What are ways of comparing soil moisture dynamics of rain gardens?

Sandra Albro and I completed installation of monitoring equipment in three experimental parcels and one control parcel in Gary’s Aetna neighborhood on April 19, 2016.  The first month’s monitoring data are now available.  Estimation of depth-weighted soil moisture (m3/m3) is the simplest way to visually compare the dynamics of soil moisture (Figure 1).  The control parcel has the highest water content followed by experimental parcels 2, 1, and 3.  Our weather station in Aetna is temporarily offline, but we can obtain precipitation data from the U.S. Weather Station in Gary (Figure 2).  In general, the spikes in soil moisture correspond to storm events.  Because all of the rain gardens have identical soil mixtures, their variation in water content must reflect differences in local catchment area and the effects of neighboring tree cover.  With predominantly sandy soils, it is not too surprising to observe the rapid decline in soil moisture following precipitation events, but the control parcel (lacking a rain garden) does appear to have a slower rate of decline in soil moisture while maintaining a higher water content.  To move beyond a qualitative comparison of these changes, we will need a quantitative method to find local minima and maxima in a soil moisture time series and determine the slope of soil moisture decline from one local maximum to the next local minimum.  We would also need the method to find the increment of soil moisture from a local minimum to the next local maximum.  Documenting this method and some preliminary results is the purpose of this blog entry.

Figure 1.  Patterns of weighted average soil moisture at four sampling locations in Gary.  Parcels are as follows: 1200 Oklahoma (Gary E1), 1035 Oklahoma (Gary E2), 1252 Dakota (Gary E3), and 910 Idaho (Gary C1).

Figure 2.  Precipitation patterns for U.S. Weather Service station KGYY in Gary, Indiana.

Finding local extrema (maximum and minimum values) in a time series is the first step in analysis.  Although it would be possible to identify these values visually, a better way is to find a quantitative, rule-based method.  Although I could find no such method in available R libraries, I did find relatively simple algorithm (findpeaks) by Timothée Poisot that uses the diff function in R (https://rtricks.wordpress.com/2009/05/03/an-algorithm-to-find-local-extrema-in-a-vector/).  This algorithm requires specification of a parameter for the analysis interval (with a value between 1 and the length of the entire time series).  I tried a range of values from 0.3% to 25% of the data range and did not find a reasonable match to the observed major spikes and troughs.  Lower parameter values performed better, but the noisy time series tend to produce too many local extrema.  The time series in Figure 1 consist of data reported at 5-minute intervals and thus consists of a mixture signals of process operating at different time scales.

A useful algorithm for smoothing noisy time series is the Loess function in R.  It is an algorithm for fitting local polynomial regressions.  The loess function in R also has a parameter for the size of the sampling interval of the time series.  To smooth out diurnal variations, I set the value of this parameter to a value that equaled the number of samplings per day (288 at 5-minute intervals).  The span parameter for the loess function is thus the ratio of the number of samples per day to the number of days in the time series being analyzed.  With the loess fitted time series, the identification of extrema by the findpeaks function was far less sensitive to values of the sampling interval, and the location of peaks and troughs varied asymptotically with decreasing size of the sampling interval.  I programmed this method in an R script (Slope_Calc.R), and the method correctly identified nearly all of the extrema in the time series for weighted average soil moisture data from the Gary experimental parcel 1 (Figure 3).

Figure 3.  Results from the R script (Slope_Calc.R) to find local minima and maxima in the soil moisture time series from the Gary experimental parcel 1.  The black line represents the raw data, the blue line the loess fitted regression, the blue points are the local minima, and the red points are the local maxima.

Once the method estimates local minima and maxima, it then estimates the slope of the observed decline from a local maximum to the next local minimum.  Assuming that the loss of soil moisture is an exponential function, the method applies a linear regression to the log10 values of soil moisture.  The method also calculates the increment in soil moisture from a local minimum to the next local maximum.  Figure 4 shows the resulting slope patterns.

Figure 4.  Semi-logarithmic plot of the estimated slopes (red lines) from a linear regression of log transformed of raw data between local maxima and the next local minimum for the weighted average soil moisture data from the Gary experimental parcel 1.  Other lines and points are the same as in Figure 3.

Although the method does correctly estimate the timing of the local maxima and minima, the smoothing functions do have some problems.  On May 13, for example the slope is positive because there was a short interval in which the analysis failed on a smoothing artifact.  The amplitude of the local maximum is also lower than the observed data.  However, the amplitude better matches the loess fitted time series.  Overall, it appears that the method does capture the basic characteristics of the soil moisture dynamics and, thus, appears to be a reasonable method for comparing experimental parcels and controls.  Figure 5 shows an interesting relationship between the magnitude of the soil moisture increment and slope of the subsequent decline in soil moisture.

