An Introduction to MODIS Land Products With Emphasis on LAI & FPAR Products

Transcription

1 An Introduction to MODIS Land Products With Emphasis on LAI & FPAR Products Ranga Myneni Boston University Rene Magritte ( ) The Beautiful Season 1/24 This goal of this lecture is to introduce you to the MODIS Land products, with emphasis on LAI and FPAR products. 1

2 Motivation The following two statistics are sufficient to justify monitoring our planet - 47% increase in global population (4.45 billion in 1980 to 6.53 billion in 2006) 12% increase in atmospheric carbon dioxide concentration (340 ppmv in 1980 to over 380 ppmv in 2006) 2/24 We are in living in a time of great changes. M other Earth is currently experiencing an unprecendent radiative forcing and the resulting changes are beginning to be visible against the background of natural variability. There are many good reasons why it is in our interest to monitor our planet, I will give you two rather dramatic stats. One, an increase in global population of about 47% just in the past 25 odd years. Man s impact on Earth is thus bound to be felt! Second, a 12% increase in the concentration of atmospheric carbon dioxide in the past 25 years - this is some indication of M an s desire to improve his or her life at the expense of the environment. 2

3 What Have We Learned? Some examples, [1] Greening of the Northern Latitudes [2] Climate driven increases in NPP Jackson Pollock ( ) Autumn Rhythm (Number 30) (1950) [3] Seasonality in Amazon rainforests 3/24 Vegetation is the only terrestrial primary producer, that means, no plants, no food. We are all parasites. We can nicely monitor our primary producers with satellite data. We have been doing that for the past 30 odd years. What have we learned? Quite a bit, actually. I will present you three examples, from research that I have been personally involved. 3

4 [1] Greening of the Northern Latitudes ANOMALY TEMPERATURE NDVI NORTH AMERICA (40N~70N) ANOMALY EURASIA (40N~70N) TEMPERATURE NDVI Analyses of satellite greenness index data sets indicate a significant greening of the northern latitudes since the early 1980s. This greening trend show a strong correlation with the pronounced warming of surface temperatures observed in these regions, and the spatio-temporal pattern of greening coincides with the major modes of global climate variability, namely, the El Nino Southern Oscillation and the Arctic Oscillation. 4/24 Analyses of satellite greenness index data from the AVHRR sensor series indicate a significant greening of the northern latitudes since the early 1980s. This greening trend show a strong correlation with the pronounced warming of surface temperatures observed in these regions, and the spatio-temporal pattern of greening coincides with the major modes of global climate variability, namely, the El Nino Southern Oscillation and the Arctic Oscillation. 4

5 [2] Climate Driven Increases in Global NPP Recent results indicate that global changes in climate have eased several critical climatic constraints to plant growth, such that net primary production increased 6% (3.4 petagrams of carbon over 18 years) globally. The largest increase was in tropical ecosystems. Amazon rain forests accounted for 42% of the global increase in net primary production, owing mainly to decreased cloud cover and the resulting increase in solar radiation. 5/24 M ore recently, the work of Dr. Rama Nemani of NASA Ames Research Center, indicates that global changes in climate have eased several critical climatic constraints to plant growth, such that the terrestrial net primary production increased by 6% (3.4 petagrams of carbon over 18 years) globally. The largest increase was in tropical ecosystems. Amazon rain forests accounted for 42% of the global increase in net primary production, owing mainly to decreased cloud cover and the resulting increase in solar radiation. 5

