Effect of soil temperature and soil water content on fine root turnover rate in a California mixed conifer ecosystem



[1] Measurement of fine root production and turnover rate, the reciprocal of mean life span of a root population, is crucial to the understanding of the carbon cycle of an ecosystem as fine roots account for up to 30% of global terrestrial net primary production. Our goal was to characterize fine root production, mortality, standing crop, and turnover rate in a Mediterranean climate. Using simulations, we established that our sampling interval must be less than monthly to keep the turnover rate error to less than 10%. Adhering to this interval, we measured fine root turnover rate by mark-recapture modeling methods and compared predicted with observed turnover rates. The best selected model indicated that these rates were a function of diameter, length, soil temperature, and soil water content. Turnover rate increased with decreasing diameter and length and increasing soil temperature and soil water content. We found a yearly pattern of hysteresis between fine root production, mortality, and turnover rate relative to soil temperature. This was explained by soil temperature-moisture hysteresis using our best selected model. Production and turnover rate were greater in spring to early summer when both soil temperature and soil moisture were high, resulting in a seasonal variation of belowground net primary production. We suggest that this behavior could be a result of fine roots' strategy to cope with a limited growing season of a semiarid Mediterranean climate.

1. Introduction

[2] Fine root (<2 mm in diameter) production, mortality, decomposition, and turnover are major processes regulating carbon (C) flux, C budget, C allocation, and C sequestration in terrestrial ecosystems, as fine roots account for 10∼30% of net primary production (NPP) in temperate forest ecosystems and up to 65∼80% of NPP in temperate grassland ecosystems [Schulze et al., 1996; Jackson et al., 1997, and references therein; Matamala et al., 2003; Norby et al., 2004; Frank, 2007]. Here, we evaluate the importance of high frequency fine root observations and in situ environmental data collection for predicting turnover rate in a mixed conifer forest ecosystem in Southern California using a mark-recapture modeling approach. The advantage of our approach is that it allows us to evaluate simultaneously multiple factors that vary with space and time.

[3] We quantified fine root production, mortality, standing crop and turnover rate using frequent minirhizotron (MR) observations (median sampling interval of every 3 days). In this study, turnover rate is defined as the reciprocal of mean life span of a fine root population [Matamala et al., 2003]. Before going into further analyses, establishing an appropriate MR observation interval is important because actual root production and mortality may occur very rapidly [Stevens et al., 2002; Tingey et al., 2003; Stewart and Frank, 2008]. Without this process, it is possible to underestimate production, mortality, turnover rate and NPP [Tingey et al., 2003]. Based on a simulation study, Johnson et al. [2001] found a simple formula that predicts the percentage of missed fine roots based on the ratio of the sampling interval to the half-life of the roots. With our frequent MR observations, we can compare our results with their predicted percentage of missed fine roots.

[4] Previous studies have correlated fine root turnover rate (survivorship) with time-independent factors such as root characteristics and experimentally manipulated factors. Root characteristics are easily quantifiable with MR and image analyses [Johnson et al., 2001]. For example, fine root turnover rate has been correlated with root diameter [e.g., Tierney and Fahey, 2001; Wells and Eissenstat, 2001; Joslin et al., 2006], root order [e.g., Eissenstat et al., 2000; Majdi et al., 2001], root pigmentation [e.g., Anderson et al., 2003], depth [e.g., Hendrick and Pregitzer, 1992; Eissenstat et al., 2000], seasonal cohorts [e.g., Wells et al., 2002; Ruess et al., 2003], root age [e.g., Gaudinski et al., 2001; Wells et al., 2002; Ruess et al., 2003], MR tube installation disturbance [e.g., Joslin and Wolfe, 1999], pruning [Anderson et al., 2003], fertilization [e.g., Guo et al., 2008b], animal browsing [Ruess et al., 1998], mean growing season temperature [Hendrick and Pregitzer, 1993], and irrigation [e.g., Anderson et al., 2003]. Eissenstat and Yanai [1997] analyzed relationships between root survivorship and above mentioned factors using a cost-benefit analysis, where root survivorship increases if the benefit of nutrients and water uptake exceeds the cost of root construction, maintenance, and protection.

