Analyzing an ecohydrological drought by model-data integration モデル－データ統合による生態水文学的干ばつの解析 Yohei Sawada Ph.D Candidate River and Environmental Engineering Laboratory, Department of Civil Engineering, the University of Tokyo 2016/1/22 Final Defense

1. Introduction (Chapter 1)

1.1. Motivation [Shukla et al., 2015] California drought 2.2 million dollar total economic loss [Howitt et al., 2014] Australian Millennium drought [van Dijk et al., 2013] The worst record since European settlement of Australia [Gergis et al., 2012] Amazon drought [Lewis., 2011] Response of largest carbon sink to drought is uncertain [e.g., Morton et al., 2014; Samanta et al., 2010] Horn of Africa & Sahel drought [Anderson et al., 2013] The food shortages affected 31 million people [e.g., Boyd et al., 2013] Future change in aridity is very uncertain. [e.g., Dai, 2011; Roderick et al., 2015]

1.2. The new concept: “ecohydrological drought” Natural climate variability Precipitation deficiency, high temperature etc… Soil water deficiency Plant water stress, reduced biomass and yield Reduced stream flow, inflow to reservoirs, Groundwater deficiency, …… Economic, Social, and Environmental impacts [from National Drought Mitigation Center, University of Nebraska-Lincoln, USA] See also [Mishra and Singh, 2010] Meteorological Ecological Hydrological  Relationship between ecological and hydrological processes is important for analyzing drought process.

1.3.1. Reviews of previous contributions (1) - remote sensing - Microwave satellite obs. Surface soil moisture [Njoku and Li, 1999; Owe et al., 2001; Paloscia et al., 2001; Koike et al., 2004] [Shibata et al., 2003] Vegetation water content [Paloscia and Pampaloni, 1988, 1992; Jackson and Schmugge, 1991; Njoku and Chan, 2006; Fujii et al., 2009] [Jones et al. 2011] Visible-infrared obs. Leaf Area Index [Myneni et al., 1997a, 1997b; Samanta et al., 2010] [Fang et al., 2013] Issue 1: Microwave vegetation observation has not been thoroughly evaluated in the field. Issue 2: Most of researches have used these two dataset as if they are looking at the same thing.

1.3.2. Reviews of previous contributions (2) - land surface modeling - hydrology Ecology Bucket model [Manabe, 1969] Sophisticated vertical transfer of Energy & Water e.g., SiB2 [Sellers et al., 1996] , BATS [Dickinson et al., 1986] Runoff scheme [Takata et al., 2003; Lawrence et al., 2011] River discharge [Oki and Sud, 1998] Catchment-scale high resolution modeling [Wang et al., 2009a, 2009b, 2009c] Dynamic Vegetation model [Foley et al., 1996; Cox et al., 2000; Smith et al., 2001; Krinner et al., 2005; Sato et al., 2007] Ecohydrological modeling [Rodriguez-Iturbe and Porporato, 2004; Ivanov et al., 2008, 2012; Fatichi et al., 2012; Niu et al., 2014; Sawada and Koike, 2013] Issue 3: No ecohydrological model has been used for the real world drought application.

1.3.3. Reviews of previous contributions (3) - land data assimilation - obs obs Initial Condition Parameter Optimization t Soil moisture, LAI  Even if the model were perfect, we cannot forecast very well without good initial conditions and model parameters What is the purpose of data assimilation??

1.3.3. Reviews of previous contributions (3) - land data assimilation - Initial Condition Parameter Optimization Hydrological Ecological Dual-pass data assimilation system Yang et al. [2007, 2009]: SiB2 & AMSR-E Reichle and Koster [2005]: Catchment LSM & SMMR Li et al. [2012]: NOAA LSM & AMSR-E Jia et al. [2013]: CLM3 & AMSR-E Su et al. [2013]: ECMWF IFS & ASCAT Jarlan et al. [2008]: ECMWF CTESSEL & MODIS Bandara et al. [2015] : JULES & SMOS Livneh and Lettenmaier [2012]: ULM & MODIS Stockli et al. [2008, 2011]: simple phenology model & MODIS Bateni et al. [2014]: SEB & SEVIRI Bacour et al. [2015]: ORCHIDEE & MODIS Issue 4: No study realizes dual-pass data assimilation system to simultaneously improve the skill of simulating both hydrological and ecological processes. Issue 5: There are few application studies of these types of LDAS.

1.4. Goal Develop the toolkit to analyze ecohydrological drought without any boundaries between hydrology and ecology. Provide new contributions to understand, quantify, monitor, and predict severe droughts.

