The effect of soil moisture perturbations on Indian Monsoon Depressions in a numerical weather prediction model

Indian monsoon depressions (MDs) are synoptic-scale cyclonic systems that propagate across peninsular India three or four times per monsoon season. They are responsible for the majority of rainfall in agrarian north India, thus constraining precipitation estimates is of high importance. Here, we use a case study from August 2014 to explore the relationship between varying soil moisture and the resulting track and structure of an incident MD using the Met Ofﬁce Uniﬁed Model. We use this case study with the view to increasing un-derstanding of the general impact of soil moisture perturbations on monsoon depressions. It is found that increasing soil moisture in the monsoon trough region results in deeper inland penetration and a more developed structure – e.g. a warmer core in the mid-troposphere and a stronger bimodal potential vorticity core in the middle/lower troposphere – with more precipitation, and a structure that in general more closely resembles that found in depressions over the ocean, indicating that soil moisture may enhance the convective mechanism that drives depressions over land. This experiment also shows that these changes are most signiﬁcant when the depression is deep, and negligible when it is weakening. Increasing soil moisture in the sub-Himalayan arable zone , a region with large irrigation coverage, also caused deeper inland penetration and some feature enhancement in the upper troposphere but no signiﬁcant changes were found in the track heading or lower-tropospheric structure.

orientation are normalised such that the centre lies at the origin and the heading is up the page; 38 land-only data were used. As asserted, there is not much spatial similarity between the extrema 39 of precipitation and Bowen ratio -indicating that if we are to believe previous work suggesting a 40 link between MD behaviour and underlying soil moisture, it may be a more subtle feedback, or 41 work on a finer spatial scale, than that suggested by Eltahir (1998). The caveat here is that surface 42 fluxes are an entirely modelled product in ERA-I, and so have substantial uncertainty; however 43 this is at least partially addressed by the similarity of composite MD precipitation between ERA-I 44 and TRMM, and the fact that most rainfall near the centre of a depression is stratiform in nature 45 (Hunt et al. 2016b). To date, a number of studies have shown that assimilation of soil moisture, 46 or better initial representation of it, improves the forecast of monsoon depressions in mesoscale 47 models (Chandrasekar et al. 2007;Vinod Kumar et al. 2007;Chandrasekar et al. 2008;Rajesh 48 and Pattnaik 2016). Further, it has been shown that inland soil moisture is capable not only of 49 extending the duration of tropical cyclones (Andersen and Shepherd 2017), but in some cases of 50 allowing them to re-intensify (Kellner et al. 2012). 51 Soil moisture is one of the meteorological variables subject to greatest change with respect to the 52 progression of the Indian monsoon, largely due to its correlation with accumulated precipitation. 53 The NOAA CPC reanalysis soil moisture climatology (Van den Dool et al. 2003) and the ESA CCI 54 satellite-derived soil moisture climatology (Liu et al. 2011(Liu et al. , 2012Wagner et al. 2012) for India for 55 April, June, August, and September are given in Fig. 2(a) and Fig. 2(b) respectively and show 56 a clear northwestward advance through most of the season: some areas in the monsoon trough 57 have September soil moisture more than double that of June. Naïvely, then, we might expect MD 58 tracks to penetrate deeper inland later into the monsoon season, given the expected influence of 59 antecedent soil moisture on the development of MDs. Fig. 3 shows the mean MD track for each 60 month  from the track datasets of Hunt et al. (2016a) and Hurley and Boos (2015) 61 respectively; note that the MD tracks have been extended to include parts where the depression is 62 strictly in a monsoon low regime (that is to say, the surface winds are below 8.5 m s −1 ). There 63 is some weak evidence here to suggest that not only do MDs tend to progress further inland later 64 in the season, they also seem more likely to have over-land genesis. This should be taken with Office use The Met Office Global and Regional Ensemble Prediction System (MOGREPS;Bowler et al. 2008) to generate ensemble NWP runs; given that this was designed specifically for the UM, 118 we aim to make our ensemble generation as similar as possible. MOGREPS uses two distinct 119 stochastic physics schemes: random parameters (RP) and stochastic kinetic energy backscatter 120 (SKEB). The former uses the premise that many parameters in the various parameterisations in 121 the UM are tuned to empirical values that appear to give the best representation of the relevant 122 process, these can be periodically varied at differing frequencies between physically reasonable 123 values to produce a spread of forecasts; the latter reintroduces kinetic energy lost through poor 124 representation of the mechanisms by which small-scale processes cascade energy to larger scales 125 (Shutts 2005). Initial tests suggested that using SKEB perturbations tended to artificially weaken 126 MDs and cause them to have much shorter tracks. Thus in our study we used a stochastic perturbed 127 tendencies (SPT) scheme which simply randomly perturbs the summation of tendencies from all 128 parameterisations in the model (Buizza et al. 1999).