[5] It is becoming increasingly important to establish how environmental factors may affect fine root turnover rate because of expected regional climate changes [Diffenbaugh et al., 2008]. Effects of soil temperature (T) and soil water content (θ) on soil CO2 efflux, which is an important part of C cycles in the soil, are well studied [e.g., Arneth et al., 1998; Davidson et al., 1998; Delire and Foley, 1999; Hirano et al., 2003; Grünzweig et al., 2009; Zhou et al., 2009]. To elucidate effects of environmental factors on turnover rate of fine roots, we continuously collected environmental data (every 5–15 min) adjacent to each of the MR tubes. Our data set is unique in that we can evaluate both time-independent fine root characteristics and time-dependent environmental factors.

[6] Fine root turnover rate (survivorship) is very often correlated to the factors of interest using statistical analyses (e.g., Kaplan-Meier survival analysis), where fine roots are pooled into groups by season or by year as if they were born on the same date [e.g., Norby et al., 2004]. Pooling of fine roots is necessary because it is difficult to obtain adequate numbers of fine roots born on the same date under natural conditions. This approach assumes that roots born on various dates within the same season or year experience the same environmental conditions over the course of their life span. This is an unlikely assumption because environmental variables change with time, especially under natural conditions. It is also statistically inappropriate to correlate time-dependent variables with survivorship after determining survivorship without them [Cooch and White, 2006]. For these reasons, we modeled fine root survivorship using mark-recapture methods [Lebreton et al., 1992]. Ruess et al. [1998, 2003] used this approach to study various factors (time, site, age, cohort, season, growing season and animal browsing) affecting fine root turnover rate in Alaskan Taiga forests, but time-dependent environmental variables were not included. Since fine root growth and decomposition are biochemical processes, they should vary with T and θ, both of which vary highly over time. Our approach allowed us to examine the relationship between turnover rate and T and θ, as well as root characteristics, in detail.

[7] Our specific objectives were (1) to determine how sampling interval affects the estimate of fine root production, mortality, standing crop, and turnover rate, (2) to characterize temporal variation in production, mortality, and turnover rate, and (3) to quantify the effects of root characteristics, T, and θ on turnover rate.

2. Materials and Methods

2.1. Study Site

[8] All observations were made at the University of California James San Jacinto Mountains Reserve (http://www.jamesreserve.edu) located in the San Jacinto Mountains in Riverside County, California USA (Elevation: 1,650 m, Longitude: 33.8078°N, Latitude: 116.7771°W). The site contains an old-growth stand of California semiarid mixed conifer forest with various age groups of trees; it has never been commercially logged. Dominant tree species are ponderosa pine (Pinus ponderosa C. Lawson), sugar pine (P. lambertiana Douglas), incense cedar (Calocedrus decurrens (Torr.) Florin), canyon live oak (Quercus chrysolepis Liebm.), and deciduous black oak (Q. kelloggii Newberry). Manzanita bush (Arctostaphylos glandulosa Eastw.) is the dominant understory shrub. There are few herbs and grasses at the study site. The climate is Mediterranean, typically with cool, wet winters and hot dry summers. Daily average air temperature varies from −5°C in winter to 23°C in midsummer. The area receives the majority of precipitation during winter with unpredictable monsoonal rains during midsummer.

2.2. Data Collection

[9] In the fall of 2004, we laid out a 76 m long transect with a network of sensors: 10 microclimate stations, 10 soil sensor arrays, and 15 minirhizotron (MR) tubes [Allen et al., 2007; Vargas and Allen, 2008a, 2008b; Rundel et al., 2009] (see also auxiliary material Figure S1). In this study we used image data collected with six MR tubes and environmental data measured with four sets of belowground sensors located in the midsection of the transect (Figure S1). For the first and second objectives of our study, we used the image data between March 2005 and August 2008, after approximately 6 months of stabilization; for the third objective of our study, we used the image and environmental data between October 2005 and August 2008 after approximately 1 year of stabilization.

[10] Within 30 cm of each minirhizotron (MR) tube, we buried soil temperature sensors (ONSET®, S-TMB-M006) and soil water content sensors (ONSET®, S-SMA-M005) at three depths (2 cm, 8 cm and 16 cm). Data were recorded every 5–15 min with a data logger (ONSET®, H21-001).