6 [3] Leaf Area Seasonality in Amazon Rainforests a Leaf Area Index Precipitation (mm/mo) Recent analysis of five years of MODIS LAI data suggest seasonal swings in green leaf area of about 25% in a majority of the Amazon rainforests. That is, leaf area equivalent to nearly 28% the size of South America appears and disappears each year in the Amazon. This seasonal cycle is timed to the seasonality of solar radiation in a manner that is suggestive of anticipatory and opportunistic patterns of net leaf flushing during the light rich dry season and net leaf abscission during the cloudy wet season. 6/ Solar Radiation (W/m Some recent work I have done, in collaboration with my student, Dr. Wenze Yang and the MODIS team, from analysis of five years of MODIS Leaf Area Index data, suggests seasonal swings in green leaf area of about 25% in a majority of the Amazon rainforests. That is, leaf area equivalent to nearly 28% the size of South America appears and disappears each year in the Amazon. This seasonal cycle is timed to the seasonality of solar radiation in a manner that is suggestive of anticipatory and opportunistic patterns of net leaf flushing during the light rich dry season and net leaf abscission during the cloudy wet season. These heretofore unknown seasonal swings in leaf area are critical to initiation of the transition from dry to wet season, seasonal carbon balance between photosynthetic gains and respiratory losses, and litterfall nutrient cycling in moist tropical forests. 6

7 MODIS MODIS: The MODerate resolution Imaging Spectroradiometer is an instrument on board NASA s Terra and Aqua platforms for remote sensing of the Earth s atmosphere, oceans and land surface. LAUNCH: The Terra platform was launched on December 18th, 1999 and the Aqua platform on May 4th, FEATURES: The MODIS instrument has a swath width of 2,330 km, orbit height of 705 km, and produces global coverage every one to two days. MODIS measures reflected solar and emitted thermal radiation in 36 spectral bands and at 3 different spatial resolutions (250 m, 500 m and 1000 m). 7/24 Okay, that is some background material. Let me get to the main theme of my presentation. The MODerate resolution Imaging Spectroradiometer (MODIS) is an instrument on board NASA s Terra and Aqua platforms for remote sensing of the Earth s atmosphere, oceans and land surface. The Terra platform was launched on December 18th, 1999 and the Aqua platform on May 4th, The MODIS instrument has a swath width of 2,330 km, orbit height of 705 km, and produces global coverage every one to two days. MODIS measures reflected solar and emitted thermal radiation in 36 spectral bands and at 3 different spatial resolutions (250 m, 500 m and 1000 m). 7

8 MODIS Land Products Energy Balance Product Suite Surface Reflectance Land Surface Temperature BRDF/Albedo Snow Cover Vegetation Parameters Suite Vegetation Indices (EVI, NDVI) LAI/FPAR NPP/PSN Land Cover Land Use Suite Land Cover Vegetation Continuous Fields Vegetation Cover Change Fire and Burned Area 8/24 This is a list of MODIS Land Products. These products are generated operationally and available to all for free. In this course, we shall focus on the Vegetation Parameters Suite, especially LAI, FPAR and NPP. 8

9 LAI and FPAR The MODIS Land team is responsible for the development of algorithms for operationally producing 16 geophysical data products and their validation. LAI: one sided green leaf area per unit ground area in broadleaf canopies, and half the total needle surface area per unit ground area in coniferous canopies. FPAR: fraction of photosynthetically active radiation ( µm) absorbed by the vegetation. These products are useful in studies of the exchange of fluxes of energy, mass (e.g., water and CO 2 ) and momentum between the surface and atmosphere. 9/24 LAI is defined as the one-sided green leaf area per unit ground area in broadleaf canopies and as half the total needle surface area per unit ground area in coniferous canopies. FPAR is the fraction of photosynthetically active radiation absorbed by the vegetation.these products are useful in studies of the exchange of fluxes of energy, mass (e.g., water and CO2) and momentum between the surface and atmosphere. 9

10 MODIS LAI and FPAR Products Terra LAI/FPAR (MOD15A2 product): Collection 3-- Coverage: November 2000 December 2002 [1 km and 8 day] Collection 4-- Coverage: March 2000 present [1 km and 8 day] Collection 5-- Began June 30th, 2006 [1 km and 8 day] Aqua LAI/FPAR (MYD15A2 product): Collection 3-- Coverage: July December 2003 [1 km and 8 day] Collection 4-- Coverage: July present [1 km and 8 day] Collection 5-- Began June 30th, 2006 [1 km and 8 day] Terra and Aqua MODIS Combined Products in Collection 5 [1 km, 4 and 8 day] The Standard MODIS Products are available from the EDC DAAC Boston University LAI/FPAR (MOD15_BU): 1- and 4- km monthly best quality composites of Collection 4 MOD15A2 product Available from LAI/FPAR PI website, 10/24 MODIS product versions are called Collections. Collection 3 is the first significant processing of MODIS data into products and covered the 26 month period from November 2000 to December Collection 4 represents the most comprehensive set of MODIS products and contains the entire time series of data starting from February 2000 to the present. Collection 5 started recently, June 30th, In addition to the standard MODIS LAI/FPAR products, which can be obtained free of charge via the web from the EDC Data Center, my group also produces quality enhanced and user-friendly versions of the standard products, which can be obtained from my web site. These products are enormously popular, in spite of my difficult personality! 10