129
In our ensemble, we must also attempt to represent uncertainties in the analyses that are used 130 to initialise the model. In MOGREPS this is typically done by applying an ensemble transform 131 Kalman filter (ETKF; Bishop et al. 2001) to a previous ensemble run, assimilating observations to 132 assess where perturbations will have the largest impact. As operational ensemble analyses were 133 not readily available for our case study, we opted to simulate the uncertainty by adding white 134 noise of amplitude 0.5 K to boundary layer potential temperature. Sensitivity tests determined 135 that this gave a realistic spread of MD tracks from a short initialisation without suppressing the 136 development and progression of the depression. For each sub-experiment, which are differentiated 137 by varying soil moisture in the same region, a ten-member ensemble was used; for each ensemble 138 member, a random seed was used such that across each experiment each ensemble was generated 139 via the same set of pseudorandom parameters to allow intercomparability. As discussed in the Introduction, two case study experiments are proposed to explore the sen-142 sitivity of duration and heading respectively to underlying soil moisture. Fig. 4 shows the masks 143 used to set up the soil moisture ancillary files: the red polygon covers much of South Asia, the 144 green polygon covers the typical monsoon trough region, and the orange covers the sub-Himalayan 145 arable land that is becoming increasingly intensively irrigated and farmed. In each instance, the 146 soil moisture control (perturbations to which will be used in the experiments) is the August clima-147 tology as computed from a fully coupled high-resolution climate simulation in the UM. This was 148 chosen to reduce spin-up/resolution issues that could be introduced by using a climatology from, 149 e.g., either of the datasets in Fig. 2. This is the current method used for soil moisture initialisation 150 of the MetUM in operational NWP mode.

151
For the first experiment (hereafter: trough zone), soil moisture in the monsoon trough region 152 (the green polygon in Fig. 4) -in which MD tracks are typically entirely embedded -was altered 153 to 1%, 80%, 100% (control), 120%, and 500% of its August climatological value. The 500% 154 value unsurprisingly gives significant oversaturation across much of the region, where this was 155 the case, soil moisture values at these locations were set to their saturation values; in reality, this 156 scaling is achievable only over the dry northwest, and the average saturation value over the trough are likely to have the biggest impact. Values of soil moisture approaching 1% of the August 164 climatology could be found in an extremely dry pre-monsoon period, but we remind the reader 165 that the purpose of this experiment is to test the effect of soil moisture contrast in the region, not 166 necessarily to replicate a physical event.  can compute the mean speeds and durations for the 1%, 80%, 100%, 120%, and 500% ensembles, 200 the mean propagation speeds are: 3.7, 3.7, 3.7, 3.9 and 3.9 m s −1 respectively, with corresponding 201 mean durations of 3.7, 4.3, 4.4, 4.2, and 4.3 days. Applying a significance test, we find that the 202 mean ensemble speeds for the two wettest cases (500% and 120%) are significantly different from 203 the drier ones, and that the mean duration for the driest case (1%) is significantly different from 204 the four wetter ones.

205
The arable zone experiment was set up to determine to what extent moisture changes in relatively 206 distant soil could affect the steering of a contemporaneous MD. Recall that for this experiment, 207 the soil moisture over South Asia was set to 1% of the climatology, and to the value specified (1%, 208 50%, 100%, or 500%) of the climatology in the sub-Himalayan belt. The results from this exper-209 iment are presented in Fig. 5(b) in an identical fashion to those from the trough zone experiment.