[11] MR tubes were made of transparent cellulose acetate butyrate, measuring 5.5 cm in outer diameter and 76 cm in length. Approximately the top 15 cm of each tube was exposed above the ground, and the exposed area was painted white, wrapped with a duct tape, and capped with a PVC fitting to keep inside of the tube dark. The burial angle varied from 26 to 40 degrees from the horizon, and vertical depth varied from 28 to 45 cm. Images of fine roots were taken using a MR camera system (Bartz Technology, Santa Barbara, California, USA), and approximately 70,800 images were collected from the six MR tubes. Each MR tube had a column of frames inscribed lengthwise with green paint as a reference so that a picture of the same area was acquired in every session. Each frame was 1.3 cm wide and 0.9 cm long, and each tube had approximately 65 frames. Movie clips showing how fine roots grow and perish with time are available at http://ccb.ucr.edu/lab_projects.html.

[12] We followed fates of 710 individual fine roots found in these images in chronological order. Typically, a fine root goes through a “light color/live” phase, followed by a “dark color/decomposition” phase, and then finally disappears [Hendrick and Pregitzer, 1992]. We recognized these three stages of the root and recorded each stage accordingly. We defined “life span” of a fine root as the duration between the appearance of a root and the disappearance of the root. Thus, precisely speaking, our definition of “life span” includes the “dark color/decomposition” phase of the root. However, the majority (∼77%) of the fine roots disappeared less than 1 week after the last observation of the “light color/live” phase, suggesting the “dark color/decomposition” phase was very short due to rapid decomposition. Therefore, we could not always identify and record the “dark color/decomposition” phase. Furthermore, since the “light color/live” phase was usually much longer than 1 week, the largest portion of observed life span was the “live” phase of the root. Therefore, our definition of life span is practically the same as “actual” life span of roots. In addition, fine roots that appear to be dead (“dark color/decomposition” phase) can remain functional for carbon, water and nutrient transport. Even live roots of woody plants contain suberized or dead cortical layers that play an active role in a plant's physiology [McKenzie and Peterson, 1995]. These dark colored roots are still intact, visible forms of carbon that are not decomposed. Therefore, the “dark/decomposition” phase should be included in the life span. Many researchers [e.g., Hendrick and Pregitzer, 1992; Norby et al., 2004] use this definition of life span (birth to disappearance). A fine root was presumed to be generated halfway through the first observed session and the previous session; similarly it was presumed to have perished halfway through the last observed session and the following session.

[13] The area within the green viewing frame was digitally extracted using program ImageJ (available at http://rsb.info.nih.gov/ij/index.html), and a computer generated 10 × 8 reference grid was laid over the top of the extracted image to count the number of intersecting fine roots with the grid. The count was then converted to length [Newman, 1966; Marsh, 1971]. Since fine roots typically grew quickly, maintained a constant length for a long steady state period, and perished in a relatively short time, the value at the steady state was used as the length for analysis. Fine root diameter was measured by using the “Straight Line Tool” in ImageJ. The finest root diameter we could visually identify was five pixels on the computer monitor, which is approximately 135 μm. Volumes were calculated from the measured diameters and lengths assuming fine roots are cylindrically shaped. Vertical depth from the ground surface to a frame on a tube was calculated from the burial angle of the tube and the slope of the transect using a simple trigonometric relationship.

2.3. Production, Mortality, and Turnover Rate

[14] To calculate production and mortality rate (grams of fine roots per square meter of ground per year), the total volume of the fine roots produced (perished) during the entire observation period were converted to mass using the specific density of fine roots (0.31 g cm−3 [Gill and Jackson, 2000]). The resulting mass was standardized to a square meter of ground [Johnson et al., 2001; Norby et al., 2004] by

equation image

We assumed that the field depth was 2 mm and the soil depth was 0.45 m. The resulting value was divided by the number of observation years (∼3.4 years). Standing crop (grams of fine roots per square meter of ground) was also calculated using the same assumption.

[15] Monthly survival rates were calculated by dividing the number of fine roots surviving between month i and month i + 1 by the number of fine roots that existed in month i. Assuming that survivorship follows an exponential decay, resulting survival rates were converted to turnover rates using

equation image

A constant of 12 emerges from the conversion from monthly to yearly.

