Wednesday, April 6, 2022

Draft Groundwater-Surface Water Model of the Ventura River Watershed

Date: April 1, 2022

To: Kevin DeLano, SWRCB

From: Paul Jenkin, Surfrider Foundation

RE: Comments on SWRCB Draft VRW GW-SW Model Report

The State Water Resources Control Board (State Water Board) Division of Water Rights and Los Angeles Regional Water Quality Control Board (collectively, the Water Boards) have published the Draft Groundwater-Surface Water Model of the Ventura River Watershed (VRW GW-SW Model) and Draft Model Documentation Report for the Groundwater- Surface Water Model of the Ventura River Watershed (VRW GW-SW Model Documentation Report).

This modeling effort has been accomplished with the SWRCB team and consultants with the goal of creating a tool that may be used in making water management decisions within the Ventura River watershed. To this end, we have reviewed the document and provide the following comments to help raise important questions and direct further refinement of the model. Quotes from the document are italicized.

First of all, and most importantly, it is recognized that all models are approximations, and the results of those approximations are heavily dependent on the available data. The report discusses model limitations:

All models and model results are subject to uncertainty, including model framework uncertainty due to incomplete scientific understanding of the system and necessary system simplifications, and model input uncertainty due to data measurement errors and data gaps (U.S. EPA, 2009). However, California Department of Water Resources (DWR, 2016b) states:

While models are, by definition, a simplification of a more complex reality, they have proven to be useful tools over several decades for addressing a range of groundwater problems and supporting the decision-making process. Models can be useful tools for estimating the potential hydrologic effects of proposed water management activities.

In order for the model to be a “useful tool”, it must accurately represent the physical reality it is attempting to replicate. In this case the modelers have been transparent in the limitations of the model and their manipulation of certain variables in order to calibrate the model to achieve a reasonable match with the field data. However, in generating output that appears to be more accurate in certain circumstances while not in others, this artificial calibration may give a false sense of the model’s accuracy and hence its usefulness in “estimating the potential hydrologic effects of proposed water management activities.”

At this point it is clear that the model is still a draft, and additional calibration will be required to more accurately match the physical systems being modeled. This will be important for implementing various management scenarios in the model, otherwise the errors in the model will outweigh any small changes in climate or water management.

The intent of this review is to provide constructive feedback such that the model may be improved upon to further its usefulness in the watershed. The following observations are based upon review of the report and some limited use of the visualization tool. (The visualization tool does not have full capabilities in older versions of Excel). These comments are focused on the main stem of the Ventura River and headwaters in Matilija Creek.

The Model underestimates the presence of surface flow

A fundamental question in assessing the accuracy of groundwater/surface water interaction is;

How well does the GW/SW Model represent rising groundwater, particularly in the “Live Reach” upstream of Foster Park?

In this gaining reach, the town of “Casitas Springs” was named for the perennial flows, i.e. “springs” along this reach of the Ventura River. However, the model calibration/validation results predict greater dryness than the field observations for this section of the river.

Fig 5.26 shows surface flow Wet-Dry Mapping comparing the model predictions with data measured in the field. This figure shows that the model predicts drying in the live reach when it was not observed during WY 2009 and 2010.



The model also predicts a greater spacial extent of drying, particularly upstream of the San Antonio creek confluence and downstream of the Foster Park gage. It is possible that this is an artifact of the model segments which may not coincide with the actual physical geographic boundaries. However, these boundaries are marked by bedrock outcrops that force groundwater to the surface and are integral to the representation of physical processes.

This anomaly in the model appears to carry into the Unimpaired Flow scenario which predicts drying in the “wet reach” in the absence of any pumping or diversion. This result is unexpected given the geology and historic record of perennial flows in this reach.


The sensitivity analysis discusses the response of the model to coefficient changes and reveals the importance of groundwater coefficients especially in areas of rising groundwater. The report states that:

The modeled wet-dry mapping is primarily a result of the groundwater level calibration coupled with streambed conductivities and widths, that together determine the extent of the gaining and losing reaches through the SFR package. During the calibration process, the wet-dry mapping was assessed and adjustments were made to the streambed conductivities to achieve better match. These adjustments were relatively minor; larger adjustments would feed back into the groundwater and surface water calibrations, requiring additional iterations. (p162)

This suggests that streambed conductivity was the sole coefficient adjusted to try to achieve a match with the data. Meanwhile, the model results derive from the complex relationship between all the other surface and groundwater inputs from the top of the watershed down, each of which is subject to assumptions and potential error. As suggested, additional iterations will be required to better calibrate the model.