210
In the absence of a control run, the concave hull given is for the "100%" ensemble plume. While it 211 may seem contrived to have such extremely dry soil over almost the entire peninsula for the sake of

216
An initial overview of Fig. 5(b) suggests two broad characteristics: firstly, that the spread of en-217 semble mean terminations is smaller than in the trough zone experiment -this is almost certainly 218 attributable to the altered soil area both having a smaller area and being further away, and thus 219 being less influential; secondly, that all the average tracks are shorter than in the previous exper-220 iment -plausibly due to a larger area of desiccation than in the 1% trough zone sub-experiment 221 resulting in even less water being available over the peninsula, bearing in mind that MDs draw 222 moisture in from distances of up to 1000 km (Hunt et al. 2016a). We also note that whilst there is 223 a perfect rank correlation between soil moisture fractional change and mean termination latitude, 224 the mean track for the 100% sub-experiment is longer than that for the 500% ensemble. Repeating 225 the termination point significance analysis carried out for the trough zone experiment, we find that 226 the three wettest sub-experiments have mean track termination points significantly different from 227 the driest (1%), but not from each other, at a 95% confidence level. Having established that soil moisture changes, both local and distant, are capable of significantly 230 altering the track of a passing MD, we will now examine the differing synoptic structure that these 231 12 changes cause and attempt to bring the discussion to its conclusion. The largest contrast was seen 232 in the trough zone experiment, so we shall start the discussion there. Fig. 6 shows longitude-height 233 cross-sections through 500%-minus-1% composite variables from the trough zone experiment. We 234 will briefly note here that structural changes of similar shape are found by comparing composites 235 arising from smaller changes in soil moisture, but with varying losses in magnitude, and hence,   For comparison, the equivalent figure to Fig. 7(a) for the arable zone experiment is Fig. 7(b).

263
Here, the consequence of increased soil moisture is largely confined to the north of the MD as for the 500%-minus-1% difference composites. These are given for potential vorticity, relative 272 humidity, and temperature in Fig. 8. It is clear (and unsurprising) that the effect of changing 273 arable zone soil moisture is felt substantially less by the MD than changing trough zone soil 274 moisture, since the arable zone soil moisture perturbation is some distance from the MD core.

275
The most prominent effect of wetting the soil there is to set up a wet, cool boundary layer; this, 15 and -as can be seen from Fig. 5 -remained there for a little longer thereafter (eroding it further, as seen in Fig. 9).

303
Related to CAPE, but not shown, is convective inhibition (CIN). Changes in soil moisture have 304 been shown to affect CIN (e.g. Clark and Arritt 1995), which typically reaches minimum mag-305 nitude just ahead of the depression centre (Hunt et al. 2016a). Applying the same analysis that 306 we did for CAPE, we find that in the 1% case, CIN is significantly much more negative (less con-307 ducive to convection) and that this extreme is much longer lasting in the vicinity of the centre when 308 compared to the other cases. The remaining cases did not differ significantly from each other.

309
Second from top in Fig. 9 is the mean total precipitable water in the area surrounding the MD 310 centre. This field is less variable than CAPE but still displays a clear maximum across all sub-311 experiments at approximately 60% of the MD lifetime before rapidly falling away. As with max-312 imum CAPE, there is significant correlation between trough soil moisture and mean total precip-313 itable water as well as a significant difference between the values of the extreme sub-experiments 314 during the middle period where the MD is at its strongest, followed by a complete loss of corre-315 lation, significance, and predictability after this point; although unlike CAPE, the correlation and 316 significance are retained during spin-up. Second from bottom is the mean lower/mid-tropospheric 317 temperature anomaly (averaged 850-400 hPa), here the picture is much the same as for total pre-318 cipitable water, although the correlation is no longer significant at the 95% confidence level, and 319 the ensemble spread does not widen as much during lysis. Finally, at the bottom is maximum 320 relative vorticity in the lower troposphere (900-800 hPa); whilst this is an inherently variable field, 321 and consequently although there is arguably some correlation between it and soil moisture during 322 the period of maximum intensity, it is not significant, nor is the difference between the two ex-323 treme sub-experiments significant more than occasionally. That having been said, any semblance 324 of correlation vanishes, as with the other fields, during the dissipation phase. we have framed these tests in the context of a single MD, significant differences have emerged 336 between the ensembles due to the imposition of soil moisture anomalies; we hope that this will 337 motivate further study of other events to explore the climatological relationship between MDs and 338 soil moisture. 339 We found that both the structure and propagation of the MD was significantly sensitive to 340 changes in soil moisture in the monsoon trough region: wetter conditions there caused a strength-341 ening of the MD with increased central PV and a warmer thermal core, as well as a more pro-342 nounced westward axial tilt. Such cases were also found to travel further inland before dissipating.