2.4. Effect of Sampling Interval

[16] To evaluate how sampling interval might affect our estimates of production, mortality, standing crop and turnover rate, we first pooled all fine roots observed with the six MRs and then filtered their life span times with various hypothetical observation intervals. If both production and mortality events of a fine root were between two hypothetical observation dates, we assumed that the fine root was missed by the observer and it was excluded from the data. Turnover rate and its 95% confidence interval was then calculated by fitting the filtered longevity times to an exponential decay function using MATLAB® R2008b (function: expfit). The fine roots still alive on the last observation date were censored. We tested for statistical differences between our actual sampling frequency and simulated ones using the Kaplan-Meier rank test (SAS®, proc lifetest).

2.5. Turnover Rate Estimation Using Mark-Recapture Modeling

[17] We split the whole data set into approximately two equal parts by randomly selecting from the fine roots pooled across the six MR tubes. The first half (314 fine roots out of 710) was used for model construction, and the other half (386 fine roots) was used for model validation. We built the history of each fine root by denoting “1” for its live phase and “0” for its dead (disappearance) phase. Recapture probabilities were fixed to 1 because the same fine root in the same viewing frame on a MR tube was followed in chronological order, and its fate was known.

[18] The created data set was used to evaluate effects of time-dependent factors (T and θ) as well as time-independent root characteristics (root diameter (D), root length (L), root depth (Dp), root volume (V), and season of cohorts). Although we measured T and θ at three different depths (2 cm, 8 cm and 16 cm), only T and θ at 8 cm were used for analyses because values at the three depths were highly autocorrelated. Averages of T and θ between two observation dates were calculated from the continuously measured environmental data. Based on date of birth, three seasonal cohorts were defined as follows: winter cohorts, November–February; spring cohorts, March–June; and summer cohorts, July–October. These three seasons approximately corresponded to cold and dry winter, warm and humid spring and hot and mostly dry summer but with occasional monsoonal rain pulses.

[19] All models we examined were a linear combination of covariates. Since diameter, length, and volume had a highly skewed distribution (Figure S2), these variables were transformed using an inverse function (1/X). Our most general model of survival rate (ϕ) is expressed as

equation image

where logit(ϕ) = exp (ϕ)/{1 + exp(ϕ)}, X1, X2, X3, … are covariates and β0,β1,β2, … are regression coefficients. Model selection among competing models was based on second-order Akaike's Information Criterion (AICc). Parameter estimation and model selection were accomplished using Program MARK ver. 5.1 [Cooch and White, 2006]. We tested the goodness of fit of the general model (survival rate as a function of time and all factors considered) using RELEASE Test 3 procedures described by Cooch and White [2006] to test a key assumption of our mark-recapture models; namely, that all ϕ have the same value across treatment groups excluding the effects of covariates. Predicted turnover rates as functions of various covariates were calculated based on the best selected model. Ninety-five percent confidence intervals were calculated for survivorship estimates using the Delta method in Program MARK [Cooch and White, 2006]. For a presentation purpose, one covariate of our interest was varied while the rest of the covariates were set to the averages of the validation data set.

2.6. Model Validation

[20] Five hundred fine roots were randomly selected from the validation data set with replacement. Since a turnover rate is a property of a group of roots, not a property of an individual root, we cannot directly validate our model of individual survivorship. Instead, fine roots were grouped by month and the observed monthly turnover rate was calculated as described in Section 2.3. The predicted monthly turnover rate was estimated based on the selected model using the monthly average values of T and θ and median values of D and L for the 500 roots. A regression analysis was conducted to calculate the slope, intercept and R2 values for each sample draw. This process was repeated 500 times and 95% confidence intervals for regression parameters were calculated by fitting them to a normal distribution.

3. Results

3.1. Effect of Sampling Interval

[21] Estimates of all parameters examined (number of fine roots, mean standing crop, production, mortality, and turnover rate) decreased with increasing sampling interval (Figure 1 and Table 1). As sampling interval increased, estimated number of fine roots decreased every week by 1.4% (r2 = 0.945), mean standing crop by 0.5% (r2 = 0.737), production by 0.6% (r2 = 0.595), mortality by 1.1% (r2 = 0.595), and turnover rate by 1.9% (r2 = 0.947), owing to shorter life span times being missed by increased sampling intervals. Turnover rate was the most sensitive to sampling interval changes and was highly predictable. A log-rank test showed that there was a significant difference between the turnover rate of “as is” (all data without sampling) and those of sampling intervals of 5 weeks and greater. Since one of the six MR tubes (tube 9) was not in a steady state until the second year of study (Figure S3), we reanalyzed the data by excluding this tube. We found that turnover rate loss after excluding tube 9 was comparable to the one for all six tubes (2.1%, r2 = 0.951). Decreases in mean standing crop, production and mortality were higher after excluding tube 9 than the ones for all six tubes (data not shown).