Hydraulic conductivity

The hydraulic conductivity constant, Kx/y, determines the rate of groundwater flow downhill through the basin. As shown in Figure 5.11a, the main stem Ventura River is modeled with a hydraulic conductivity coefficient an order of magnitude above other regions in the watershed (i.e. 1250 ft/day vs 150 ft/day). This assumption was based on a single test done at Foster Park, and applied to the entire Upper Ventura River Groundwater Basin upstream.


Given a similar geology, why is the Upper Ventura River Groundwater basin conductivity an order of magnitude great than the Ojai basin?

When estimates of hydraulic parameters are available for the regions of the modeled physical hydrogeologic system, the corresponding values of those parameters in the model should be similar, but do not have to be identical. There are two reasons for this. First, the estimates themselves have associated errors, often of an order of magnitude. Second, when these estimates are based on hydraulic tests, the volume of soil or rock stressed by the test is often smaller than the volume in the model for which the parameter applies. In that case, the input hydraulic conductivity or transmissivityrequired to calibrate the model is often larger than the measured value due to the scale effect. (p209)

Hydraulic conductivity can be hard to estimate but has a significant influence on the groundwater model. The assumption of homogeneous high conductivity throughout the groundwater basin underlying the main stem Ventura River results in the model exhibiting rapid underflow through this basin which in turn has a significant effect on predicted groundwater levels. Additional monitoring wells and hydrogeologic testing are needed to improve understanding of the transmissivity of this basin.

Calibrating to Flow Gage Data:

A major concern is that;

Uncertainty from PRMS-portion of the simulation will propagate and influence groundwater recharge estimates.

Flow gage data is perhaps the most important parameter used in calibration of the groundwater- surface water model. As acknowledged in the report, the accuracy of flow data decreases at lower flows. One reason for this is that the majority of the stream gages were installed and maintained by the Ventura County Flood Control District as part of their ALERT flood warning system. These gages were neither intended nor maintained to provide low flow data, as their purpose was for flood control. The two possible exceptions are the USGS gages at Foster Park (608) and Matilija Creek (602) which have historically received more regular re-staging during the dry months, but even at Gage 608 measurements during the low flow periods (i.e., summers) are of Poor quality with errors anticipated to be greater than 8% (p158)
Although the USGS gages provide the most accurate flow data, the model calibration relied heavily on the County flow gages. The resulting modeling error statistics show that the greatest errors occur at the USGS gages, Matilija Creek (602) and Foster Park (608).

The report discussion of calibration and error analyses reveal the difficulty in achieving good correlation for all sites over the full range of water year types.

Agreement of the model results during low-flow periods (i.e., summer and fall) is generally good, with monthly average flows being well predicted down to approximately 1 cfs at most locations. Exceptions are noted in years with higher summer flows (e.g., 1995, 1998, 2005, and 2006) where the model underpredicts the summer flows. Adjusting calibration parameters to correct these years resulted in poorer prediction of the summer flows in the more typical and lower flow years. Since the lower flows are more critical for many drivers in the watershed (e.g., fish passage), model calibration was focused on obtaining better predictions in the lower flow years. The effect of this on the summer volume errors is discussed in more detail in Section 5.4.1.2. (p184)



The mean errors in Table 5.6 are negative at all gage locations, indicating a general bias in the model (i.e., a consistent underestimation of flows at all locations). Additional evaluation of the mean and RMS errors indicates that the largest summer streamflow errors are during the wet years with higher stream flows. This is a result of prioritizing the accuracy of years with lower flows during the calibration process, since the lower flow years are critical with respect to many of the project goals (e.g., evaluation of fish passage). (p198) 

 Relative summer volume errors (as percentages) are misleadingly high due to low measured flow rates and are poor metrics for assessing calibration performance for ephemeral systems that can result in a zero in the denominator. Absolute errors (i.e., in cfs) are more appropriate. For example, at Foster Park (Gage 608) a relative error of - 47.1% corresponds to mean and root-mean-square (RMS) errors of only -3.5 cfs and 5.9 cfs, respectively. Additionally, these flow rate errors are dominated by high runoff years. Excluding the six years with >50,000 AF of run-off at Foster Park results in the mean and RMS errors decreasing to -1.4 cfs and 2.6 cfs, respectively. Furthermore, in the Very Dry and Dry years the mean and RMS errors decrease to -0.3 cfs and 1.3 cfs, respectively. (p200)


Note that although errors of 1-3 cfs sound minimal, this is actually very significant in a system that is now often running dry.