1.5. Structure of the study Ecosystem resilience to multi- year Australian drought (Chapter 6) Quantifying Ecohydrological drought in a semi-arid river basin (Chapter 7) Ecohydrogical drought forecast in a data-scarce region (Chapter 8) Ecohydrological Modeling • WEB-DHM-Veg (catchment) • EcoHydro-SiB (point) Satellite Observation • Microwave Optical Fusion Approach (MiOFA) Model-Data Fusion • Coupled Land and Vegetation Data Assimilation System (CLVDAS) Eco-Hydrological Drought Analysis Toolkit (EHDAT) Basin-Scale [Sawada and Koike, 2013] Point-Scale (Chapter 4 & 5) Field-Experiment (Chapter 2) Footprint-Scale (Chapter 3) Tool Development Verification Application Objective Understand Drought Quantify Drought Monitor Drought Predict Drought

2. Microwave and optical fusion approach (Chapter 2 & 3)

2.1. Microwave and Optical Fusion Approach (MiOFA) Microwave Vegetation Optical Depth Optical Leaf Area Index effect Roughness : R Content Water Vegetation : VWC * R VWC b VOD     The spatial variabilities of roughness and vegetation cannot be estimated independently” [Njoku and Chan, 2006] y LAI y LAI VWC / 1 ) / exp(    bias R LAI y b VOD   * Optical leaf area index is sensitive to photosynthetic activity while microwave VOD is sensitive to whole aboveground vegetation water content.

2.2. Field Experiment @ Tanashi Brightness temperature (multi-frequency, multi-polarization) Meteorological forcings (rain, radiation, wind speed,…) Land surface properties (soil moisture, temperature, roughness,…) Vegetation properties (Biomass amount, vegetation water content, leaf area, ……) → The set of observations has been carried out once in 1-2 weeks.

2.3. Results (1) VWC retrieval by 6.9GHz with elimination of surface roughness effects VWC retrieval by 6.9GHz without elimination of surface roughness effects  By including LAI information, we can reasonably estimate vegetation water content by using C-band and X-band. And species dependency is minimal.

2.4. Field Survey @ Yanco observation site, Australia [Piles et al., 2014] Soil moisture observation network Destructive vegetation sampling  We validate MiOFA using satellite data (AMSR2 and MODIS) and footprint-scale in-situ observations.

2.5. Results (2) – Satellite application @ Australia Black: before roughness correction, Blue: after roughness correction, Green: LPRM, Purple: JAXA standard, Red: in-situ observation  Roughness correction positively impacts to the soil moisture retrieval. RMSE LPRM: 0.199 JAXA: 0.073 MiOFA: 0.070 Correlation coefficient LPRM: 0.372 JAXA: 0.299 MiOFA: 0.441

3. A Land data assimilation system for simultaneous simulation of soil moisture and vegetation dynamics (Chapter 4 & 5)

3.1. Ecohydrological model: EcoHydro-SiB hydro-SiB [Wang et al., 2009] Dynamic Vegetation Model (DVM) ＋ • Based on SiB2 [Sellers et al., 1996] • Including precise hydrological scheme [Wang et al., 2009] with Dynamic vegetation model. • Multi-layer transpiration • Modified V-G water retention curve • Multi-layer soil parameter distribution  Even if the model were perfect, we cannot forecast very well without good initial conditions and model parameters  How can we get initial conditions and the unknown parameters in the ungauged area?? : GOAL

3.2. Coupled Land and Vegetation Data Assimilation System (CLVDAS) EcoHydro-SiB Soil moisture Vegetation(LAI) Temperature Forward-RTM Estimated TB Core-Model Pass0: Parameter Selection Parameter ensemble Core-Model TB TB TB TB Sensitivity analysis of each parameter Pass1: Parameter Optimization Parameter Core-Model Estimated TB Satellite observed TB Schuffled Complex Evolution COST Pass2: Data Assimilation ~1year Soil Moisture, LAI ensemble Core-Model Estimated TB ~5days Satellite observed TB COST Genetic Particle Filter

3.3.1. Results (1) Parameter Selection Parameter Sensitivity to TBs (18.7GHz Horizontal) Blue：West Africa (Hot and dry) Orange：Mongolia (cold and dry) Gray：California (US) (temperate) Hydrological parameters （e.g., hydraulic conductivity） Ecological parameters （e.g. Maximum efficiency of Rubisco） → In dry area, we can improve the performance by tuning only hydrological parameters → We can reduce the number of the calibrated parameters by using this method.

3.3.2. Results (2) Parameter Optimization @ West Africa Calibration Validation LAI Surface Soil Moisture Green: Optimized, Black：Default, Red：Observed, Yellow：Observed (Microwave VOD (NASA LPRM))  Optimization improves the skill of estimating surface soil moisture and vegetation dynamics at the same time.