Figure 1.

Effect of sampling interval on estimated number of fine roots (open circles, dotted line), mean standing crop (red), production (green), mortality (blue), and turnover rate (solid circles, solid line). All values are expressed as relative percentages of the values of “As Is” (all data without sampling). Best fit lines are shown by solid or dotted lines, and their constants were set to 100.

Table 1. Effect of Hypothetical Sampling Interval on Mean Standing Crop, Production, Mortality, and Turnover Ratea
Sampling Interval (Week)Number of Fine RootsMean Standing Crop (g m−2)Production (g m−2 yr−1)Mortality (g m−2 yr−1)Turnover Rate (yr−1)p Value for log Rank Test
  • a

    Turnover rate was calculated by fitting fine root longevities with an exponential decay curve. Mean standing crop, production, and mortality were standardized to a square meter of ground. The last column indicates the p value for the log rank test between “As Is” (all data without sampling) and each hypothetical sampling interval.

As Is710211.1174.191.10.98n/a

3.2. Production, Mortality, Standing Crop, and Turnover Rate

[22] Production, mortality and mean standing crop were highly correlated (p < 0.05), but none of these were correlated to turnover rate (Table S1). The overall turnover rate (pooled across all six tubes) was 0.98 yr−1, which is typical for temperate forest ecosystems [Gill and Jackson, 2000].

[23] Temporal variation in production and mortality were also high (Figure 2). Production was greater in May–June, when both T and θ were high, than for any other months. Seasonal variation in mortality was somewhat similar to production, but it was not as pronounced as production. Turnover rate was also high during April–July when both T and θ were high. When θ decreased in August and September, turnover rate suddenly plunged even when T was still high. These finding suggested both environmental variables, T and θ, played a significant role in production and turnover rate.

Figure 2.

Trends of (a) monthly production (solid blue circles) and mortality (solid red circles) and (b) monthly average soil temperature (open red circles), soil water content (open blue circles), and turnover rate (gray bars).

3.3. Survivorship Estimation Using Mark-Recapture Modeling

[24] Goodness of fit tests confirmed that basic assumptions of mark-recapture model were not violated for this data set (p = 0.999). Since models with ΔAICc values < 2–5 are similar in their fit to the data [Burnham and Anderson, 1998], we selected the model whose survival rate was a function of covariates that were common to most of the best performing models; namely diameter, length, T and θ (parameters in model 2 in Table 2). We did not include depth because the ΔAICc values of models with depth were never more than 2 from the similarly parameterized models without depth. Survival rate was therefore modeled as

equation image

Model 4 (Table 2) was the most parsimonious among potential candidates, but its support was only 1/6 (based on AICc weight ratio) of the selected model (model 2). Turnover rates predicted by the selected model are plotted against diameter, length, T, and θ (Figures 3a, 3b, 3c, and 3d, respectively). Predicted turnover rate decreased with increasing diameter and length (Figures 3a and 3b, respectively), and it increased rapidly with increasing T and θ (Figures 3c and 3d, respectively). In terms of range of the values observed in our study, T and θ had the greatest influence on the turnover rate, followed by diameter. Length had the least influence on the turnover rate.

Figure 3.

Estimated turnover rates (solid lines) as a function of (a) diameter, (b) length, (c) soil temperature, and (d) soil water content, based on the best selected model. Dotted lines show 95% confidence interval.

Table 2. Competing Models Sorted in the Increasing Order of AICca
Model IDModelAICcDelta AICcAICc WeightsNum. Par.
  • a

    All models are based on the 321 randomly selected fine roots. Model ID is for explanatory purpose only. Num. Par., number of parameters in the model; ϕ, survival rate; InvD, inverse of fine root diameter; InvL, inverse of fine root length; Dp, depth; T, soil temperature; θ, soil water content; season, season of cohorts.

  • b

    The most general model.

  • c

    The best model.