Because of difficulties modeling varied flow regimes, the model was calibrated to low flow gage data. However, because gage accuracy decreases at lower flows, this strategy may introduce a significant error. It may be prudent to first calibrate the model to the more accurate gage data at Matilija Creek (602) and Foster Park (608) for moderate flow years. In theory the modeling coefficients should not vary significantly under different flow regimes. Developing a model that behaves well in “normal” conditions and then “stress testing” and fine tuning the model so that it also performs well in dry conditions would result in a more robust representation of the physical system. Calibration to dry year flow data results in the dry bias seen in the Wet-Dry mapping discussed above.

The Matilija Creek at Matilija Hot Springs (602) USGS gage is perhaps the most accurate stream flow gage in the watershed as it is controlled with a concrete weir and regularly staged. This gage may be more appropriate for model sensitivity and calibration than the North Fork Matilija gage.

The model underestimates summer flow volumes (June-Sept) at Gage 602 by 42%. Any error in this reach, which is the primary inflow to the mainstem Ventura River during the hot summer months, will be amplified in the downstream groundwater/surface water model.


Are diversions in Matilija Creek adequately accounted for?

The Visualization Tool for Gage 603A/Upper Matilija Creek reveals that the model Base Case tracks closely with Unimpaired flow for the dry months, while the gage often reads lower. The model overpredicts flows at Gage 603. There are irrigated lands and residential parcels in the canyon all of which draw from the creek or shallow wells. Most of these appear to be included in the model. Have the groundwater and surface water diversions in Matilija Canyon been adequately estimated in the model?





Are diversions in the Kennedy Reach adequately accounted for?

This reach has a number of wells and at least two surface diversions that collectively have the capacity to pump at a rate greater than average streamflow input to shallow alluvium in this reach during the summer months. However, the model predicts very little effect (0.2-0.7 cfs) on surface flows from eliminating pumping and diversion in the Unimpaired Flow condition.




At Gage 607 (Figure 5.16), the measurement indicates flows decrease to zero in most years; whereas, the model typically decreases to approximately 1 cfs in most years. The discrepancy is likely due to not including details of the hydraulic structure related to the Robles diversion (i.e., the embankment and gate structure that blocks the Ventura River) in the VRW GSFLOW Model. This structure would result in pooling of water and additional streamflow losses (through infiltration) upstream of the diversion structure. While these local details are not fully captured in the VRW GSFLOW Model, the additional water passing the diversion location during low flows infiltrates and is lost from the stream shortly downstream.

Comparing the Visualization Tool Base Case with Gage 607 data shows no clear trend in the errors, but as noted the river often goes dry upstream of Robles Diversion. A dry river within the shallow alluvium of the Kennedy Reach while the model predicts 1cfs or greater indicates that closer attention to pumping and diversion in this reach may be helpful.

It is important to be able to accurately model the flows entering the Ventura River from the upper watershed, especially during summer months, as this affects the water balance in the groundwater basin downstream during times of high irrigation demand. However, the surface flow modeling is not consistent with the gage data. The model generally overpredicts flows at Gage 603, underpredicts flows downstream at Gage 602, and underpredicts drying at Gage 607. These observations describe inconsistencies in modeled stream flows that should be resolved in order to provide a more accurate assessment of inflows into the groundwater basin.


Assumptions for Arundo donax:

The model assumes that the extent of riparian vegetation is fixed in time. This neglects the effects of Arundo eradication efforts, reduction in vegetation following storm events in wet years, and potentially increased ET following wet years as vegetation reestablishes. This limitation would primarily affect dry season low-flow periods.


According to the County of Ventura, over 270 acres of Arundo donax have been removed from an area encompassing 1,200 acres of the watershed encompassing Matilija Canyon and the Upper Ventura River during the period 2006 to the present. Most of this was in Matilija Canyon as shown in red in Figure 4.10.
Discussion during the workshop indicated that the model applies ET rate of 24 ft/yr for Arundo, with reference to CalIPC 2011. (For comparison, turf grass exhibits ET of 3-4 AF per acre.). Using a conservative estimate for Arundo ET of 20AF /acre, this would theoretically yield 270acres X 20AFY = 5400 AFY. This translates into 5400AFY/365days = 7 cfs. (An equivalent removal of turf grass would yield around 1 cfs)

It is important to note that 7 cfs is a significant flow augmentation in a system that often experiences flows of 2-3 cfs or less, yet stream gage observations do not reflect any substantial change in flow in the years following arundo removal.

