Hypertemporal and Hyperspectral Remote Sensing Applications for Regional Water Quality Assessments Aditya Singh Department of Agricultural and Biological Engineering University of Florida, Gainesville
Background Pressing issues: Biodiversity loss, anthropogenic disturbances, climate • change etc… • Increasing pressures on ecosystem service provisioning Remote sensing an important tool historically • Allows regional assessments extrapolations from field- • based studies • Synoptic, repeatable measurements Continuing need for new tools and techniques for the • most pressing issues
Optical remote sensing: tradeoffs in scale/resolution Multispectral space-borne: Landscape scale Composition, disturbance, phenology … $ Hyperspectral airborne: Hypersp p Landsca Landscape, field, plot scale a Composition, biochemistry, function, disturbance $$$$ Hyperspectral UAS, mobile: Hypers p Plot, canopy Plot, ca a Pl Plant, c Plant, canopy biochemistry, function t , $$ - $$ -$$$ $$$ $ Hyperspectral contact: Hypersp p Leaf/pl Leaf/plant scale Leaf bio Leaf biochemistry, function Leaf bio o $$-$$$ $$-$$$
Organization Methodological developments in satellite remote sensing • Landscape-scale nutrient cycling, crop production Imaging spectroscopy Mapping foliar biochemical, morphological • and metabolic traits and their uncertainties. Filling gaps, ongoing research Desktop spectroscopy, mobile and • airborne remote sensing platforms Contact spectroscopy Deriving foliar biochemical and • morphological traits
Organization Methodological developments in satellite remote sensing • Landscape-scale nutrient cycling, crop production Imaging spectroscopy Mapping foliar biochemical, morphological • and metabolic traits and their uncertainties. Filling gaps, ongoing research Desktop spectroscopy, mobile and • airborne remote sensing platforms Contact spectroscopy Deriving foliar biochemical and • morphological traits
Landscape dynamics, satellite imagery, and water quality Predict baseflow water quality (NO 3 -N,SRP), one year in advance Get data from previous studies, 315 watersheds in Wisconsin • Obtain MODIS data, derive vegetation indices, organize by seasons • Build PLSR models • Predict across the entire state •
Results: Nitrate-N mg/L
Advancing to continuous-time models: The Chesapeake Bay Issues: hypoxia, loss of loss of aquatic vegetation… Forest ~60%
Study area: Chesapeake Bay watershed 10 Years, 9 Watersheds, Monthly Nitrate-N loads Determine: • what influences water quality and where? • when are those influences most strong?
Method: Functional Linear Models (FLMs) • Functional models: – Relate observations to functions of (…classically, time-varying) predictors: – OLS: FLM: – Flexible: responses can also be functions (FL concurrent models). Log(NO 3 -N) mg/L/ha – ‘ Concurrent ’: responses at time ‘ t ’ are functions of predictors at the same redictors at the same time . – Interpretation simple Raw Fourier approx. • similar to OLS models – Beta coefficients are also functional → Structurally down-scalable. down scalable
Spatial variables: Landcover (NLCD 2006) Watershed characteristics Fixed Fixed http://www.horizon-systems.com/nhdplus/ • Landcover • Ws characteristics http://www.mrlc.gov/index.php • N. Deposition • Precipitation • NDVI • Disturbance Total Atm. N deposition (NADP) Precipitation (PRISM) Disturbance, NDVI (MODIS) Annual Monthly 8-day http://www.prism.oregonstate.edu/ http://nadp.sws.uiuc.edu/ https://lpdaac.usgs.gov/
Results: Summer Spring Summer flushing uptake flushing Intercept Direct inputs In-stream Forest Shorter processing functional type flowpath
Results: Model matches both intra- and inter-annual variations well Observed Predicted Also see: Eshleman et al. 2013 (ES&T)
Results: Pixel-wise / watershed averaged predictions:
Optical remote sensing: Issues of scale and resolution Methodological developments in satellite remote sensing • Landscape-scale nutrient cycling, crop production Imaging spectroscopy Mapping foliar biochemical, morphological • and metabolic traits and their uncertainties. Filling gaps, ongoing research Desktop spectroscopy, mobile and • airborne remote sensing platforms Contact spectroscopy Deriving foliar biochemical and • morphological traits