343
Further, we found that these changes were greatest (among variables associated with MD strength: 344 CAPE, TPW, mid-tropospheric temperature, and lower-tropospheric vorticity) during the period 345 when the MD is most intense, and that varying soil moisture has no noticeable effect on the MD 346 during its spin-down.
In the other experiment, soil across South Asia was kept desiccated while moisture in the sub-

348
Himalayan arable zone was varied. This had a lesser effect on both the structure and track of the 349 case study, although some significant differences persisted: tracks in the wetter cases terminated 350 later, and there was some weak strengthening of the MD in the middle and upper troposphere. 351 We also noted that in the wetter trough zone experiments, the ensemble composite MD became where ρ w is the density of water, ∆z is the thickness of the layer, and Θ u is the liquid water 381 concentration (for the sake of this discussion, we neglect frozen water, although it is catered for in 382 the scheme). This is subject to the transport equation: where subscript n denotes the layer, W n and W n−1 the diffusion terms in the layer and that immedi-384 ately below it, and E n is the evapotranspiration (including interaction with roots). The evapotran-385 spiration function is controlled by land usage and vegetation data embedded in JULES, whereas 386 the diffusion terms are prescribed by the Darcy equation: where K is the hydraulic conductivity and Ψ is the soil water suction function. Within MOSES 388 these are respectively described by the Clapp-Hornberger relationships (Clapp and Hornberger 389 1978): where Ψ s , K s and b are empirical constants that can be set on model initialisation. For this study,

392
the default values used operationally by the Met Office were used.

393
There are then two boundary conditions: at the surface, the flux (aside from evaporation) is 394 computed as the summation of canopy throughfall, snowmelt, and surface runoff; underneath the 395 bottom (Nth) layer, the drainage (W N ) is set to equal the hydraulic conductivity.

396
Finally, the evaporation to the atmosphere from soil at the surface is given by: where f a is the tile saturation fraction (e.g. 1 for ice, lake, ocean, 0 for dry rock), ρ is the density LIST OF FIGURES   FIG. 6. Differences in selected fields of the composite mean ensembles for the 500% and 1% (the former minus the latter) trough zone experiment. The composite is normalised such that its centre lies at the origin, but no rotation is carried out; these are then presented as a height-longitude cross section (at zero latitude). Greyed areas indicate the difference between the sub-experiment composites was not met at the 95% significance level according to a 10,000 member bootstrap test. The selected fields are: (a) potential vorticity (10 −7 K m 2 kg −1 s −1 ), (b) relative humidity (%), and (c) temperature (K). White lines on each subfigure indicate the zero contour.  9. Selected fields as a function of normalised depression lifetime for the trough experiment, with the soil moisture changes coloured thus: 1% -red, 80% -yellow, 120% -green, 500% -blue. From top to bottom, they are: the maximum CAPE (J kg −1 ) found in the advance quadrant 4 of the MD; mean total precipitable water (mm); mean temperature anomaly (K) between 850 and 400 hPa; and maximum relative vorticity (10 −5 s −1 ).
The thick, solid lines represent the ensemble average, with the thinner, dashed lines representing the ensemble minimum and maximum values. Each is computed over a box of side length 250 km centred on the MD centre.