3.3.3. Results (3) Data Assimilation @ Yanco, AUS Grey: Open loop Blue: Genetic Particle Filter Red: observation  We can dramatically reduce the uncertainty of LAI estimation We can inversely estimate root-zone soil moisture from the observation of vegetation dynamics on land surface. Leaf Area Index 10-15cm Soil Moisture

4. Ecosystem resilience to the “Millennium Drought” in southeast Australia (2001-2009) (Chapter 6)

4.1.1. Satellite observations of ecosystem responses to the “Millennium Drought” (1)  Strong ecosystem resilience to millennium drought has been detected [e.g., Yang et al., 2015] and an existing ecosystem model failed to simulate it [van Dijk et al., 2013].  LAI did not significantly decrease but microwave VOD did, why??

4.1.2. Satellite observations of ecosystem responses to the “Millennium Drought” (2) Rain Leaf Area Index Vegetation Optical Depth Statistically significant pixels  LAI did not significantly decrease but microwave VOD did, why?? Linear trend from 1993 to 2009

4.2. Hypothesis non-photosynthetic active photosynthetic active WET DRY Optical leaf area index is sensitive to photosynthetic activity while microwave VOD is sensitive to whole aboveground vegetation water content.  This interpretation of the difference between LAI and VOD can be found in the previous studies [Jones et al., 2013] .  Structured vegetation is generally vulnerable to droughts [e.g., Rowland et al., 2015] Vulnerable to drought Strong to drought Note: This conceptual model represents ‘lumped’ ecosystem

4.3. Numerical experiment We run EcoHydro-SiB with two different settings WET NPP DRY Carbon allocation Ecosystem traits e.g., root distribution NPP NPP NPP CTL: fixed allocation strategy & traits DYN: dynamic allocation strategy & traits Note: This numerical model represents ‘lumped’ ecosystem

4.4.1. Results (1) LAI LAI time series Red: CTL, Blue: DYN, Green: Satellite CTL DYN Linear trend from 1993 to 2009 Statistically significant pixels  Dynamic carbon allocation and ecosystem traits bring LAI resilience

4.4.2. Results (2) aboveground biomass Aboveground biomass time series Red: CTL, Blue: DYN, CTL DYN Linear trend from 1993 to 2009 Statistically significant pixels  Dynamic carbon allocation and ecosystem traits does not bring resilience of total aboveground biomass.  This corresponds to VOD observations.

4.5. Summary of mechanism leaf stem A1 A2 B1 B2 A1 A2 A3 A4 strong vulnerable A2 B2 A1 B1 strong vulnerable × × A2 A1 A3 A4 WET DRY Total carbon Ecosystem traits

5. Modeling hydrologic and ecologic responses using a new ecohydrological model for identification of droughts (Chapter 7)

5.1. Strategy of ecohydrological drought quantification Ecohydrological Model In-situ observed rainfall JRA25 reanalysis [Onogi et al., 2007] LAI Soil Moisture Groundwater River discharge Satellite LAI (AVHRR) Calibration & Validation In-situ river discharge Calibration & Validation Agricultural Drought Index Nationwide crop production & Reports about past droughts Validation Hydrological Drought Index Drought Analysis Drought Indices - Standardized Anomaly Index (SA index) – [Jaranilla-Sanchez et al., 2011]

5.2.1. Results (1) Agricultural Drought Index Drought indices (SA index) Green:simulated annual peak LAI and Orange:nationwide crop production  The drought index calculated from the model-estimated annual peak of leaf area index correlates well with the drought index from nationwide annual crop production. R =0.89 Drought

5.2.2. Results (2) Ecohydrological drought analysis on 1988-1989 drought Drought indices Blue: River discharge Gray: Groundwater level Green: Leaf Area Index Drought Agricultural Drought Hydrological Drought  Historic agricultural droughts predominantly occurred prior to hydrological droughts and the hydrological drought lasted much longer, even after crop production has recovered.

6. Towards ecohydrological drought monitoring and prediction using a land data assimilation system: a case study on Horn of Africa drought (2010-2011)

6.1. Horn of Africa drought [FAO, 2011] [Anderson et al., 2012]  We cannot have the access to many ground observations to develop the drought prediction system.  Previous studies used satellite observations [Anderson et al., 2012] , numerical simulation [Sheffield et al., 2014] to build drought monitoring system in this region.