Optical remote sensing: Issues of spectral resolution Solar spectral irradiance at sea level H 2 O O 2 B O 2 A
Optical remote sensing: Issues of spectral resolution MODIS Terra, Landsat 7 Spectral sampling: Multispectral sensors
Optical remote sensing: Issues of spectral resolution AVIRIS-C Spectral sampling: Hyperspectral sensors
Spectroscopy Chlorophyll NPQ Biologically important absorption features Chlorophyll Nitrogen SLA Phenolics SLA Photochemistry e - Transport Phenolics Photochemistry Nitrogen Phenolics
Foliar biochemistry from spectroscopy 4 ears (200 4 years (2008 4 ye 08- 08 8-2011), 8 20 2 011) ), 237 plots, 6 ), 37 plots, 6 23 6 states, 6 tates st s, 36 species, 7 Traits s, 6 spe 36 spe p ec e ies, 1 N) δ 15 15 (N%, LMA, C%, Lignin%, Cellulose% , Fiber%, %, % δ
Leaf Image Image Methods level level level Analysis & Analysis & prediction prediction
Leaf Leaf Image Methods level level level Analysis & Analysis & prediction prediction
Leaf Leaf Image Methods level level level Analysis & prediction
Foliar traits from imaging spectroscopy Partial least squares regression • Chemometric method designed to handle high-dimensional, multicollinear data • 50/50 Jackknife to get model uncertainties LMA
Foliar traits from imaging spectroscopy PLSR model results, 25/75 Cal/Val, 500 × randomized Jackknife, 237 plots Observed Predicted Singh et al. (2015) Ecological Applications
Trait Uncertainty Trait Uncertainty Results Trait maps Savage River State Forest MD
2007 Emergent patterns NLCD 2011 R:G:B = N:Lignin:LMA 2007 2009 N% mean N% uncertainty
What can we use maps of foliar biochemistry for? Methodological developments in satellite remote sensing • Landscape-scale nutrient cycling, crop production Imaging spectroscopy Mapping foliar biochemical, morphological • and metabolic traits and their uncertainties. Filling gaps, ongoing research Desktop spectroscopy, mobile and • airborne remote sensing platforms Contact spectroscopy Deriving foliar biochemical and • morphological traits
Water quality as a function of foliar traits • 250 Watersheds across Wisconsin • NO3-N , SRP Data from MODIS, AVIRIS, NLCD • Latent variable Manifest variable Source C : N AVIRIS Foliar retention↓ Lignin : N AVIRIS TC Wetness index MODIS Wetland retention ↓ % Water NLCD Runoff from % Agriculture NLCD ag/pasture↑ % Pasture NLCD Foliar N % AVIRIS External inputs ↑ Atm. Nitrogen dep. N. Dep log(NO 3 -N) Field Water quality
Method: Proposed PLS-path model Retention due to foliar recalcitrance Foliar Retention Retention in wetlands Water Wetland retention Quality External Runoff inputs Runoff from Fertilizer*, Atm. Dep agricultural land and pastures
Results: Fitted path model Retention due to foliar recalcitrance Foliar Retention Retention in wetlands Water Wetland -0.270*** retention Quality External 0.020 ns Runoff inputs Runoff from Fertilizer*, Atm. Dep agricultural land and pastures 31
N. Minnesota Flambeau SF Path model: Mapping the ‘foliar recalcitrance’ latent variable Baraboo Hills Conif./wetlands Decid./wetlands Agricultural
Path model: Results, Comparing forests and agriculture → Significant differences between mechanisms
Ongoing and future research Methodological developments in satellite remote sensing • Landscape-scale nutrient cycling, crop production Imaging spectroscopy Mapping foliar biochemical, morphological • and metabolic traits and their uncertainties. Filling gaps, ongoing research Desktop spectroscopy, mobile and • airborne remote sensing platforms Contact spectroscopy Deriving foliar biochemical and • morphological traits
Research in progress: UAS spectroscopy • Parallel system being built at UF • Headwall Photonics NanoHyperspec (400-1000nm) imaging spectrometer, Thermal • Will be used to estimate ET at the canopy scale
Conclusion ● Remote sensing and spectroscopy powerful tools for assessing ecological responses to stress, at multiple scales. ● Combined with coordinated field surveys and analysis techniques, can help answer basic and applied questions in ecosystem sciences. ● In combination with process-based models, spatial estimates of ecosystem attributes can help inform responses to environmental change. ● Field-scale instrumentation and UASs can enable better characterization of entire ecosystems across space and time.
Recommend
More recommend