1bϕ (InvD + InvL + Dp + T + θ)1893.2340.0000.5526
2cϕ (InvD + InvL + T + θ)1894.3751.1410.3125
3ϕ (InvD + Dp + T + θ)1896.9673.7330.0855
4ϕ (InvD + T + θ)1898.1014.8670.0484
5ϕ (InvD + InvL + Dp + T)1906.78513.5510.0015
6ϕ (InvD + InvL + T)1907.90014.6660.0004
7ϕ (InvD + Dp + T)1911.24818.0140.0004
8ϕ (InvL + Dp + T + θ)1934.79141.5570.0005
9ϕ (InvD + InvL + Dp + θ)1935.04041.8070.0005
10ϕ (season + InvD + InvL + Dp)1939.49646.2620.0006

3.4. Validation

[25] The average regression line between the observed monthly turnover rates and predicted ones was very close to the 1:1 ratio line (Figure 4, slope = 1.072 with 95% C.I. = [1.060, 1.084], intercept = −0.028 with 95% C.I. = [−0.040, −0.015]). However, the R2 value was somewhat low (R2 = 0.338 with 95% C.I. = [0.332, 0.344]). While these results demonstrate a strong effect of T and θ on temporal variation in root turnover rate, the apparent overdispersion suggested that other unmeasured factors also played a significant role.

Figure 4.

Observed monthly turnover rates (dots), best fit line (solid line), and 1:1 ratio line (dotted line) as a function of predicted monthly turnover rates. Predicted turnover rates were estimated from the best selected model. Best fit line is the average of 500 repeats.

3.5. Hysteresis

[26] One of our main goals was to model temporal variation in turnover rate as a function of environmental covariates, T and θ. T and θ are highly correlated because, in Southern California, we receive the majority of rain in winter and soil dries up slowly with increasing T from spring to later summer (Figure 2a), yet a close examination of T and θ shows a more complicated relationship. When monthly average θ are plotted against monthly average T for each year, a pattern of hysteresis emerges (Figure S4a). Typically, T reaches maximum in July or August and θ reaches minimum in September. θ remains low until December because there is not enough precipitation in summer and fall to fully recharge the soil layer. Similarly when monthly average production and mortality are plotted against monthly average T for each year, we also find a pattern of hysteresis (Figures S4b and S4c), though hysteresis of mortality was less pronounced than production.

[27] θ, production, mortality, and observed turnover rate were averaged and plotted by calendar month to demonstrate how the T − θ hysteresis loop might affect production and mortality and turnover rate (Figure 5a). Production, mortality and observed turnover rate as well as θ showed a hysteresis pattern as a function of T. Our selected model predicted this hysteresis pattern well (Figure 5b). As T increases between February and August, θ slowly decreases with time. Because of relatively high T and θ, production rapidly increases with time during these months. Similarly, turnover rate reaches its maximum in May or June when both T and θ are high. In July or August, T reaches maximum and θ does minimum. By this time, water potential is at or below the wilting point and production declines rapidly. Turnover rate also declines fast in these months. In September and October, although T is as high as or higher than in May or June, production remains low because of lack of water. Although there is a plenty of water in the soil during the rainy season between December and March, T is too low for extensive production to occur. Turnover rate also remains low during these months. This process is repeated yearly, although with slightly different transition dates between phases.

Figure 5.

Hysteresis relationships of (a) production (red), mortality (green), and soil water content (blue) and (b) observed (red) and predicted (blue) turnover rates, as a function of soil temperature. All data points are averages by calendar month. Observed monthly turnover rates were calculated from all of the 710 fine roots we observed. Arrows indicate the progression of time from January (J) to December (D).

4. Discussion

[28] Sampling intervals vary from weekly to quarterly depending on researchers and their study goals [see Johnson et al., 2001, and references therein]. Tingey et al. [2003] studied the effects of sampling interval on cumulative production and mortality with Pseudotsuga menziesii and Tilia cordata. They found that 24% and 35% of the fine root production in P. menziesii and T. cordata, respectively, were missed by their 8 week sampling interval. Johnson et al. [2001] simulated the percentage of fine roots potentially missed with increasing sampling intervals, and they came up with a simple formula that predicts percentage of missed fine roots as a function of sampling interval, assuming the exponential decay rule of root survivorships and constant production and mortality over time. Our result (1.4% missed per week) was higher than their predicted estimate (∼0.8% missed per week). This was possibly because our data did not satisfy their assumption of constant production and mortality over time, as we found that production and mortality were highly seasonal.