6.2. Strategy of ecohydrological drought forecast 2003 2004 2005 2006 2007 2008 2009 2010 2011 Observed meteorological forcings Microwave land observation CLVDAS (Reanalysis) CLVDAS (Perfect Prediction) Drought CLVDAS (Ensemble Stream Prediction) CLVDAS (Real Predicion) NOAA GFDL Meteorological forecast

6.3.1. Results (1) 2010-2011 drought in reanalysis LAI anomaly of 2010-2011 droughts in “reanalysis”.

6.3.2. Results (2) 2010-2011 drought in reanalysis  Root-zone soil moisture and LAI have longer memory of the past precipitation deficit than surface soil moisture. Time series of anomaly Surface soil moisture Root-zone Soil moisture Leaf Area Index

6.3.3. Results (3) Predictions: starting from 1 Sep 2010 Gray: Climatorogy Green: Horn of Africa drought (reanalysis) CLVDAS (Perfect Prediction) Leaf Area Index timeseries CLVDAS (Ensemble Stream Prediction) CLVDAS (Real Predicion)

6.3.3. Results (3) Predictions: starting from 1 Oct 2010 CLVDAS (Perfect Prediction) Leaf Area Index timeseries CLVDAS (Ensemble Stream Prediction) CLVDAS (Real Predicion) Gray: Climatorogy Green: Horn of Africa drought (reanalysis)

6.3.3. Results (3) Predictions: starting from 1 Jan 2011 CLVDAS (Perfect Prediction) Leaf Area Index timeseries CLVDAS (Ensemble Stream Prediction) CLVDAS (Real Predicion) Gray: Climatorogy Green: Horn of Africa drought (reanalysis)

6.3.3. Results (3) Predictions: starting from 1 Mar 2011 CLVDAS (Perfect Prediction) Leaf Area Index timeseries CLVDAS (Ensemble Stream Prediction) CLVDAS (Real Predicion) Gray: Climatorogy Green: Horn of Africa drought (reanalysis)

6.3.3. Results (3) Predictions: starting from 1 May 2011 CLVDAS (Perfect Prediction) Leaf Area Index timeseries CLVDAS (Ensemble Stream Prediction) CLVDAS (Real Predicion) Gray: Climatorogy Green: Horn of Africa drought (reanalysis)

7. Concluding Remarks (Chapter 9)

7.1. Publication list Chapter 2: Sawada, Y., H. Tsutsui, T. Koike, M. Rasmy, R. Seto, and H. Fujii (2016), A field verification of an algorithm for retrieving vegetation water content from passive microwave observations, IEEE Trans. Geosci. Remote Sens. In press. Chapter 3: Sawada, Y., T. Koike, K. Aida, K. Toride, and J. P. Walker, Fusing microwave and optical satellite observations to simultaneously retrieve surface soil moisture, vegetation water content, and surface soil roughness, Geophys. Res. Lett., submitted. Chapter 4: 澤田洋平, 小池俊雄 (2014), マイクロ波衛星観測輝度温度データを用いた水文－陸上生態系結合モデルのパラメータ最適化, 土木 学会論文集 B1 (水工学), 70(4). Sawada, Y. and T. Koike (2014), Simultaneous estimation of both hydrological and ecological parameters in an eco- hydrological model assimilating microwave signal, J. Geophys. Res. Atmos., 119, 8839-8857, doi:10.1002/2014JD021536. Chapter 5: Sawada, Y., T. Koike, and J. P. Walker (2015), A land data assimilation system for simultaneous simulation of soil moisture and vegetation dynamics, J. Geophys. Res. Atmos., 120, doi:10.1002/2014JD022895. Chapter 6: Sawada, Y. and T. Koike, Ecosystem resilience to the Millennium Drought in southeast Australia (2001-2009), J. Geophys. Res. Biogeosci., in preparation Chapter 7: Sawada, Y., T. Koike, and P. A. Jaranilla-Sanchez (2014), Modeling hydrologic and ecologic responses using a new eco- hydrological model for identification of droughts, Water Resour. Res., 50, 6214-6235, doi:10.1002/2013WR014847. Chapter 8: Sawada, Y. and T. Koike, Towards ecohydrological drought monitoring and prediction using a land data assimilation system: a case study on the Horn of Africa drought (2010-2011), J. Geophys. Res. Atmos. submitted

7.2. Research Contributions a) Our Microwave and Optical Fusion Approach (MiOFA) is the first field-validated algorithm which can objectively estimate the effect of soil surface roughness. b) Coupled Land and Vegetation Data Assimilation System (CLVDAS) is the first satellite-based LDAS which can simultaneously improve the skill of simulating both soil moisture and vegetation dynamics. c) We explicitly use the difference of optical and microwave vegetation observations to analyze the drought process for the first time. We focus on changes in ecosystem structure and their traits to understand the ecosystem responses to drought for the first time. d) We provide the new drought quantification strategy by explicitly calculating ecosystem damage as well as hydrological deficits. e) We develop the first drought monitoring and prediction framework which can predict ecosystem damages using LDAS applicable to the data scarce regions  These contributions open the door to the new perspective to analyze droughts without any barriers between hydrology and ecology. Thank You !!