11 MODIS LAI and FPAR Production 11/24 The algorithm finds LAI and FPAR values given sun and view directions, Bidirectional Reflectance Factor (BRF) for each MODIS band, band uncertainties, and six biome land cover classes. The algorithm are executed on a daily basis. The daily LAI/FPAR values are composited over an eight day period. These 1 km 8-day products are made available to the user. The processing is done by NASA. The products are distributed from the EDC Data Center. This flow chart shows one of the main themes of my talk. The closed loop between product analysis, validation and algorithm refinement - key elements for improving product quality in each successive re-processing. 11

12 MODIS LAI and FPAR Processing 12/24 This flow chart shows the latest processing scheme - Collection 5. The LAI/FPAR algorithm is executed on 8 days of Terra and Aqua MODIS daily reflectance data to produce 16 daily LAI/FPAR retrievals. These retrievals are then composited to yield four different products - (1) Terra 8-day, (2) Aqua 8-day, (3) Combined 4-day and (4) Combined 8-day. 12

13 MODIS LAI and FPAR Algorithm Main RT Algorithm Back-Up Algorithm Saturation Best quality retrievals are obtained with the Main RT-based algorithm. In case of high LAI (>5) surface reflectances have low sensitivity to LAI (saturation domain). Only retrievals by main algorithm (with or without saturation) are recommended for application studies. In case of main algorithm failure, the back-up algorithm (LAI/FPAR-NDVI empirical relationships) is employed. Such retrievals have generally low reliability and are not recommended for application studies, including validation. 13/24 Actually, the M ODIS LAI/FPAR algorithm is two algorithms. The main algorithm is based on the principles of radiative transfer in vegetation canopies. The retrieval technique compares observed and modeled reflectances and outputs the LAI and FPAR values. The algorithm, however, may fail if input reflectance data uncertainties are greater than preset threshold values in the algorithm or due to deficiencies in model formulation which result in incorrect simulated reflectances. In all such cases, the retrievals are generated by a back-up algorithm based on biome-specific empirical relationships between the Normalized Difference Vegetation Index (NDVI) and LAI/FPAR. 13

14 EXAMPLES (a) (b) (c) (d) (e) (f) LAI Fill Fill FPAR 0.00 QC Main Saturation Back-Up Bare Fill 14/24 These are global maps of LAI, FPAR and Quality Control (QC) generated from Collection 3 (panels a, c, e) and Collection 4 (b, d, f) data sets for Julian date in year 2002 (August 5-13, 2002). You can clearly see that there are many more main algorithm retrievals in Collection 4 compared to Collection 3. 14

15 Algorithm Refinements Main algorithm retrievals increased from 55 percent in Collection 3 to 67 percent in Collection 4 as a result algorithm refinements. The impact can be seen even in globally averaged LAI values. 15/24 Left Panel: The retrieval index (RI) is defined as the ratio of the number of pixels with LAI and FPAR retrieved by the main algorithm to the total number of retrievals by both the main and back-up algorithms. This index does not indicate retrieval quality but rather the success rate of the main algorithm. The retrieval index shows a stable but seasonal pattern through the five years of MODIS operations. The annual average retrieval index increased from 55% in Collection 3 to 67% in Collection 4 because of the closed loop between product assessment, validation and algorithm refinements, I spoke of earlier. The retrieval index can be as low as 40-50% during the boreal winter time mainly due to poor quality surface reflectances and as high as 65-80% during the boreal summer time. Right Panel: The time series of LAI from the Terra MODIS sensor exhibit the seasonal cycle as expected, with a boreal winter time minimum of about 1.3 to 1.5 LAI and a boreal summer time maximum of about 1.8 to 2.5 for the Collections 3 and 4) processing. The difference between the two Collections is due to two reasons (1) LAI overestimation in the non-woody stemming from a mismatch between simulated reflectances evaluated from algorithm LUT entries and MODIS reflectances in Collection 3, and (2) significant misclassification of grasses and cereal crop pixels into broadleaf crop pixels in the biome map used in Collection 3 processing. 15