[29] Stewart and Frank [2008] reported a life span <3 days in a temperate grassland, while Stevens et al. [2002] found fine roots with a life span <3 weeks consisted of 24% of the total root length in a pine woodland. The shortest life span time observed in our study was 8 days; however, life spans of <3 weeks consisted of only 3% of the total number of fine roots. Our median sampling interval was every 3 days and included daily observations made during a 2 week period in the most active months for production in spring. This short sampling interval should have allowed us to detect rapid turnover if it existed.

[30] The question of the optimal sampling interval required depends in part on how much turnover error is acceptable. For example, in order to keep the error to less than 10%, roots must be observed at least every month, judging from the sampling interval simulation. Clearly, quarterly sampling would greatly underestimate turnover rate (∼25%). Of course, optimal sampling interval also depends on the ecosystem and its constituent plant species.

[31] Effects of root characteristics (e.g., diameter, length, depth, root order, seasonal cohorts, and mycorrhizal association) on root turnover rates have been well studied. Larger diameter fine roots are known to survive longer than smaller ones [Eissenstat et al., 2000; Tierney and Fahey, 2001; Wells and Eissenstat, 2001; King et al., 2002]. Our results also support this conclusion. However, because diameter is generally correlated to root order [Majdi et al., 2001; Pregitzer et al., 2002; Guo et al., 2008a, 2008b] and functionality also varies with root order [Eissenstat et al., 2000], root order rather than diameter can be the true driving factor. Either way, diameter can be used as a proxy for predicting turnover rate.

[32] Since length is measured within a confined frame, it is proportional to root density (root mass or length per area of soil). Our best model predicts that turnover rate decreases with increasing length (root density), though the effect of length is much smaller than the others (Figure 3b). Wells and Eissenstat [2001] reported decreased survivorship of apple roots with increasing number of neighboring roots in the same viewing frame of an MR tube in one out of 2 years of observations. In contrast, Anderson et al. [2003] reported no effect of number of neighboring roots on survivorship of grape roots. We found that root volume was not as good a predictor as diameter and/or length when volume was evaluated as a factor (data not shown). This suggests that smaller diameter and higher density is more favorable for the efficient uptake of nutrients [see also Barber and Silberbush, 1984]. We propose that the effect of root density is not as important as diameter, but more important than volume, and that this may be true especially in both ecto- and arbuscular mycorrhizal dominant mixed coniferous forests [Cui and Caldwell, 1996a, 1996b].

[33] Many studies have shown that turnover rate decreases with increasing depth [Majdi et al., 2001; Wells et al., 2002; Anderson et al., 2003], which has been attributed to the higher stability of T and θ in deeper soil. However, these studies did not separate effects of T and θ from one another, as both varied with depth. Our best model excluded depth as a factor. Since T and θ are highly correlated with depth, it is possible to explain the effect of depth by T and θ alone. It is also plausible that due to a relatively shallow soil (<45 cm) of our study site we cannot see the effect of depth on turnover rate, even if it exists.

[34] Effects of seasonal cohorts have been also well documented [Ruess et al., 1998; Tierney and Fahey, 2001; Wells et al., 2002; Ruess et al., 2003]. However, we found that cohort season was a much poorer predictor than the combined effect of temperature and moisture (Table 2, compare model 1 versus 10). This suggests that life span might be variable depending on environment conditions rather than being predetermined by the initial conditions present on the roots' birth date.

[35] Our best model predicted that survivorship decreased with increasing T. Forbes et al. [1997] reported decreased survivorship with increasing temperature for rye grass roots, while King et al. [1999] also found a similar pattern for trembling aspen. Gill and Jackson [2000] observed a similar relationship on a global scale, and they listed three reasons for this relationship: (1) increased maintenance respiration, (2) increased nutrient mineralization, and (3) increased pathogen and herbivore activities with increasing T. All three are likely causes in our study site.

[36] Effects of θ or irrigation on survivorship are inconsistent and vary among systems [see Anderson et al., 2003, and references therein]. Bauerle et al. [2008a, 2008b] found that the survival rate of grape vine roots was independent of θ because of internal hydraulic redistribution during the night. On a global scale, Gill and Jackson [2000] reported that root turnover rates increased with increasing annual mean T, but were independent of annual mean precipitation in various ecosystems. Stewart and Frank [2008] made a similar finding in a temperate grassland. Our best model included both T and θ as factors and the interaction of these two factors predicted the hysteresis loop of turnover rate relative to T.

