , Coll, C.; Caselles, V.; Schmugge, T. Estimation of Land Suface Emissivity Diffeences in the Split Window Channels of AVHRR. Remote Sensing of Envionment, Vol 48, Elsevie, New Yok, pp , Fanning, D. S.; Rabenhost M. C.; Bigham J. M. Colos of Acid Sulfate Soils, SSSA Special Publication, No. 31. Madison, WI, Lozano-Gacia, D. F.; Fenandez, R. N.; Johannsen, C. J. Assessment of Regional Biomass-Soil Relationships using Vegetation Indexes, IEEE Tansactions on Geoscience and Remote Sensing, v. 29, p , Munsell, A. H. A Colo Notation. Geo H. Ellis Co., 114p Munsell Soil Colo Chats. New Windson: Kollmogen Instuments-Macbeth Division, Naisimha Rao, P. V.; Venkataatnam, L.; Kishna Rao, P. V.; Ramana, K. V. Relation between Zone Soil Moistue and NDVI of Vegetated Fields. Intenational Jounal of Remote Sensing, v. 14, p , Pinty, B., Vestaete, M. M. GEMI: a non-linea index to monito global vegetation fom satellites. Vegetation, v. 101, p , Qi, J.; Chehebouni, A.; Huete, A. R.; Ke, Y. H., Soooshian, S. Modified Soil Adjusted Vegetation Index (MSAVI). Remote Sensing of Envionment, v. 48, Elsevie, New Yok, p ,

8 Richadson, J. L. and Daniels R. B. Statigaphic and Hydaulic Influences on Soil Colo Development. SSSA Special Publication, n. 31. Madison, WI, Schultz, P. A.; Halpet, M. S. Global Coelation of Tempeatue, NDVI and Pecipitation. Advances in Space Reseach, v. 13, p , Schulze, D. G.; Nagel, J. L.; Van Scoyoc, G. E.; Hendeson, T. L.; Baumgadne, M. F.; Stott, D. E. Significance of Oganic Matte in Detemining Soil Colo. SSSA Special Publication, n. 31. Madison, WI, Schwetmann, U. Relations Between Ion Oxides, Soil Colo, and Soil Fomation. SSSA Special Publication, n. 31. Madison, WI, Singh, D.; Helin, I.; Beoi, J. P.; Silva, E. F.; Meielles, M. S. An appoach to coelate NDVI with soil colou fo eosion pocess using NOAA/AVHRR data. Advances in Space Reseach, v. 33, p , Sobino, J. A.; Raissouni, N. Towad Remote Sensing Methods fo Land Cove Dinamic Monitoing. Application to Maocco. Int. J. Remote Sensing, v. 21, p , Sobino, J.A.; Raissouni, N.; Zhao-Liang Li. A Compaative Study of Land Suface Emissivity Retieval fom NOAA Data. Remote Sensing of Envionment, v. 75, Elsevie, New Yok, p , Soil Suvey Staff. Examination and Desciption of Soils in the Field. Soil Suvey Manual, n. 1, USDA-SCS, Washington, DC, Soil Suvey Staff. Soil Taxonomy. USDA Handbook No. 436, 2nd Edition. U.S. Govenment Pinting Office, Washington, DC, Uliviei, C.; Castonouvo, M. M.; Fancioni, R.; Cadillo, A. A Split Window Algoithm fo Estimating Land Suface fom Satellites. COSPAR, 27 Aug.-5 Sept., Washington DC,