Figure 5.  Plot of slope (1/s) versus soil moisture increment (m3/m3) from the data in Figure 4.  Omitted is the spurious slope from the peak on May 13.  The blue line is the regression line with an r value of -0.82 (p<0.001).

The loss of soil moisture from a rain garden or control plot is a complex function of processes operating at different time scales.  Evapotranspiration and deep infiltration are likely dominant processes in the Aetna neighborhood.  These two process depend on additional factors such as soil temperature, atmospheric relative humidity, vegetation cover and type, and soil porosity.  The results in Figure 5 suggest that it might be possible to find patterns associated with various driving variables in different parts of the time series.  Additionally, the profile of these responses may be useful in comparing water retention by experimental and control plots in different seasons. 

How do rain gardens process runoff and rainfall?

Through the Vacant to Vibrant Project, we have been able to obtain preliminary performance data for three rain gardens in Cleveland.  The data are from a rain garden in the Cleveland Botanical Garden and two Vacant to Vibrant parcels in Cleveland’s Woodland Hills neighborhood.  The remote monitoring station at 10607 Hulda Avenue (figure 1) consisted of an Onset RX3000 series data logger with cellular communication, soil moisture probes (Onset Part Number S-SMC) deployed at 3 cm, 10 cm, and 20 cm and a soil temperature probe (Onset Part Number S-TMB) deployed at 3 cm in the primary rain garden.   At the other two experimental parcels (10611 Crestwood Avenue and 10411 Shale Avenue), we installed buried Onset H21 data logger and deployed a soil temperature probe at 3 cm and soil moisture probes at 3 cm, 10 cm, and 20 cm in the primary rain garden.  The H21 data logger in the Shale experimental parcel failed due to water damage.  Monitoring stations in the Hulda and Crestwood parcels, however, provided continuous data logging from July 7, 2015 through 2016.  In June 2014, we installed the H21 data logger and three soil moisture and one soil temperature sensors (using the same design as with the Shale and Crestwood parcels) in the experimental rain garden of the Cleveland Botanical Garden.   We have been monitoring the Cleveland Botanical Garden rain garden continuously since June 2014 and have used it for various experimental and calibration studies.

Figure 1.  Remote logging station in rain garden at 10607 Hulda Avenue.

Complementing the rain garden monitoring, we have also obtained estimates of rainfall from three sources: archival from U. S. National Weather Service for the KCLE station and from two weather stations.  One weather station was in the Woodland Hills neighborhood at the Green Corps Woodland Avenue Farm.  It is a remote weather stations consisting of an Onset RX3000 series data logger with cellular communication and rain gauge (Onset Part Number S-RGB), solar radiometer (Onset Part Number S-LIB), temperature and relative humidity (Onset Part Number S-THB), and wind speed sensors (Onset Part Number S-WSA).  The Woodland Hills weather station was active from August 5, 2015 to November 5, 2015.  We also have weather data available from an H21 data logger equipped weather station at the Cleveland Botanical Garden.  We installed this weather station in June 2014 using the same sensor set as the Woodland Hills weather station.

The results reported here are for the summer of 2015 (July 7 to August 27).  Figure 2 shows the depth-weighted soil moisture for each of the three monitored rain gardens.  The dynamics of water content in each rain garden reflect the differences in patterns of soil temperature (Figure 3).  The CBG rain garden had the highest soil water content and lowest soil temperature and the Crestwood rain garden had the lowest soil water content and highest soil temperature.  The Hulda rain garden was intermediate for both soil water content and soil temperature.

We calibrated all soil moisture sensors prior to deployment.  Although we detected bias in the sets of sensors used for each of the rain gardens, the biases were not consistent across types of calibration media (air, water, and damp potting soil mixture).  Therefore, we did not try to analyze gradients of soil moisture within each of the rain gardens.  Rather, we compared patterns of change in soil moisture at each of the sampled depths (3 cm in Figure 4, 10 cm in Figure 5, and 20 cm in Figure 6).   The patterns are similar to those for the weighted average soil moisture measurements (Figure 1), and they reflect the patterns of precipitation (Figure 7) observed at various weather stations.  In general, however, the rain garden at the Cleveland Botanical Garden seemed to have much higher rainfall than measured at the other two weather stations; indicating that the higher water content of the rain garden at the Cleveland Botanical Garden is the result of lower soil temperature and higher precipitation.  Figure 8 shows the relationship between rainfall measured at the Cleveland Botanical Garden and at the KCLE station of the U.S. Weather service.  These two locations are 13.5 miles apart and the residual variability between the two stations reflects the regional variation in intensity of rainfall events.

