ROSES Proposals
PACE-EMIT data fusion for functional characteristics of agricultural crops
PI: Dave Schimel - JPLCo-Is: Amy Braverman (JPL); Margaret Johnson (JPL); Jouni Susiluoto (JPL); Fred Huemmrich (GSFC); Shawn Serbin (GSFC); Phil Townsend (University of Wisconsin); Steven Mirsky (USDA-ARS); Jyoti Jennewein (USDA-ARS)
Hyperspectral geophysical products for land systems contain considerable information about vegetation state and condition, growth potential, and growth environment. Planned spaceborne imaging spectrometers tend to have appropriately small pixel sizes, 30 to 60 meters ground sampling distance, which capture relatively pure end members for vegetation composition and condition at bi-weekly or longer intervals. However, user needs elicitations, including those done for LANDSAT-next and SBG, also emphasize the value of repeat frequency, a conclusion emphasized by the studies done for the SHIFT field campaign conducted by the SBG team.
PACE fills a gap in the international Program of Record, providing VNIR spectroscopy with a high revisit frequency, but at coarse spatial resolution (1 km) This allows revisit similar to MODIS, which has proven value, while enabling a measurement of a set of critical additional variables that can characterize vegetation response to rapidly evolving conditions, such as effects of pest/pathogen disturbances and drought, as well as nutrient deficiency in crops. On the other hand, EMIT observes at fine spatial resolution (60 m) but much less frequently. The full potential is unknown, though chlorophyll and other pigment indices have been demonstrated. The near-coincidence of PACE and EMIT opens several important possibilities, and we will develop methods to achieve these new products through the following activities:
We will port target SBG/EMIT algorithms to PACE. The SBG team has developed a series of advanced algorithms for vegetation through the SBG-VSWIR team and the SBG precursor project, SISTER. The algorithms have been developed to be 'instrument agnostic' wherever possible and are readily adaptable to additional spectrometers. We will employ the existing framework and datasets built to support these efforts to generate PACE-derived versions of the vegetation products. We will compare these to the products derived from other imaging spectrometers, and identify additional data sources (as needed) that can provide necessary information to account for the lack of shortwave infrared channels on PACE. Key vegetation characteristics will relate to plant nutrients (e.g., nitrogen), nutritional quality (e.g., lignin), and water, overall greenness and stress (e.g. pigments such as chlorophyll), and, more speculatively, plant defensive compounds.
We will use co-located EMIT and PACE scenes over agricultural landscapes, we will upscale in-situ cal/val data to test PACE retrievals and tune PACE algorithms, if need be, informed by crop foliar measurements. We will use leaf composition data collected operationally from the USDA's Long Term Agricultural Research (LTAR) Network and other available data, including the NEON agricultural sites. We will initially focus on wheat, then extending measurements to other major row crops and rangeland as resources permit. The USDA’s Sustainable Agricultural Systems Laboratory (Beltsville) will provide access to canopy data from ongoing collections at LTAR sites.
We will create a broader suite of prototype products exploiting synergies between high-temporal-resolution/low-spatial resolution PACE data and low-temporal-resolution/high-spatial-resolution EMIT data. To achieve this, we use data fusion techniques developed at JPL to efficiently learn how to optimally combine these heterogenous data sets to robustly estimate underlying quantities (reflectances) necessary for retrieving crop phenology or other quantities of interest. Crucially, our data fusion method automatically provides uncertainties associated with these estimates. We will use EMIT to bridge in situ to PACE spatial resolution. This activity will build on algorithm development by JPL and GSFC for SBG, EMIT, and PACE. The objective is to generate temporally continuous products that are required by users, in recognition that there will always be gaps in direct retrievals from satellite due to atmospheric conditions and repeat intervals.
We will assess the computational and logistical issues for the large-scale, low-latency production of fused products that will eventually be produced at an assigned DAAC.
PACE fills a gap in the international Program of Record, providing VNIR spectroscopy with a high revisit frequency, but at coarse spatial resolution (1 km) This allows revisit similar to MODIS, which has proven value, while enabling a measurement of a set of critical additional variables that can characterize vegetation response to rapidly evolving conditions, such as effects of pest/pathogen disturbances and drought, as well as nutrient deficiency in crops. On the other hand, EMIT observes at fine spatial resolution (60 m) but much less frequently. The full potential is unknown, though chlorophyll and other pigment indices have been demonstrated. The near-coincidence of PACE and EMIT opens several important possibilities, and we will develop methods to achieve these new products through the following activities:
We will port target SBG/EMIT algorithms to PACE. The SBG team has developed a series of advanced algorithms for vegetation through the SBG-VSWIR team and the SBG precursor project, SISTER. The algorithms have been developed to be 'instrument agnostic' wherever possible and are readily adaptable to additional spectrometers. We will employ the existing framework and datasets built to support these efforts to generate PACE-derived versions of the vegetation products. We will compare these to the products derived from other imaging spectrometers, and identify additional data sources (as needed) that can provide necessary information to account for the lack of shortwave infrared channels on PACE. Key vegetation characteristics will relate to plant nutrients (e.g., nitrogen), nutritional quality (e.g., lignin), and water, overall greenness and stress (e.g. pigments such as chlorophyll), and, more speculatively, plant defensive compounds.
We will use co-located EMIT and PACE scenes over agricultural landscapes, we will upscale in-situ cal/val data to test PACE retrievals and tune PACE algorithms, if need be, informed by crop foliar measurements. We will use leaf composition data collected operationally from the USDA's Long Term Agricultural Research (LTAR) Network and other available data, including the NEON agricultural sites. We will initially focus on wheat, then extending measurements to other major row crops and rangeland as resources permit. The USDA’s Sustainable Agricultural Systems Laboratory (Beltsville) will provide access to canopy data from ongoing collections at LTAR sites.
We will create a broader suite of prototype products exploiting synergies between high-temporal-resolution/low-spatial resolution PACE data and low-temporal-resolution/high-spatial-resolution EMIT data. To achieve this, we use data fusion techniques developed at JPL to efficiently learn how to optimally combine these heterogenous data sets to robustly estimate underlying quantities (reflectances) necessary for retrieving crop phenology or other quantities of interest. Crucially, our data fusion method automatically provides uncertainties associated with these estimates. We will use EMIT to bridge in situ to PACE spatial resolution. This activity will build on algorithm development by JPL and GSFC for SBG, EMIT, and PACE. The objective is to generate temporally continuous products that are required by users, in recognition that there will always be gaps in direct retrievals from satellite due to atmospheric conditions and repeat intervals.
We will assess the computational and logistical issues for the large-scale, low-latency production of fused products that will eventually be produced at an assigned DAAC.