16 Main vs. Backup Retrievals! = " LAI LAI Main LAI Backup Near linear relationship between delta LAI and main algorithm LAI. Backup algorithm underestimates LAI over dense broadleaf forests. 16/24 This figure shows the difference between main and back-up algorithm retrievals over broadleaf forests in the northern latitudes (blue) and the tropics (red). As these forests are dense, most of the true LAI values (main algorithm retrievals) are greater than 5 (the histogram indicated by the dashed lines). As the difference in LAI values between the main and back-up algorithms increases linearly, we conclude that the back-up algorithm retrievals are in significant error, most likely due to clouds. 16

17 Snow and Cloud Conditions Few main algorithm retrievals under snow or cloud conditions. Spurious seasonality due to snow. The difference between LAI retrievals under cloud-free and cloudy condition is not negligible. 17/24 There are fewer main algorithm retrievals when the pixels are contaminated with either clouds or snow (Top Panels). This is to be expected as the main algorithm fails under such conditions. The difference in LAI values between cloud contaminated pixels and clear pixels is significant (Bottom Right Panel). Therefore, it is important to examine the cloud quality flag accompanying the data sets. 17

18 From Product Assessment & Validation to Algorithm Refinement Steps Analyze the products Identify anomalies Trace the cause of anomalies Refine the operational algorithm Sources of uncertainties Landcover data (KONZ, Alpilles) Surface reflectance (AGRO, KONZ) Model (HARV) For example, Collection 3 anomalies include Summer product LAI higher than in situ LAI in herbaceous vegetation Too few main algorithm retrievals during summer Seasonality differences between main and backup algorithms 18/24 The process of improving p roduct quality is illustrated here. We perform comprehensive analysis of the product to diagnose anomalies. We then trace the sources of these anomalies. These sources could be due to (1) input data quality and (2) algorithm deficiencies identified through field validation activities. For example, we found that the retrieved Collection 3 summer LAI is than field measured LAI in non-woody vegetation. We therefore went back to the designing board and fine tuned the Look-Up-Tables for these biomes, implemented these in Collection 4 re-processing, and then checked the Collection 4 LAI with field data again to make sure that the identified anomalies have been fixed. 18

19 Global Validation Sites Flakaliden Shrubs Savannas Needleleaf Forests Grasses/ Cereal Crops Broadleaf Crops Broadleaf Forests 19/24 Let me talk a little bit more about validation. The MODIS LAI data have been validated against ground measurements of leaf area in a host of vegetation types across the globe. Validation is a very expensive activity. Therefore, collaboration between all interested international partners is of great value. This slide shows some of the validation sites. 19

20 Validation Procedure Field sampling representative of LAI spatial distribution and dynamic range within each major land cover type at a validation site. Development of a transfer function between field measured LAI and high resolution satellite data to generate a reference LAI map over an extended area. Comparison of MODIS LAI with the aggregated reference LAI map at patch scale in view of geo-location and pixel shift uncertainties. 20/24 The validation protocol that we have developed may be summarized as follows. A direct comparison between ground measurements and corresponding satellite-based products is not recommended because of scale-mismatch, geolocation errors and vegetation heterogeneity at the resolution of MODIS data. Thus, an intermediate step that involves a fine resolution map of the variable of interest is introduced. This map is generated with field data and high resolution satellite data (ETM+, SPOT, ASTER, etc.) When aggregated to the MODIS resolution, this map serves as the ground-truth. Therefore, the validation of moderate resolution LAI products includes these steps. (1) ground sampling of vegetation variables during field campaigns, (2) generation of a fine resolution map of the variables and (3) comparison of the aggregated fine resolution map with satellite-based products. 20