[37] High turnover rate when both T and θ are high is puzzling. We would expect turnover rate to be low when both T and θ are high [e.g., Ruess et al., 1998, 2003] or independent of θ because of internal hydraulic redistribution [Bauerle et al., 2008a, 2008b]. To eliminate the possibility that high T and θ accelerated decomposition and consequently increased turnover rate, we reanalyzed the data using only the “light color/live” phase of life spans. Results were very similar to those included both the “light color/live” and “dark color/decomposition” phases and the best model still included both T and θ (Table S2). Huang and Nobel [1992] observed that Agave deserti in a Southern California desert produced short-lived, ephemeral, fine roots after a pulse of rain. Similarly, Reynolds et al. [1999] reported that a small proportion of desert shrub roots in a warm desert was short lived. To cope with water stress in arid environments, producing ephemeral, short-lived, fine roots during a short wet season may be a possible strategy. We suggest that, in contrast to inferences from global-scale observations, θ plays a significant and variable role in the regulation of turnover rates in arid climates.

[38] Belowground NPP (g of Carbon per unit area per year) is a function of both production and root turnover rate. Because of the hysteresis of production and turnover rate relative to T, belowground NPP is greater between spring and early summer than between midsummer and winter in our study site. For the same reason, we expect wetter years have a greater belowground NPP than drier years. Similar studies in other climate conditions will be needed to generalize across ecosystem types.

[39] Previous studies at the James Reserve showed that soil CO2 efflux also exhibit seasonal and diurnal hysteresis patterns [Vargas and Allen, 2008b] and that seasonal variation in CO2 production is significantly influenced by EM rhizomorph production, T, and θ [Hasselquist et al., 2010]. A similar hysteresis was also found in a neotropical rain forest [Vargas and Allen, 2008c]. It is very likely that the hysteresis of turnover rate contributes to the soil CO2 efflux hysteresis, though we do not know the exact percentage of the contribution yet and it must be addressed in future studies.

[40] Pregitzer et al. [2000] observed that an increase in root production rate seemed to correspond roughly to leaf duration, and that an increase in mortality rate corresponded to canopy senescence in the fall. We observed an increase in both root production and mortality rates between March and August, which roughly corresponds to the growing season of the study site, though annual average values varied year to year. Our observations are consistent with other studies that demonstrated seasonality and interannual variation in production and mortality rates [Ruess et al., 1998; Tierney and Fahey, 2001; Ruess et al., 2003; Norby et al., 2004].

[41] Our somewhat low R2 value between the predicted and the observed turnover rates suggested that T and θ alone cannot fully explain temporal variation in the turnover rate. Ruess et al. [1998] found that aboveground herbivory increased fine root turnover rate belowground. Since aboveground growth, belowground growth and senescence are tightly connected [e.g., Poorter and Nagel, 2000], aboveground phenology should be included in the future studies of root turnover rate. In Southern California, because the growing season is between spring and early summer when there is enough moisture in the soil, it is difficult to differentiate between moisture and aboveground phenology. However, even aboveground phenology should be ultimately attributed to aboveground abiotic factors such as air temperature, relative humidity, precipitation, PAR, etc.

[42] We have demonstrated links between time-dependent environmental variables (T and θ) and fine root turnover rate in a Mediterranean climate. Studies such as ours that link environmental variables and NPP are urgently needed in various climate types to predict how anticipated local and global climate changes may affect C flux, C budget, C allocation, and C sequestration.


[43] The authors would like to thank Hector Estrada-Medina, Darrel Jenerette, Alisha Glass, Laurel Goode, Niles Hasselquist, Einav Mayzlish-Gati, Ayesha Sirajuddin, William Swenson, and Rodrigo Vargas for image data collection and discussions. Chris Glover, Amadou Camara, Kayleen Leong, Norman Ho, and Jasmin Quon helped us with processing image data. CENS engineers Mike Taggart and Tom Unwin helped develop and maintain the sensor network. Special thanks go to the James Reserve. This research was funded by the National Science Foundation (EF-0410408 and CRR-0120778).