Issues in Developing and Validating Satellite Land Surface Temperature Product

Yunyue (Bob) Yu NOAA/NESDIS, Center for Satellite Applications and Research

Outlines

• LST Basics

• Issues in LST algorithm development

• Issues in LST product validation

• Summary

2 LST Basics

Definition: Land Surface Temperature (LST) is the mean radiative skin temperature derived from thermal radiation of all objects comprising the surface, as measured by remote sensing ground-viewing or satellite instruments. Benefits: . plays a key role in describing the physics of land-surface processes on regional and global scales . provides a globally consistent record from satellite of clear-sky, radiative temperatures of the Earth’s surface . provides a crucial constraint on surface energy balances, particularly in moisture-limited states . provides a metric of surface state when combined with vegetation parameters and soil moisture, and is related to the driving of vegetation phenology . an important source of information for deriving surface air temperature in regions with sparse measurement stations

Target Requirement: Horizontal resolution – 1 km, Temporal resolution – 1 h, Accuracy – 1 K

Current : MODIS/VIIRS : H = 1 km, T = Daily, A = 1.4 K, Uncertainty = 2.4 K GOES Imager : H = 4 km (2 km) , T = 15 min/1 h, A = 1.4 K, Uncertainty = 2.4 K 3 LST Basics ---- LST available at STAR

4 Basics of Satellite LST Retrieval

LST algorithm is inherited from Sea Surface Temperature nidar satellite dR

a a Top of Atmosphere (TOA) radiance: R θ

Split Window Technique: LST ∝ , (T11-), (T11-T12)(secθ - 1) Earth Surface

a = R cosθ dR = R − a = a(secθ −1)

Difference to SST: . Surface Emissivity and its variation . Elevation difference atmosphere . Anisotropic . Spatial Variation

. Temporal Variation Earth surface

Is(ν) : contributed from surface emission . In-situ data availability  Is(ν) : contributed from upwelling radiance  5 Is(ν) : contributed from reflected downwelling radiance Issues in LST Algorithm Development ---- Temporal and Spatial Variations

LST can vary significantly in space and in time temporal scale : ~ minutes spatial scale : ~ meters 6 Due to surface emission anisotropic feature, LST observed from satellite may be significantly different from one view angle to another.

Case Study LST data: Split-window algorithm using GOES-E and GOES-W Imagers data. Time Period: The year 2008 Location: at one SURFRAD station (Desert Rock, NV ; 36.63ºN, 116.02ºW) Data pairs: 2096 simultaneous observations View zenith of GOES-8: 60.140 LST difference at one SURFRAD station, observed from View zenith of GOES-10: 46.810 7 GOES-E and GOES-W. • Considerable seasonal emissivity variation over Emissivity Impact to LST some surface types • Considerable emissivity variation within one type

Jan. Apr. Jul. Oct.

Woody Savannas (2012)

Grassland(2011)

Closed Shrub lands (2012)8

Emissivity Variation Example from a Modified Geometric Projection model . Vegetation surface: Green needle Forest ; Background: bare soil

. Mean Emissivity=(ε1+ε2)/2, where ε1 and ε2 are the spectral emissivity at MODIS bands 31 and 32, respectively. This image cannot currently be displayed.

Top-left: LST Error (given fixed emissivity assumption).

Bottom-left: LST and Modeled Emissivity Variability (over- and understory)

Emissivity LUT dimension: vegetation type, cover%, LAI, soil background, view zenith

Simulation Example: LST diff. (upper figure) is the difference of LSTs retrieved using a variable emissivity calculated from the MGP model and using a fixed emissivity. No solar zenith dependent is observed. The apparent emissivity varies from about 0.9560 to 0.9680 (solid line, lower figure). Parameter setting: case 1, 30% veg coverage, LAI=1. The Simulation result is calculated from atmospheric profile 9 IPR06065 of NOAA88. LST algorithm is by Sobrino et al. (1994), modified with path correction. BT difference for atmospheric correction Split-window algorithm feature: brightness temperature (BT) difference at 11 and 12 µm is used for atmospheric correction. It is the SST heritage. However, the BT difference can be very different over land. Additional measure is needed.

BT difference at daytime

Land Surface Gulf of Mexico Left: Significant BT differences over land and sea water surface. The BT Australia difference is much Ocean Surface smaller over sea surface 10 LST Validation Difficulties Anemometer Thermometer – In Situ data limitation Down-looking PIR on the tower At 8-m from ground • Measurement difficulty • Emissivity Directional impact • Cloud contamination impact

– Spatial and temporal variations Down-looking PIR at 8 meter height from the ground • Spot vs pixel difference UP-looking PIR Diffuse Radiometer • Accurate match-up process (in space and in time) – Others (e.g., angle effect)

Surface heterogeneity is shown in a 4km x 4km Google map (1km x 1km, in the center box) around the Bondville station area 11 Issues in LST Product Validation ---- In-situ Validation Impact of Ground Data Fluctuation

The ground LST estimate can be fluctuated significantly, resulting big match-up uncertainty ( ~6K)

12 Issues in LST Product Validation ---- In-situ Validation

A case study of in-situ data comparison in Africa Suggested for matchup (Gobabeb, Namibia)

Sensor Location

VIIRS LST

*the Africa site data provided by Frank Goettsche (KIT & EUMETSAT Land SAF), through MODIS v5 LST LST validation collaboration 13 Issues in LST Product Validation ---- Cross-Satellite Comparison

2013362 A Case Study of LPEATE Global Browse Images of Day LST

NPP VIIRS

Global images are generated by courtesy of NASA/GSFC Aqua MODIS LPEATE 14 Issues in LST Product Validation ---- Cross-Satellite Comparison

Impact of time difference in cross-satellite comparison

About 25 min difference between VIIRS and MODIS

15 Cross-Satellite Comparison

Cross-satellite Comparison: dataset difference VIIRS - MODIS

MODIS_LST Date: 4/19/2014 VIIRS/MODIS - MODIS

Cross-satellite LST comparison is used in VIIRS

LST evaluation.

Caution: Time difference is

VIIRS - LST a significant impact; VIIRS/MODIS LST VIIRS/MODIS granule level comparison is needed. 16 MODIS LST Issues in LST Product Validation ---- Granule level Cross-Satellite Comparison

17 Summary

Issues in LST Algorithm Development • Emissivity sensitivity –– emissivity explicit algorithm is more appropriate; emissivity sensitivity study is needed • Spatial heterogeneity –– Higher spatial resolution, less retrieval uncertainty • Temporal variation –– Time information is necessary for gridded LST product • Atmospheric difference –– stratified algorithm coefficients • Cloud contamination –– need of cloud filter development for LST production

Issues in LST Product Validation • In-situ validation – Spatial heterogeneity: spot-pixel difference –– site characterization – Temporal variation: time match restriction – In-situ LST estimation: quality control of in-situ data, reliable emissivity data • Cross-satellite comparison – Data gridding: consistent aggregation process – Time match – BRDF consideration (constraints) 18