Figure 2.  Patterns of change in weighted average soil moisture at three rain gardens:  Cleveland Botanical Garden demonstration rain garden and in Woodland Hills rain gardens at 10611 Crestwood Avenue and 10411 Hulda Avenue.

Figure 3.  Patterns of change in surface soil temperature at three rain gardens:  Cleveland Botanical Garden demonstration rain garden and in Woodland Hills rain gardens at 10611 Crestwood Avenue and 10411 Hulda Avenue.

Figure 4.  Patterns of change in soil moisture at 3 cm depth in three rain gardens:  Cleveland Botanical Garden demonstration rain garden and in Woodland Hills rain gardens at 10611 Crestwood Avenue and 10411 Hulda Avenue.

Figure 5.  Patterns of change in soil moisture at 10 cm depth in three rain gardens:  Cleveland Botanical Garden demonstration rain garden and in Woodland Hills rain gardens at 10611 Crestwood Avenue and 10411 Hulda Avenue.

Figure 6.  Patterns of change in soil moisture at 20 cm depth in three rain gardens:  Cleveland Botanical Garden demonstration rain garden and in Woodland Hills rain gardens at 10611 Crestwood Avenue and 10411 Hulda Avenue.

Figure 7.  Comparison of measured daily rainfall at three locations (U.S. Weather Service at KCLE, project weather station in Woodland Hills, and project weather station at the rain garden in the Cleveland Botanical Garden).  The Woodland Hills weather station was not deployed until August 5, 2015.

Figure 8.  Comparison of daily rainfall estimates from KCLE archives and measurements at the Cleveland Botanical Garden weather station for the period July 7 to August 27, 2015 with regression line (correlation coefficient, 0.583, p<0.001).

Rationale for Monitoring and Analysis of Green Infrastructure in the Vacant to Vibrant Project

Monitoring and analysis of stormwater capture by green infrastructure in the experimental parcels in each of 3 cities is one of the main tasks of the Vacant to Vibrant Project.  Specifically, we planned to monitor in-site soil stormwater capture and filtration by continuously measuring soil moisture at 5 cm, 10 cm, and 20 cm soil depths and soil temperature at 5 cm at one site in each parcel. We planned to use existing LIDAR datasets to interpret soil moisture in light of local topography and drainage patterns.  We have now completed preliminary analysis of the method for using soil moisture to assess the effectiveness of GI installations in altering stormwater runoff.

Figures 1 and 2 present soil moisture data collected at CWRU’s University Farm at two sites (Forest and Meadow) for 2011 and 2012.  These data were collected with the same type of instruments that we propose using to monitor parcels.  The contrast in patterns of variability at the two sites for the two years is instructive because of the contrast in total annual precipitation between years.  Total precipitation was 65.32 in for 2011 and 44.62 in for 2012.  Furthermore, we anticipate that the experimental parcels will function more like the forest site and the control sites will resemble the meadow site.  Both of these sites at University Farms have similar slopes, but have different soil compaction and plant communities.

Figure 1.  Pattern of variation of soil moisture at two sites in University Farms (Case Western Reserve University, Hunting Valley, Ohio) for 2011.

Figure 2.  Pattern of variation of soil moisture at two sites in University Farms (Case Western Reserve University, Hunting Valley, Ohio) for 2012.

These data demonstrate substantial differences in water retention and processing by the two sites.  The simplest comparison is the level of soil moisture saturation.  These soils have a saturation value of 0.35 m^3/m³.  The forest site averaged 58% of saturated soil moisture in 2012 and 66% in 2011.  In contrast, the meadow site averaged 81% of saturation in 2012 and 93% saturation in 2011.  Thus, the forest site’s mean saturation was 0.23 and 0.27 less than the meadow site’s for 2012 and 2011.  In general, higher soil moisture saturation results in greater runoff from a storm event.  The more saturated soils of the meadow site have significantly less capacity to absorb and process individual storm events, particularly in the spring of both 2011 and 2012 and especially in the summer of 2011, when the Cleveland area received nearly double its annual mean precipitation.

The monitoring design for soil moisture monitoring in the Vacant to Vibrant project will provide better estimates of retention and processing of storm water than the monitoring summarized in Figures 1 and 2.  The data in these two figures is from a single, vertically positioned 10 cm probe.  The proposed monitoring will use three probes with horizontal placement at three depths (5, 10, and 20 cm) below the soil surface.  This placement will provide a profile of soil moisture through the root zone and thereby will yield a more accurate assessment of soil moisture dynamics.