SWAT best modeling practices, are we getting it right?

Size: px
Start display at page:

Download "SWAT best modeling practices, are we getting it right?"

Transcription

1 SWAT best modeling practices, are we getting it right? Dr. Indrajeet Chaubey Drs. Cibin Raj, K.P. Sudheer Purdue University (Chaubey) Presented as a keynote address at the SWAT China Conference Beijing Normal University July 27,

2 SWAT a truly global model One of the most widely used watershed simulation models 2700 peer reviewed publications About 450 publications/year >1 paper/day on SWAT applications Numerous graduate students/postdocs/researchers using SWAT to answer their research questions 2

3 What makes SWAT popular A comprehensive model with physical process representation of crop growth, hydrology and water quality Model uses readily available data (especially US) User friendly interface (ArcSWAT, QSWAT), toolboxes for data preparation and data access Well documented user manual and theoretical manual SWAT workshops and online tutorials Active SWAT user groups Well-written and open-source code: easy to understand and refer with the theoretical manual Helpful SWAT development team 3

4 Need for SWAT quality control SWAT research and results are widely used in environmental decision making SWAT is made very simple: a SWAT freshman can develop a basic running model within a week; there are potentials for misuse, misrepresentation, and misinterpretation SWAT community has a responsibility to ensure we apply the model and evaluate results appropriately We need to have a strict quality control in model development, interpretation, and reporting 4

5 Reproducibility of SWAT results Reproducibility of laboratory experiments has received much attention and is expected of any published study However, reproducibility of simulation experiments is still not widely demanded by the scientific community Structural uncertainty of the model: two different models setups with same data can give very different results Many user options: stream density, number of sub-basins, HRU thresholds Guidelines on user options are often not possible: have to be problem specific Considerable information may be lost with HRU threshold 5

6 Information lost in HRU delineation Her et al., 2015; Trans ASABE 6

7 7

8 Measured data has uncertainty Input data uncertainty: Measurement uncertainty associated DEM, land use, soil, weather data (precipitation, temperature etc.) Uncertainty in data used for model evaluation and calibration e.g. Stream flow (smaller uncertainty); water quality (larger uncertainty) Spatial variability in weather attributes: data is limited and also model is only capable to handle data resolution at a sub-basin scale Parameter uncertainty: Many parameters cannot be directly measured or are empirically based in SWAT 8

9 How are we accounting for model uncertainty (structural, input, parameter)? Calibrate model 9

10 Calibration should not be a solution for accounting uncertainties Calibration of model is for minimizing parameter uncertainty We need to reduce structural uncertainty of model by careful model set up Strauch et al. 2015, SWAT conference proposed a method to identify HRU thresholds with minimum error HRU and sub-basin threshold should be defined based on the application of the model, e.g. minimum land use threshold in a land use change study is recommended e.g. Chiang et al TASABE 53(5) 10

11 Default land management practices in model may not be sufficient for your study area Management practices are very critical in hydrology and water quality simulation in model Tile drain in Midwest US: tile drain contributes more water and nutrients to streams in Midwest US Pasture, forest management: default management practices are a good start but users need to explore local data about the watershed Plant and Kill operation of crops every year is not appropriate for perennials and trees 11

12 Crop growth representation is critical for hydrology and water quality 12

13 Systematic approach for improving bioenergy crops in SWAT SWAT requires about 25 crop growth parameters Miscanthus and upland switchgrass needed to be included in the SWAT crop database Identify Parameters One at a Time Sensitivity analysis Measure/estimate sensitive parameters Data collected from experimental fields (WQFS/TPAC): Biomass, leaf area index, crop height, harvest efficiency Improve SWAT crop growth algorithm Validate crop growth model Check SWAT simulation of perennial grasses and modify if required Validate energy crop simulations of SWAT with measured data from WQFS/TPAC Trybula et al, 2015; GCB bioenergy

14 Better growth and nutrient uptake representation by revised SWAT Nutrients stored in below ground biomass not considered About 100 kg N/ha & 30 kg P/ha stored Affect nutrient uptake process water quality estimations impacted 14

15 Model improvements are now incorporated in the official SWAT model: Version 612 ( 15

16 SWAT enhancements are required for continuing growth of the model Precipitation Source HRU SurQ (source HRU) Sed (source HRU) Nutrient (source HRU) SurQ (source HRU) Sed (source HRU minus VFS trapped) Nutrient (source HRU minus VFS trapped) Precipitation VFS Area (Conceptual) Precipitation Biomass yield SurQ (source HRU) SurQ (infiltrated) SurQ (source HRU minus VFS infiltrated) SurQ (VFSHRU) Source HRU Sed (source HRU) Sed (trapped) Sed (source HRU minus VFS trapped) Sed (VFS HRU) Nutrient (source HRU) Nutrient (source HRU minus VFS trapped) Nutrient (trapped) Nutrient (VFS HRU) VFS HRU Cibin et al, 2015; SWAT Purdue conference 16

17 Validation of model processes at small watershed scale is needed VFS improvements in the model validated using paired watershed measured data Field measured Control no VFS watershed VFS watersheds No VFS SWAT simulation With VFS (Default) With VFS (New) Runoff (mm) Sediment (Mg/ha) TN (kg/ha) TP (kg/ha) NO3 (kg/ha)

18 Soil Moisture is a Key State Variable Global Water Fresh Water Surface Water 2% 98% Fresh Oceans 69% 1% 30% Surface Groundwater Glaciers 3% Plants/Animals Rivers Marshes Atmosphere Soil Moisture Lakes Permafrost (Shiklomanov 1993 ) Data assimilation: Lack of monitoring data Model uncertainty Approaches not fully developed Land surface models Remote sensing data

19 Splitting one soil layer to two with same soil properties changes hydrology simulations Soil moisture from remote sensing data is for a depth up to 5cm Added a layer of soil at 5 cm depth with same properties as second layer This could affect vadose zone hydrology Need to improve the subsurface water movement will help in reproducibility

20 Relative Diff. (mm) Splitting one soil layer to two with same soil properties increases flows and decreases ET

21 Model reporting Some of the key information should be reported in manuscripts to make model simulations also reproducible Input data used Number of sub-basins and HRUs, thresholds used Calibration methodology Default and calibrated parameters Cibin et al., 2016; Global Change Biology-Bioenergy Supplemental information 21

22 Conclusions Calibration should not be a solution for accounting all model uncertainties Measurement uncertainty in input data and calibration data should be acknowledged during model evaluation We should strive to make our simulation experiments reproducible Explore SWAT process representation: contribute to model development and validation 22

23 Thematic Issue of Environmental Modelling and Software You are encouraged to submit an extended abstract for Thematic issue on Environmental Modeling and Software Extended abstract will be reviewed by Associate Editors If accepted, deadline to submit the full length paper is October 1, 2016 Please your abstract to 23