This suggests that the model overestimates ET losses from the upper watershed, which would reduce predicted instream flows. This may be a source of error in the predicted flows at Matilija Creek at Matilija Hot Springs (602).


Matilija Dam:

additional errors in the model results may result from uncertainties in estimating release volumes from Matilija Reservoir (Section 3.5.1).

 If the SWRCB proposes to run a scenario for dam removal, this aspect of the model will need close attention. It is important to note that Matilija Dam spillway elevation has been set at 1095 and operated as run of the river except in cases of releases for diversion at Robles. The model documentation and discussions are unclear as to the assumptions made on the operation of Matilija Dam. A modeling scenario for dam removal will primarily depend on predicted changes in ET for the reservoir reach upstream. Dam removal will convert a large area of riparian and lacustrine habitat to upland habitat, with an associated decrease in ET. Additional changes will occur in the main stem Ventura River from renewed sand, gravel, and cobble supply which will alter the characteristics of the stream bed and alluvium. (This also applies to a post-Thomas Fire scenario.)

When constructed in 1947, Matilija Reservoir originally had an active storage volume of 7,000 AF. However, due to sedimentation and lowering (or ‘notching’) of the dam, the active volume has decreased substantially as indicated Table 3.1. Over the modeling period, the active storage volume at full pool decreased from 930 AF to 270 AF. This changing storage capacity during the modeling period is not implemented into the model. In the model the elevations of the lake cells were lowered to create a volume of 1,503 AF with a spillway elevation of 1,095 ft based on information on Ventura County Public Works Agency website. Although this is not consistent with information in Table 3.1 the effect in the model is primarily to increase dead storage with anticipated negligible effects on streamflow.

Outflows from Matilija Reservoir were modeled as a combination of overflows over the dam crest and specified releases (Section 3.5.1), each into a downstream stream segment representing the dam spillway. (p80)

Historically, the CMWD would release water from Matilija Reservoir to enable additional diversions downstream through the Robles Canal to Lake Casitas. Information on these releases is limited and had to be estimated for the modeling period.
Reservoir elevation data were available from July 2003 onwards and these were used with the stage-storage information from 2002 (Table 3.1) to estimate daily releasevolumes. Prior to July 2003 the releases were estimated by correlating streamflow data from Gage 602B (downstream of Matilija Reservoir) and Gage 604 (North Fork Matilija Creek, and not subject to releases) to identify periods of releases and estimate release rates. These estimates were capped at 150 cubic feet per second (cfs), based on outlet capacity. The resulting releases implemented into the model are plotted in Figure 3.7. (p84)

 

Other Comments:

Gage 604, No Frk Matilija Creek at Matilija Hot Springs

Matilija Hot Springs is located on Matilija Creek below the dam. This gage is actually located less than a mile upstream of the Matilija Creek/NF Matilija Creek confluence under a bridge over Hwy 33. The correct stream gage nomenclature per VCWPD is:

Matilija Creek at Matilija Hot Springs (602) 
North Fork Matilija Creek (604)



Conclusions and Recommendations:

The Draft Groundwater-Surface Water Model of the Ventura River Watershed is a good first draft representation of a complex and dynamic physical system. The disclosure of uncertainties in the modeling process and the errors matching stream flow data suggest that further work is required to create a model that is useful for assessing the various scenarios that have been proposed.

The model must first be able to accurately represent changes in the watershed resulting from the highly variable precipitation during the study period. Robust model performance over a wide range of rainfall and flows is required before an assessment of any future changes can be evaluated.

Because of the inherent error in low flow stream gage data, it is suggested to first calibrate the model to the best available stream flow data for the moderate years and then work to match the high and low flow years. In this manner the physical coefficients may be established to ensure confidence that simulated model scenarios produce useful results.

For example, a climate change scenario would include incremental variations in precipitation and ambient temperature. It is essential that the model can accurately predict the extremes of the recent past before it can project into the future. The current assumptions regarding Arundo donax should be closely examined for ET in Matilija Canyon and effect on streamflow. If the Matilija Dam removal scenario is pursued, close coordination with the Matilija Dam Ecosystem Restoration Project (MDERP) technical team is recommended. The current assumptions for pumping and diversion should be closely examined so that any future water management scenarios may be adequately assessed.

Future development of the model should also include a robust monitoring network including additional stream gages and dedicated monitoring wells to better understand the nature of surface and subsurface flows and provide data to feed back into the model.


References:

State Water Board Instream Flows - Ventura River