21 Validation Example Site: Ruokolahti, Finland Measurements Dates: June 14-21, 2000 Land Cover Type: Coniferous forests Results: Comparison of aggregated high-resolution LAI map and corresponding MODIS LAI retrievals suggests satisfactory behavior of the MODIS LAI algorithm although variation in MODIS LAI product is higher than expected. Collection 4 MODIS LAI match field measurements within 0.5LAI (meet specifications) Publication: Wang et al. (2004). Evaluation of the MODIS LAI algorithm at a coniferous forest site in Finland. Remote Sensing of Environment, 91, S. Reflectances, 1x1 km Patches, 1x1 km LAI, 1x1 km LAI LAI, 10x10 km 21/24 This slide illustrates the validation protocol, using our analysis for a coniferous forest site in Ruokolahti, Finland. First, ETM + (30 m) surface reflectances are segmented to identify the vegetation patches in a 1x1 km region, or some such similar small region, which should be chosen to represent the larger picture of vegetation. Second, field LAI measurements are carried out in each of the patches. Third, a transfer function is developed between fine resolution satellite data (ETM + based Reduced Simple Ratio, in this case) and field measured LAI. Fourth, using this transfer, a fine resolution LAI map is derived for a larger region, say 10x10 km. This fine resolution LAI map is then aggregated to 1km resolution and serves as the ground-truth for validation of moderate resolution satellite-based LAI products, MODIS in our case. 21

22 Validation Results, 0 C4 MODIS LAI & $ ) RMSE=0.66LAI!"#"$%$&'"(")%$& * & "#")%+, *-./#)%00123 ) $ & , Field LAI Grasses & Cereal Crops Shrubs Broadleaf Crops Savannas Broadleaf Forests Needleleaf Forests b1-b4 all one-one b1-b4,b6 Linear (one-one) Linear (all) 22/24 This slide presents a comparison of Collection 4 MODIS LAI with field measurements. Field data from 29 different sites representing all the six biomes referenced by the algorithm were used to generate this plot. From this summary slide, we infer that the M ODIS LAI values may overestimate field measurements by about 12% in dense vegetation. The MODIS values can explain 87% of the variability of LAI between sites, and on average, MODIS values will be in error in their estimation by 0.66 LAI. 22

23 Collection 5 MODIS LAI & FPAR Products LAI/FPAR algorithm was optimized to improve consistency of RT-simulated and MODIS surface reflectances -- LUTs for all biomes were recalculated based on new Stochastic RT model -- Special focus of refinements was on woody vegetation. 6-biome LC was replaced with 8-biome LC and separate LUTs were developed to account for variability spectral signatures of evergreen and deciduous subclasses of broadleaf and needle leaf forests. -- Terra and Aqua combined 8- and 4-day products were prepared for production Extensive refinements have been implemented for MODIS surface reflectances product (bug fixes, conservative cloud mask, improvements of geolocation accuracy, aerosol retrievals only over non-cloudy pixels, inclusion of polarization effect). Due to significant amount of changes science testing and production of Collection 5 was delayed (Started on June 20th 2006). 23/24 Based on our comprehensive analysis of Collection 4 M ODIS LAI/FPAR product, and on a less than comprehensive amount of validation work, we revised the algorithm for Collection 5 processing, which began on June 30th, The changes include (1) New Look-Up-Tables for all biomes based on Stochastic Radiative Transfer principles which better accounts for spatial heterogeneity and foliage clumping, (2) split the forest biome classes into evergreen and deciduous broadleaf and needle leaf classes, resulting in 8 biomes in Collection 5 as opposed to six biomes in Collections 3 and 4, (3) Terra- Aqua Combined 8-day and 4-day products. 23

24 User needs should drive the program. Lessons Learned Scientists should be involved intimately - from instrument design, algorithm development, implementation, data system design and data processing. Plan on processing the data in real time, at least for the critical products. Plan on multiple re-processings - this should be based on feedback from validation. Make data available freely through the web. Data should be accompanied by quality information and documentation. The goal should be to serve the user! 24/24 24